Fix critical bugs and issues found in alternator code review

Co-authored-by: nyh <584227+nyh@users.noreply.github.com>

.github/workflows/call_backport_with_jira.yaml

@@ -1,53 +0,0 @@
-name: Backport with Jira Integration
-
-on:
-  push:
-    branches:
-      - master
-      - next-*.*
-      - branch-*.*
-  pull_request_target:
-    types: [labeled, closed]
-    branches: 
-      - master
-      - next
-      - next-*.*
-      - branch-*.*
-
-jobs:
-  backport-on-push:
-    if: github.event_name == 'push'
-    uses: scylladb/github-automation/.github/workflows/backport-with-jira.yaml@main
-    with:
-      event_type: 'push'
-      base_branch: ${{ github.ref }}
-      commits: ${{ github.event.before }}..${{ github.sha }}
-    secrets:
-      gh_token: ${{ secrets.AUTO_BACKPORT_TOKEN }}
-      jira_auth: ${{ secrets.USER_AND_KEY_FOR_JIRA_AUTOMATION }}
-
-  backport-on-label:
-    if: github.event_name == 'pull_request_target' && github.event.action == 'labeled'
-    uses: scylladb/github-automation/.github/workflows/backport-with-jira.yaml@main
-    with:
-      event_type: 'labeled'
-      base_branch: refs/heads/${{ github.event.pull_request.base.ref }}
-      pull_request_number: ${{ github.event.pull_request.number }}
-      head_commit: ${{ github.event.pull_request.base.sha }}
-      label_name: ${{ github.event.label.name }}
-      pr_state: ${{ github.event.pull_request.state }}
-    secrets:
-      gh_token: ${{ secrets.AUTO_BACKPORT_TOKEN }}
-      jira_auth: ${{ secrets.USER_AND_KEY_FOR_JIRA_AUTOMATION }}
-
-  backport-chain:
-    if: github.event_name == 'pull_request_target' && github.event.action == 'closed' && github.event.pull_request.merged == true
-    uses: scylladb/github-automation/.github/workflows/backport-with-jira.yaml@main
-    with:
-      event_type: 'chain'
-      base_branch: refs/heads/${{ github.event.pull_request.base.ref }}
-      pull_request_number: ${{ github.event.pull_request.number }}
-      pr_body: ${{ github.event.pull_request.body }}
-    secrets:
-      gh_token: ${{ secrets.AUTO_BACKPORT_TOKEN }}
-      jira_auth: ${{ secrets.USER_AND_KEY_FOR_JIRA_AUTOMATION }}

.github/workflows/iwyu.yaml

@@ -14,7 +14,8 @@ env:
   CLEANER_DIRS: test/unit exceptions alternator api auth cdc compaction db dht gms index lang message mutation mutation_writer node_ops raft redis replica service
   SEASTAR_BAD_INCLUDE_OUTPUT_PATH: build/seastar-bad-include.log
 
-permissions: {}
+permissions:
+  contents: read
 
 # cancel the in-progress run upon a repush
 concurrency:
@@ -34,8 +35,6 @@ jobs:
       - uses: actions/checkout@v4
         with:
           submodules: true
-      - run: |
-          sudo dnf -y install clang-tools-extra
       - name: Generate compilation database
         run: |
           cmake                                         \

.github/workflows/trigger-scylla-ci.yaml

@@ -9,57 +9,16 @@ on:
 
 jobs:
   trigger-jenkins:
-    if: (github.event_name == 'issue_comment' && github.event.comment.user.login != 'scylladbbot') || github.event.label.name == 'conflicts'
+    if: (github.event.comment.user.login != 'scylladbbot' && contains(github.event.comment.body, '@scylladbbot') && contains(github.event.comment.body, 'trigger-ci')) || github.event.label.name == 'conflicts'
     runs-on: ubuntu-latest
     steps:
-      - name: Verify Org Membership
-        id: verify_author
-        env:
-          EVENT_NAME: ${{ github.event_name }}
-          PR_AUTHOR: ${{ github.event.pull_request.user.login }}
-          PR_ASSOCIATION: ${{ github.event.pull_request.author_association }}
-          COMMENT_AUTHOR: ${{ github.event.comment.user.login }}
-          COMMENT_ASSOCIATION: ${{ github.event.comment.author_association }}
-        shell: bash
-        run: |
-          if [[ "$EVENT_NAME" == "pull_request_target" ]]; then
-            AUTHOR="$PR_AUTHOR"
-            ASSOCIATION="$PR_ASSOCIATION"
-          else
-            AUTHOR="$COMMENT_AUTHOR"
-            ASSOCIATION="$COMMENT_ASSOCIATION"
-          fi
-          ORG="scylladb"
-          if gh api "/orgs/${ORG}/members/${AUTHOR}" --silent 2>/dev/null; then
-            echo "member=true" >> $GITHUB_OUTPUT
-          else
-            echo "::warning::${AUTHOR} is not a member of ${ORG}; skipping CI trigger."
-            echo "member=false" >> $GITHUB_OUTPUT
-          fi
-
-      - name: Validate Comment Trigger
-        if: github.event_name == 'issue_comment'
-        id: verify_comment
-        env:
-          COMMENT_BODY: ${{ github.event.comment.body }}
-        shell: bash
-        run: |
-          CLEAN_BODY=$(echo "$COMMENT_BODY" | grep -v '^[[:space:]]*>')
-
-          if echo "$CLEAN_BODY" | grep -qi '@scylladbbot' && echo "$CLEAN_BODY" | grep -qi 'trigger-ci'; then
-            echo "trigger=true" >> $GITHUB_OUTPUT
-          else
-            echo "trigger=false" >> $GITHUB_OUTPUT
-          fi
-
       - name: Trigger Scylla-CI-Route Jenkins Job
-        if: steps.verify_author.outputs.member == 'true' && (github.event_name == 'pull_request_target' || steps.verify_comment.outputs.trigger == 'true')
         env:
           JENKINS_USER: ${{ secrets.JENKINS_USERNAME }}
           JENKINS_API_TOKEN: ${{ secrets.JENKINS_TOKEN }}
           JENKINS_URL: "https://jenkins.scylladb.com"
-          PR_NUMBER: "${{ github.event.issue.number || github.event.pull_request.number }}"
-          PR_REPO_NAME: "${{ github.event.repository.full_name }}"
         run: |
+          PR_NUMBER=${{ github.event.issue.number }}
+          PR_REPO_NAME=${{ github.event.repository.full_name }}
           curl -X POST "$JENKINS_URL/job/releng/job/Scylla-CI-Route/buildWithParameters?PR_NUMBER=$PR_NUMBER&PR_REPO_NAME=$PR_REPO_NAME" \
-            --user "$JENKINS_USER:$JENKINS_API_TOKEN" --fail
+          --user "$JENKINS_USER:$JENKINS_API_TOKEN" --fail -i -v

SCYLLA-VERSION-GEN

@@ -78,7 +78,7 @@ fi
 
 # Default scylla product/version tags
 PRODUCT=scylla
-VERSION=2026.1.1
+VERSION=2026.2.0-dev
 
 if test -f version
 then

alternator/conditions.cc

@@ -244,7 +244,10 @@ static bool is_set_of(const rjson::value& type1, const rjson::value& type2) {
 
 // Check if two JSON-encoded values match with the CONTAINS relation
 bool check_CONTAINS(const rjson::value* v1, const rjson::value& v2, bool v1_from_query, bool v2_from_query) {
-    if (!v1) {
+    if (!v1 || !v1->IsObject() || v1->MemberCount() == 0) {
+        return false;
+    }
+    if (!v2.IsObject() || v2.MemberCount() == 0) {
         return false;
     }
     const auto& kv1 = *v1->MemberBegin();

alternator/consumed_capacity.cc

@@ -45,7 +45,7 @@ bool consumed_capacity_counter::should_add_capacity(const rjson::value& request)
 }
 
 void consumed_capacity_counter::add_consumed_capacity_to_response_if_needed(rjson::value& response) const noexcept {
-    if (_should_add_to_reponse) {
+    if (_should_add_to_response) {
         auto consumption = rjson::empty_object();
         rjson::add(consumption, "CapacityUnits", get_consumed_capacity_units());
         rjson::add(response, "ConsumedCapacity", std::move(consumption));
@@ -53,7 +53,9 @@ void consumed_capacity_counter::add_consumed_capacity_to_response_if_needed(rjso
 }
 
 static uint64_t calculate_half_units(uint64_t unit_block_size, uint64_t total_bytes, bool is_quorum) {
-    uint64_t half_units = (total_bytes + unit_block_size -1) / unit_block_size; //divide by unit_block_size and round up
+    // Avoid potential integer overflow when total_bytes is close to UINT64_MAX
+    // by using division with modulo instead of addition before division
+    uint64_t half_units = total_bytes / unit_block_size + (total_bytes % unit_block_size != 0 ? 1 : 0);
 
     if (is_quorum) {
         half_units *= 2;

alternator/consumed_capacity.hh

@@ -28,9 +28,9 @@ namespace alternator {
 class consumed_capacity_counter {
 public:
     consumed_capacity_counter() = default;
-    consumed_capacity_counter(bool should_add_to_reponse) : _should_add_to_reponse(should_add_to_reponse){}
+    consumed_capacity_counter(bool should_add_to_response) : _should_add_to_response(should_add_to_response){}
     bool operator()() const noexcept {
-        return _should_add_to_reponse;
+        return _should_add_to_response;
     }
 
     consumed_capacity_counter& operator +=(uint64_t bytes);
@@ -44,7 +44,7 @@ public:
     uint64_t _total_bytes = 0;
     static bool should_add_capacity(const rjson::value& request);
 protected:
-    bool _should_add_to_reponse = false;
+    bool _should_add_to_response = false;
 };
 
 class rcu_consumed_capacity_counter : public consumed_capacity_counter {

alternator/executor.cc

@@ -834,11 +834,13 @@ future<> executor::fill_table_size(rjson::value &table_description, schema_ptr s
             total_size = co_await _ss.estimate_total_sstable_volume(schema->id(), service::storage_service::ignore_errors::yes);
             const auto expiry = std::chrono::seconds{ _proxy.data_dictionary().get_config().alternator_describe_table_info_cache_validity_in_seconds() };
             // Note: we don't care when the notification of other shards will finish, as long as it will be done
-            // it's possible to get into race condition (next DescribeTable comes to other shard, that new shard doesn't have
-            // the size yet, so it will calculate it again) - this is not a problem, because it will call cache_newly_calculated_size_on_all_shards
-            // with expiry, which is extremely unlikely to be exactly the same as the previous one, all shards will keep the size coming with expiry that is further into the future.
-            // In case of the same expiry, some shards will have different size, which means DescribeTable will return different values depending on the shard
-            // which is also fine, as the specification doesn't give precision guarantees of any kind.
+            // A race condition is possible: if a DescribeTable request arrives on a different shard before
+            // that shard receives the cached size, it will recalculate independently. This is acceptable because:
+            // 1. Both calculations will cache their results with an expiry time
+            // 2. Expiry times are unlikely to be identical, so eventually all shards converge to the most recent value
+            // 3. Even if expiry times match, different shards may briefly return different table sizes
+            // 4. This temporary inconsistency is acceptable per DynamoDB specification, which doesn't guarantee
+            //    exact precision for DescribeTable size information
             co_await cache_newly_calculated_size_on_all_shards(schema, total_size, expiry);
         }
     }

alternator/streams.cc

@@ -491,7 +491,7 @@ future<executor::request_return_type> executor::describe_stream(client_state& cl
 
     if (!opts.enabled()) {
         rjson::add(ret, "StreamDescription", std::move(stream_desc));
-        return make_ready_future<executor::request_return_type>(rjson::print(std::move(ret)));
+        co_return rjson::print(std::move(ret));
     }
 
     // TODO: label
@@ -502,123 +502,121 @@ future<executor::request_return_type> executor::describe_stream(client_state& cl
     // filter out cdc generations older than the table or now() - cdc::ttl (typically dynamodb_streams_max_window - 24h)
     auto low_ts = std::max(as_timepoint(schema->id()), db_clock::now() - ttl);
 
-    return _sdks.cdc_get_versioned_streams(low_ts, { normal_token_owners }).then([db, shard_start, limit, ret = std::move(ret), stream_desc = std::move(stream_desc)] (std::map<db_clock::time_point, cdc::streams_version> topologies) mutable {
+    std::map<db_clock::time_point, cdc::streams_version> topologies = co_await _sdks.cdc_get_versioned_streams(low_ts, { normal_token_owners });
+    auto e = topologies.end();
+    auto prev = e;
+    auto shards = rjson::empty_array();
 
-        auto e = topologies.end();
-        auto prev = e;
-        auto shards = rjson::empty_array();
+    std::optional<shard_id> last;
 
-        std::optional<shard_id> last;
+    auto i = topologies.begin();
+    // if we're a paged query, skip to the generation where we left of.
+    if (shard_start) {
+        i = topologies.find(shard_start->time);
+    }
 
-        auto i = topologies.begin();
-        // if we're a paged query, skip to the generation where we left of.
-        if (shard_start) {
-            i = topologies.find(shard_start->time);
-        }
+    // for parent-child stuff we need id:s to be sorted by token
+    // (see explanation above) since we want to find closest
+    // token boundary when determining parent.
+    // #7346 - we processed and searched children/parents in
+    // stored order, which is not necessarily token order,
+    // so the finding of "closest" token boundary (using upper bound)
+    // could give somewhat weird results.
+    static auto token_cmp = [](const cdc::stream_id& id1, const cdc::stream_id& id2) {
+        return id1.token() < id2.token();
+    };
 
-        // for parent-child stuff we need id:s to be sorted by token
-        // (see explanation above) since we want to find closest
-        // token boundary when determining parent.
-        // #7346 - we processed and searched children/parents in
-        // stored order, which is not necessarily token order,
-        // so the finding of "closest" token boundary (using upper bound)
-        // could give somewhat weird results.
-        static auto token_cmp = [](const cdc::stream_id& id1, const cdc::stream_id& id2) {
-            return id1.token() < id2.token();
-        };
+    // #7409 - shards must be returned in lexicographical order,
+    // normal bytes compare is string_traits<int8_t>::compare.
+    // thus bytes 0x8000 is less than 0x0000. By doing unsigned
+    // compare instead we inadvertently will sort in string lexical.
+    static auto id_cmp = [](const cdc::stream_id& id1, const cdc::stream_id& id2) {
+        return compare_unsigned(id1.to_bytes(), id2.to_bytes()) < 0;
+    };
+
+    // need a prev even if we are skipping stuff
+    if (i != topologies.begin()) {
+        prev = std::prev(i);
+    }
+
+    for (; limit > 0 && i != e; prev = i, ++i) {
+        auto& [ts, sv] = *i;
+
+        last = std::nullopt;
+
+        auto lo = sv.streams.begin();
+        auto end = sv.streams.end();
 
         // #7409 - shards must be returned in lexicographical order,
-        // normal bytes compare is string_traits<int8_t>::compare.
-        // thus bytes 0x8000 is less than 0x0000. By doing unsigned
-        // compare instead we inadvertently will sort in string lexical.
-        static auto id_cmp = [](const cdc::stream_id& id1, const cdc::stream_id& id2) {
-            return compare_unsigned(id1.to_bytes(), id2.to_bytes()) < 0;
-        };
+        std::sort(lo, end, id_cmp);
 
-        // need a prev even if we are skipping stuff
-        if (i != topologies.begin()) {
-            prev = std::prev(i);
+        if (shard_start) {
+            // find next shard position
+            lo = std::upper_bound(lo, end, shard_start->id, id_cmp);
+            shard_start = std::nullopt;
         }
 
-        for (; limit > 0 && i != e; prev = i, ++i) {
-            auto& [ts, sv] = *i;
+        if (lo != end && prev != e) {
+            // We want older stuff sorted in token order so we can find matching
+            // token range when determining parent shard.
+            std::stable_sort(prev->second.streams.begin(), prev->second.streams.end(), token_cmp);
+        }
+
+        auto expired = [&]() -> std::optional<db_clock::time_point> {
+            auto j = std::next(i);
+            if (j == e) {
+                return std::nullopt;
+            }
+            // add this so we sort of match potential 
+            // sequence numbers in get_records result.
+            return j->first + confidence_interval(db);
+        }();
+
+        while (lo != end) {
+            auto& id = *lo++;
+
+            auto shard = rjson::empty_object();
+
+            if (prev != e) {
+                auto& pids = prev->second.streams;
+                auto pid = std::upper_bound(pids.begin(), pids.end(), id.token(), [](const dht::token& t, const cdc::stream_id& id) {
+                    return t < id.token();
+                });
+                if (pid != pids.begin()) {
+                    pid = std::prev(pid);
+                }
+                if (pid != pids.end()) {
+                    rjson::add(shard, "ParentShardId", shard_id(prev->first, *pid));
+                }
+            }
+
+            last.emplace(ts, id);
+            rjson::add(shard, "ShardId", *last);
+            auto range = rjson::empty_object();
+            rjson::add(range, "StartingSequenceNumber", sequence_number(utils::UUID_gen::min_time_UUID(ts.time_since_epoch())));
+            if (expired) {
+                rjson::add(range, "EndingSequenceNumber", sequence_number(utils::UUID_gen::min_time_UUID(expired->time_since_epoch())));
+            }
+
+            rjson::add(shard, "SequenceNumberRange", std::move(range));
+            rjson::push_back(shards, std::move(shard));
+            
+            if (--limit == 0) {
+                break;
+            }
 
             last = std::nullopt;
-
-            auto lo = sv.streams.begin();
-            auto end = sv.streams.end();
-
-            // #7409 - shards must be returned in lexicographical order,
-            std::sort(lo, end, id_cmp);
-
-            if (shard_start) {
-                // find next shard position
-                lo = std::upper_bound(lo, end, shard_start->id, id_cmp);
-                shard_start = std::nullopt;
-            }
-
-            if (lo != end && prev != e) {
-                // We want older stuff sorted in token order so we can find matching
-                // token range when determining parent shard.
-                std::stable_sort(prev->second.streams.begin(), prev->second.streams.end(), token_cmp);
-            }
-
-            auto expired = [&]() -> std::optional<db_clock::time_point> {
-                auto j = std::next(i);
-                if (j == e) {
-                    return std::nullopt;
-                }
-                // add this so we sort of match potential 
-                // sequence numbers in get_records result.
-                return j->first + confidence_interval(db);
-            }();
-
-            while (lo != end) {
-                auto& id = *lo++;
-
-                auto shard = rjson::empty_object();
-
-                if (prev != e) {
-                    auto& pids = prev->second.streams;
-                    auto pid = std::upper_bound(pids.begin(), pids.end(), id.token(), [](const dht::token& t, const cdc::stream_id& id) {
-                        return t < id.token();
-                    });
-                    if (pid != pids.begin()) {
-                        pid = std::prev(pid);
-                    }
-                    if (pid != pids.end()) {
-                        rjson::add(shard, "ParentShardId", shard_id(prev->first, *pid));
-                    }
-                }
-
-                last.emplace(ts, id);
-                rjson::add(shard, "ShardId", *last);
-                auto range = rjson::empty_object();
-                rjson::add(range, "StartingSequenceNumber", sequence_number(utils::UUID_gen::min_time_UUID(ts.time_since_epoch())));
-                if (expired) {
-                    rjson::add(range, "EndingSequenceNumber", sequence_number(utils::UUID_gen::min_time_UUID(expired->time_since_epoch())));
-                }
-
-                rjson::add(shard, "SequenceNumberRange", std::move(range));
-                rjson::push_back(shards, std::move(shard));
-                
-                if (--limit == 0) {
-                    break;
-                }
-
-                last = std::nullopt;
-            }
         }
+    }
 
-        if (last) {
-            rjson::add(stream_desc, "LastEvaluatedShardId", *last);
-        }
+    if (last) {
+        rjson::add(stream_desc, "LastEvaluatedShardId", *last);
+    }
 
-        rjson::add(stream_desc, "Shards", std::move(shards));
-        rjson::add(ret, "StreamDescription", std::move(stream_desc));
-            
-        return make_ready_future<executor::request_return_type>(rjson::print(std::move(ret)));
-    });
+    rjson::add(stream_desc, "Shards", std::move(shards));
+    rjson::add(ret, "StreamDescription", std::move(stream_desc));
+        
+    co_return rjson::print(std::move(ret));
 }
 
 enum class shard_iterator_type {
@@ -898,172 +896,169 @@ future<executor::request_return_type> executor::get_records(client_state& client
     auto command = ::make_lw_shared<query::read_command>(schema->id(), schema->version(), partition_slice, _proxy.get_max_result_size(partition_slice),
             query::tombstone_limit(_proxy.get_tombstone_limit()), query::row_limit(limit * mul));
 
-    co_return co_await _proxy.query(schema, std::move(command), std::move(partition_ranges), cl, service::storage_proxy::coordinator_query_options(default_timeout(), std::move(permit), client_state)).then(
-            [this, schema, partition_slice = std::move(partition_slice), selection = std::move(selection), start_time = std::move(start_time), limit, key_names = std::move(key_names), attr_names = std::move(attr_names), type, iter, high_ts] (service::storage_proxy::coordinator_query_result qr) mutable {       
-        cql3::selection::result_set_builder builder(*selection, gc_clock::now());
-        query::result_view::consume(*qr.query_result, partition_slice, cql3::selection::result_set_builder::visitor(builder, *schema, *selection));
+    service::storage_proxy::coordinator_query_result qr = co_await _proxy.query(schema, std::move(command), std::move(partition_ranges), cl, service::storage_proxy::coordinator_query_options(default_timeout(), std::move(permit), client_state));
+    cql3::selection::result_set_builder builder(*selection, gc_clock::now());
+    query::result_view::consume(*qr.query_result, partition_slice, cql3::selection::result_set_builder::visitor(builder, *schema, *selection));
 
-        auto result_set = builder.build();
-        auto records = rjson::empty_array();
+    auto result_set = builder.build();
+    auto records = rjson::empty_array();
 
-        auto& metadata = result_set->get_metadata();
+    auto& metadata = result_set->get_metadata();
 
-        auto op_index = std::distance(metadata.get_names().begin(), 
-            std::find_if(metadata.get_names().begin(), metadata.get_names().end(), [](const lw_shared_ptr<cql3::column_specification>& cdef) {
-                return cdef->name->name() == op_column_name;
-            })
-        );
-        auto ts_index = std::distance(metadata.get_names().begin(), 
-            std::find_if(metadata.get_names().begin(), metadata.get_names().end(), [](const lw_shared_ptr<cql3::column_specification>& cdef) {
-                return cdef->name->name() == timestamp_column_name;
-            })
-        );
-        auto eor_index = std::distance(metadata.get_names().begin(), 
-            std::find_if(metadata.get_names().begin(), metadata.get_names().end(), [](const lw_shared_ptr<cql3::column_specification>& cdef) {
-                return cdef->name->name() == eor_column_name;
-            })
-        );
+    auto op_index = std::distance(metadata.get_names().begin(), 
+        std::find_if(metadata.get_names().begin(), metadata.get_names().end(), [](const lw_shared_ptr<cql3::column_specification>& cdef) {
+            return cdef->name->name() == op_column_name;
+        })
+    );
+    auto ts_index = std::distance(metadata.get_names().begin(), 
+        std::find_if(metadata.get_names().begin(), metadata.get_names().end(), [](const lw_shared_ptr<cql3::column_specification>& cdef) {
+            return cdef->name->name() == timestamp_column_name;
+        })
+    );
+    auto eor_index = std::distance(metadata.get_names().begin(), 
+        std::find_if(metadata.get_names().begin(), metadata.get_names().end(), [](const lw_shared_ptr<cql3::column_specification>& cdef) {
+            return cdef->name->name() == eor_column_name;
+        })
+    );
 
-        std::optional<utils::UUID> timestamp;
-        auto dynamodb = rjson::empty_object();
-        auto record = rjson::empty_object();
-        const auto dc_name = _proxy.get_token_metadata_ptr()->get_topology().get_datacenter();
+    std::optional<utils::UUID> timestamp;
+    auto dynamodb = rjson::empty_object();
+    auto record = rjson::empty_object();
+    const auto dc_name = _proxy.get_token_metadata_ptr()->get_topology().get_datacenter();
 
-        using op_utype = std::underlying_type_t<cdc::operation>;
+    using op_utype = std::underlying_type_t<cdc::operation>;
 
-        auto maybe_add_record = [&] {
-            if (!dynamodb.ObjectEmpty()) {
-                rjson::add(record, "dynamodb", std::move(dynamodb));
-                dynamodb = rjson::empty_object();
-            }
-            if (!record.ObjectEmpty()) {
-                rjson::add(record, "awsRegion", rjson::from_string(dc_name));
-                rjson::add(record, "eventID", event_id(iter.shard.id, *timestamp));
-                rjson::add(record, "eventSource", "scylladb:alternator");
-                rjson::add(record, "eventVersion", "1.1");
-                rjson::push_back(records, std::move(record));
-                record = rjson::empty_object();
-                --limit;
-            }
-        };
+    auto maybe_add_record = [&] {
+        if (!dynamodb.ObjectEmpty()) {
+            rjson::add(record, "dynamodb", std::move(dynamodb));
+            dynamodb = rjson::empty_object();
+        }
+        if (!record.ObjectEmpty()) {
+            rjson::add(record, "awsRegion", rjson::from_string(dc_name));
+            rjson::add(record, "eventID", event_id(iter.shard.id, *timestamp));
+            rjson::add(record, "eventSource", "scylladb:alternator");
+            rjson::add(record, "eventVersion", "1.1");
+            rjson::push_back(records, std::move(record));
+            record = rjson::empty_object();
+            --limit;
+        }
+    };
 
-        for (auto& row : result_set->rows()) {
-            auto op = static_cast<cdc::operation>(value_cast<op_utype>(data_type_for<op_utype>()->deserialize(*row[op_index])));
-            auto ts = value_cast<utils::UUID>(data_type_for<utils::UUID>()->deserialize(*row[ts_index]));
-            auto eor = row[eor_index].has_value() ? value_cast<bool>(boolean_type->deserialize(*row[eor_index])) : false;
+    for (auto& row : result_set->rows()) {
+        auto op = static_cast<cdc::operation>(value_cast<op_utype>(data_type_for<op_utype>()->deserialize(*row[op_index])));
+        auto ts = value_cast<utils::UUID>(data_type_for<utils::UUID>()->deserialize(*row[ts_index]));
+        auto eor = row[eor_index].has_value() ? value_cast<bool>(boolean_type->deserialize(*row[eor_index])) : false;
 
-            if (!dynamodb.HasMember("Keys")) {
-                auto keys = rjson::empty_object();
-                describe_single_item(*selection, row, key_names, keys);
-                rjson::add(dynamodb, "Keys", std::move(keys));
-                rjson::add(dynamodb, "ApproximateCreationDateTime", utils::UUID_gen::unix_timestamp_in_sec(ts).count());
-                rjson::add(dynamodb, "SequenceNumber", sequence_number(ts));
-                rjson::add(dynamodb, "StreamViewType", type);
-                // TODO: SizeBytes
-            }
-
-            /**
-             * We merge rows with same timestamp into a single event.
-             * This is pretty much needed, because a CDC row typically
-             * encodes ~half the info of an alternator write. 
-             * 
-             * A big, big downside to how alternator records are written
-             * (i.e. CQL), is that the distinction between INSERT and UPDATE
-             * is somewhat lost/unmappable to actual eventName. 
-             * A write (currently) always looks like an insert+modify
-             * regardless whether we wrote existing record or not. 
-             * 
-             * Maybe RMW ops could be done slightly differently so 
-             * we can distinguish them here...
-             * 
-             * For now, all writes will become MODIFY.
-             * 
-             * Note: we do not check the current pre/post
-             * flags on CDC log, instead we use data to 
-             * drive what is returned. This is (afaict)
-             * consistent with dynamo streams
-             */
-            switch (op) {
-            case cdc::operation::pre_image:
-            case cdc::operation::post_image:
-            {
-                auto item = rjson::empty_object();
-                describe_single_item(*selection, row, attr_names, item, nullptr, true);
-                describe_single_item(*selection, row, key_names, item);
-                rjson::add(dynamodb, op == cdc::operation::pre_image ? "OldImage" : "NewImage", std::move(item));
-                break;
-            }
-            case cdc::operation::update:
-                rjson::add(record, "eventName", "MODIFY");
-                break;
-            case cdc::operation::insert:
-                rjson::add(record, "eventName", "INSERT");
-                break;
-            case cdc::operation::service_row_delete:
-            case cdc::operation::service_partition_delete:
-            {
-                auto user_identity = rjson::empty_object();
-                rjson::add(user_identity, "Type", "Service");
-                rjson::add(user_identity, "PrincipalId", "dynamodb.amazonaws.com");
-                rjson::add(record, "userIdentity", std::move(user_identity));
-                rjson::add(record, "eventName", "REMOVE");
-                break;
-            }
-            default:
-                rjson::add(record, "eventName", "REMOVE");
-                break;
-            }
-            if (eor) {
-                maybe_add_record();
-                timestamp = ts;
-                if (limit == 0) {
-                    break;
-                }
-            }
+        if (!dynamodb.HasMember("Keys")) {
+            auto keys = rjson::empty_object();
+            describe_single_item(*selection, row, key_names, keys);
+            rjson::add(dynamodb, "Keys", std::move(keys));
+            rjson::add(dynamodb, "ApproximateCreationDateTime", utils::UUID_gen::unix_timestamp_in_sec(ts).count());
+            rjson::add(dynamodb, "SequenceNumber", sequence_number(ts));
+            rjson::add(dynamodb, "StreamViewType", type);
+            // TODO: SizeBytes
         }
 
-        auto ret = rjson::empty_object();
-        auto nrecords = records.Size();
-        rjson::add(ret, "Records", std::move(records));
-
-        if (nrecords != 0) {
-            // #9642. Set next iterators threshold to > last
-            shard_iterator next_iter(iter.table, iter.shard, *timestamp, false);
-            // Note that here we unconditionally return NextShardIterator,
-            // without checking if maybe we reached the end-of-shard. If the
-            // shard did end, then the next read will have nrecords == 0 and
-            // will notice end end of shard and not return NextShardIterator.
-            rjson::add(ret, "NextShardIterator", next_iter);
-            _stats.api_operations.get_records_latency.mark(std::chrono::steady_clock::now() - start_time);
-            return make_ready_future<executor::request_return_type>(rjson::print(std::move(ret)));
+        /**
+         * We merge rows with same timestamp into a single event.
+         * This is pretty much needed, because a CDC row typically
+         * encodes ~half the info of an alternator write. 
+         * 
+         * A big, big downside to how alternator records are written
+         * (i.e. CQL), is that the distinction between INSERT and UPDATE
+         * is somewhat lost/unmappable to actual eventName. 
+         * A write (currently) always looks like an insert+modify
+         * regardless whether we wrote existing record or not. 
+         * 
+         * Maybe RMW ops could be done slightly differently so 
+         * we can distinguish them here...
+         * 
+         * For now, all writes will become MODIFY.
+         * 
+         * Note: we do not check the current pre/post
+         * flags on CDC log, instead we use data to 
+         * drive what is returned. This is (afaict)
+         * consistent with dynamo streams
+         */
+        switch (op) {
+        case cdc::operation::pre_image:
+        case cdc::operation::post_image:
+        {
+            auto item = rjson::empty_object();
+            describe_single_item(*selection, row, attr_names, item, nullptr, true);
+            describe_single_item(*selection, row, key_names, item);
+            rjson::add(dynamodb, op == cdc::operation::pre_image ? "OldImage" : "NewImage", std::move(item));
+            break;
         }
-
-        // ugh. figure out if we are and end-of-shard
-        auto normal_token_owners = _proxy.get_token_metadata_ptr()->count_normal_token_owners();
-
-        return _sdks.cdc_current_generation_timestamp({ normal_token_owners }).then([this, iter, high_ts, start_time, ret = std::move(ret)](db_clock::time_point ts) mutable {
-            auto& shard = iter.shard;            
-
-            if (shard.time < ts && ts < high_ts) {
-                // The DynamoDB documentation states that when a shard is
-                // closed, reading it until the end has NextShardIterator
-                // "set to null". Our test test_streams_closed_read
-                // confirms that by "null" they meant not set at all.
-            } else {
-                // We could have return the same iterator again, but we did
-                // a search from it until high_ts and found nothing, so we
-                // can also start the next search from high_ts.
-                // TODO: but why? It's simpler just to leave the iterator be.
-                shard_iterator next_iter(iter.table, iter.shard, utils::UUID_gen::min_time_UUID(high_ts.time_since_epoch()), true);
-                rjson::add(ret, "NextShardIterator", iter);
+        case cdc::operation::update:
+            rjson::add(record, "eventName", "MODIFY");
+            break;
+        case cdc::operation::insert:
+            rjson::add(record, "eventName", "INSERT");
+            break;
+        case cdc::operation::service_row_delete:
+        case cdc::operation::service_partition_delete:
+        {
+            auto user_identity = rjson::empty_object();
+            rjson::add(user_identity, "Type", "Service");
+            rjson::add(user_identity, "PrincipalId", "dynamodb.amazonaws.com");
+            rjson::add(record, "userIdentity", std::move(user_identity));
+            rjson::add(record, "eventName", "REMOVE");
+            break;
+        }
+        default:
+            rjson::add(record, "eventName", "REMOVE");
+            break;
+        }
+        if (eor) {
+            maybe_add_record();
+            timestamp = ts;
+            if (limit == 0) {
+                break;
             }
-            _stats.api_operations.get_records_latency.mark(std::chrono::steady_clock::now() - start_time);
-            if (is_big(ret)) {
-                return make_ready_future<executor::request_return_type>(make_streamed(std::move(ret)));
-            }
-            return make_ready_future<executor::request_return_type>(rjson::print(std::move(ret)));
-        });
-    });
+        }
+    }
+
+    auto ret = rjson::empty_object();
+    auto nrecords = records.Size();
+    rjson::add(ret, "Records", std::move(records));
+
+    if (nrecords != 0) {
+        // #9642. Set next iterators threshold to > last
+        shard_iterator next_iter(iter.table, iter.shard, *timestamp, false);
+        // Note that here we unconditionally return NextShardIterator,
+        // without checking if maybe we reached the end-of-shard. If the
+        // shard did end, then the next read will have nrecords == 0 and
+        // will notice end end of shard and not return NextShardIterator.
+        rjson::add(ret, "NextShardIterator", next_iter);
+        _stats.api_operations.get_records_latency.mark(std::chrono::steady_clock::now() - start_time);
+        co_return rjson::print(std::move(ret));
+    }
+
+    // ugh. figure out if we are and end-of-shard
+    auto normal_token_owners = _proxy.get_token_metadata_ptr()->count_normal_token_owners();
+
+    db_clock::time_point ts = co_await _sdks.cdc_current_generation_timestamp({ normal_token_owners });
+    auto& shard = iter.shard;
+
+    if (shard.time < ts && ts < high_ts) {
+        // The DynamoDB documentation states that when a shard is
+        // closed, reading it until the end has NextShardIterator
+        // "set to null". Our test test_streams_closed_read
+        // confirms that by "null" they meant not set at all.
+    } else {
+        // We could have return the same iterator again, but we did
+        // a search from it until high_ts and found nothing, so we
+        // can also start the next search from high_ts.
+        // TODO: but why? It's simpler just to leave the iterator be.
+        shard_iterator next_iter(iter.table, iter.shard, utils::UUID_gen::min_time_UUID(high_ts.time_since_epoch()), true);
+        rjson::add(ret, "NextShardIterator", iter);
+    }
+    _stats.api_operations.get_records_latency.mark(std::chrono::steady_clock::now() - start_time);
+    if (is_big(ret)) {
+        co_return make_streamed(std::move(ret));
+    }
+    co_return rjson::print(std::move(ret));
 }
 
 bool executor::add_stream_options(const rjson::value& stream_specification, schema_builder& builder, service::storage_proxy& sp) {

alternator/ttl.cc

@@ -767,7 +767,7 @@ static future<bool> scan_table(
                 // by tasking another node to take over scanning of the dead node's primary
                 // ranges. What we do here is that this node will also check expiration
                 // on its *secondary* ranges - but only those whose primary owner is down.
-                auto tablet_secondary_replica = tablet_map.get_secondary_replica(*tablet, erm->get_topology()); // throws if no secondary replica
+                auto tablet_secondary_replica = tablet_map.get_secondary_replica(*tablet); // throws if no secondary replica
                 if (tablet_secondary_replica.host == my_host_id && tablet_secondary_replica.shard == this_shard_id()) {
                     if (!gossiper.is_alive(tablet_primary_replica.host)) {
                         co_await scan_tablet(*tablet, proxy, abort_source, page_sem, expiration_stats, scan_ctx, tablet_map);

api/storage_service.cc

@@ -515,15 +515,6 @@ void set_sstables_loader(http_context& ctx, routes& r, sharded<sstables_loader>&
         auto sstables = parsed.GetArray() |
             std::views::transform([] (const auto& s) { return sstring(rjson::to_string_view(s)); }) |
             std::ranges::to<std::vector>();
-        apilog.info("Restore invoked with following parameters: keyspace={}, table={}, endpoint={}, bucket={}, prefix={}, sstables_count={}, scope={}, primary_replica_only={}",
-                    keyspace,
-                    table,
-                    endpoint,
-                    bucket,
-                    prefix,
-                    sstables.size(),
-                    scope,
-                    primary_replica_only);
         auto task_id = co_await sst_loader.local().download_new_sstables(keyspace, table, prefix, std::move(sstables), endpoint, bucket, scope, primary_replica_only);
         co_return json::json_return_type(fmt::to_string(task_id));
     });

audit/audit.cc

@@ -209,15 +209,11 @@ future<> audit::stop_audit() {
     });
 }
 
-audit_info_ptr audit::create_audit_info(statement_category cat, const sstring& keyspace, const sstring& table) {
+audit_info_ptr audit::create_audit_info(statement_category cat, const sstring& keyspace, const sstring& table, bool batch) {
     if (!audit_instance().local_is_initialized()) {
         return nullptr;
     }
-    return std::make_unique<audit_info>(cat, keyspace, table);
-}
-
-audit_info_ptr audit::create_no_audit_info() {
-    return audit_info_ptr();
+    return std::make_unique<audit_info>(cat, keyspace, table, batch);
 }
 
 future<> audit::start(const db::config& cfg) {
@@ -267,18 +263,21 @@ future<> audit::log_login(const sstring& username, socket_address client_ip, boo
 }
 
 future<> inspect(shared_ptr<cql3::cql_statement> statement, service::query_state& query_state, const cql3::query_options& options, bool error) {
-    cql3::statements::batch_statement* batch = dynamic_cast<cql3::statements::batch_statement*>(statement.get());
-    if (batch != nullptr) {
+    auto audit_info = statement->get_audit_info();
+    if (!audit_info) {
+        return make_ready_future<>();
+    }
+    if (audit_info->batch()) {
+        cql3::statements::batch_statement* batch = static_cast<cql3::statements::batch_statement*>(statement.get());
         return do_for_each(batch->statements().begin(), batch->statements().end(), [&query_state, &options, error] (auto&& m) {
             return inspect(m.statement, query_state, options, error);
         });
     } else {
-        auto audit_info = statement->get_audit_info();
-        if (bool(audit_info) && audit::local_audit_instance().should_log(audit_info)) {
+        if (audit::local_audit_instance().should_log(audit_info)) {
             return audit::local_audit_instance().log(audit_info, query_state, options, error);
         }
+        return make_ready_future<>();
     }
-    return make_ready_future<>();
 }
 
 future<> inspect_login(const sstring& username, socket_address client_ip, bool error) {

audit/audit.hh

@@ -75,11 +75,13 @@ class audit_info final {
     sstring _keyspace;
     sstring _table;
     sstring _query;
+    bool _batch;
 public:
-    audit_info(statement_category cat, sstring keyspace, sstring table)
+    audit_info(statement_category cat, sstring keyspace, sstring table, bool batch)
         : _category(cat)
         , _keyspace(std::move(keyspace))
         , _table(std::move(table))
+        , _batch(batch)
     { }
     void set_query_string(const std::string_view& query_string) {
         _query = sstring(query_string);
@@ -89,6 +91,7 @@ public:
     const sstring& query() const { return _query; }
     sstring category_string() const;
     statement_category category() const { return _category; }
+    bool batch() const { return _batch; }
 };
 
 using audit_info_ptr = std::unique_ptr<audit_info>;
@@ -126,8 +129,7 @@ public:
     }
     static future<> start_audit(const db::config& cfg, sharded<locator::shared_token_metadata>& stm, sharded<cql3::query_processor>& qp, sharded<service::migration_manager>& mm);
     static future<> stop_audit();
-    static audit_info_ptr create_audit_info(statement_category cat, const sstring& keyspace, const sstring& table);
-    static audit_info_ptr create_no_audit_info();
+    static audit_info_ptr create_audit_info(statement_category cat, const sstring& keyspace, const sstring& table, bool batch = false);
     audit(locator::shared_token_metadata& stm,
           cql3::query_processor& qp,
           service::migration_manager& mm,

auth/standard_role_manager.cc

@@ -52,13 +52,6 @@ static const class_registrator<
         ::service::migration_manager&,
         cache&> registration("org.apache.cassandra.auth.CassandraRoleManager");
 
-struct record final {
-    sstring name;
-    bool is_superuser;
-    bool can_login;
-    role_set member_of;
-};
-
 static db::consistency_level consistency_for_role(std::string_view role_name) noexcept {
     if (role_name == meta::DEFAULT_SUPERUSER_NAME) {
         return db::consistency_level::QUORUM;
@@ -67,13 +60,13 @@ static db::consistency_level consistency_for_role(std::string_view role_name) no
     return db::consistency_level::LOCAL_ONE;
 }
 
-static future<std::optional<record>> find_record(cql3::query_processor& qp, std::string_view role_name) {
+future<std::optional<standard_role_manager::record>> standard_role_manager::legacy_find_record(std::string_view role_name) {
     const sstring query = seastar::format("SELECT * FROM {}.{} WHERE {} = ?",
-            get_auth_ks_name(qp),
+            get_auth_ks_name(_qp),
             meta::roles_table::name,
             meta::roles_table::role_col_name);
 
-    const auto results = co_await qp.execute_internal(
+    const auto results = co_await _qp.execute_internal(
             query,
             consistency_for_role(role_name),
             internal_distributed_query_state(),
@@ -93,8 +86,25 @@ static future<std::optional<record>> find_record(cql3::query_processor& qp, std:
                         : role_set())});
 }
 
-static future<record> require_record(cql3::query_processor& qp, std::string_view role_name) {
-    return find_record(qp, role_name).then([role_name](std::optional<record> mr) {
+future<std::optional<standard_role_manager::record>> standard_role_manager::find_record(std::string_view role_name) {
+    if (legacy_mode(_qp)) {
+        return legacy_find_record(role_name);
+    }
+    auto name = sstring(role_name);
+    auto role = _cache.get(name);
+    if (!role) {
+        return make_ready_future<std::optional<record>>(std::nullopt);
+    }
+    return make_ready_future<std::optional<record>>(std::make_optional(record{
+        .name = std::move(name),
+        .is_superuser = role->is_superuser,
+        .can_login = role->can_login,
+        .member_of = role->member_of
+    }));
+}
+
+future<standard_role_manager::record> standard_role_manager::require_record(std::string_view role_name) {
+    return find_record(role_name).then([role_name](std::optional<record> mr) {
         if (!mr) {
             throw nonexistant_role(role_name);
         }
@@ -386,7 +396,7 @@ standard_role_manager::alter(std::string_view role_name, const role_config_updat
         return fmt::to_string(fmt::join(assignments, ", "));
     };
 
-    return require_record(_qp, role_name).then([this, role_name, &u, &mc](record) {
+    return require_record(role_name).then([this, role_name, &u, &mc](record) {
         if (!u.is_superuser && !u.can_login) {
             return make_ready_future<>();
         }
@@ -620,18 +630,17 @@ standard_role_manager::revoke(std::string_view revokee_name, std::string_view ro
     });
 }
 
-static future<> collect_roles(
-        cql3::query_processor& qp,
+future<> standard_role_manager::collect_roles(
         std::string_view grantee_name,
         bool recurse,
         role_set& roles) {
-    return require_record(qp, grantee_name).then([&qp, &roles, recurse](record r) {
-        return do_with(std::move(r.member_of), [&qp, &roles, recurse](const role_set& memberships) {
-            return do_for_each(memberships.begin(), memberships.end(), [&qp, &roles, recurse](const sstring& role_name) {
+    return require_record(grantee_name).then([this, &roles, recurse](standard_role_manager::record r) {
+        return do_with(std::move(r.member_of), [this, &roles, recurse](const role_set& memberships) {
+            return do_for_each(memberships.begin(), memberships.end(), [this, &roles, recurse](const sstring& role_name) {
                 roles.insert(role_name);
 
                 if (recurse) {
-                    return collect_roles(qp, role_name, true, roles);
+                    return collect_roles(role_name, true, roles);
                 }
 
                 return make_ready_future<>();
@@ -646,7 +655,7 @@ future<role_set> standard_role_manager::query_granted(std::string_view grantee_n
     return do_with(
             role_set{sstring(grantee_name)},
             [this, grantee_name, recurse](role_set& roles) {
-        return collect_roles(_qp, grantee_name, recurse, roles).then([&roles] { return roles; });
+        return collect_roles(grantee_name, recurse, roles).then([&roles] { return roles; });
     });
 }
 
@@ -706,27 +715,21 @@ future<role_set> standard_role_manager::query_all(::service::query_state& qs) {
 }
 
 future<bool> standard_role_manager::exists(std::string_view role_name) {
-    return find_record(_qp, role_name).then([](std::optional<record> mr) {
+    return find_record(role_name).then([](std::optional<record> mr) {
         return static_cast<bool>(mr);
     });
 }
 
 future<bool> standard_role_manager::is_superuser(std::string_view role_name) {
-    return require_record(_qp, role_name).then([](record r) {
+    return require_record(role_name).then([](record r) {
         return r.is_superuser;
     });
 }
 
 future<bool> standard_role_manager::can_login(std::string_view role_name) {
-    if (legacy_mode(_qp)) {
-       const auto r = co_await require_record(_qp, role_name);
-       co_return r.can_login;
-    }
-    auto role = _cache.get(sstring(role_name));
-    if (!role) {
-        throw nonexistant_role(role_name);
-    }
-    co_return role->can_login;
+    return require_record(role_name).then([](record r) {
+        return r.can_login;
+    });
 }
 
 future<std::optional<sstring>> standard_role_manager::get_attribute(std::string_view role_name, std::string_view attribute_name, ::service::query_state& qs) {

auth/standard_role_manager.hh

@@ -90,6 +90,12 @@ public:
 
 private:
     enum class membership_change { add, remove };
+    struct record final {
+        sstring name;
+        bool is_superuser;
+        bool can_login;
+        role_set member_of;
+    };
 
     future<> create_legacy_metadata_tables_if_missing() const;
 
@@ -107,6 +113,14 @@ private:
     future<> legacy_modify_membership(std::string_view role_name, std::string_view grantee_name, membership_change);
 
     future<> modify_membership(std::string_view role_name, std::string_view grantee_name, membership_change, ::service::group0_batch& mc);
+
+    future<std::optional<record>> legacy_find_record(std::string_view role_name);
+    future<std::optional<record>> find_record(std::string_view role_name);
+    future<record> require_record(std::string_view role_name);
+    future<> collect_roles(
+            std::string_view grantee_name,
+            bool recurse,
+            role_set& roles);
 };
 
 } // namespace auth

compaction/compaction.cc

@@ -48,7 +48,6 @@
 #include "mutation/mutation_fragment_stream_validator.hh"
 #include "utils/assert.hh"
 #include "utils/error_injection.hh"
-#include "utils/chunked_vector.hh"
 #include "utils/pretty_printers.hh"
 #include "readers/multi_range.hh"
 #include "readers/compacting.hh"
@@ -612,23 +611,23 @@ private:
     }
 
     // Called in a seastar thread
-    utils::chunked_vector<dht::partition_range>
+    dht::partition_range_vector
     get_ranges_for_invalidation(const std::vector<sstables::shared_sstable>& sstables) {
         // If owned ranges is disengaged, it means no cleanup work was done and
         // so nothing needs to be invalidated.
         if (!_owned_ranges) {
-            return {};
+            return dht::partition_range_vector{};
         }
-        auto owned_ranges = dht::to_partition_ranges_chunked(*_owned_ranges).get();
+        auto owned_ranges = dht::to_partition_ranges(*_owned_ranges, utils::can_yield::yes);
 
         auto non_owned_ranges = sstables
                 | std::views::transform([] (const sstables::shared_sstable& sst) {
             seastar::thread::maybe_yield();
             return dht::partition_range::make({sst->get_first_decorated_key(), true},
                                               {sst->get_last_decorated_key(), true});
-        })      | std::ranges::to<utils::chunked_vector<dht::partition_range>>();
+        })      | std::ranges::to<dht::partition_range_vector>();
 
-        return dht::subtract_ranges(*_schema, std::move(non_owned_ranges), std::move(owned_ranges)).get();
+        return dht::subtract_ranges(*_schema, non_owned_ranges, std::move(owned_ranges)).get();
     }
 protected:
     compaction(compaction_group_view& table_s, compaction_descriptor descriptor, compaction_data& cdata, compaction_progress_monitor& progress_monitor, use_backlog_tracker use_backlog_tracker)
@@ -719,8 +718,8 @@ protected:
 
     compaction_completion_desc
     get_compaction_completion_desc(std::vector<sstables::shared_sstable> input_sstables, std::vector<sstables::shared_sstable> output_sstables) {
-        auto ranges = get_ranges_for_invalidation(input_sstables);
-        return compaction_completion_desc{std::move(input_sstables), std::move(output_sstables), std::move(ranges)};
+        auto ranges_for_for_invalidation = get_ranges_for_invalidation(input_sstables);
+        return compaction_completion_desc{std::move(input_sstables), std::move(output_sstables), std::move(ranges_for_for_invalidation)};
     }
 
     // Tombstone expiration is enabled based on the presence of sstable set.

compaction/compaction_descriptor.hh

@@ -16,7 +16,6 @@
 #include "sstables/sstable_set.hh"
 #include "compaction_fwd.hh"
 #include "mutation_writer/token_group_based_splitting_writer.hh"
-#include "utils/chunked_vector.hh"
 
 namespace compaction {
 
@@ -39,7 +38,7 @@ struct compaction_completion_desc {
     // New, fresh SSTables that should be added to SSTable set, replacing the old ones.
     std::vector<sstables::shared_sstable> new_sstables;
     // Set of compacted partition ranges that should be invalidated in the cache.
-    utils::chunked_vector<dht::partition_range> ranges_for_cache_invalidation;
+    dht::partition_range_vector ranges_for_cache_invalidation;
 };
 
 // creates a new SSTable for a given shard

compaction/compaction_manager.cc

@@ -778,7 +778,6 @@ compaction_manager::get_incremental_repair_read_lock(compaction::compaction_grou
         cmlog.debug("Get get_incremental_repair_read_lock for {} started", reason);
     }
     compaction::compaction_state& cs = get_compaction_state(&t);
-    auto gh = cs.gate.hold();
     auto ret = co_await cs.incremental_repair_lock.hold_read_lock();
     if (!reason.empty()) {
         cmlog.debug("Get get_incremental_repair_read_lock for {} done", reason);
@@ -792,7 +791,6 @@ compaction_manager::get_incremental_repair_write_lock(compaction::compaction_gro
         cmlog.debug("Get get_incremental_repair_write_lock for {} started", reason);
     }
     compaction::compaction_state& cs = get_compaction_state(&t);
-    auto gh = cs.gate.hold();
     auto ret = co_await cs.incremental_repair_lock.hold_write_lock();
     if (!reason.empty()) {
         cmlog.debug("Get get_incremental_repair_write_lock for {} done", reason);
@@ -1521,9 +1519,7 @@ future<> compaction_manager::maybe_wait_for_sstable_count_reduction(compaction_g
             | std::views::transform(std::mem_fn(&sstables::sstable::run_identifier))
             | std::ranges::to<std::unordered_set>());
     };
-    const auto threshold = utils::get_local_injector().inject_parameter<size_t>("set_sstable_count_reduction_threshold")
-        .value_or(size_t(std::max(schema->max_compaction_threshold(), 32)));
-
+    const auto threshold = size_t(std::max(schema->max_compaction_threshold(), 32));
     auto count = co_await num_runs_for_compaction();
     if (count <= threshold) {
         cmlog.trace("No need to wait for sstable count reduction in {}: {} <= {}",
@@ -1538,7 +1534,9 @@ future<> compaction_manager::maybe_wait_for_sstable_count_reduction(compaction_g
     auto& cstate = get_compaction_state(&t);
     try {
         while (can_perform_regular_compaction(t) && co_await num_runs_for_compaction() > threshold) {
-            co_await cstate.compaction_done.wait();
+            co_await cstate.compaction_done.wait([this, &t] {
+                return !can_perform_regular_compaction(t);
+            });
         }
     } catch (const broken_condition_variable&) {
         co_return;
@@ -2389,8 +2387,6 @@ future<> compaction_manager::remove(compaction_group_view& t, sstring reason) no
     if (!c_state.gate.is_closed()) {
         auto close_gate = c_state.gate.close();
         co_await stop_ongoing_compactions(reason, &t);
-        // Wait for users of incremental repair lock (can be either repair itself or maintenance compactions).
-        co_await c_state.incremental_repair_lock.write_lock();
         co_await std::move(close_gate);
     }
 

configure.py

@@ -730,6 +730,28 @@ vector_search_tests = set([
     'test/vector_search/rescoring_test'
 ])
 
+vector_search_validator_bin = 'vector-search-validator/bin/vector-search-validator'
+vector_search_validator_deps = set([
+    'test/vector_search_validator/build-validator',
+    'test/vector_search_validator/Cargo.toml',
+    'test/vector_search_validator/crates/validator/Cargo.toml',
+    'test/vector_search_validator/crates/validator/src/main.rs',
+    'test/vector_search_validator/crates/validator-scylla/Cargo.toml',
+    'test/vector_search_validator/crates/validator-scylla/src/lib.rs',
+    'test/vector_search_validator/crates/validator-scylla/src/cql.rs',
+])
+
+vector_store_bin = 'vector-search-validator/bin/vector-store'
+vector_store_deps = set([
+    'test/vector_search_validator/build-env',
+    'test/vector_search_validator/build-vector-store',
+])
+
+vector_search_validator_bins = set([
+    vector_search_validator_bin,
+    vector_store_bin,
+])
+
 wasms = set([
     'wasm/return_input.wat',
     'wasm/test_complex_null_values.wat',
@@ -763,7 +785,7 @@ other = set([
     'iotune',
 ])
 
-all_artifacts = apps | cpp_apps | tests | other | wasms
+all_artifacts = apps | cpp_apps | tests | other | wasms | vector_search_validator_bins
 
 arg_parser = argparse.ArgumentParser('Configure scylla', add_help=False, formatter_class=argparse.ArgumentDefaultsHelpFormatter)
 arg_parser.add_argument('--out', dest='buildfile', action='store', default='build.ninja',
@@ -795,6 +817,9 @@ arg_parser.add_argument('--c-compiler', action='store', dest='cc', default='clan
                         help='C compiler path')
 arg_parser.add_argument('--compiler-cache', action='store', dest='compiler_cache', default='auto',
                         help='Compiler cache to use: auto (default, prefers sccache), sccache, ccache, none, or a path to a binary')
+# Workaround for https://github.com/mozilla/sccache/issues/2575
+arg_parser.add_argument('--sccache-rust', action=argparse.BooleanOptionalAction, default=False,
+                        help='Use sccache for rust code (if sccache is selected as compiler cache). Doesn\'t work with distributed builds.')
 add_tristate(arg_parser, name='dpdk', dest='dpdk', default=False,
                         help='Use dpdk (from seastar dpdk sources)')
 arg_parser.add_argument('--dpdk-target', action='store', dest='dpdk_target', default='',
@@ -925,8 +950,7 @@ scylla_core = (['message/messaging_service.cc',
                 'utils/crypt_sha512.cc',
                 'utils/logalloc.cc',
                 'utils/large_bitset.cc',
-                'utils/buffer_input_stream.cc',
-                'utils/limiting_data_source.cc',
+                'test/lib/limiting_data_source.cc',
                 'utils/updateable_value.cc',
                 'message/dictionary_service.cc',
                 'utils/directories.cc',
@@ -1535,6 +1559,7 @@ scylla_perfs = ['test/perf/perf_alternator.cc',
                 'test/perf/perf_fast_forward.cc',
                 'test/perf/perf_row_cache_update.cc',
                 'test/perf/perf_simple_query.cc',
+                'test/perf/perf_cql_raw.cc',
                 'test/perf/perf_sstable.cc',
                 'test/perf/perf_tablets.cc',
                 'test/perf/tablet_load_balancing.cc',
@@ -2383,7 +2408,7 @@ def write_build_file(f,
     # If compiler cache is available, prefix the compiler with it
     cxx_with_cache = f'{compiler_cache} {args.cxx}' if compiler_cache else args.cxx
     # For Rust, sccache is used via RUSTC_WRAPPER environment variable
-    rustc_wrapper = f'RUSTC_WRAPPER={compiler_cache} ' if compiler_cache and 'sccache' in compiler_cache else ''
+    rustc_wrapper = f'RUSTC_WRAPPER={compiler_cache} ' if compiler_cache and 'sccache' in compiler_cache and args.sccache_rust else ''
     f.write(textwrap.dedent('''\
         configure_args = {configure_args}
         builddir = {outdir}
@@ -2560,10 +2585,11 @@ def write_build_file(f,
               description = RUST_LIB $out
             ''').format(mode=mode, antlr3_exec=args.antlr3_exec, fmt_lib=fmt_lib, test_repeat=args.test_repeat, test_timeout=args.test_timeout, rustc_wrapper=rustc_wrapper, **modeval))
         f.write(
-            'build {mode}-build: phony {artifacts} {wasms}\n'.format(
+            'build {mode}-build: phony {artifacts} {wasms} {vector_search_validator_bins}\n'.format(
                 mode=mode,
-                artifacts=str.join(' ', ['$builddir/' + mode + '/' + x for x in sorted(build_artifacts - wasms)]),
+                artifacts=str.join(' ', ['$builddir/' + mode + '/' + x for x in sorted(build_artifacts - wasms - vector_search_validator_bins)]),
                 wasms = str.join(' ', ['$builddir/' + x for x in sorted(build_artifacts & wasms)]),
+                vector_search_validator_bins=str.join(' ', ['$builddir/' + x for x in sorted(build_artifacts & vector_search_validator_bins)]),
             )
         )
         if profile_recipe := modes[mode].get('profile_recipe'):
@@ -2593,7 +2619,7 @@ def write_build_file(f,
                 continue
             profile_dep = modes[mode].get('profile_target', "")
 
-            if binary in other or binary in wasms:
+            if binary in other or binary in wasms or binary in vector_search_validator_bins:
                 continue
             srcs = deps[binary]
             # 'scylla'
@@ -2704,10 +2730,11 @@ def write_build_file(f,
         )
 
         f.write(
-            'build {mode}-test: test.{mode} {test_executables} $builddir/{mode}/scylla {wasms}\n'.format(
+            'build {mode}-test: test.{mode} {test_executables} $builddir/{mode}/scylla {wasms} {vector_search_validator_bins} \n'.format(
                 mode=mode,
                 test_executables=' '.join(['$builddir/{}/{}'.format(mode, binary) for binary in sorted(tests)]),
                 wasms=' '.join([f'$builddir/{binary}' for binary in sorted(wasms)]),
+                vector_search_validator_bins=' '.join([f'$builddir/{binary}' for binary in sorted(vector_search_validator_bins)]),
             )
         )
         f.write(
@@ -2875,6 +2902,19 @@ def write_build_file(f,
             'build compiler-training: phony {}\n'.format(' '.join(['{mode}-compiler-training'.format(mode=mode) for mode in default_modes]))
     )
 
+    f.write(textwrap.dedent(f'''\
+        rule build-vector-search-validator
+            command = test/vector_search_validator/build-validator $builddir
+        rule build-vector-store
+            command = test/vector_search_validator/build-vector-store $builddir
+        '''))
+    f.write(
+            'build $builddir/{vector_search_validator_bin}: build-vector-search-validator {}\n'.format(' '.join([dep for dep in sorted(vector_search_validator_deps)]), vector_search_validator_bin=vector_search_validator_bin)
+    )
+    f.write(
+            'build $builddir/{vector_store_bin}: build-vector-store {}\n'.format(' '.join([dep for dep in sorted(vector_store_deps)]), vector_store_bin=vector_store_bin)
+    )
+
     f.write(textwrap.dedent(f'''\
         build dist-unified-tar: phony {' '.join([f'$builddir/{mode}/dist/tar/{scylla_product}-unified-{scylla_version}-{scylla_release}.{arch}.tar.gz' for mode in default_modes])}
         build dist-unified: phony dist-unified-tar
@@ -3112,7 +3152,7 @@ def configure_using_cmake(args):
         settings['CMAKE_CXX_COMPILER_LAUNCHER'] = compiler_cache
         settings['CMAKE_C_COMPILER_LAUNCHER'] = compiler_cache
         # For Rust, sccache is used via RUSTC_WRAPPER
-        if 'sccache' in compiler_cache:
+        if 'sccache' in compiler_cache and args.sccache_rust:
             settings['Scylla_RUSTC_WRAPPER'] = compiler_cache
 
     if args.date_stamp:

cql3/functions/vector_similarity_fcts.cc

@@ -10,41 +10,9 @@
 #include "types/types.hh"
 #include "types/vector.hh"
 #include "exceptions/exceptions.hh"
-#include <span>
-#include <bit>
 
 namespace cql3 {
 namespace functions {
-
-namespace detail {
-
-std::vector<float> extract_float_vector(const bytes_opt& param, size_t dimension) {
-    if (!param) {
-        throw exceptions::invalid_request_exception("Cannot extract float vector from null parameter");
-    }
-
-    const size_t expected_size = dimension * sizeof(float);
-    if (param->size() != expected_size) {
-        throw exceptions::invalid_request_exception(
-            fmt::format("Invalid vector size: expected {} bytes for {} floats, got {} bytes",
-                       expected_size, dimension, param->size()));
-    }
-
-    std::vector<float> result;
-    result.reserve(dimension);
-
-    bytes_view view(*param);
-    for (size_t i = 0; i < dimension; ++i) {
-        // read_simple handles network byte order (big-endian) conversion
-        uint32_t raw = read_simple<uint32_t>(view);
-        result.push_back(std::bit_cast<float>(raw));
-    }
-
-    return result;
-}
-
-} // namespace detail
-
 namespace {
 
 // The computations of similarity scores match the exact formulas of Cassandra's (jVector's) implementation to ensure compatibility.
@@ -54,14 +22,14 @@ namespace {
 
 // You should only use this function if you need to preserve the original vectors and cannot normalize
 // them in advance.
-float compute_cosine_similarity(std::span<const float> v1, std::span<const float> v2) {
+float compute_cosine_similarity(const std::vector<data_value>& v1, const std::vector<data_value>& v2) {
     double dot_product = 0.0;
     double squared_norm_a = 0.0;
     double squared_norm_b = 0.0;
 
     for (size_t i = 0; i < v1.size(); ++i) {
-        double a = v1[i];
-        double b = v2[i];
+        double a = value_cast<float>(v1[i]);
+        double b = value_cast<float>(v2[i]);
 
         dot_product += a * b;
         squared_norm_a += a * a;
@@ -69,7 +37,7 @@ float compute_cosine_similarity(std::span<const float> v1, std::span<const float
     }
 
     if (squared_norm_a == 0 || squared_norm_b == 0) {
-        return std::numeric_limits<float>::quiet_NaN();
+        throw exceptions::invalid_request_exception("Function system.similarity_cosine doesn't support all-zero vectors");
     }
 
     // The cosine similarity is in the range [-1, 1].
@@ -78,12 +46,12 @@ float compute_cosine_similarity(std::span<const float> v1, std::span<const float
     return (1 + (dot_product / (std::sqrt(squared_norm_a * squared_norm_b)))) / 2;
 }
 
-float compute_euclidean_similarity(std::span<const float> v1, std::span<const float> v2) {
+float compute_euclidean_similarity(const std::vector<data_value>& v1, const std::vector<data_value>& v2) {
     double sum = 0.0;
 
     for (size_t i = 0; i < v1.size(); ++i) {
-        double a = v1[i];
-        double b = v2[i];
+        double a = value_cast<float>(v1[i]);
+        double b = value_cast<float>(v2[i]);
 
         double diff = a - b;
         sum += diff * diff;
@@ -97,12 +65,12 @@ float compute_euclidean_similarity(std::span<const float> v1, std::span<const fl
 
 // Assumes that both vectors are L2-normalized.
 // This similarity is intended as an optimized way to perform cosine similarity calculation.
-float compute_dot_product_similarity(std::span<const float> v1, std::span<const float> v2) {
+float compute_dot_product_similarity(const std::vector<data_value>& v1, const std::vector<data_value>& v2) {
     double dot_product = 0.0;
 
     for (size_t i = 0; i < v1.size(); ++i) {
-        double a = v1[i];
-        double b = v2[i];
+        double a = value_cast<float>(v1[i]);
+        double b = value_cast<float>(v2[i]);
         dot_product += a * b;
     }
 
@@ -168,15 +136,13 @@ bytes_opt vector_similarity_fct::execute(std::span<const bytes_opt> parameters)
         return std::nullopt;
     }
 
-    // Extract dimension from the vector type
-    const auto& type = static_cast<const vector_type_impl&>(*arg_types()[0]);
-    size_t dimension = type.get_dimension();
+    const auto& type = arg_types()[0];
+    data_value v1 = type->deserialize(*parameters[0]);
+    data_value v2 = type->deserialize(*parameters[1]);
+    const auto& v1_elements = value_cast<std::vector<data_value>>(v1);
+    const auto& v2_elements = value_cast<std::vector<data_value>>(v2);
 
-    // Optimized path: extract floats directly from bytes, bypassing data_value overhead
-    std::vector<float> v1 = detail::extract_float_vector(parameters[0], dimension);
-    std::vector<float> v2 = detail::extract_float_vector(parameters[1], dimension);
-
-    float result = SIMILARITY_FUNCTIONS.at(_name)(v1, v2);
+    float result = SIMILARITY_FUNCTIONS.at(_name)(v1_elements, v2_elements);
     return float_type->decompose(result);
 }
 

cql3/functions/vector_similarity_fcts.hh

@@ -11,7 +11,6 @@
 #include "native_scalar_function.hh"
 #include "cql3/assignment_testable.hh"
 #include "cql3/functions/function_name.hh"
-#include <span>
 
 namespace cql3 {
 namespace functions {
@@ -20,7 +19,7 @@ static const function_name SIMILARITY_COSINE_FUNCTION_NAME = function_name::nati
 static const function_name SIMILARITY_EUCLIDEAN_FUNCTION_NAME = function_name::native_function("similarity_euclidean");
 static const function_name SIMILARITY_DOT_PRODUCT_FUNCTION_NAME = function_name::native_function("similarity_dot_product");
 
-using similarity_function_t = float (*)(std::span<const float>, std::span<const float>);
+using similarity_function_t = float (*)(const std::vector<data_value>&, const std::vector<data_value>&);
 extern thread_local const std::unordered_map<function_name, similarity_function_t> SIMILARITY_FUNCTIONS;
 
 std::vector<data_type> retrieve_vector_arg_types(const function_name& name, const std::vector<shared_ptr<assignment_testable>>& provided_args);
@@ -34,14 +33,5 @@ public:
     virtual bytes_opt execute(std::span<const bytes_opt> parameters) override;
 };
 
-namespace detail {
-
-// Extract float vector directly from serialized bytes, bypassing data_value overhead.
-// This is an internal API exposed for testing purposes.
-// Vector<float, N> wire format: N floats as big-endian uint32_t values, 4 bytes each.
-std::vector<float> extract_float_vector(const bytes_opt& param, size_t dimension);
-
-} // namespace detail
-
 } // namespace functions
 } // namespace cql3

cql3/prepared_statements_cache.hh

@@ -105,7 +105,6 @@ public:
     static const std::chrono::minutes entry_expiry;
 
     using key_type = prepared_cache_key_type;
-    using pinned_value_type = cache_value_ptr;
     using value_type = checked_weak_ptr;
     using statement_is_too_big = typename cache_type::entry_is_too_big;
 
@@ -117,14 +116,9 @@ public:
         : _cache(size, entry_expiry, logger)
     {}
 
-    template <typename LoadFunc>
-    future<pinned_value_type> get_pinned(const key_type& key, LoadFunc&& load) {
-        return _cache.get_ptr(key.key(), [load = std::forward<LoadFunc>(load)] (const cache_key_type&) { return load(); });
-    }
-
     template <typename LoadFunc>
     future<value_type> get(const key_type& key, LoadFunc&& load) {
-        return get_pinned(key, std::forward<LoadFunc>(load)).then([] (cache_value_ptr v_ptr) {
+        return _cache.get_ptr(key.key(), [load = std::forward<LoadFunc>(load)] (const cache_key_type&) { return load(); }).then([] (cache_value_ptr v_ptr) {
             return make_ready_future<value_type>((*v_ptr)->checked_weak_from_this());
         });
     }

cql3/query_processor.cc

@@ -697,7 +697,7 @@ future<::shared_ptr<cql_transport::messages::result_message::prepared>>
 query_processor::prepare(sstring query_string, const service::client_state& client_state, cql3::dialect d) {
     try {
         auto key = compute_id(query_string, client_state.get_raw_keyspace(), d);
-        auto prep_entry = co_await _prepared_cache.get_pinned(key, [this, &query_string, &client_state, d] {
+        auto prep_ptr = co_await _prepared_cache.get(key, [this, &query_string, &client_state, d] {
                 auto prepared = get_statement(query_string, client_state, d);
                 prepared->calculate_metadata_id();
                 auto bound_terms = prepared->statement->get_bound_terms();
@@ -711,13 +711,13 @@ query_processor::prepare(sstring query_string, const service::client_state& clie
                 return make_ready_future<std::unique_ptr<statements::prepared_statement>>(std::move(prepared));
             });
 
-        co_await utils::get_local_injector().inject(
-                "query_processor_prepare_wait_after_cache_get",
-                utils::wait_for_message(std::chrono::seconds(60)));
-  
-        auto msg = ::make_shared<result_message::prepared::cql>(prepared_cache_key_type::cql_id(key), std::move(prep_entry),
+        const auto& warnings = prep_ptr->warnings;
+        const auto msg = ::make_shared<result_message::prepared::cql>(prepared_cache_key_type::cql_id(key), std::move(prep_ptr),
                     client_state.is_protocol_extension_set(cql_transport::cql_protocol_extension::LWT_ADD_METADATA_MARK));
-        co_return std::move(msg);
+        for (const auto& w : warnings) {
+            msg->add_warning(w);
+        }
+        co_return ::shared_ptr<cql_transport::messages::result_message::prepared>(std::move(msg));
     } catch(typename prepared_statements_cache::statement_is_too_big&) {
         throw prepared_statement_is_too_big(query_string);
     }

cql3/query_result_printer.hh

@@ -1,20 +0,0 @@
-/*
- * Copyright 2025-present ScyllaDB
- */
-
-/*
- * SPDX-License-Identifier: LicenseRef-ScyllaDB-Source-Available-1.0
- */
-
-#pragma once
-
-#include <ostream>
-
-namespace cql3 {
-
-class result;
-
-void print_query_results_text(std::ostream& os, const result& result);
-void print_query_results_json(std::ostream& os, const result& result);
-
-} // namespace cql3

cql3/result_set.cc

@@ -9,10 +9,8 @@
  */
 
 #include <cstdint>
-#include "types/json_utils.hh"
 #include "utils/assert.hh"
 #include "utils/hashers.hh"
-#include "utils/rjson.hh"
 #include "cql3/result_set.hh"
 
 namespace cql3 {
@@ -197,85 +195,4 @@ make_empty_metadata() {
     return empty_metadata_cache;
 }
 
-void print_query_results_text(std::ostream& os, const cql3::result& result) {
-    const auto& metadata = result.get_metadata();
-    const auto& column_metadata = metadata.get_names();
-
-    struct column_values {
-        size_t max_size{0};
-        sstring header_format;
-        sstring row_format;
-        std::vector<sstring> values;
-
-        void add(sstring value) {
-            max_size = std::max(max_size, value.size());
-            values.push_back(std::move(value));
-        }
-    };
-
-    std::vector<column_values> columns;
-    columns.resize(column_metadata.size());
-
-    for (size_t i = 0; i < column_metadata.size(); ++i) {
-        columns[i].add(column_metadata[i]->name->text());
-    }
-
-    for (const auto& row : result.result_set().rows()) {
-        for (size_t i = 0; i < row.size(); ++i) {
-            if (row[i]) {
-                columns[i].add(column_metadata[i]->type->to_string(linearized(managed_bytes_view(*row[i]))));
-            } else {
-                columns[i].add("");
-            }
-        }
-    }
-
-    std::vector<sstring> separators(columns.size(), sstring());
-    for (size_t i = 0; i < columns.size(); ++i) {
-        auto& col_values = columns[i];
-        col_values.header_format = seastar::format(" {{:<{}}} ", col_values.max_size);
-        col_values.row_format = seastar::format(" {{:>{}}} ", col_values.max_size);
-        for (size_t c = 0; c < col_values.max_size; ++c) {
-            separators[i] += "-";
-        }
-    }
-
-    for (size_t r = 0; r < result.result_set().rows().size() + 1; ++r) {
-        std::vector<sstring> row;
-        row.reserve(columns.size());
-        for (size_t i = 0; i < columns.size(); ++i) {
-            const auto& format = r == 0 ? columns[i].header_format : columns[i].row_format;
-            row.push_back(fmt::format(fmt::runtime(std::string_view(format)), columns[i].values[r]));
-        }
-        fmt::print(os, "{}\n", fmt::join(row, "|"));
-        if (!r) {
-            fmt::print(os, "-{}-\n", fmt::join(separators, "-+-"));
-        }
-    }
-}
-
-void print_query_results_json(std::ostream& os, const cql3::result& result) {
-    const auto& metadata = result.get_metadata();
-    const auto& column_metadata = metadata.get_names();
-
-    rjson::streaming_writer writer(os);
-
-    writer.StartArray();
-    for (const auto& row : result.result_set().rows()) {
-        writer.StartObject();
-        for (size_t i = 0; i < row.size(); ++i) {
-            writer.Key(column_metadata[i]->name->text());
-            if (!row[i] || row[i]->empty()) {
-                writer.Null();
-                continue;
-            }
-            const auto value = to_json_string(*column_metadata[i]->type, *row[i]);
-            const auto type = to_json_type(*column_metadata[i]->type, *row[i]);
-            writer.RawValue(value, type);
-        }
-        writer.EndObject();
-    }
-    writer.EndArray();
-}
-
 }

cql3/statements/describe_statement.cc

@@ -23,7 +23,6 @@
 #include "index/vector_index.hh"
 #include "schema/schema.hh"
 #include "service/client_state.hh"
-#include "service/paxos/paxos_state.hh"
 #include "types/types.hh"
 #include "cql3/query_processor.hh"
 #include "cql3/cql_statement.hh"
@@ -330,19 +329,6 @@ future<std::vector<description>> table(const data_dictionary::database& db, cons
                 "*/",
                 *table_desc.create_statement);
 
-        table_desc.create_statement = std::move(os).to_managed_string();
-    } else if (service::paxos::paxos_store::try_get_base_table(name)) {
-        // Paxos state table is internally managed by Scylla and it shouldn't be exposed to the user.
-        // The table is allowed to be described as a comment to ease administrative work but it's hidden from all listings.
-        fragmented_ostringstream os{};
-
-        fmt::format_to(os.to_iter(),
-                "/* Do NOT execute this statement! It's only for informational purposes.\n"
-                "   A paxos state table is created automatically when enabling LWT on a base table.\n"
-                "\n{}\n"
-                "*/",
-                *table_desc.create_statement);
-
         table_desc.create_statement = std::move(os).to_managed_string();
     }
     result.push_back(std::move(table_desc));
@@ -378,7 +364,7 @@ future<std::vector<description>> table(const data_dictionary::database& db, cons
 future<std::vector<description>> tables(const data_dictionary::database& db, const lw_shared_ptr<keyspace_metadata>& ks, std::optional<bool> with_internals = std::nullopt) {
     auto& replica_db = db.real_database();
     auto tables = ks->tables() | std::views::filter([&replica_db] (const schema_ptr& s) {
-        return !cdc::is_log_for_some_table(replica_db, s->ks_name(), s->cf_name()) && !service::paxos::paxos_store::try_get_base_table(s->cf_name());
+        return !cdc::is_log_for_some_table(replica_db, s->ks_name(), s->cf_name());
     }) | std::ranges::to<std::vector<schema_ptr>>();
     std::ranges::sort(tables, std::ranges::less(), std::mem_fn(&schema::cf_name));
 

cql3/statements/raw/batch_statement.hh

@@ -50,8 +50,8 @@ public:
 protected:
     virtual audit::statement_category category() const override;
     virtual audit::audit_info_ptr audit_info() const override {
-        // We don't audit batch statements. Instead we audit statements that are inside the batch.
-        return audit::audit::create_no_audit_info();
+        constexpr bool batch = true;
+        return audit::audit::create_audit_info(category(), sstring(), sstring(), batch);
     }
 };
 

db/batchlog_manager.cc

@@ -55,21 +55,8 @@ int32_t batchlog_shard_of(db_clock::time_point written_at) {
     return hash & ((1ULL << batchlog_shard_bits) - 1);
 }
 
-bool is_batchlog_v1(const schema& schema) {
-    return schema.cf_name() == system_keyspace::BATCHLOG;
-}
-
 std::pair<partition_key, clustering_key>
 get_batchlog_key(const schema& schema, int32_t version, db::batchlog_stage stage, int32_t batchlog_shard, db_clock::time_point written_at, std::optional<utils::UUID> id) {
-    if (is_batchlog_v1(schema)) {
-        if (!id) {
-            on_internal_error(blogger, "get_batchlog_key(): key for batchlog v1 requires batchlog id");
-        }
-        auto pkey = partition_key::from_single_value(schema, {serialized(*id)});
-        auto ckey = clustering_key::make_empty();
-        return std::pair(std::move(pkey), std::move(ckey));
-    }
-
     auto pkey = partition_key::from_exploded(schema, {serialized(version), serialized(int8_t(stage)), serialized(batchlog_shard)});
 
     std::vector<bytes> ckey_components;
@@ -98,14 +85,6 @@ mutation get_batchlog_mutation_for(schema_ptr schema, managed_bytes data, int32_
     auto cdef_data = schema->get_column_definition(to_bytes("data"));
     m.set_cell(ckey, *cdef_data, atomic_cell::make_live(*cdef_data->type, timestamp, std::move(data)));
 
-    if (is_batchlog_v1(*schema)) {
-        auto cdef_version = schema->get_column_definition(to_bytes("version"));
-        m.set_cell(ckey, *cdef_version, atomic_cell::make_live(*cdef_version->type, timestamp, serialized(version)));
-
-        auto cdef_written_at = schema->get_column_definition(to_bytes("written_at"));
-        m.set_cell(ckey, *cdef_written_at, atomic_cell::make_live(*cdef_written_at->type, timestamp, serialized(now)));
-    }
-
     return m;
 }
 
@@ -143,10 +122,9 @@ mutation get_batchlog_delete_mutation(schema_ptr schema, int32_t version, db_clo
 const std::chrono::seconds db::batchlog_manager::replay_interval;
 const uint32_t db::batchlog_manager::page_size;
 
-db::batchlog_manager::batchlog_manager(cql3::query_processor& qp, db::system_keyspace& sys_ks, gms::feature_service& fs, batchlog_manager_config config)
+db::batchlog_manager::batchlog_manager(cql3::query_processor& qp, db::system_keyspace& sys_ks, batchlog_manager_config config)
         : _qp(qp)
         , _sys_ks(sys_ks)
-        , _fs(fs)
         , _replay_timeout(config.replay_timeout)
         , _replay_rate(config.replay_rate)
         , _delay(config.delay)
@@ -322,206 +300,149 @@ future<> db::batchlog_manager::maybe_migrate_v1_to_v2() {
     });
 }
 
-namespace {
-
-using clock_type = db_clock::rep;
-
-struct replay_stats {
-    std::optional<db_clock::time_point> min_too_fresh;
-    bool need_cleanup = false;
-};
-
-} // anonymous namespace
-
-static future<db::all_batches_replayed> process_batch(
-        cql3::query_processor& qp,
-        db::batchlog_manager::stats& stats,
-        db::batchlog_manager::post_replay_cleanup cleanup,
-        utils::rate_limiter& limiter,
-        schema_ptr schema,
-        std::unordered_map<int32_t, replay_stats>& replay_stats_per_shard,
-        const db_clock::time_point now,
-        db_clock::duration replay_timeout,
-        std::chrono::seconds write_timeout,
-        const cql3::untyped_result_set::row& row) {
-    const bool is_v1 = db::is_batchlog_v1(*schema);
-    const auto stage = is_v1 ? db::batchlog_stage::initial : static_cast<db::batchlog_stage>(row.get_as<int8_t>("stage"));
-    const auto batch_shard = is_v1 ? 0 : row.get_as<int32_t>("shard");
-    auto written_at = row.get_as<db_clock::time_point>("written_at");
-    auto id = row.get_as<utils::UUID>("id");
-    // enough time for the actual write + batchlog entry mutation delivery (two separate requests).
-    auto timeout = replay_timeout;
-
-    if (utils::get_local_injector().is_enabled("skip_batch_replay")) {
-        blogger.debug("Skipping batch replay due to skip_batch_replay injection");
-        co_return db::all_batches_replayed::no;
-    }
-
-    auto data = row.get_blob_unfragmented("data");
-
-    blogger.debug("Replaying batch {} from stage {} and batch shard {}", id, int32_t(stage), batch_shard);
-
-    utils::chunked_vector<mutation> mutations;
-    bool send_failed = false;
-
-    auto& shard_written_at = replay_stats_per_shard.try_emplace(batch_shard, replay_stats{}).first->second;
-
-    try {
-        utils::chunked_vector<std::pair<canonical_mutation, schema_ptr>> fms;
-        auto in = ser::as_input_stream(data);
-        while (in.size()) {
-            auto fm = ser::deserialize(in, std::type_identity<canonical_mutation>());
-            const auto tbl = qp.db().try_find_table(fm.column_family_id());
-            if (!tbl) {
-                continue;
-            }
-            if (written_at <= tbl->get_truncation_time()) {
-                continue;
-            }
-            schema_ptr s = tbl->schema();
-            if (s->tombstone_gc_options().mode() == tombstone_gc_mode::repair) {
-                timeout = std::min(timeout, std::chrono::duration_cast<db_clock::duration>(s->tombstone_gc_options().propagation_delay_in_seconds()));
-            }
-            fms.emplace_back(std::move(fm), std::move(s));
-        }
-
-        if (now < written_at + timeout) {
-            blogger.debug("Skipping replay of {}, too fresh", id);
-
-            shard_written_at.min_too_fresh = std::min(shard_written_at.min_too_fresh.value_or(written_at), written_at);
-
-            co_return db::all_batches_replayed::no;
-        }
-
-        auto size = data.size();
-
-        for (const auto& [fm, s] : fms) {
-            mutations.emplace_back(fm.to_mutation(s));
-            co_await coroutine::maybe_yield();
-        }
-
-        if (!mutations.empty()) {
-            const auto ttl = [written_at]() -> clock_type {
-                /*
-                * Calculate ttl for the mutations' hints (and reduce ttl by the time the mutations spent in the batchlog).
-                * This ensures that deletes aren't "undone" by an old batch replay.
-                */
-                auto unadjusted_ttl = std::numeric_limits<gc_clock::rep>::max();
-                warn(unimplemented::cause::HINT);
-#if 0
-                for (auto& m : *mutations) {
-                    unadjustedTTL = Math.min(unadjustedTTL, HintedHandOffManager.calculateHintTTL(mutation));
-                }
-#endif
-                return unadjusted_ttl - std::chrono::duration_cast<gc_clock::duration>(db_clock::now() - written_at).count();
-            }();
-
-            if (ttl > 0) {
-                // Origin does the send manually, however I can't see a super great reason to do so.
-                // Our normal write path does not add much redundancy to the dispatch, and rate is handled after send
-                // in both cases.
-                // FIXME: verify that the above is reasonably true.
-                co_await limiter.reserve(size);
-                stats.write_attempts += mutations.size();
-                auto timeout = db::timeout_clock::now() + write_timeout;
-                if (cleanup) {
-                    co_await qp.proxy().send_batchlog_replay_to_all_replicas(mutations, timeout);
-                } else {
-                    co_await qp.proxy().send_batchlog_replay_to_all_replicas(std::move(mutations), timeout);
-                }
-            }
-        }
-    } catch (data_dictionary::no_such_keyspace& ex) {
-        // should probably ignore and drop the batch
-    } catch (const data_dictionary::no_such_column_family&) {
-        // As above -- we should drop the batch if the table doesn't exist anymore.
-    } catch (...) {
-        blogger.warn("Replay failed (will retry): {}", std::current_exception());
-        // timeout, overload etc.
-        // Do _not_ remove the batch, assuning we got a node write error.
-        // Since we don't have hints (which origin is satisfied with),
-        // we have to resort to keeping this batch to next lap.
-        if (is_v1 || !cleanup || stage == db::batchlog_stage::failed_replay) {
-            co_return db::all_batches_replayed::no;
-        }
-        send_failed = true;
-    }
-
-    auto& sp = qp.proxy();
-
-    if (send_failed) {
-        blogger.debug("Moving batch {} to stage failed_replay", id);
-        auto m = get_batchlog_mutation_for(schema, mutations, netw::messaging_service::current_version, db::batchlog_stage::failed_replay, written_at, id);
-        co_await sp.mutate_locally(m, tracing::trace_state_ptr(), db::commitlog::force_sync::no);
-    }
-
-    // delete batch
-    auto m = get_batchlog_delete_mutation(schema, netw::messaging_service::current_version, stage, written_at, id);
-    co_await qp.proxy().mutate_locally(m, tracing::trace_state_ptr(), db::commitlog::force_sync::no);
-
-    shard_written_at.need_cleanup = true;
-
-    co_return db::all_batches_replayed(!send_failed);
-}
-
-future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches_v1(post_replay_cleanup) {
-    db::all_batches_replayed all_replayed = all_batches_replayed::yes;
-    // rate limit is in bytes per second. Uses Double.MAX_VALUE if disabled (set to 0 in cassandra.yaml).
-    // max rate is scaled by the number of nodes in the cluster (same as for HHOM - see CASSANDRA-5272).
-    auto throttle = _replay_rate / _qp.proxy().get_token_metadata_ptr()->count_normal_token_owners();
-    utils::rate_limiter limiter(throttle);
-
-    auto schema = _qp.db().find_schema(system_keyspace::NAME, system_keyspace::BATCHLOG);
-
-    std::unordered_map<int32_t, replay_stats> replay_stats_per_shard;
-
-    // Use a stable `now` across all batches, so skip/replay decisions are the
-    // same across a while prefix of written_at (across all ids).
-    const auto now = db_clock::now();
-
-    auto batch = [this, &limiter, schema, &all_replayed, &replay_stats_per_shard, now] (const cql3::untyped_result_set::row& row) mutable -> future<stop_iteration> {
-        all_replayed = all_replayed && co_await process_batch(_qp, _stats, post_replay_cleanup::no, limiter, schema, replay_stats_per_shard, now, _replay_timeout, write_timeout, row);
-        co_return stop_iteration::no;
-    };
-
-    co_await with_gate(_gate, [this, &all_replayed, batch = std::move(batch)] () mutable -> future<> {
-        blogger.debug("Started replayAllFailedBatches");
-        co_await utils::get_local_injector().inject("add_delay_to_batch_replay", std::chrono::milliseconds(1000));
-
-        auto schema = _qp.db().find_schema(system_keyspace::NAME, system_keyspace::BATCHLOG);
-
-        co_await _qp.query_internal(
-                format("SELECT * FROM {}.{} BYPASS CACHE", system_keyspace::NAME, system_keyspace::BATCHLOG),
-                db::consistency_level::ONE,
-                {},
-                page_size,
-                batch);
-
-        blogger.debug("Finished replayAllFailedBatches with all_replayed: {}", all_replayed);
-    });
-
-    co_return all_replayed;
-}
-
-future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches_v2(post_replay_cleanup cleanup) {
+future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches(post_replay_cleanup cleanup) {
     co_await maybe_migrate_v1_to_v2();
 
+    typedef db_clock::rep clock_type;
+
     db::all_batches_replayed all_replayed = all_batches_replayed::yes;
     // rate limit is in bytes per second. Uses Double.MAX_VALUE if disabled (set to 0 in cassandra.yaml).
     // max rate is scaled by the number of nodes in the cluster (same as for HHOM - see CASSANDRA-5272).
     auto throttle = _replay_rate / _qp.proxy().get_token_metadata_ptr()->count_normal_token_owners();
-    utils::rate_limiter limiter(throttle);
+    auto limiter = make_lw_shared<utils::rate_limiter>(throttle);
 
     auto schema = _qp.db().find_schema(system_keyspace::NAME, system_keyspace::BATCHLOG_V2);
 
+    struct replay_stats {
+        std::optional<db_clock::time_point> min_too_fresh;
+        bool need_cleanup = false;
+    };
+
     std::unordered_map<int32_t, replay_stats> replay_stats_per_shard;
 
     // Use a stable `now` across all batches, so skip/replay decisions are the
     // same across a while prefix of written_at (across all ids).
     const auto now = db_clock::now();
 
-    auto batch = [this, cleanup, &limiter, schema, &all_replayed, &replay_stats_per_shard, now] (const cql3::untyped_result_set::row& row) mutable -> future<stop_iteration> {
-        all_replayed = all_replayed && co_await process_batch(_qp, _stats, cleanup, limiter, schema, replay_stats_per_shard, now, _replay_timeout, write_timeout, row);
+    auto batch = [this, cleanup, limiter, schema, &all_replayed, &replay_stats_per_shard, now] (const cql3::untyped_result_set::row& row) -> future<stop_iteration> {
+        const auto stage = static_cast<batchlog_stage>(row.get_as<int8_t>("stage"));
+        const auto batch_shard = row.get_as<int32_t>("shard");
+        auto written_at = row.get_as<db_clock::time_point>("written_at");
+        auto id = row.get_as<utils::UUID>("id");
+        // enough time for the actual write + batchlog entry mutation delivery (two separate requests).
+        auto timeout = _replay_timeout;
+
+        if (utils::get_local_injector().is_enabled("skip_batch_replay")) {
+            blogger.debug("Skipping batch replay due to skip_batch_replay injection");
+            all_replayed = all_batches_replayed::no;
+            co_return stop_iteration::no;
+        }
+
+        auto data = row.get_blob_unfragmented("data");
+
+        blogger.debug("Replaying batch {} from stage {} and batch shard {}", id, int32_t(stage), batch_shard);
+
+        utils::chunked_vector<mutation> mutations;
+        bool send_failed = false;
+
+        auto& shard_written_at = replay_stats_per_shard.try_emplace(batch_shard, replay_stats{}).first->second;
+
+        try {
+            utils::chunked_vector<std::pair<canonical_mutation, schema_ptr>> fms;
+            auto in = ser::as_input_stream(data);
+            while (in.size()) {
+                auto fm = ser::deserialize(in, std::type_identity<canonical_mutation>());
+                const auto tbl = _qp.db().try_find_table(fm.column_family_id());
+                if (!tbl) {
+                    continue;
+                }
+                if (written_at <= tbl->get_truncation_time()) {
+                    continue;
+                }
+                schema_ptr s = tbl->schema();
+                if (s->tombstone_gc_options().mode() == tombstone_gc_mode::repair) {
+                    timeout = std::min(timeout, std::chrono::duration_cast<db_clock::duration>(s->tombstone_gc_options().propagation_delay_in_seconds()));
+                }
+                fms.emplace_back(std::move(fm), std::move(s));
+            }
+
+            if (now < written_at + timeout) {
+                blogger.debug("Skipping replay of {}, too fresh", id);
+
+                shard_written_at.min_too_fresh = std::min(shard_written_at.min_too_fresh.value_or(written_at), written_at);
+
+                co_return stop_iteration::no;
+            }
+
+            auto size = data.size();
+
+            for (const auto& [fm, s] : fms) {
+                mutations.emplace_back(fm.to_mutation(s));
+                co_await coroutine::maybe_yield();
+            }
+
+            if (!mutations.empty()) {
+                const auto ttl = [written_at]() -> clock_type {
+                    /*
+                    * Calculate ttl for the mutations' hints (and reduce ttl by the time the mutations spent in the batchlog).
+                    * This ensures that deletes aren't "undone" by an old batch replay.
+                    */
+                    auto unadjusted_ttl = std::numeric_limits<gc_clock::rep>::max();
+                    warn(unimplemented::cause::HINT);
+#if 0
+                    for (auto& m : *mutations) {
+                        unadjustedTTL = Math.min(unadjustedTTL, HintedHandOffManager.calculateHintTTL(mutation));
+                    }
+#endif
+                    return unadjusted_ttl - std::chrono::duration_cast<gc_clock::duration>(db_clock::now() - written_at).count();
+                }();
+
+                if (ttl > 0) {
+                    // Origin does the send manually, however I can't see a super great reason to do so.
+                    // Our normal write path does not add much redundancy to the dispatch, and rate is handled after send
+                    // in both cases.
+                    // FIXME: verify that the above is reasonably true.
+                    co_await limiter->reserve(size);
+                    _stats.write_attempts += mutations.size();
+                    auto timeout = db::timeout_clock::now() + write_timeout;
+                    if (cleanup) {
+                        co_await _qp.proxy().send_batchlog_replay_to_all_replicas(mutations, timeout);
+                    } else {
+                        co_await _qp.proxy().send_batchlog_replay_to_all_replicas(std::move(mutations), timeout);
+                    }
+                }
+            }
+        } catch (data_dictionary::no_such_keyspace& ex) {
+            // should probably ignore and drop the batch
+        } catch (const data_dictionary::no_such_column_family&) {
+            // As above -- we should drop the batch if the table doesn't exist anymore.
+        } catch (...) {
+            blogger.warn("Replay failed (will retry): {}", std::current_exception());
+            all_replayed = all_batches_replayed::no;
+            // timeout, overload etc.
+            // Do _not_ remove the batch, assuning we got a node write error.
+            // Since we don't have hints (which origin is satisfied with),
+            // we have to resort to keeping this batch to next lap.
+            if (!cleanup || stage == batchlog_stage::failed_replay) {
+                co_return stop_iteration::no;
+            }
+            send_failed = true;
+        }
+
+        auto& sp = _qp.proxy();
+
+        if (send_failed) {
+            blogger.debug("Moving batch {} to stage failed_replay", id);
+            auto m = get_batchlog_mutation_for(schema, mutations, netw::messaging_service::current_version, batchlog_stage::failed_replay, written_at, id);
+            co_await sp.mutate_locally(m, tracing::trace_state_ptr(), db::commitlog::force_sync::no);
+        }
+
+        // delete batch
+        auto m = get_batchlog_delete_mutation(schema, netw::messaging_service::current_version, stage, written_at, id);
+        co_await _qp.proxy().mutate_locally(m, tracing::trace_state_ptr(), db::commitlog::force_sync::no);
+
+        shard_written_at.need_cleanup = true;
+
         co_return stop_iteration::no;
     };
 
@@ -580,10 +501,3 @@ future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches
 
     co_return all_replayed;
 }
-
-future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches(post_replay_cleanup cleanup) {
-    if (_fs.batchlog_v2) {
-        return replay_all_failed_batches_v2(cleanup);
-    }
-    return replay_all_failed_batches_v1(cleanup);
-}

db/batchlog_manager.hh

@@ -27,12 +27,6 @@ class query_processor;
 
 } // namespace cql3
 
-namespace gms {
-
-class feature_service;
-
-} // namespace gms
-
 namespace db {
 
 class system_keyspace;
@@ -55,11 +49,6 @@ class batchlog_manager : public peering_sharded_service<batchlog_manager> {
 public:
     using post_replay_cleanup = bool_class<class post_replay_cleanup_tag>;
 
-    struct stats {
-        uint64_t write_attempts = 0;
-    };
-
-
 private:
     static constexpr std::chrono::seconds replay_interval = std::chrono::seconds(60);
     static constexpr uint32_t page_size = 128; // same as HHOM, for now, w/out using any heuristics. TODO: set based on avg batch size.
@@ -67,13 +56,14 @@ private:
 
     using clock_type = lowres_clock;
 
-    stats _stats;
+    struct stats {
+        uint64_t write_attempts = 0;
+    } _stats;
 
     seastar::metrics::metric_groups _metrics;
 
     cql3::query_processor& _qp;
     db::system_keyspace& _sys_ks;
-    gms::feature_service& _fs;
     db_clock::duration _replay_timeout;
     uint64_t _replay_rate;
     std::chrono::milliseconds _delay;
@@ -94,14 +84,12 @@ private:
 
     future<> maybe_migrate_v1_to_v2();
 
-    future<all_batches_replayed> replay_all_failed_batches_v1(post_replay_cleanup cleanup);
-    future<all_batches_replayed> replay_all_failed_batches_v2(post_replay_cleanup cleanup);
     future<all_batches_replayed> replay_all_failed_batches(post_replay_cleanup cleanup);
 public:
     // Takes a QP, not a distributes. Because this object is supposed
     // to be per shard and does no dispatching beyond delegating the the
     // shard qp (which is what you feed here).
-    batchlog_manager(cql3::query_processor&, db::system_keyspace& sys_ks, gms::feature_service& fs, batchlog_manager_config config);
+    batchlog_manager(cql3::query_processor&, db::system_keyspace& sys_ks, batchlog_manager_config config);
 
     // abort the replay loop and return its future.
     future<> drain();
@@ -114,7 +102,7 @@ public:
         return _last_replay;
     }
 
-    const stats& get_stats() const {
+    const stats& stats() const {
         return _stats;
     }
 private:

db/commitlog/commitlog.cc

@@ -1986,13 +1986,13 @@ future<> db::commitlog::segment_manager::replenish_reserve() {
             }
             continue;
         } catch (shutdown_marker&) {
+            _reserve_segments.abort(std::current_exception());
             break;
         } catch (...) {
             clogger.warn("Exception in segment reservation: {}", std::current_exception());
         }
         co_await sleep(100ms);
     }
-    _reserve_segments.abort(std::make_exception_ptr(shutdown_marker()));
 }
 
 future<std::vector<db::commitlog::descriptor>>

db/config.cc

@@ -621,25 +621,6 @@ db::config::config(std::shared_ptr<db::extensions> exts)
     * @GroupDescription: Provides an overview of the group.
     */
     /**
-    * @Group Ungrouped properties
-    */
-    , background_writer_scheduling_quota(this, "background_writer_scheduling_quota", value_status::Deprecated, 1.0,
-        "max cpu usage ratio (between 0 and 1) for compaction process. Not intended for setting in normal operations. Setting it to 1 or higher will disable it, recommended operational setting is 0.5.")
-    , auto_adjust_flush_quota(this, "auto_adjust_flush_quota", value_status::Deprecated, false,
-        "true: auto-adjust memtable shares for flush processes")
-    , memtable_flush_static_shares(this, "memtable_flush_static_shares", liveness::LiveUpdate, value_status::Used, 0,
-        "If set to higher than 0, ignore the controller's output and set the memtable shares statically. Do not set this unless you know what you are doing and suspect a problem in the controller. This option will be retired when the controller reaches more maturity.")
-    , compaction_static_shares(this, "compaction_static_shares", liveness::LiveUpdate, value_status::Used, 0,
-        "If set to higher than 0, ignore the controller's output and set the compaction shares statically. Do not set this unless you know what you are doing and suspect a problem in the controller. This option will be retired when the controller reaches more maturity.")
-    , compaction_max_shares(this, "compaction_max_shares", liveness::LiveUpdate, value_status::Used, default_compaction_maximum_shares,
-        "Set the maximum shares of regular compaction to the specific value. Do not set this unless you know what you are doing and suspect a problem in the controller. This option will be retired when the controller reaches more maturity.")
-    , compaction_enforce_min_threshold(this, "compaction_enforce_min_threshold", liveness::LiveUpdate, value_status::Used, false,
-        "If set to true, enforce the min_threshold option for compactions strictly. If false (default), Scylla may decide to compact even if below min_threshold.")
-    , compaction_flush_all_tables_before_major_seconds(this, "compaction_flush_all_tables_before_major_seconds", value_status::Used, 86400,
-        "Set the minimum interval in seconds between flushing all tables before each major compaction (default is 86400)."
-        "This option is useful for maximizing tombstone garbage collection by releasing all active commitlog segments."
-        "Set to 0 to disable automatic flushing all tables before major compaction.")
-    /**
     * @Group Initialization properties
     * @GroupDescription The minimal properties needed for configuring a cluster.
     */
@@ -1291,7 +1272,7 @@ db::config::config(std::shared_ptr<db::extensions> exts)
     , ignore_dead_nodes_for_replace(this, "ignore_dead_nodes_for_replace", value_status::Used, "", "List dead nodes to ignore for replace operation using a comma-separated list of host IDs. E.g., scylla --ignore-dead-nodes-for-replace 8d5ed9f4-7764-4dbd-bad8-43fddce94b7c,125ed9f4-7777-1dbn-mac8-43fddce9123e")
     , override_decommission(this, "override_decommission", value_status::Deprecated, false, "Set true to force a decommissioned node to join the cluster (cannot be set if consistent-cluster-management is enabled).")
     , enable_repair_based_node_ops(this, "enable_repair_based_node_ops", liveness::LiveUpdate, value_status::Used, true, "Set true to use enable repair based node operations instead of streaming based.")
-    , allowed_repair_based_node_ops(this, "allowed_repair_based_node_ops", liveness::LiveUpdate, value_status::Used, "replace,removenode,rebuild", "A comma separated list of node operations which are allowed to enable repair based node operations. The operations can be bootstrap, replace, removenode, decommission and rebuild.")
+    , allowed_repair_based_node_ops(this, "allowed_repair_based_node_ops", liveness::LiveUpdate, value_status::Used, "replace,removenode,rebuild,bootstrap,decommission", "A comma separated list of node operations which are allowed to enable repair based node operations. The operations can be bootstrap, replace, removenode, decommission and rebuild.")
     , enable_compacting_data_for_streaming_and_repair(this, "enable_compacting_data_for_streaming_and_repair", liveness::LiveUpdate, value_status::Used, true, "Enable the compacting reader, which compacts the data for streaming and repair (load'n'stream included) before sending it to, or synchronizing it with peers. Can reduce the amount of data to be processed by removing dead data, but adds CPU overhead.")
     , enable_tombstone_gc_for_streaming_and_repair(this, "enable_tombstone_gc_for_streaming_and_repair", liveness::LiveUpdate, value_status::Used, false,
             "If the compacting reader is enabled for streaming and repair (see enable_compacting_data_for_streaming_and_repair), allow it to garbage-collect tombstones."
@@ -1394,6 +1375,10 @@ db::config::config(std::shared_ptr<db::extensions> exts)
             "Start killing reads after their collective memory consumption goes above $normal_limit * $multiplier.")
     , reader_concurrency_semaphore_cpu_concurrency(this, "reader_concurrency_semaphore_cpu_concurrency", liveness::LiveUpdate, value_status::Used, 2,
             "Admit new reads while there are less than this number of requests that need CPU.")
+    , reader_concurrency_semaphore_preemptive_abort_factor(this, "reader_concurrency_semaphore_preemptive_abort_factor", liveness::LiveUpdate, value_status::Used, 0.3,
+            "Admit new reads while their remaining time is more than this factor times their timeout times when arrived to a semaphore. Its vale means\n"
+            "* <= 0.0 means new reads will never get rejected during admission\n"
+            "* >= 1.0 means new reads will always get rejected during admission\n")
     , view_update_reader_concurrency_semaphore_serialize_limit_multiplier(this, "view_update_reader_concurrency_semaphore_serialize_limit_multiplier", liveness::LiveUpdate, value_status::Used, 2,
             "Start serializing view update reads after their collective memory consumption goes above $normal_limit * $multiplier.")
     , view_update_reader_concurrency_semaphore_kill_limit_multiplier(this, "view_update_reader_concurrency_semaphore_kill_limit_multiplier", liveness::LiveUpdate, value_status::Used, 4,
@@ -1602,6 +1587,25 @@ db::config::config(std::shared_ptr<db::extensions> exts)
         "Sets the maximum difference in percentages between the most loaded and least loaded nodes, below which the load balancer considers nodes balanced.")
     , minimal_tablet_size_for_balancing(this, "minimal_tablet_size_for_balancing", liveness::LiveUpdate, value_status::Used, service::default_target_tablet_size / 100,
         "Sets the minimal tablet size for the load balancer. For any tablet smaller than this, the balancer will use this size instead of the actual tablet size.")
+    /**
+    * @Group Ungrouped properties
+    */
+    , background_writer_scheduling_quota(this, "background_writer_scheduling_quota", value_status::Deprecated, 1.0,
+        "max cpu usage ratio (between 0 and 1) for compaction process. Not intended for setting in normal operations. Setting it to 1 or higher will disable it, recommended operational setting is 0.5.")
+    , auto_adjust_flush_quota(this, "auto_adjust_flush_quota", value_status::Deprecated, false,
+        "true: auto-adjust memtable shares for flush processes")
+    , memtable_flush_static_shares(this, "memtable_flush_static_shares", liveness::LiveUpdate, value_status::Used, 0,
+        "If set to higher than 0, ignore the controller's output and set the memtable shares statically. Do not set this unless you know what you are doing and suspect a problem in the controller. This option will be retired when the controller reaches more maturity.")
+    , compaction_static_shares(this, "compaction_static_shares", liveness::LiveUpdate, value_status::Used, 0,
+        "If set to higher than 0, ignore the controller's output and set the compaction shares statically. Do not set this unless you know what you are doing and suspect a problem in the controller. This option will be retired when the controller reaches more maturity.")
+    , compaction_max_shares(this, "compaction_max_shares", liveness::LiveUpdate, value_status::Used, default_compaction_maximum_shares,
+        "Set the maximum shares of regular compaction to the specific value. Do not set this unless you know what you are doing and suspect a problem in the controller. This option will be retired when the controller reaches more maturity.")
+    , compaction_enforce_min_threshold(this, "compaction_enforce_min_threshold", liveness::LiveUpdate, value_status::Used, false,
+        "If set to true, enforce the min_threshold option for compactions strictly. If false (default), Scylla may decide to compact even if below min_threshold.")
+    , compaction_flush_all_tables_before_major_seconds(this, "compaction_flush_all_tables_before_major_seconds", value_status::Used, 86400,
+        "Set the minimum interval in seconds between flushing all tables before each major compaction (default is 86400)."
+        "This option is useful for maximizing tombstone garbage collection by releasing all active commitlog segments."
+        "Set to 0 to disable automatic flushing all tables before major compaction.")
     , default_log_level(this, "default_log_level", value_status::Used, seastar::log_level::info, "Default log level for log messages")
     , logger_log_level(this, "logger_log_level", value_status::Used, {}, "Map of logger name to log level. Valid log levels are 'error', 'warn', 'info', 'debug' and 'trace'")
     , log_to_stdout(this, "log_to_stdout", value_status::Used, true, "Send log output to stdout")

db/config.hh

@@ -185,13 +185,6 @@ public:
      * All values and documentation taken from
      * http://docs.datastax.com/en/cassandra/2.1/cassandra/configuration/configCassandra_yaml_r.html
      */
-    named_value<double> background_writer_scheduling_quota;
-    named_value<bool> auto_adjust_flush_quota;
-    named_value<float> memtable_flush_static_shares;
-    named_value<float> compaction_static_shares;
-    named_value<float> compaction_max_shares;
-    named_value<bool> compaction_enforce_min_threshold;
-    named_value<uint32_t> compaction_flush_all_tables_before_major_seconds;
     named_value<sstring> cluster_name;
     named_value<sstring> listen_address;
     named_value<sstring> listen_interface;
@@ -446,6 +439,7 @@ public:
     named_value<uint32_t> reader_concurrency_semaphore_serialize_limit_multiplier;
     named_value<uint32_t> reader_concurrency_semaphore_kill_limit_multiplier;
     named_value<uint32_t> reader_concurrency_semaphore_cpu_concurrency;
+    named_value<float> reader_concurrency_semaphore_preemptive_abort_factor;
     named_value<uint32_t> view_update_reader_concurrency_semaphore_serialize_limit_multiplier;
     named_value<uint32_t> view_update_reader_concurrency_semaphore_kill_limit_multiplier;
     named_value<uint32_t> view_update_reader_concurrency_semaphore_cpu_concurrency;
@@ -612,6 +606,14 @@ public:
     named_value<float> size_based_balance_threshold_percentage;
     named_value<uint64_t> minimal_tablet_size_for_balancing;
 
+    named_value<double> background_writer_scheduling_quota;
+    named_value<bool> auto_adjust_flush_quota;
+    named_value<float> memtable_flush_static_shares;
+    named_value<float> compaction_static_shares;
+    named_value<float> compaction_max_shares;
+    named_value<bool> compaction_enforce_min_threshold;
+    named_value<uint32_t> compaction_flush_all_tables_before_major_seconds;
+
     static const sstring default_tls_priority;
 private:
     template<typename T>

db/hints/internal/hint_endpoint_manager.cc

@@ -158,7 +158,7 @@ void hint_endpoint_manager::cancel_draining() noexcept {
     _sender.cancel_draining();
 }
 
-hint_endpoint_manager::hint_endpoint_manager(const endpoint_id& key, fs::path hint_directory, manager& shard_manager)
+hint_endpoint_manager::hint_endpoint_manager(const endpoint_id& key, fs::path hint_directory, manager& shard_manager, scheduling_group send_sg)
     : _key(key)
     , _shard_manager(shard_manager)
     , _store_gate("hint_endpoint_manager")
@@ -169,7 +169,7 @@ hint_endpoint_manager::hint_endpoint_manager(const endpoint_id& key, fs::path hi
     // Approximate the position of the last written hint by using the same formula as for segment id calculation in commitlog
     // TODO: Should this logic be deduplicated with what is in the commitlog?
     , _last_written_rp(this_shard_id(), std::chrono::duration_cast<std::chrono::milliseconds>(runtime::get_boot_time().time_since_epoch()).count())
-    , _sender(*this, _shard_manager.local_storage_proxy(), _shard_manager.local_db(), _shard_manager.local_gossiper())
+    , _sender(*this, _shard_manager.local_storage_proxy(), _shard_manager.local_db(), _shard_manager.local_gossiper(), send_sg)
 {}
 
 hint_endpoint_manager::hint_endpoint_manager(hint_endpoint_manager&& other)

db/hints/internal/hint_endpoint_manager.hh

@@ -63,7 +63,7 @@ private:
     hint_sender _sender;
 
 public:
-    hint_endpoint_manager(const endpoint_id& key, std::filesystem::path hint_directory, manager& shard_manager);
+    hint_endpoint_manager(const endpoint_id& key, std::filesystem::path hint_directory, manager& shard_manager, scheduling_group send_sg);
     hint_endpoint_manager(hint_endpoint_manager&&);
     ~hint_endpoint_manager();
 

db/hints/internal/hint_sender.cc

@@ -122,7 +122,7 @@ const column_mapping& hint_sender::get_column_mapping(lw_shared_ptr<send_one_fil
     return cm_it->second;
 }
 
-hint_sender::hint_sender(hint_endpoint_manager& parent, service::storage_proxy& local_storage_proxy,replica::database& local_db, const gms::gossiper& local_gossiper) noexcept
+hint_sender::hint_sender(hint_endpoint_manager& parent, service::storage_proxy& local_storage_proxy,replica::database& local_db, const gms::gossiper& local_gossiper, scheduling_group sg) noexcept
     : _stopped(make_ready_future<>())
     , _ep_key(parent.end_point_key())
     , _ep_manager(parent)
@@ -130,7 +130,7 @@ hint_sender::hint_sender(hint_endpoint_manager& parent, service::storage_proxy&
     , _resource_manager(_shard_manager._resource_manager)
     , _proxy(local_storage_proxy)
     , _db(local_db)
-    , _hints_cpu_sched_group(_db.get_streaming_scheduling_group())
+    , _hints_cpu_sched_group(sg)
     , _gossiper(local_gossiper)
     , _file_update_mutex(_ep_manager.file_update_mutex())
 {}

db/hints/internal/hint_sender.hh

@@ -120,7 +120,7 @@ private:
     std::multimap<db::replay_position, lw_shared_ptr<std::optional<promise<>>>> _replay_waiters;
 
 public:
-    hint_sender(hint_endpoint_manager& parent, service::storage_proxy& local_storage_proxy, replica::database& local_db, const gms::gossiper& local_gossiper) noexcept;
+    hint_sender(hint_endpoint_manager& parent, service::storage_proxy& local_storage_proxy, replica::database& local_db, const gms::gossiper& local_gossiper, scheduling_group sg) noexcept;
     ~hint_sender();
 
     /// \brief A constructor that should be called from the copy/move-constructor of hint_endpoint_manager.

db/hints/manager.cc

@@ -142,7 +142,7 @@ future<> directory_initializer::ensure_rebalanced() {
 }
 
 manager::manager(service::storage_proxy& proxy, sstring hints_directory, host_filter filter, int64_t max_hint_window_ms,
-        resource_manager& res_manager, sharded<replica::database>& db)
+        resource_manager& res_manager, sharded<replica::database>& db, scheduling_group sg)
     : _hints_dir(fs::path(hints_directory) / fmt::to_string(this_shard_id()))
     , _host_filter(std::move(filter))
     , _proxy(proxy)
@@ -150,6 +150,7 @@ manager::manager(service::storage_proxy& proxy, sstring hints_directory, host_fi
     , _local_db(db.local())
     , _draining_eps_gate(seastar::format("hints::manager::{}", _hints_dir.native()))
     , _resource_manager(res_manager)
+    , _hints_sending_sched_group(sg)
 {
     if (utils::get_local_injector().enter("decrease_hints_flush_period")) {
         hints_flush_period = std::chrono::seconds{1};
@@ -415,7 +416,7 @@ hint_endpoint_manager& manager::get_ep_manager(const endpoint_id& host_id, const
 
     try {
         std::filesystem::path hint_directory = hints_dir() / (_uses_host_id ? fmt::to_string(host_id) : fmt::to_string(ip));
-        auto [it, _] = _ep_managers.emplace(host_id, hint_endpoint_manager{host_id, std::move(hint_directory), *this});
+        auto [it, _] = _ep_managers.emplace(host_id, hint_endpoint_manager{host_id, std::move(hint_directory), *this, _hints_sending_sched_group});
         hint_endpoint_manager& ep_man = it->second;
 
         manager_logger.trace("Created an endpoint manager for {}", host_id);

db/hints/manager.hh

@@ -133,6 +133,7 @@ private:
 
     hint_stats _stats;
     seastar::metrics::metric_groups _metrics;
+    scheduling_group _hints_sending_sched_group;
 
     // We need to keep a variant here. Before migrating hinted handoff to using host ID, hint directories will
     // still represent IP addresses. But after the migration, they will start representing host IDs.
@@ -155,7 +156,7 @@ private:
 
 public:
     manager(service::storage_proxy& proxy, sstring hints_directory, host_filter filter,
-            int64_t max_hint_window_ms, resource_manager& res_manager, sharded<replica::database>& db);
+            int64_t max_hint_window_ms, resource_manager& res_manager, sharded<replica::database>& db, scheduling_group sg);
 
     manager(const manager&) = delete;
     manager& operator=(const manager&) = delete;

db/row_cache.cc

@@ -24,12 +24,11 @@
 #include "readers/forwardable.hh"
 #include "readers/nonforwardable.hh"
 #include "cache_mutation_reader.hh"
-#include "partition_snapshot_reader.hh"
+#include "replica/partition_snapshot_reader.hh"
 #include "keys/clustering_key_filter.hh"
 #include "utils/assert.hh"
 #include "utils/updateable_value.hh"
 #include "utils/labels.hh"
-#include "utils/chunked_vector.hh"
 
 namespace cache {
 
@@ -846,7 +845,7 @@ mutation_reader row_cache::make_nonpopulating_reader(schema_ptr schema, reader_p
             cache_entry& e = *i;
             upgrade_entry(e);
             tracing::trace(ts, "Reading partition {} from cache", pos);
-            return make_partition_snapshot_flat_reader<false, dummy_accounter>(
+            return replica::make_partition_snapshot_reader<false, dummy_accounter>(
                     schema,
                     std::move(permit),
                     e.key(),
@@ -1216,10 +1215,10 @@ future<> row_cache::invalidate(external_updater eu, const dht::decorated_key& dk
 }
 
 future<> row_cache::invalidate(external_updater eu, const dht::partition_range& range, cache_invalidation_filter filter) {
-    return invalidate(std::move(eu), utils::chunked_vector<dht::partition_range>({range}), std::move(filter));
+    return invalidate(std::move(eu), dht::partition_range_vector({range}), std::move(filter));
 }
 
-future<> row_cache::invalidate(external_updater eu, utils::chunked_vector<dht::partition_range>&& ranges, cache_invalidation_filter filter) {
+future<> row_cache::invalidate(external_updater eu, dht::partition_range_vector&& ranges, cache_invalidation_filter filter) {
     return do_update(std::move(eu), [this, ranges = std::move(ranges), filter = std::move(filter)] mutable {
         return seastar::async([this, ranges = std::move(ranges), filter = std::move(filter)] {
             auto on_failure = defer([this] () noexcept {

db/row_cache.hh

@@ -17,7 +17,6 @@
 #include "utils/histogram.hh"
 #include "mutation/partition_version.hh"
 #include "utils/double-decker.hh"
-#include "utils/chunked_vector.hh"
 #include "db/cache_tracker.hh"
 #include "readers/empty.hh"
 #include "readers/mutation_source.hh"
@@ -458,7 +457,7 @@ public:
     // mutation source made prior to the call to invalidate().
     future<> invalidate(external_updater, const dht::decorated_key&);
     future<> invalidate(external_updater, const dht::partition_range& = query::full_partition_range, cache_invalidation_filter filter = [] (const auto&) { return true; });
-    future<> invalidate(external_updater, utils::chunked_vector<dht::partition_range>&&, cache_invalidation_filter filter = [] (const auto&) { return true; });
+    future<> invalidate(external_updater, dht::partition_range_vector&&, cache_invalidation_filter filter = [] (const auto&) { return true; });
 
     // Evicts entries from cache.
     //

db/schema_applier.cc

@@ -1139,17 +1139,14 @@ future<> schema_applier::finalize_tables_and_views() {
     // was already dropped (see https://github.com/scylladb/scylla/issues/5614)
     for (auto& dropped_view : diff.tables_and_views.local().views.dropped) {
         auto s = dropped_view.get();
-        co_await _ss.local().on_cleanup_for_drop_table(s->id());
         co_await replica::database::cleanup_drop_table_on_all_shards(sharded_db, _sys_ks, true, diff.table_shards[s->id()]);
     }
     for (auto& dropped_table : diff.tables_and_views.local().tables.dropped) {
         auto s = dropped_table.get();
-        co_await _ss.local().on_cleanup_for_drop_table(s->id());
         co_await replica::database::cleanup_drop_table_on_all_shards(sharded_db, _sys_ks, true, diff.table_shards[s->id()]);
     }
     for (auto& dropped_cdc : diff.tables_and_views.local().cdc.dropped) {
         auto s = dropped_cdc.get();
-        co_await _ss.local().on_cleanup_for_drop_table(s->id());
         co_await replica::database::cleanup_drop_table_on_all_shards(sharded_db, _sys_ks, true, diff.table_shards[s->id()]);
     }
 

db/schema_tables.cc

@@ -105,7 +105,7 @@ namespace {
         schema_builder::register_schema_initializer([](schema_builder& builder) {
             if (builder.ks_name() == schema_tables::NAME) {
                 // all schema tables are group0 tables
-                builder.set_is_group0_table();
+                builder.set_is_group0_table(true);
             }
         });
 }

db/system_keyspace.cc

@@ -87,15 +87,31 @@ namespace {
         static const std::unordered_set<sstring> tables = {
             schema_tables::SCYLLA_TABLE_SCHEMA_HISTORY,
             system_keyspace::BROADCAST_KV_STORE,
+            system_keyspace::CDC_GENERATIONS_V3,
             system_keyspace::RAFT,
             system_keyspace::RAFT_SNAPSHOTS,
             system_keyspace::RAFT_SNAPSHOT_CONFIG,
             system_keyspace::GROUP0_HISTORY,
             system_keyspace::DISCOVERY,
+            system_keyspace::TABLETS,
+            system_keyspace::TOPOLOGY,
+            system_keyspace::TOPOLOGY_REQUESTS,
             system_keyspace::LOCAL,
             system_keyspace::PEERS,
+            system_keyspace::SCYLLA_LOCAL,
             system_keyspace::COMMITLOG_CLEANUPS,
+            system_keyspace::SERVICE_LEVELS_V2,
+            system_keyspace::VIEW_BUILD_STATUS_V2,
+            system_keyspace::CDC_STREAMS_STATE,
+            system_keyspace::CDC_STREAMS_HISTORY,
+            system_keyspace::ROLES,
+            system_keyspace::ROLE_MEMBERS,
+            system_keyspace::ROLE_ATTRIBUTES,
+            system_keyspace::ROLE_PERMISSIONS,
             system_keyspace::CDC_LOCAL,
+            system_keyspace::DICTS,
+            system_keyspace::VIEW_BUILDING_TASKS,
+            system_keyspace::CLIENT_ROUTES,
         };
         if (builder.ks_name() == system_keyspace::NAME && tables.contains(builder.cf_name())) {
             builder.enable_schema_commitlog();
@@ -127,7 +143,7 @@ namespace {
                 system_keyspace::REPAIR_TASKS,
             };
             if (builder.ks_name() == system_keyspace::NAME && tables.contains(builder.cf_name())) {
-                builder.set_is_group0_table();
+                builder.set_is_group0_table(true);
             }
         });
 }
@@ -1698,9 +1714,7 @@ std::unordered_set<dht::token> decode_tokens(const set_type_impl::native_type& t
     std::unordered_set<dht::token> tset;
     for (auto& t: tokens) {
         auto str = value_cast<sstring>(t);
-        if (str != dht::token::from_sstring(str).to_sstring()) {
-            on_internal_error(slogger, format("decode_tokens: invalid token string '{}'", str));
-        }
+        SCYLLA_ASSERT(str == dht::token::from_sstring(str).to_sstring());
         tset.insert(dht::token::from_sstring(str));
     }
     return tset;
@@ -3177,7 +3191,7 @@ future<service::topology> system_keyspace::load_topology_state(const std::unorde
                     };
                 }
             } else if (must_have_tokens(nstate)) {
-                on_internal_error(slogger, format(
+                on_fatal_internal_error(slogger, format(
                         "load_topology_state: node {} in {} state but missing ring slice", host_id, nstate));
             }
         }
@@ -3259,7 +3273,7 @@ future<service::topology> system_keyspace::load_topology_state(const std::unorde
             // Currently, at most one node at a time can be in transitioning state.
             if (!map->empty()) {
                 const auto& [other_id, other_rs] = *map->begin();
-                on_internal_error(slogger, format(
+                on_fatal_internal_error(slogger, format(
                     "load_topology_state: found two nodes in transitioning state: {} in {} state and {} in {} state",
                     other_id, other_rs.state, host_id, nstate));
             }
@@ -3317,7 +3331,8 @@ future<service::topology> system_keyspace::load_topology_state(const std::unorde
                 format("SELECT count(range_end) as cnt FROM {}.{} WHERE key = '{}' AND id = ?",
                         NAME, CDC_GENERATIONS_V3, cdc::CDC_GENERATIONS_V3_KEY),
                 gen_id.id);
-            if (!gen_rows || gen_rows->empty()) {
+            SCYLLA_ASSERT(gen_rows);
+            if (gen_rows->empty()) {
                 on_internal_error(slogger, format(
                     "load_topology_state: last committed CDC generation time UUID ({}) present, but data missing", gen_id.id));
             }

db/system_keyspace.hh

@@ -215,8 +215,6 @@ public:
     static constexpr auto BUILT_VIEWS = "built_views";
     static constexpr auto SCYLLA_VIEWS_BUILDS_IN_PROGRESS = "scylla_views_builds_in_progress";
     static constexpr auto CDC_LOCAL = "cdc_local";
-    static constexpr auto CDC_TIMESTAMPS = "cdc_timestamps";
-    static constexpr auto CDC_STREAMS = "cdc_streams";
 
     // auth
     static constexpr auto ROLES = "roles";

db/view/view.cc

@@ -23,6 +23,7 @@
 
 #include <seastar/core/future-util.hh>
 #include <seastar/core/coroutine.hh>
+#include <seastar/coroutine/all.hh>
 #include <seastar/coroutine/maybe_yield.hh>
 #include <flat_map>
 
@@ -65,6 +66,7 @@
 #include "mutation/timestamp.hh"
 #include "utils/assert.hh"
 #include "utils/small_vector.hh"
+#include "view_builder.hh"
 #include "view_info.hh"
 #include "view_update_checks.hh"
 #include "types/list.hh"
@@ -930,7 +932,8 @@ bool view_updates::can_skip_view_updates(const clustering_or_static_row& update,
     const row& existing_row = existing.cells();
     const row& updated_row = update.cells();
 
-    return std::ranges::all_of(_base->regular_columns(), [this, &updated_row, &existing_row] (const column_definition& cdef) {
+    const bool base_has_nonexpiring_marker = update.marker().is_live() && !update.marker().is_expiring();
+    return std::ranges::all_of(_base->regular_columns(), [this, &updated_row, &existing_row, base_has_nonexpiring_marker] (const column_definition& cdef) {
         const auto view_it = _view->columns_by_name().find(cdef.name());
         const bool column_is_selected = view_it != _view->columns_by_name().end();
 
@@ -938,29 +941,49 @@ bool view_updates::can_skip_view_updates(const clustering_or_static_row& update,
         // as part of its PK, there are NO virtual columns corresponding to the unselected columns in the view.
         // Because of that, we don't generate view updates when the value in an unselected column is created
         // or changes.
-        if (!column_is_selected) {
+        if (!column_is_selected && _base_info.has_base_non_pk_columns_in_view_pk) {
             return true;
         }
 
-        // We cannot skip if the value was created or deleted
+        //TODO(sarna): Optimize collections case - currently they do not go under optimization
+        if (!cdef.is_atomic()) {
+            return false;
+        }
+
+        // We cannot skip if the value was created or deleted, unless we have a non-expiring marker
         const auto* existing_cell = existing_row.find_cell(cdef.id);
         const auto* updated_cell = updated_row.find_cell(cdef.id);
         if (existing_cell == nullptr || updated_cell == nullptr) {
-            return existing_cell == updated_cell;
+            return existing_cell == updated_cell || (!column_is_selected && base_has_nonexpiring_marker);
         }
-
-        if (!cdef.is_atomic()) {
-            return existing_cell->as_collection_mutation().data == updated_cell->as_collection_mutation().data;
-        }
-
         atomic_cell_view existing_cell_view = existing_cell->as_atomic_cell(cdef);
         atomic_cell_view updated_cell_view = updated_cell->as_atomic_cell(cdef);
 
         // We cannot skip when a selected column is changed
-        if (view_it->second->is_view_virtual()) {
-            return atomic_cells_liveness_equal(existing_cell_view, updated_cell_view);
+        if (column_is_selected) {
+            if (view_it->second->is_view_virtual()) {
+                return atomic_cells_liveness_equal(existing_cell_view, updated_cell_view);
+            }
+            return compare_atomic_cell_for_merge(existing_cell_view, updated_cell_view) == 0;
         }
-        return compare_atomic_cell_for_merge(existing_cell_view, updated_cell_view) == 0;
+
+        // With non-expiring row marker, liveness checks below are not relevant
+        if (base_has_nonexpiring_marker) {
+            return true;
+        }
+
+        if (existing_cell_view.is_live() != updated_cell_view.is_live()) {
+            return false;
+        }
+
+        // We cannot skip if the change updates TTL
+        const bool existing_has_ttl = existing_cell_view.is_live_and_has_ttl();
+        const bool updated_has_ttl = updated_cell_view.is_live_and_has_ttl();
+        if (existing_has_ttl || updated_has_ttl) {
+            return existing_has_ttl == updated_has_ttl && existing_cell_view.expiry() == updated_cell_view.expiry();
+        }
+
+        return true;
     });
 }
 
@@ -1728,7 +1751,7 @@ static endpoints_to_update get_view_natural_endpoint_vnodes(
         std::vector<std::reference_wrapper<const locator::node>> base_nodes,
         std::vector<std::reference_wrapper<const locator::node>> view_nodes,
         locator::endpoint_dc_rack my_location,
-        const bool network_topology,
+        const locator::network_topology_strategy* network_topology,
         replica::cf_stats& cf_stats) {
     using node_vector = std::vector<std::reference_wrapper<const locator::node>>;
     node_vector base_endpoints, view_endpoints;
@@ -1881,7 +1904,7 @@ endpoints_to_update get_view_natural_endpoint(
         locator::host_id me,
         const locator::effective_replication_map_ptr& base_erm,
         const locator::effective_replication_map_ptr& view_erm,
-        const bool network_topology,
+        const locator::abstract_replication_strategy& replication_strategy,
         const dht::token& base_token,
         const dht::token& view_token,
         bool use_tablets,
@@ -1889,6 +1912,7 @@ endpoints_to_update get_view_natural_endpoint(
     auto& topology = base_erm->get_token_metadata_ptr()->get_topology();
     auto& view_topology = view_erm->get_token_metadata_ptr()->get_topology();
     auto& my_location = topology.get_location(me);
+    auto* network_topology = dynamic_cast<const locator::network_topology_strategy*>(&replication_strategy);
 
     auto resolve = [&] (const locator::topology& topology, const locator::host_id& ep, bool is_view) -> const locator::node& {
         if (auto* np = topology.find_node(ep)) {
@@ -1922,7 +1946,7 @@ endpoints_to_update get_view_natural_endpoint(
                 // view pairing as the leaving base replica.
                 // note that the recursive call will not recurse again because leaving_base is in base_nodes.
                 auto leaving_base = it->get().host_id();
-                return get_view_natural_endpoint(leaving_base, base_erm, view_erm, network_topology, base_token,
+                return get_view_natural_endpoint(leaving_base, base_erm, view_erm, replication_strategy, base_token,
                         view_token, use_tablets, cf_stats);
             }
         }
@@ -2018,9 +2042,7 @@ future<> view_update_generator::mutate_MV(
         wait_for_all_updates wait_for_all)
 {
     auto& ks = _db.find_keyspace(base->ks_name());
-    const bool uses_tablets = ks.uses_tablets();
-    const bool uses_nts = dynamic_cast<const locator::network_topology_strategy*>(&ks.get_replication_strategy()) != nullptr;
-    // The object pointed by `ks` may disappear after preeemption. It should not be touched again after this comment.
+    auto& replication = ks.get_replication_strategy();
     std::unordered_map<table_id, locator::effective_replication_map_ptr> erms;
     auto get_erm = [&] (table_id id) {
         auto it = erms.find(id);
@@ -2039,8 +2061,8 @@ future<> view_update_generator::mutate_MV(
     co_await max_concurrent_for_each(view_updates, max_concurrent_updates, [&] (frozen_mutation_and_schema mut) mutable -> future<> {
         auto view_token = dht::get_token(*mut.s, mut.fm.key());
         auto view_ermp = erms.at(mut.s->id());
-        auto [target_endpoint, no_pairing_endpoint] = get_view_natural_endpoint(me, base_ermp, view_ermp, uses_nts, base_token, view_token,
-                uses_tablets, cf_stats);
+        auto [target_endpoint, no_pairing_endpoint] = get_view_natural_endpoint(me, base_ermp, view_ermp, replication, base_token, view_token,
+                ks.uses_tablets(), cf_stats);
         auto remote_endpoints = view_ermp->get_pending_replicas(view_token);
         auto memory_units = seastar::make_lw_shared<db::timeout_semaphore_units>(pending_view_update_memory_units.split(memory_usage_of(mut)));
         if (no_pairing_endpoint) {
@@ -2218,12 +2240,20 @@ void view_builder::setup_metrics() {
 }
 
 future<> view_builder::start_in_background(service::migration_manager& mm, utils::cross_shard_barrier barrier) {
+    auto step_fiber = make_ready_future<>();
     try {
         view_builder_init_state vbi;
         auto fail = defer([&barrier] mutable { barrier.abort(); });
-        // Guard the whole startup routine with a semaphore,
-        // so that it's not intercepted by `on_drop_view`, `on_create_view`
-        // or `on_update_view` events.
+        // Semaphore usage invariants:
+        // - One unit of _sem serializes all per-shard bookkeeping that mutates view-builder state
+        //   (_base_to_build_step, _built_views, build_status, reader resets).
+        // - The unit is held for the whole operation, including the async chain, until the state
+        //   is stable for the next operation on that shard.
+        // - Cross-shard operations acquire _sem on shard 0 for the duration of the broadcast.
+        //   Other shards acquire their own _sem only around their local handling; shard 0 skips
+        //   the local acquire because it already holds the unit from the dispatcher.
+        // Guard the whole startup routine with a semaphore so that it's not intercepted by
+        // `on_drop_view`, `on_create_view`, or `on_update_view` events.
         auto units = co_await get_units(_sem, view_builder_semaphore_units);
         // Wait for schema agreement even if we're a seed node.
         co_await mm.wait_for_schema_agreement(_db, db::timeout_clock::time_point::max(), &_as);
@@ -2244,8 +2274,10 @@ future<> view_builder::start_in_background(service::migration_manager& mm, utils
         _mnotifier.register_listener(this);
         co_await calculate_shard_build_step(vbi);
         _current_step = _base_to_build_step.begin();
-        // Waited on indirectly in stop().
-        (void)_build_step.trigger();
+
+        // If preparation above fails, run_in_background() is not invoked, just
+        // the start_in_background() emits a warning into logs and resolves
+        step_fiber = run_in_background();
     } catch (...) {
         auto ex = std::current_exception();
         auto ll = log_level::error;
@@ -2260,10 +2292,12 @@ future<> view_builder::start_in_background(service::migration_manager& mm, utils
         }
         vlogger.log(ll, "start aborted: {}", ex);
     }
+
+    co_await std::move(step_fiber);
 }
 
 future<> view_builder::start(service::migration_manager& mm, utils::cross_shard_barrier barrier) {
-    _started = start_in_background(mm, std::move(barrier));
+    _step_fiber = start_in_background(mm, std::move(barrier));
     return make_ready_future<>();
 }
 
@@ -2273,12 +2307,12 @@ future<> view_builder::drain() {
     }
     vlogger.info("Draining view builder");
     _as.request_abort();
-    co_await std::move(_started);
     co_await _mnotifier.unregister_listener(this);
     co_await _vug.drain();
     co_await _sem.wait();
     _sem.broken();
-    co_await _build_step.join();
+    _build_step.broken();
+    co_await std::move(_step_fiber);
     co_await coroutine::parallel_for_each(_base_to_build_step, [] (std::pair<const table_id, build_step>& p) {
         return p.second.reader.close();
     });
@@ -2647,63 +2681,59 @@ static bool should_ignore_tablet_keyspace(const replica::database& db, const sst
     return db.features().view_building_coordinator && db.has_keyspace(ks_name) && db.find_keyspace(ks_name).uses_tablets();
 }
 
-future<> view_builder::dispatch_create_view(sstring ks_name, sstring view_name) {
-    if (should_ignore_tablet_keyspace(_db, ks_name)) {
-        return make_ready_future<>();
-    }
-    return with_semaphore(_sem, view_builder_semaphore_units, [this, ks_name = std::move(ks_name), view_name = std::move(view_name)] () mutable {
-        // This runs on shard 0 only; seed the global rows before broadcasting.
-        return handle_seed_view_build_progress(ks_name, view_name).then([this, ks_name = std::move(ks_name), view_name = std::move(view_name)] () mutable {
-            return container().invoke_on_all([ks_name = std::move(ks_name), view_name = std::move(view_name)] (view_builder& vb) mutable {
-                return vb.handle_create_view_local(std::move(ks_name), std::move(view_name));
-            });
-        });
-    });
+future<view_builder::view_builder_units> view_builder::get_or_adopt_view_builder_lock(view_builder_units_opt units) {
+    co_return units ? std::move(*units) : co_await get_units(_sem, view_builder_semaphore_units);
 }
 
-future<> view_builder::handle_seed_view_build_progress(sstring ks_name, sstring view_name) {
+future<> view_builder::dispatch_create_view(sstring ks_name, sstring view_name) {
+    if (should_ignore_tablet_keyspace(_db, ks_name)) {
+        co_return;
+    }
+
+    auto units = co_await get_or_adopt_view_builder_lock(std::nullopt);
+    co_await handle_seed_view_build_progress(ks_name, view_name);
+
+    co_await coroutine::all(
+        [this, ks_name, view_name, units = std::move(units)] mutable -> future<> {
+            co_await handle_create_view_local(ks_name, view_name, std::move(units)); },
+        [this, ks_name, view_name] mutable -> future<> {
+            co_await container().invoke_on_others([ks_name = std::move(ks_name), view_name = std::move(view_name)] (view_builder& vb) mutable -> future<> {
+                return vb.handle_create_view_local(ks_name, view_name, std::nullopt); }); });
+}
+
+future<> view_builder::handle_seed_view_build_progress(const sstring& ks_name, const sstring& view_name) {
     auto view = view_ptr(_db.find_schema(ks_name, view_name));
     auto& step = get_or_create_build_step(view->view_info()->base_id());
     return _sys_ks.register_view_for_building_for_all_shards(view->ks_name(), view->cf_name(), step.current_token());
 }
 
-future<> view_builder::handle_create_view_local(sstring ks_name, sstring view_name){
-    if (this_shard_id() == 0) { 
-        return handle_create_view_local_impl(std::move(ks_name), std::move(view_name));
-    } else {
-        return with_semaphore(_sem, view_builder_semaphore_units, [this, ks_name = std::move(ks_name), view_name = std::move(view_name)] () mutable {
-            return handle_create_view_local_impl(std::move(ks_name), std::move(view_name));
-        });
-    }
-}
-
-future<> view_builder::handle_create_view_local_impl(sstring ks_name, sstring view_name) {
+future<> view_builder::handle_create_view_local(const sstring& ks_name, const sstring& view_name, view_builder_units_opt units) {
+    [[maybe_unused]] auto sem_units = co_await get_or_adopt_view_builder_lock(std::move(units));
     auto view = view_ptr(_db.find_schema(ks_name, view_name));
     auto& step = get_or_create_build_step(view->view_info()->base_id());
-    return when_all(step.base->await_pending_writes(), step.base->await_pending_streams()).discard_result().then([this, &step] {
-        return flush_base(step.base, _as);
-    }).then([this, view, &step] () {
+    try {
+        co_await coroutine::all(
+            [&step] -> future<> {
+                co_await step.base->await_pending_writes(); },
+            [&step] -> future<> {
+                co_await step.base->await_pending_streams(); });
+        co_await flush_base(step.base, _as);
+    
         // This resets the build step to the current token. It may result in views currently
         // being built to receive duplicate updates, but it simplifies things as we don't have
         // to keep around a list of new views to build the next time the reader crosses a token
         // threshold.
-        return initialize_reader_at_current_token(step).then([this, view, &step] () mutable {
-            return add_new_view(view, step);
-        }).then_wrapped([this, view] (future<>&& f) {
-            try {
-                f.get();
-            } catch (abort_requested_exception&) {
-                vlogger.debug("Aborted while setting up view for building {}.{}", view->ks_name(), view->cf_name());
-            } catch (raft::request_aborted&) {
-                vlogger.debug("Aborted while setting up view for building {}.{}", view->ks_name(), view->cf_name());
-            } catch (...) {
-                vlogger.error("Error setting up view for building {}.{}: {}", view->ks_name(), view->cf_name(), std::current_exception());
-            }
+        co_await initialize_reader_at_current_token(step);
+        co_await add_new_view(view, step);
+    } catch (abort_requested_exception&) {
+        vlogger.debug("Aborted while setting up view for building {}.{}", view->ks_name(), view->cf_name());
+    } catch (raft::request_aborted&) {
+        vlogger.debug("Aborted while setting up view for building {}.{}", view->ks_name(), view->cf_name());
+    } catch (...) {
+        vlogger.error("Error setting up view for building {}.{}: {}", view->ks_name(), view->cf_name(), std::current_exception());
+    }
 
-            // Waited on indirectly in stop().
-            static_cast<void>(_build_step.trigger());
-        });
-    });
+    _build_step.signal();
 }
 
 void view_builder::on_create_view(const sstring& ks_name, const sstring& view_name) {
@@ -2740,62 +2770,55 @@ void view_builder::on_update_view(const sstring& ks_name, const sstring& view_na
 
 future<> view_builder::dispatch_drop_view(sstring ks_name, sstring view_name) {
     if (should_ignore_tablet_keyspace(_db, ks_name)) {
-        return make_ready_future<>();
+        co_return;
     }
 
-    return with_semaphore(_sem, view_builder_semaphore_units, [this, ks_name = std::move(ks_name), view_name = std::move(view_name)] () mutable {
-        // This runs on shard 0 only; broadcast local cleanup before global cleanup.
-        return container().invoke_on_all([ks_name, view_name] (view_builder& vb) mutable {
-            return vb.handle_drop_view_local(std::move(ks_name), std::move(view_name));
-        }).then([this, ks_name = std::move(ks_name), view_name = std::move(view_name)] () mutable {
-            return handle_drop_view_global_cleanup(std::move(ks_name), std::move(view_name));
-        });
-    });
+    auto units = co_await get_or_adopt_view_builder_lock(std::nullopt);
+
+    co_await coroutine::all(
+        [this, ks_name, view_name, units = std::move(units)] mutable -> future<> {
+            co_await handle_drop_view_local(ks_name, view_name, std::move(units)); },
+        [this, ks_name, view_name] mutable -> future<> {
+            co_await container().invoke_on_others([ks_name = std::move(ks_name), view_name = std::move(view_name)] (view_builder& vb) mutable -> future<> {
+                return vb.handle_drop_view_local(ks_name, view_name, std::nullopt); });});
+    co_await handle_drop_view_global_cleanup(ks_name, view_name);
 }
 
-future<> view_builder::handle_drop_view_local(sstring ks_name, sstring view_name) {
-    if (this_shard_id() == 0) { 
-        return handle_drop_view_local_impl(std::move(ks_name), std::move(view_name));
-    } else {
-        return with_semaphore(_sem, view_builder_semaphore_units, [this, ks_name = std::move(ks_name), view_name = std::move(view_name)] () mutable {
-            return handle_drop_view_local_impl(std::move(ks_name), std::move(view_name));
-        });
-    }
-}
-
-future<> view_builder::handle_drop_view_local_impl(sstring ks_name, sstring view_name) {
+future<> view_builder::handle_drop_view_local(const sstring& ks_name, const sstring& view_name, view_builder_units_opt units) {
+    [[maybe_unused]] auto sem_units = co_await get_or_adopt_view_builder_lock(std::move(units));
     vlogger.info0("Stopping to build view {}.{}", ks_name, view_name);
-    // The view is absent from the database at this point, so find it by brute force.
-    ([&, this] {
-        for (auto& [_, step] : _base_to_build_step) {
-            if (step.build_status.empty() || step.build_status.front().view->ks_name() != ks_name) {
-                continue;
-            }
-            for (auto it = step.build_status.begin(); it != step.build_status.end(); ++it) {
-                if (it->view->cf_name() == view_name) {
-                    _built_views.erase(it->view->id());
-                    step.build_status.erase(it);
-                    return;
-                }
+
+    for (auto& [_, step] : _base_to_build_step) {
+        if (step.build_status.empty() || step.build_status.front().view->ks_name() != ks_name) {
+            continue;
+        }
+        for (auto it = step.build_status.begin(); it != step.build_status.end(); ++it) {
+            if (it->view->cf_name() == view_name) {
+                _built_views.erase(it->view->id());
+                step.build_status.erase(it);
+                co_return;
             }
         }
-    })();
-    return make_ready_future<>();  
+    }
 }
 
-future<> view_builder::handle_drop_view_global_cleanup(sstring ks_name, sstring view_name) {
+future<> view_builder::handle_drop_view_global_cleanup(const sstring& ks_name, const sstring& view_name) {
     if (this_shard_id() != 0) {
-        return make_ready_future<>();
+        co_return;
     }
     vlogger.info0("Starting view global cleanup {}.{}", ks_name, view_name);
-    return when_all_succeed(
-                _sys_ks.remove_view_build_progress_across_all_shards(ks_name, view_name),
-                _sys_ks.remove_built_view(ks_name, view_name),
-                remove_view_build_status(ks_name, view_name))
-                    .discard_result()
-                    .handle_exception([ks_name, view_name] (std::exception_ptr ep) {
-        vlogger.warn("Failed to cleanup view {}.{}: {}", ks_name, view_name, ep);
-    });
+    
+    try {
+        co_await coroutine::all(
+            [this, &ks_name, &view_name] -> future<>  {
+                co_await _sys_ks.remove_view_build_progress_across_all_shards(ks_name, view_name); },
+            [this, &ks_name, &view_name] -> future<>  {
+                co_await _sys_ks.remove_built_view(ks_name, view_name); },
+            [this, &ks_name, &view_name] -> future<>  {
+                co_await remove_view_build_status(ks_name, view_name); });
+    } catch (...) {
+        vlogger.warn("Failed to cleanup view {}.{}: {}", ks_name, view_name, std::current_exception());
+    }
 }
 
 void view_builder::on_drop_view(const sstring& ks_name, const sstring& view_name) {
@@ -2809,14 +2832,15 @@ void view_builder::on_drop_view(const sstring& ks_name, const sstring& view_name
     }));
 }
 
-future<> view_builder::do_build_step() {
-    // Run the view building in the streaming scheduling group
-    // so that it doesn't impact other tasks with higher priority.
-    seastar::thread_attributes attr;
-    attr.sched_group = _db.get_streaming_scheduling_group();
-    return seastar::async(std::move(attr), [this] {
+future<> view_builder::run_in_background() {
+    return seastar::async([this] {
         exponential_backoff_retry r(1s, 1min);
-        while (!_base_to_build_step.empty() && !_as.abort_requested()) {
+        while (!_as.abort_requested()) {
+            try {
+                _build_step.wait([this] { return !_base_to_build_step.empty(); }).get();
+            } catch (const seastar::broken_condition_variable&) {
+                return;
+            }
             auto units = get_units(_sem, view_builder_semaphore_units).get();
             ++_stats.steps_performed;
             try {

db/view/view.hh

@@ -303,7 +303,7 @@ endpoints_to_update get_view_natural_endpoint(
     locator::host_id node,
     const locator::effective_replication_map_ptr& base_erm,
     const locator::effective_replication_map_ptr& view_erm,
-    const bool network_topology,
+    const locator::abstract_replication_strategy& replication_strategy,
     const dht::token& base_token,
     const dht::token& view_token,
     bool use_tablets,

db/view/view_builder.hh

@@ -11,13 +11,13 @@
 #include "query/query-request.hh"
 #include "service/migration_listener.hh"
 #include "service/raft/raft_group0_client.hh"
-#include "utils/serialized_action.hh"
 #include "utils/cross-shard-barrier.hh"
 #include "replica/database.hh"
 
 #include <seastar/core/abort_source.hh>
 #include <seastar/core/future.hh>
 #include <seastar/core/semaphore.hh>
+#include <seastar/core/condition-variable.hh>
 #include <seastar/core/sharded.hh>
 #include <seastar/core/shared_future.hh>
 #include <seastar/core/shared_ptr.hh>
@@ -104,6 +104,12 @@ class view_update_generator;
  *            redo the missing step, for simplicity.
  */
 class view_builder final : public service::migration_listener::only_view_notifications, public seastar::peering_sharded_service<view_builder> {
+    //aliasing for semaphore units that will be used throughout the class
+    using view_builder_units = semaphore_units<named_semaphore_exception_factory>;
+
+    //aliasing for optional semaphore units that will be used throughout the class
+    using view_builder_units_opt = std::optional<view_builder_units>;
+
     /**
      * Keeps track of the build progress for a particular view.
      * When the view is built, next_token == first_token.
@@ -168,14 +174,24 @@ class view_builder final : public service::migration_listener::only_view_notific
     reader_permit _permit;
     base_to_build_step_type _base_to_build_step;
     base_to_build_step_type::iterator _current_step = _base_to_build_step.end();
-    serialized_action _build_step{std::bind(&view_builder::do_build_step, this)};
+    condition_variable _build_step;
     static constexpr size_t view_builder_semaphore_units = 1;
     // Ensures bookkeeping operations are serialized, meaning that while we execute
     // a build step we don't consider newly added or removed views. This simplifies
     // the algorithms. Also synchronizes an operation wrt. a call to stop().
+    // Semaphore usage invariants:
+    // - One unit of _sem serializes all per-shard bookkeeping that mutates view-builder state
+    //   (_base_to_build_step, _built_views, build_status, reader resets).
+    // - The unit is held for the whole operation, including the async chain, until the state
+    //   is stable for the next operation on that shard.
+    // - Cross-shard operations acquire _sem on shard 0 for the duration of the broadcast.
+    //   Other shards acquire their own _sem only around their local handling; shard 0 skips
+    //   the local acquire because it already holds the unit from the dispatcher.
+    // Guard the whole startup routine with a semaphore so that it's not intercepted by
+    // `on_drop_view`, `on_create_view`, or `on_update_view` events.
     seastar::named_semaphore _sem{view_builder_semaphore_units, named_semaphore_exception_factory{"view builder"}};
     seastar::abort_source _as;
-    future<> _started = make_ready_future<>();
+    future<> _step_fiber = make_ready_future<>();
     // Used to coordinate between shards the conclusion of the build process for a particular view.
     std::unordered_set<table_id> _built_views;
     // Used for testing.
@@ -262,19 +278,18 @@ private:
     void setup_shard_build_step(view_builder_init_state& vbi, std::vector<system_keyspace_view_name>, std::vector<system_keyspace_view_build_progress>);
     future<> calculate_shard_build_step(view_builder_init_state& vbi);
     future<> add_new_view(view_ptr, build_step&);
-    future<> do_build_step();
+    future<> run_in_background();
     void execute(build_step&, exponential_backoff_retry);
     future<> maybe_mark_view_as_built(view_ptr, dht::token);
     future<> mark_as_built(view_ptr);
     void setup_metrics();
     future<> dispatch_create_view(sstring ks_name, sstring view_name);
     future<> dispatch_drop_view(sstring ks_name, sstring view_name);
-    future<> handle_seed_view_build_progress(sstring ks_name, sstring view_name);
-    future<> handle_create_view_local(sstring ks_name, sstring view_name);
-    future<> handle_drop_view_local(sstring ks_name, sstring view_name);
-    future<> handle_create_view_local_impl(sstring ks_name, sstring view_name);
-    future<> handle_drop_view_local_impl(sstring ks_name, sstring view_name);
-    future<> handle_drop_view_global_cleanup(sstring ks_name, sstring view_name);
+    future<> handle_seed_view_build_progress(const sstring& ks_name, const sstring& view_name);
+    future<> handle_create_view_local(const sstring& ks_name, const sstring& view_name, view_builder_units_opt units);
+    future<> handle_drop_view_local(const sstring& ks_name, const sstring& view_name, view_builder_units_opt units);
+    future<> handle_drop_view_global_cleanup(const sstring& ks_name, const sstring& view_name);
+    future<view_builder_units> get_or_adopt_view_builder_lock(view_builder_units_opt units);
 
     template <typename Func1, typename Func2>
     future<> write_view_build_status(Func1&& fn_group0, Func2&& fn_sys_dist) {

db/view/view_building_worker.cc

@@ -242,7 +242,7 @@ future<> view_building_worker::create_staging_sstable_tasks() {
                 utils::UUID_gen::get_time_UUID(), view_building_task::task_type::process_staging, false,
                 table_id, ::table_id{}, {my_host_id, sst_info.shard}, sst_info.last_token
             };
-            auto mut = co_await _group0.client().sys_ks().make_view_building_task_mutation(guard.write_timestamp(), task);
+            auto mut = co_await _sys_ks.make_view_building_task_mutation(guard.write_timestamp(), task);
             cmuts.emplace_back(std::move(mut));
         }
     }
@@ -386,7 +386,6 @@ future<> view_building_worker::update_built_views() {
         auto schema = _db.find_schema(table_id);
         return std::make_pair(schema->ks_name(), schema->cf_name());
     };
-    auto& sys_ks = _group0.client().sys_ks();
 
     std::set<std::pair<sstring, sstring>> built_views;
     for (auto& [id, statuses]: _vb_state_machine.views_state.status_map) {
@@ -395,22 +394,22 @@ future<> view_building_worker::update_built_views() {
         }
     }
 
-    auto local_built = co_await sys_ks.load_built_views() | std::views::filter([&] (auto& v) {
+    auto local_built = co_await _sys_ks.load_built_views() | std::views::filter([&] (auto& v) {
         return !_db.has_keyspace(v.first) || _db.find_keyspace(v.first).uses_tablets();
     }) | std::ranges::to<std::set>();
 
     // Remove dead entries
     for (auto& view: local_built) {
         if (!built_views.contains(view)) {
-            co_await sys_ks.remove_built_view(view.first, view.second);
+            co_await _sys_ks.remove_built_view(view.first, view.second);
         }
     }
 
     // Add new entries
     for (auto& view: built_views) {
         if (!local_built.contains(view)) {
-            co_await sys_ks.mark_view_as_built(view.first, view.second);
-            co_await sys_ks.remove_view_build_progress_across_all_shards(view.first, view.second);
+            co_await _sys_ks.mark_view_as_built(view.first, view.second);
+            co_await _sys_ks.remove_view_build_progress_across_all_shards(view.first, view.second);
         }
     }
 }

db/virtual_tables.cc

@@ -1345,8 +1345,8 @@ public:
 
 private:
     static schema_ptr build_schema() {
-        auto id = generate_legacy_id(system_keyspace::NAME, system_keyspace::CDC_TIMESTAMPS);
-        return schema_builder(system_keyspace::NAME, system_keyspace::CDC_TIMESTAMPS, std::make_optional(id))
+        auto id = generate_legacy_id(system_keyspace::NAME, "cdc_timestamps");
+        return schema_builder(system_keyspace::NAME, "cdc_timestamps", std::make_optional(id))
             .with_column("keyspace_name", utf8_type, column_kind::partition_key)
             .with_column("table_name", utf8_type, column_kind::partition_key)
             .with_column("timestamp", reversed_type_impl::get_instance(timestamp_type), column_kind::clustering_key)
@@ -1428,8 +1428,8 @@ public:
     }
 private:
     static schema_ptr build_schema() {
-        auto id = generate_legacy_id(system_keyspace::NAME, system_keyspace::CDC_STREAMS);
-        return schema_builder(system_keyspace::NAME, system_keyspace::CDC_STREAMS, std::make_optional(id))
+        auto id = generate_legacy_id(system_keyspace::NAME, "cdc_streams");
+        return schema_builder(system_keyspace::NAME, "cdc_streams", std::make_optional(id))
             .with_column("keyspace_name", utf8_type, column_kind::partition_key)
             .with_column("table_name", utf8_type, column_kind::partition_key)
             .with_column("timestamp", timestamp_type, column_kind::clustering_key)

debug.cc

@@ -11,5 +11,6 @@
 namespace debug {
 
 seastar::sharded<replica::database>* volatile the_database = nullptr;
+seastar::scheduling_group streaming_scheduling_group;
 
 }

debug.hh

@@ -17,7 +17,7 @@ class database;
 namespace debug {
 
 extern seastar::sharded<replica::database>* volatile the_database;
-
+extern seastar::scheduling_group streaming_scheduling_group;
 
 }
 

dht/i_partitioner.cc

@@ -352,16 +352,6 @@ dht::partition_range_vector to_partition_ranges(const dht::token_range_vector& r
     return prs;
 }
 
-future<utils::chunked_vector<dht::partition_range>> to_partition_ranges_chunked(const dht::token_range_vector& ranges) {
-    utils::chunked_vector<dht::partition_range> prs;
-    prs.reserve(ranges.size());
-    for (auto& range : ranges) {
-        prs.push_back(dht::to_partition_range(range));
-        co_await coroutine::maybe_yield();
-    }
-    co_return prs;
-}
-
 std::map<unsigned, dht::partition_range_vector>
 split_range_to_shards(dht::partition_range pr, const schema& s, const sharder& raw_sharder) {
     std::map<unsigned, dht::partition_range_vector> ret;
@@ -374,11 +364,11 @@ split_range_to_shards(dht::partition_range pr, const schema& s, const sharder& r
     return ret;
 }
 
-future<utils::chunked_vector<dht::partition_range>> subtract_ranges(const schema& schema, utils::chunked_vector<dht::partition_range> source_ranges, utils::chunked_vector<dht::partition_range> ranges_to_subtract) {
+future<dht::partition_range_vector> subtract_ranges(const schema& schema, const dht::partition_range_vector& source_ranges, dht::partition_range_vector ranges_to_subtract) {
     auto cmp = dht::ring_position_comparator(schema);
     // optimize set of potentially overlapping ranges by deoverlapping them.
-    auto ranges = dht::partition_range::deoverlap(std::move(source_ranges), cmp);
-    utils::chunked_vector<dht::partition_range> res;
+    auto ranges = dht::partition_range::deoverlap(source_ranges, cmp);
+    dht::partition_range_vector res;
     res.reserve(ranges.size() * 2);
 
     auto range = ranges.begin();

dht/i_partitioner.hh

@@ -91,7 +91,6 @@ inline token get_token(const schema& s, partition_key_view key) {
 
 dht::partition_range to_partition_range(dht::token_range);
 dht::partition_range_vector to_partition_ranges(const dht::token_range_vector& ranges, utils::can_yield can_yield = utils::can_yield::no);
-future<utils::chunked_vector<dht::partition_range>> to_partition_ranges_chunked(const dht::token_range_vector& ranges);
 
 // Each shard gets a sorted, disjoint vector of ranges
 std::map<unsigned, dht::partition_range_vector>
@@ -106,7 +105,7 @@ std::unique_ptr<dht::i_partitioner> make_partitioner(sstring name);
 // Returns a sorted and deoverlapped list of ranges that are
 // the result of subtracting all ranges from ranges_to_subtract.
 // ranges_to_subtract must be sorted and deoverlapped.
-future<utils::chunked_vector<dht::partition_range>> subtract_ranges(const schema& schema, utils::chunked_vector<dht::partition_range> ranges, utils::chunked_vector<dht::partition_range> ranges_to_subtract);
+future<dht::partition_range_vector> subtract_ranges(const schema& schema, const dht::partition_range_vector& ranges, dht::partition_range_vector ranges_to_subtract);
 
 // Returns a token_range vector split based on the given number of most-significant bits
 dht::token_range_vector split_token_range_msb(unsigned most_significant_bits);

dht/token.hh

@@ -30,31 +30,6 @@ enum class token_kind {
     after_all_keys,
 };
 
-// Represents a token for partition keys.
-// Has a disengaged state, which sorts before all engaged states.
-struct raw_token {
-    int64_t value;
-
-    /// Constructs a disengaged token.
-    raw_token() : value(std::numeric_limits<int64_t>::min()) {}
-
-    /// Constructs an engaged token.
-    /// The token must be of token_kind::key kind.
-    explicit raw_token(const token&);
-
-    explicit raw_token(int64_t v) : value(v) {};
-
-    std::strong_ordering operator<=>(const raw_token& o) const noexcept = default;
-    std::strong_ordering operator<=>(const token& o) const noexcept;
-
-    /// Returns true iff engaged.
-    explicit operator bool() const noexcept {
-        return value != std::numeric_limits<int64_t>::min();
-    }
-};
-
-using raw_token_opt = seastar::optimized_optional<raw_token>;
-
 class token {
     // INT64_MIN is not a legal token, but a special value used to represent
     // infinity in token intervals.
@@ -77,10 +52,6 @@ public:
 
     constexpr explicit token(int64_t d) noexcept : token(kind::key, normalize(d)) {}
 
-    token(raw_token raw) noexcept
-        : token(raw ? kind::key : kind::before_all_keys, raw.value)
-    { }
-
     // This constructor seems redundant with the bytes_view constructor, but
     // it's necessary for IDL, which passes a deserialized_bytes_proxy here.
     // (deserialized_bytes_proxy is convertible to bytes&&, but not bytes_view.)
@@ -252,29 +223,6 @@ public:
     }
 };
 
-inline
-raw_token::raw_token(const token& t)
-    : value(t.raw())
-{
-#ifdef DEBUG
-    assert(t._kind == token::kind::key);
-#endif
-}
-
-inline
-std::strong_ordering raw_token::operator<=>(const token& o) const noexcept {
-    switch (o._kind) {
-        case token::kind::after_all_keys:
-            return std::strong_ordering::less;
-        case token::kind::before_all_keys:
-            // before_all_keys has a raw value set to the same raw value as a disengaged raw_token, and sorts before all keys.
-            // So we can order them by just comparing raw values.
-            [[fallthrough]];
-        case token::kind::key:
-            return value <=> o._data;
-    }
-}
-
 inline constexpr std::strong_ordering tri_compare_raw(const int64_t l1, const int64_t l2) noexcept {
     if (l1 == l2) {
         return std::strong_ordering::equal;
@@ -381,17 +329,6 @@ struct fmt::formatter<dht::token> : fmt::formatter<string_view> {
     }
 };
 
-template <>
-struct fmt::formatter<dht::raw_token> : fmt::formatter<string_view> {
-    template <typename FormatContext>
-    auto format(const dht::raw_token& t, FormatContext& ctx) const {
-        if (!t) {
-            return fmt::format_to(ctx.out(), "null");
-        }
-        return fmt::format_to(ctx.out(), "{}", t.value);
-    }
-};
-
 namespace std {
 
 template<>

dist/docker/redhat/build_docker.sh

@@ -97,9 +97,7 @@ bcp LICENSE-ScyllaDB-Source-Available.md /licenses/
 
 run microdnf clean all
 run microdnf --setopt=tsflags=nodocs -y update
-run microdnf --setopt=tsflags=nodocs -y install hostname kmod procps-ng python3 python3-pip cpio
-# Extract only systemctl binary from systemd package to avoid installing the whole systemd in the container.
-run bash -rc "microdnf download systemd && rpm2cpio systemd-*.rpm | cpio -idmv ./usr/bin/systemctl && rm -rf systemd-*.rpm"
+run microdnf --setopt=tsflags=nodocs -y install hostname kmod procps-ng python3 python3-pip
 run curl -L --output /etc/yum.repos.d/scylla.repo ${repo_file_url}
 run pip3 install --no-cache-dir --prefix /usr supervisor
 run bash -ec "echo LANG=C.UTF-8 > /etc/locale.conf"
@@ -108,8 +106,6 @@ run bash -ec "cat /scylla_bashrc >> /etc/bash.bashrc"
 run mkdir -p /var/log/scylla
 run chown -R scylla:scylla /var/lib/scylla
 run sed -i -e 's/^SCYLLA_ARGS=".*"$/SCYLLA_ARGS="--log-to-syslog 0 --log-to-stdout 1 --network-stack posix"/' /etc/sysconfig/scylla-server
-# Cleanup packages not needed in the final image and clean package manager cache to reduce image size.
-run bash -rc "microdnf remove -y cpio && microdnf clean all"
 
 run mkdir -p /opt/scylladb/supervisor
 run touch /opt/scylladb/SCYLLA-CONTAINER-FILE

docs/alternator/alternator.md

@@ -142,6 +142,10 @@ want modify a non-top-level attribute directly (e.g., a.b[3].c) need RMW:
 Alternator implements such requests by reading the entire top-level
 attribute a, modifying only a.b[3].c, and then writing back a.
 
+Currently, Alternator doesn't use Tablets. That's because Alternator relies
+on LWT (lightweight transactions), and LWT is not supported in keyspaces
+with Tablets enabled.
+
 ```{eval-rst}
 .. toctree::
     :maxdepth: 2

docs/architecture/tablets.rst

@@ -187,23 +187,6 @@ You can create a keyspace with tablets enabled with the ``tablets = {'enabled':
     the keyspace schema with ``tablets = { 'enabled': false }`` or 
     ``tablets = { 'enabled': true }``.
 
-.. _keyspace-rf-rack-valid-to-enforce-rack-list:
-
-Enforcing Rack-List Replication for Tablet Keyspaces
-------------------------------------------------------------------
-
-The ``rf_rack_valid_keyspaces`` is a legacy option that ensures that all keyspaces with tablets enabled are
-:term:`RF-rack-valid <RF-rack-valid keyspace>`.
-
-Requiring every tablet keyspace to use the rack list replication factor exclusively is enough to guarantee the keyspace is
-:term:`RF-rack-valid <RF-rack-valid keyspace>`. It reduces restrictions and provides stronger guarantees compared
-to ``rf_rack_valid_keyspaces`` option.
-
-To enforce rack list in tablet keyspaces, use ``enforce_rack_list`` option. It can be set only if all tablet keyspaces use
-rack list. To ensure that, follow a procedure of :ref:`conversion to rack list replication factor <conversion-to-rack-list-rf>`.
-After that restart all nodes in the cluster, with ``enforce_rack_list`` enabled and ``rf_rack_valid_keyspaces`` disabled. Make
-sure to avoid setting or updating replication factor (with CREATE KEYSPACE or ALTER KEYSPACE) while nodes are being restarted.
-
 .. _tablets-limitations:
 
 Limitations and Unsupported Features

docs/cql/ddl.rst

@@ -200,6 +200,8 @@ for two cases. One is setting replication factor to 0, in which case the number
 The other is when the numeric replication factor is equal to the current number of replicas
 for a given datacanter, in which case the current rack list is preserved.
 
+Altering from a numeric replication factor to a rack list is not supported yet.
+
 Note that when ``ALTER`` ing keyspaces and supplying ``replication_factor``,
 auto-expansion will only *add* new datacenters for safety, it will not alter
 existing datacenters or remove any even if they are no longer in the cluster.
@@ -422,21 +424,6 @@ Altering from a rack list to a numeric replication factor is not supported.
 
 Keyspaces which use rack lists are :term:`RF-rack-valid <RF-rack-valid keyspace>` if each rack in the rack list contains at least one node (excluding :doc:`zero-token nodes </architecture/zero-token-nodes>`).
 
-.. _conversion-to-rack-list-rf:
-
-Conversion to rack-list replication factor
-~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
-
-To migrate a keyspace from a numeric replication factor to a rack-list replication factor, provide the rack-list replication factor explicitly in ALTER KEYSPACE statement. The number of racks in the list must be equal to the numeric replication factor. The replication factor can be converted in any number of DCs at once. In a statement that converts replication factor, no replication factor updates (increase or decrease) are allowed in any DC.
-
-.. code-block:: cql
-
-  CREATE KEYSPACE Excelsior
-   WITH replication = { 'class' : 'NetworkTopologyStrategy', 'dc1' : 3, 'dc2' : 1} AND tablets = { 'enabled': true };
-
-  ALTER KEYSPACE Excelsior
-   WITH replication = { 'class' : 'NetworkTopologyStrategy', 'dc1' : ['RAC1', 'RAC2', 'RAC3'], 'dc2' : ['RAC4']} AND tablets = { 'enabled': true };
-
 .. _drop-keyspace-statement:
 
 DROP KEYSPACE
@@ -1039,7 +1026,29 @@ You can enable the after-repair tombstone GC by setting the ``repair`` mode usin
 
     ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'repair'} ;
 
-The following modes are available:
+To support writes arriving out-of-order -- either due to natural delays, or user provided timestamps -- the repair mode has a propagation delay.
+Out-of-order writes present a problem for repair mode tombstone gc. Consider the following example sequence of events:
+
+1) Write ``DELETE FROM table WHERE key = K1`` arrives at the node.
+2) Repair is run.
+3) Compaction runs and garbage collects the tombstone for ``key = K1``.
+4) Write ``INSERT INTO table (key, ...) VALUES (K1, ...)`` arrives at the node with timestamp smaller than that of the delete. The tombstone for ``key = K1`` should apply to this write, but it is already garbage collected, so this data is resurrected.
+
+Propagation delay solves this problem by establishing a window before repair, where tombstones are not yet garbage collectible: a tombstone is garbage collectible if it was written before the last repair by at least the propagation delay.
+
+The value of the propagation delay can be set via the ``propagation_delay_in_seconds`` parameter:
+
+.. code-block:: cql
+
+    CREATE TABLE ks.cf (key blob PRIMARY KEY, val blob) WITH tombstone_gc = {'mode':'repair', 'propagation_delay_in_seconds': 120};
+
+.. code-block:: cql
+
+    ALTER TABLE ks.cf WITH tombstone_gc = {'mode':'repair', 'propagation_delay_in_seconds': 120};
+
+The default value of the propagation delay is 1 hour. This parameter should only be changed if your application uses user provided timestamps and writes and deletes can arrive out-of-order by more than the default 1 hour.
+
+The following tombstone gc modes are available:
 
 .. list-table::
    :widths: 20 80

docs/cql/dml/select.rst

@@ -282,7 +282,6 @@ For example::
 
 
 Vector queries also support filtering with ``WHERE`` clauses on columns that are part of the primary key.
-See :ref:`WHERE <where-clause>`.
 
 For example::
 
@@ -290,6 +289,10 @@ For example::
       WHERE user_id = 'user123'
       ORDER BY embedding ANN OF [0.1, 0.2, 0.3, 0.4] LIMIT 5;
 
+The supported operations are equal relations (``=`` and ``IN``) with restrictions as in regular ``WHERE`` clauses. See :ref:`WHERE <where-clause>`.
+
+Other filtering scenarios are currently not supported.
+
 .. note::
 
    Vector indexes are supported in ScyllaDB Cloud only in clusters that have the Vector Search feature enabled.

docs/cql/secondary-indexes.rst

@@ -140,83 +140,17 @@ Vector Index :label-note:`ScyllaDB Cloud`
    `ScyllaDB Cloud documentation <https://cloud.docs.scylladb.com/stable/vector-search/>`_.
 
 ScyllaDB supports creating vector indexes on tables, allowing queries on the table to use those indexes for efficient
-similarity search on vector data. Vector indexes can be a global index for indexing vectors per table or a local
-index for indexing vectors per partition.
+similarity search on vector data. 
 
 The vector index is the only custom type index supported in ScyllaDB. It is created using
-the ``CUSTOM`` keyword and specifying the index type as ``vector_index``. It is also possible to
-add additional columns to the index for filtering the search results. The partition column
-specified in the global vector index definition must be the vector column, and any subsequent
-columns are treated as filtering columns. The local vector index requires that the partition key
-of the base table is also the partition key of the index and the vector column is the first one
-from the following columns.
-
-Example of a simple index:
+the ``CUSTOM`` keyword and specifying the index type as ``vector_index``. Example:
 
 .. code-block:: cql
 
-      CREATE CUSTOM INDEX vectorIndex ON ImageEmbeddings (embedding)
+      CREATE CUSTOM INDEX vectorIndex ON ImageEmbeddings (embedding) 
       USING 'vector_index' 
       WITH OPTIONS = {'similarity_function': 'COSINE', 'maximum_node_connections': '16'};
 
-The vector column (``embedding``) is indexed to enable similarity search using
-a global vector index. Additional filtering can be performed on the primary key
-columns of the base table.
-
-Example of a global vector index with additional filtering:
-
-.. code-block:: cql
-
-      CREATE CUSTOM INDEX vectorIndex ON ImageEmbeddings (embedding, category, info)
-      USING 'vector_index' 
-      WITH OPTIONS = {'similarity_function': 'COSINE', 'maximum_node_connections': '16'};
-
-The vector column (``embedding``) is indexed to enable similarity search using
-a global index. Additional columns are added for filtering the search results.
-The filtering is possible on ``category``, ``info`` and all primary key columns
-of the base table.
-
-Example of a local vector index:
-
-.. code-block:: cql
-
-      CREATE CUSTOM INDEX vectorIndex ON ImageEmbeddings ((id, created_at), embedding, category, info)
-      USING 'vector_index' 
-      WITH OPTIONS = {'similarity_function': 'COSINE', 'maximum_node_connections': '16'};
-
-The vector column (``embedding``) is indexed for similarity search (a local
-index) and additional columns are added for filtering the search results. The
-filtering is possible on ``category``, ``info`` and all primary key columns of
-the base table. The columns ``id`` and ``created_at`` must be the partition key
-of the base table.
-
-Vector indexes support additional filtering columns of native data types
-(excluding counter and duration). The indexed column itself must be a vector
-column, while the extra columns can be used to filter search results.
-
-The supported types are:
-
-* ``ascii``
-* ``bigint``
-* ``blob``
-* ``boolean``
-* ``date``
-* ``decimal``
-* ``double``
-* ``float``
-* ``inet``
-* ``int``
-* ``smallint``
-* ``text``
-* ``varchar``
-* ``time``
-* ``timestamp``
-* ``timeuuid``
-* ``tinyint``
-* ``uuid``
-* ``varint``
-
-
 The following options are supported for vector indexes. All of them are optional.
 
 +------------------------------+----------------------------------------------------------------------------------------------------------+---------------+

docs/dev/reader-concurrency-semaphore.md

@@ -78,6 +78,7 @@ Permits are in one of the following states:
 * `active/await` - a previously `active/need_cpu` permit, which needs something other than CPU to proceed, it is waiting on I/O or a remote shards, other permits can be admitted while the permit is in this state, pending resource availability;
 * `inactive` - the permit was marked inactive, it can be evicted to make room for admitting more permits if needed;
 * `evicted` - a former inactive permit which was evicted, the permit has to undergo admission again for the read to resume;
+* `preemptive_aborted` - the permit timed out or was rejected during admission as it was detected the read might time out later during execution;
 
 Note that some older releases will have different names for some of these states or lack some of the states altogether:
 

docs/dev/testing.md

@@ -124,6 +124,7 @@ There are several test directories that are excluded from orchestration by `test
 - test/cql
 - test/cqlpy
 - test/rest_api
+- test/scylla_gdb
 
 This means that `test.py` will not run tests directly, but will delegate all work to `pytest`.
 That's why all these directories do not have `suite.yaml` files.

docs/features/automatic-repair.rst

@@ -3,9 +3,9 @@
 Automatic Repair
 ================
 
-Traditionally, launching :doc:`repairs </operating-scylla/procedures/maintenance/repair>` in a ScyllaDB cluster is left to an external process, typically done via `Scylla Manager <https://manager.docs.scylladb.com/stable/repair/index.html>`_.
+Traditionally, launching `repairs </operating-scylla/procedures/maintenance/repair>`_ in a ScyllaDB cluster is left to an external process, typically done via `Scylla Manager <https://manager.docs.scylladb.com/stable/repair/index.html>`_.
 
-Automatic repair offers built-in scheduling in ScyllaDB itself. If the time since the last repair is greater than the configured repair interval, ScyllaDB will start a repair for the :doc:`tablet table </architecture/tablets>` automatically.
+Automatic repair offers built-in scheduling in ScyllaDB itself. If the time since the last repair is greater than the configured repair interval, ScyllaDB will start a repair for the tablet `tablet </architecture/tablets>`_ automatically.
 Repairs are spread over time and among nodes and shards, to avoid load spikes or any adverse effects on user workloads.
 
 To enable automatic repair, add this to the configuration (``scylla.yaml``):
@@ -20,4 +20,4 @@ More featureful configuration methods will be implemented in the future.
 
 To disable, set ``auto_repair_enabled_default: false``.
 
-Automatic repair relies on :doc:`Incremental Repair </features/incremental-repair>` and as such it only works with :doc:`tablet </architecture/tablets>` tables.
+Automatic repair relies on `Incremental Repair </features/incremental-repair>`_ and as such it only works with `tablet </architecture/tablets>`_ tables.

docs/features/incremental-repair.rst

@@ -3,7 +3,7 @@
 Incremental Repair
 ==================
 
-ScyllaDB's standard :doc:`repair </operating-scylla/procedures/maintenance/repair>` process scans and processes all the data on a node, regardless of whether it has changed since the last repair. This operation can be resource-intensive and time-consuming. The Incremental Repair feature provides a much more efficient and lightweight alternative for maintaining data consistency.
+ScyllaDB's standard `repair </operating-scylla/procedures/maintenance/repair>`_ process scans and processes all the data on a node, regardless of whether it has changed since the last repair. This operation can be resource-intensive and time-consuming. The Incremental Repair feature provides a much more efficient and lightweight alternative for maintaining data consistency.
 
 The core idea of incremental repair is to repair only the data that has been written or changed since the last repair was run. It intelligently skips data that has already been verified, dramatically reducing the time, I/O, and CPU resources required for the repair operation.
 
@@ -51,7 +51,7 @@ Benefits of Incremental Repair
 *   **Reduced Resource Usage:** Consumes significantly less CPU, I/O, and network bandwidth compared to a full repair.
 *   **More Frequent Repairs:** The efficiency of incremental repair allows you to run it more frequently, ensuring a higher level of data consistency across your cluster at all times.
 
-Tables using Incremental Repair can schedule repairs in ScyllaDB itself, with :doc:`Automatic Repair </features/automatic-repair>`.
+Tables using Incremental Repair can schedule repairs in ScyllaDB itself, with `Automatic Repair </features/automatic-repair>`_.
 
 Notes
 -----

docs/getting-started/install-scylla/index.rst

@@ -10,6 +10,7 @@ Install ScyllaDB |CURRENT_VERSION|
    /getting-started/install-scylla/launch-on-azure
    /getting-started/installation-common/scylla-web-installer
    /getting-started/install-scylla/install-on-linux
+   /getting-started/installation-common/install-jmx
    /getting-started/install-scylla/run-in-docker
    /getting-started/installation-common/unified-installer
    /getting-started/installation-common/air-gapped-install
@@ -23,9 +24,9 @@ Keep your versions up-to-date. The two latest versions are supported. Also, alwa
   :id: "getting-started"
   :class: my-panel
 
-  * :doc:`Launch ScyllaDB on AWS </getting-started/install-scylla/launch-on-aws>`
-  * :doc:`Launch ScyllaDB on GCP </getting-started/install-scylla/launch-on-gcp>`
-  * :doc:`Launch ScyllaDB on Azure </getting-started/install-scylla/launch-on-azure>`
+  * :doc:`Launch ScyllaDB |CURRENT_VERSION| on AWS </getting-started/install-scylla/launch-on-aws>`
+  * :doc:`Launch ScyllaDB |CURRENT_VERSION| on GCP </getting-started/install-scylla/launch-on-gcp>`
+  * :doc:`Launch ScyllaDB |CURRENT_VERSION| on Azure </getting-started/install-scylla/launch-on-azure>`
 
 
 .. panel-box::
@@ -34,7 +35,8 @@ Keep your versions up-to-date. The two latest versions are supported. Also, alwa
   :class: my-panel
 
   * :doc:`Install ScyllaDB with Web Installer (recommended) </getting-started/installation-common/scylla-web-installer>`
-  * :doc:`Install ScyllaDB Linux Packages </getting-started/install-scylla/install-on-linux>`
+  * :doc:`Install ScyllaDB |CURRENT_VERSION| Linux Packages </getting-started/install-scylla/install-on-linux>`
+  * :doc:`Install scylla-jmx Package </getting-started/installation-common/install-jmx>`
   * :doc:`Install ScyllaDB Without root Privileges </getting-started/installation-common/unified-installer>`
   * :doc:`Air-gapped Server Installation </getting-started/installation-common/air-gapped-install>`
   * :doc:`ScyllaDB Developer Mode </getting-started/installation-common/dev-mod>`

docs/getting-started/install-scylla/install-on-linux.rst

@@ -52,7 +52,7 @@ Install ScyllaDB
             .. code-block:: console
                :substitutions:
     
-               sudo wget -O /etc/apt/sources.list.d/scylla.list https://downloads.scylladb.com/deb/debian/|UBUNTU_SCYLLADB_LIST|
+               sudo wget -O /etc/apt/sources.list.d/scylla.list http://downloads.scylladb.com/deb/debian/|UBUNTU_SCYLLADB_LIST|
 
 
         #. Install ScyllaDB packages.
@@ -94,6 +94,16 @@ Install ScyllaDB
 
                apt-get install scylla{,-server,-kernel-conf,-node-exporter,-conf,-python3,-cqlsh}=2025.3.1-0.20250907.2bbf3cf669bb-1
 
+
+        #. (Ubuntu only) Set Java 11.
+
+            .. code-block:: console
+    
+               sudo apt-get update
+               sudo apt-get install -y openjdk-11-jre-headless
+               sudo update-java-alternatives --jre-headless -s java-1.11.0-openjdk-amd64
+
+
    .. group-tab:: Centos/RHEL
 
         #. Install the EPEL repository.
@@ -125,7 +135,7 @@ Install ScyllaDB
             .. code-block:: console
                :substitutions:
     
-               sudo curl -o /etc/yum.repos.d/scylla.repo -L https://downloads.scylladb.com/rpm/centos/|CENTOS_SCYLLADB_REPO|
+               sudo curl -o /etc/yum.repos.d/scylla.repo -L http://downloads.scylladb.com/rpm/centos/|CENTOS_SCYLLADB_REPO|
 
         #. Install ScyllaDB packages.
 
@@ -133,19 +143,27 @@ Install ScyllaDB
     
                sudo yum install scylla
 
-            Running the command installs the latest official version of ScyllaDB.
-            Alternatively, you can install a specific patch version:
+            Running the command installs the latest official version of ScyllaDB Open Source.
+            Alternatively, you can to install a specific patch version:
 
             .. code-block:: console
     
                sudo yum install scylla-<your patch version>
 
-            Example: The following example shows installing ScyllaDB 2025.3.1.
+            Example: The following example shows the command to install ScyllaDB 5.2.3.
 
             .. code-block:: console
                :class: hide-copy-button
     
-               sudo yum install scylla-2025.3.1
+               sudo yum install scylla-5.2.3
+
+(Optional) Install scylla-jmx
+-------------------------------
+
+    scylla-jmx is an optional package and is not installed by default.
+    If you need JMX server, see :doc:`Install scylla-jmx Package </getting-started/installation-common/install-jmx>`.
+
+
 
 .. include:: /getting-started/_common/setup-after-install.rst
 

docs/getting-started/installation-common/install-jmx.rst

@@ -0,0 +1,78 @@
+
+======================================
+Install scylla-jmx Package
+======================================
+
+scylla-jmx is an optional package and is not installed by default.
+If you need JMX server, you can still install it from scylla-jmx GitHub page.
+
+.. tabs::
+
+   .. group-tab:: Debian/Ubuntu
+        #. Download .deb package from scylla-jmx page.
+
+            Access to https://github.com/scylladb/scylla-jmx, select latest
+            release from "releases", download a file end with ".deb".
+
+        #. (Optional) Transfer the downloaded package to the install node.
+
+            If the pc from which you downloaded the package is different from
+            the node where you install scylladb, you will need to transfer
+            the files to the node.
+
+        #. Install scylla-jmx package.
+
+            .. code-block:: console
+    
+               sudo apt install -y ./scylla-jmx_<version>_all.deb
+
+
+   .. group-tab:: Centos/RHEL
+
+        #. Download .rpm package from scylla-jmx page.
+
+            Access to https://github.com/scylladb/scylla-jmx, select latest
+            release from "releases", download a file end with ".rpm".
+
+        #. (Optional) Transfer the downloaded package to the install node.
+
+            If the pc from which you downloaded the package is different from
+            the node where you install scylladb, you will need to transfer
+            the files to the node.
+
+        #. Install scylla-jmx package.
+
+            .. code-block:: console
+    
+               sudo yum install -y ./scylla-jmx-<version>.noarch.rpm
+
+
+   .. group-tab:: Install without root privileges
+
+        #. Download .tar.gz package from scylla-jmx page.
+
+            Access to https://github.com/scylladb/scylla-jmx, select latest
+            release from "releases", download a file end with ".tar.gz".
+
+        #. (Optional) Transfer the downloaded package to the install node.
+
+            If the pc from which you downloaded the package is different from
+            the node where you install scylladb, you will need to transfer
+            the files to the node.
+
+        #. Install scylla-jmx package.
+
+            .. code:: console
+    
+                tar xpf scylla-jmx-<version>.noarch.tar.gz
+                cd scylla-jmx
+                ./install.sh --nonroot
+
+Next Steps
+-----------
+
+* :doc:`Configure ScyllaDB </getting-started/system-configuration>`
+* Manage your clusters with `ScyllaDB Manager <https://manager.docs.scylladb.com/>`_
+* Monitor your cluster and data with `ScyllaDB Monitoring <https://monitoring.docs.scylladb.com/>`_
+* Get familiar with ScyllaDB’s :doc:`command line reference guide </operating-scylla/nodetool>`.
+* Learn about ScyllaDB at `ScyllaDB University <https://university.scylladb.com/>`_

docs/getting-started/installation-common/scylla-web-installer.rst

@@ -36,8 +36,11 @@ release versions, run:
   curl -sSf get.scylladb.com/server | sudo bash -s -- --list-active-releases
 
 
-To install a non-default version, run the command with the ``--scylla-version``
-option to specify the version you want to install.
+Versions 2025.1 and Later
+==============================
+
+Run the command with the ``--scylla-version`` option to specify the version
+you want to install.
 
 **Example**
 
@@ -47,4 +50,20 @@ option to specify the version you want to install.
   curl -sSf get.scylladb.com/server | sudo bash -s -- --scylla-version |CURRENT_VERSION|
 
 
+Versions Earlier than 2025.1
+================================
+
+To install a supported version of *ScyllaDB Enterprise*, run the command with:
+
+* ``--scylla-product scylla-enterprise`` to specify that you want to install
+  ScyllaDB Entrprise.
+* ``--scylla-version`` to specify the version you want to install.
+
+For example:
+
+.. code:: console
+  
+  curl -sSf get.scylladb.com/server | sudo bash -s -- --scylla-product scylla-enterprise --scylla-version 2024.1
+
+
 .. include:: /getting-started/_common/setup-after-install.rst

docs/getting-started/installation-common/unified-installer.rst

@@ -14,35 +14,44 @@ Prerequisites
 Ensure your platform is supported by the ScyllaDB version you want to install. 
 See :doc:`OS Support </getting-started/os-support>` for information about supported Linux distributions and versions.
 
+Note that if you're on CentOS 7, only root offline installation is supported.
+
 Download and Install
 -----------------------
 
 #. Download the latest tar.gz file for ScyllaDB version (x86 or ARM) from ``https://downloads.scylladb.com/downloads/scylla/relocatable/scylladb-<version>/``.
 
-   **Example** for version 2025.1:
-   
-   - Go to https://downloads.scylladb.com/downloads/scylla/relocatable/scylladb-2025.1/
-   - Download the ``scylla-unified`` file for the patch version you want to
-     install. For example, to install 2025.1.9 (x86), download
-     ``scylla-unified-2025.1.9-0.20251010.6c539463bbda.x86_64.tar.gz``.
+   Example for version 6.1: https://downloads.scylladb.com/downloads/scylla/relocatable/scylladb-6.1/
 
 #. Uncompress the downloaded package.
 
-   **Example** for version 2025.1.9 (x86) (downloaded in the previous step):
+   The following example shows the package for ScyllaDB 6.1.1 (x86):
 
-   .. code::
+   .. code:: console
 
-    tar xvfz scylla-unified-2025.1.9-0.20251010.6c539463bbda.x86_64.tar.gz
+    tar xvfz scylla-unified-6.1.1-0.20240814.8d90b817660a.x86_64.tar.gz
 
-#. (Root offline installation only) For root offline installation on Debian-like
-   systems, two additional packages, ``xfsprogs`` and ``mdadm``, should be
-   installed to be used in RAID setup.
+#. Install OpenJDK 8 or 11.
+
+   The following example shows Java installation on a CentOS-like system:
+
+   .. code:: console
+    
+    sudo yum install -y java-11-openjdk-headless
+
+   For root offline installation on Debian-like systems, two additional packages, ``xfsprogs`` 
+   and ``mdadm``, should be installed to be used in RAID setup.
 
 #. Install ScyllaDB as a user with non-root privileges:
 
    .. code:: console
 
-    ./install.sh --nonroot
+    ./install.sh --nonroot --python3 ~/scylladb/python3/bin/python3
+
+#. (Optional) Install scylla-jmx
+
+    scylla-jmx is an optional package and is not installed by default.
+    If you need JMX server, see :doc:`Install scylla-jmx Package </getting-started/installation-common/install-jmx>`.
 
 Configure and Run ScyllaDB
 ----------------------------
@@ -72,14 +81,19 @@ Run nodetool:
 
 .. code:: console
 
-    ~/scylladb/bin/nodetool nodetool status
+    ~/scylladb/share/cassandra/bin/nodetool status
 
 Run cqlsh:
 
 .. code:: console
 
-    ~/scylladb/bin/cqlsh 
+    ~/scylladb/share/cassandra/bin/cqlsh 
 
+Run cassandra-stress:
+
+.. code:: console
+
+    ~/scylladb/share/cassandra/bin/cassandra-stress write
 
 .. note::
 
@@ -110,7 +124,7 @@ Nonroot install
 
     ./install.sh --upgrade --nonroot
 
-.. note:: The installation script does not upgrade scylla-tools. You will have to upgrade them separately. 
+.. note:: The installation script does not upgrade scylla-jmx and scylla-tools. You will have to upgrade them separately. 
 
 Uninstall
 ===========
@@ -140,4 +154,4 @@ Next Steps
 * Manage your clusters with `ScyllaDB Manager <https://manager.docs.scylladb.com/>`_
 * Monitor your cluster and data with `ScyllaDB Monitoring <https://monitoring.docs.scylladb.com/>`_
 * Get familiar with ScyllaDB’s :doc:`command line reference guide </operating-scylla/nodetool>`.
-* Learn about ScyllaDB at `ScyllaDB University <https://university.scylladb.com/>`_
+* Learn about ScyllaDB at `ScyllaDB University <https://university.scylladb.com/>`_

docs/operating-scylla/admin-tools/scylla-sstable.rst

@@ -601,11 +601,7 @@ Scrub has several modes:
 * **segregate** - Fixes partition/row/mutation-fragment out-of-order errors by segregating the output into as many SStables as required so that the content of each output SStable is properly ordered.
 * **validate** - Validates the content of the SStable, reporting any corruptions found. Writes no output SStables. In this mode, scrub has the same outcome as the `validate operation <scylla-sstable-validate-operation_>`_ - and the validate operation is recommended over scrub.
 
-Output SStables are written to the directory specified via ``--output-directory``. They will be written with the ``BIG`` format and the highest supported SStable format, with generations chosen by scylla-sstable. Generations are chosen such
-that they are unique among the SStables written by the current scrub.
-
-The output directory must be empty; otherwise, scylla-sstable will abort scrub. You can allow writing to a non-empty directory by setting the ``--unsafe-accept-nonempty-output-dir`` command line flag.
-Note that scrub will be aborted if an SStable cannot be written because its generation clashes with a pre-existing SStable in the output directory.
+Output SStables are written to the directory specified via ``--output-dir``. They will be written with the ``BIG`` format and the highest supported SStable format, with random generation.
 
 validate-checksums
 ^^^^^^^^^^^^^^^^^^
@@ -870,7 +866,7 @@ The SSTable version to be used can be overridden with the ``--version`` flag, al
 SSTables which are already on the designated version are skipped. To force rewriting *all* SSTables, use the ``--all`` flag. 
 
 Output SSTables are written to the path provided by the ``--output-dir`` flag, or to the current directory if not specified.
-This directory is expected to exist and be empty. If not empty the tool will refuse to run. This can be overridden with the ``--unsafe-accept-nonempty-output-dir`` flag.
+This directory is expected to exist.
 
 It is strongly recommended to use the system schema tables as the schema source for this command, see the :ref:`schema options <scylla-sstable-schema>` for more details.
 A schema which is good enough to read the SSTable and dump its content, may not be good enough to write its content back verbatim.
@@ -882,6 +878,25 @@ But even an altered schema which changed only the table options can lead to data
 
 The mapping of input SSTables to output SSTables is printed to ``stdout``.
 
+filter
+^^^^^^
+
+Filter the SSTable(s), including/excluding specified partitions.
+
+Similar to ``scylla sstable dump-data --partition|--partition-file``, with some notable differences:
+
+* Instead of dumping the content to stdout, the filtered content is written back to SSTable(s) on disk.
+* Also supports negative filters (keep all partitions except the those specified).
+
+The partition list can be provided either via the ``--partition`` command line argument, or via a file path passed to the the ``--partitions-file`` argument. The file should contain one partition key per line.
+Partition keys should be provided in the hex format, as produced by `scylla types serialize </operating-scylla/admin-tools/scylla-types/>`_.
+
+With ``--include``, only the specified partitions are kept from the input SSTable(s). With ``--exclude``, the specified partitions are discarded and won't be written to the output SSTable(s).
+It is possible that certain input SSTable(s) won't have any content left after the filtering. These input SSTable(s) will not have a matching output SSTable.
+
+By default, each input sstable is filtered individually. Use ``--merge`` to filter the combined content of all input sstables, producing a single output SSTable.
+
+Output sstables use the latest supported sstable format (can be changed with ``--sstable-version``).
 
 Examples
 --------

docs/operating-scylla/nodetool-commands/rebuild.rst

@@ -25,8 +25,4 @@ For Example:
 
    nodetool rebuild <source-dc-name>
 
-``nodetool rebuild`` command works only for vnode keyspaces. For tablet keyspaces, use ``nodetool cluster repair`` instead.
-
-See :doc:`Data Distribution with Tablets </architecture/tablets/>`.
-
 .. include:: nodetool-index.rst

docs/operating-scylla/procedures/cluster-management/add-dc-to-existing-dc.rst

@@ -155,6 +155,7 @@ Add New DC
       UN   54.235.9.159    109.75 KB       256     ?               39798227-9f6f-4868-8193-08570856c09a    RACK1
       UN   54.146.228.25   128.33 KB       256     ?               7a4957a1-9590-4434-9746-9c8a6f796a0c    RACK1
 
+.. TODO possibly provide additional information WRT how ALTER works with tablets
 
 #. When all nodes are up and running ``ALTER`` the following Keyspaces in the new nodes:
 
@@ -170,68 +171,26 @@ Add New DC
 
       DESCRIBE KEYSPACE mykeyspace;
 
-      CREATE KEYSPACE mykeyspace WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3};
+      CREATE KEYSPACE mykeyspace WITH replication = { 'class' : 'NetworkTopologyStrategy', '<exiting_dc>' : 3};
 
    ALTER Command
 
    .. code-block:: cql
 
-      ALTER KEYSPACE mykeyspace WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 3};
-      ALTER KEYSPACE system_distributed WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 3};
-      ALTER KEYSPACE system_traces WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 3};
+      ALTER KEYSPACE mykeyspace WITH replication = { 'class' : 'NetworkTopologyStrategy', '<exiting_dc>' : 3, <new_dc> : 3};
+      ALTER KEYSPACE system_distributed WITH replication = { 'class' : 'NetworkTopologyStrategy', '<exiting_dc>' : 3, <new_dc> : 3};
+      ALTER KEYSPACE system_traces WITH replication = { 'class' : 'NetworkTopologyStrategy', '<exiting_dc>' : 3, <new_dc> : 3};
 
    After
 
    .. code-block:: cql
 
       DESCRIBE KEYSPACE mykeyspace;
-      CREATE KEYSPACE mykeyspace WITH REPLICATION = {'class': 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 3};
-      CREATE KEYSPACE system_distributed WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 3};
-      CREATE KEYSPACE system_traces WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 3};
+      CREATE KEYSPACE mykeyspace WITH REPLICATION = {'class’: 'NetworkTopologyStrategy', <exiting_dc>:3, <new_dc>: 3};
+      CREATE KEYSPACE system_distributed WITH replication = { 'class' : 'NetworkTopologyStrategy', '<exiting_dc>' : 3, <new_dc> : 3};
+      CREATE KEYSPACE system_traces WITH replication = { 'class' : 'NetworkTopologyStrategy', '<exiting_dc>' : 3, <new_dc> : 3};
 
-   For tablet keyspaces, update the replication factor one by one:
-
-   .. code-block:: cql
-
-      DESCRIBE KEYSPACE mykeyspace2;
-
-      CREATE KEYSPACE mykeyspace2 WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3} AND tablets = { 'enabled': true };
-
-   .. code-block:: cql
-
-      ALTER KEYSPACE mykeyspace2 WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 1} AND tablets = { 'enabled': true };
-      ALTER KEYSPACE mykeyspace2 WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 2} AND tablets = { 'enabled': true };
-      ALTER KEYSPACE mykeyspace2 WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : 3, '<new_dc>' : 3} AND tablets = { 'enabled': true };
-
-   .. note::
-         If ``rf_rack_valid_keyspaces`` option is set, a tablet keyspace needs to use rack list replication factor, so that a new DC (rack) can be added. See :ref:`the conversion procedure <conversion-to-rack-list-rf>`. In this case, to add a datacenter:
-
-         Before
-
-         .. code-block:: cql
-
-            DESCRIBE KEYSPACE mykeyspace3;
-
-            CREATE KEYSPACE mykeyspace3 WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : ['<existing_rack1>', '<existing_rack2>', '<existing_rack3>']} AND tablets = { 'enabled': true };
-
-         Add all the nodes to the new datacenter and then alter the keyspace one by one:
-
-         .. code-block:: cql
-
-            ALTER KEYSPACE mykeyspace3 WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : ['<existing_rack1>', '<existing_rack2>', '<existing_rack3>'], '<new_dc>' : ['<new_rack1>']} AND tablets = { 'enabled': true };
-            ALTER KEYSPACE mykeyspace3 WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : ['<existing_rack1>', '<existing_rack2>', '<existing_rack3>'], '<new_dc>' : ['<new_rack1>', '<new_rack2>']} AND tablets = { 'enabled': true };
-            ALTER KEYSPACE mykeyspace3 WITH replication = { 'class' : 'NetworkTopologyStrategy', '<existing_dc>' : ['<existing_rack1>', '<existing_rack2>', '<existing_rack3>'], '<new_dc>' : ['<new_rack1>', '<new_rack2>', '<new_rack3>']} AND tablets = { 'enabled': true };
-
-         After
-
-         .. code-block:: cql
-
-            DESCRIBE KEYSPACE mykeyspace3;
-            CREATE KEYSPACE mykeyspace3 WITH REPLICATION = {'class': 'NetworkTopologyStrategy', '<existing_dc>' : ['<existing_rack1>', '<existing_rack2>', '<existing_rack3>'], '<new_dc>' : ['<new_rack1>', '<new_rack2>', '<new_rack3>']} AND tablets = { 'enabled': true };
-
-         Consider :ref:`upgrading rf_rack_valid_keyspaces option to enforce_rack_list option <keyspace-rf-rack-valid-to-enforce-rack-list>` to ensure all tablet keyspaces use rack lists.
-
-#. If any vnode keyspace was altered, run ``nodetool rebuild`` on each node in the new datacenter, specifying the existing datacenter name in the rebuild command.
+#. Run ``nodetool rebuild`` on each node in the new datacenter, specify the existing datacenter name in the rebuild command.
 
    For example:
 
@@ -239,7 +198,7 @@ Add New DC
 
    The rebuild ensures that the new nodes that were just added to the cluster will recognize the existing datacenters in the cluster.
 
-#. If any vnode keyspace was altered, run a full cluster repair, using :doc:`nodetool repair -pr </operating-scylla/nodetool-commands/repair>` on each node, or using `ScyllaDB Manager ad-hoc repair <https://manager.docs.scylladb.com/stable/repair>`_
+#. Run a full cluster repair, using :doc:`nodetool repair -pr </operating-scylla/nodetool-commands/repair>` on each node, or using `ScyllaDB Manager ad-hoc repair <https://manager.docs.scylladb.com/stable/repair>`_
 
 #. If you are using ScyllaDB Monitoring, update the `monitoring stack <https://monitoring.docs.scylladb.com/stable/install/monitoring_stack.html#configure-scylla-nodes-from-files>`_ to monitor it. If you are using ScyllaDB Manager, make sure you install the `Manager Agent <https://manager.docs.scylladb.com/stable/install-scylla-manager-agent.html>`_ and Manager can access the new DC.
 

docs/operating-scylla/procedures/cluster-management/cluster-platform-migration.rst

@@ -1,492 +0,0 @@
-=================================================
-Cluster Platform Migration Using Node Cycling
-=================================================
-
-This procedure describes how to migrate a ScyllaDB cluster to new instance types
-using the add-and-replace approach, which is commonly used for:
-
-* Migrating from one CPU architecture to another (e.g., x86_64 to ARM/Graviton)
-* Upgrading to newer instance types with better performance
-* Changing instance families within the same cloud provider
-
-The add-and-replace approach maintains data replication throughout the migration
-and ensures zero downtime for client applications.
-
-.. note::
-
-   This procedure does **not** change the ScyllaDB software version. All nodes
-   (both existing and new) must run the same ScyllaDB version. For software
-   version upgrades, see :doc:`Upgrade </upgrade/index>`.
-
-Overview
---------
-
-The add-and-replace migration follows these steps:
-
-#. Add new nodes (on target instance type) to the existing cluster
-#. Wait for data to stream to the new nodes
-#. Decommission old nodes (on source instance type)
-
-This approach keeps the cluster operational throughout the migration while
-maintaining the configured replication factor.
-
-Key characteristics
-===================
-
-* **Zero downtime**: Client applications continue to operate during migration
-* **Data safety**: Replication factor is maintained throughout the process
-* **Flexible**: Works with both vnodes and tablets-enabled clusters
-* **Multi-DC support**: Can migrate nodes across multiple datacenters
-
-.. warning::
-
-   Ensure your cluster has sufficient capacity during the migration. At the peak
-   of the process, your cluster will temporarily have double the number of nodes.
-
-Prerequisites
--------------
-
-Check cluster health
-====================
-
-Before starting the migration, verify that your cluster is healthy:
-
-#. Check that all nodes are in Up Normal (UN) status:
-
-   .. code-block:: shell
-
-      nodetool status
-
-   All nodes should show ``UN`` status. Do not proceed if any nodes are down.
-
-#. Ensure no streaming or repair operations are in progress:
-
-   .. code-block:: shell
-
-      nodetool netstats
-      nodetool compactionstats
-
-Plan the migration
-==================
-
-Before provisioning new instances, plan the following:
-
-**Instance type mapping**: Identify the source and target instance types.
-If your cluster uses vnodes (not tablets), consider that mismatched shard
-counts between source and target instance types can cause slower repairs.
-With tablets enabled, shard count mismatch is fully supported.
-
-**Rack assignment planning**: Each new node must be assigned to the same rack
-as the node it will replace. This maintains rack-aware topology for:
-
-* Rack-aware replication (NetworkTopologyStrategy)
-* Proper data distribution across failure domains
-* Minimizing data movement during decommission
-
-Example mapping for a 3-node cluster:
-
-.. code-block:: none
-
-   Source nodes (to be decommissioned):     Target nodes (to be added):
-   192.168.1.10 - RACK0                 →   192.168.2.10 - RACK0
-   192.168.1.11 - RACK1                 →   192.168.2.11 - RACK1
-   192.168.1.12 - RACK2                 →   192.168.2.12 - RACK2
-
-Create a backup
-===============
-
-Back up the data before starting the migration. One of the following
-methods can be used:
-
-* **ScyllaDB Manager** (recommended): Use ScyllaDB Manager to perform a
-  cluster-wide backup. See the
-  `ScyllaDB Manager documentation <https://manager.docs.scylladb.com/stable/backup/>`_
-  for details.
-
-* **Snapshots**: On each node in the cluster, create a snapshot:
-
-  .. code-block:: shell
-
-     nodetool snapshot -t pre_migration_backup
-     nodetool listsnapshots
-
-  .. note::
-
-     Snapshots are local to each node and do not protect against node or disk
-     failure. For full disaster recovery, use ScyllaDB Manager backup.
-
-
-Procedure
----------
-
-Adding new nodes
-================
-
-#. Provision new instances with the target instance type. Ensure:
-
-   * The same ScyllaDB version as existing nodes
-   * Same network configuration and security groups
-   * Appropriate storage configuration
-
-#. On each new node, configure ``/etc/scylla/scylla.yaml`` to join the existing
-   cluster:
-
-   * **cluster_name**: Must match the existing cluster name
-   * **seeds**: IP address of an existing node in the cluster (used to discover cluster topology on join)
-   * **endpoint_snitch**: Must match the existing cluster configuration
-   * **listen_address**: IP address of the new node
-   * **rpc_address**: IP address of the new node
-
-   All other cluster-wide settings (tablets configuration, encryption settings,
-   experimental features, etc.) must match the existing nodes.
-
-   .. caution::
-
-      Make sure that the ScyllaDB version on the new node is identical to the
-      version on the other nodes in the cluster. Running nodes with different
-      versions is not supported.
-
-#. If using ``GossipingPropertyFileSnitch``, configure
-   ``/etc/scylla/cassandra-rackdc.properties`` with the correct datacenter
-   and rack assignment for this node:
-
-   .. code-block:: none
-
-      dc = <datacenter-name>
-      rack = <rack-name>
-      prefer_local = true
-
-   .. warning::
-
-      Each node must have the correct rack assignment. Using the same rack for
-      all new nodes breaks rack-aware replication topology.
-
-#. Start ScyllaDB on the new node:
-
-   .. code-block:: shell
-
-      sudo systemctl start scylla-server
-
-   For Docker deployments:
-
-   .. code-block:: shell
-
-      docker exec -it <container-name> supervisorctl start scylla
-
-#. Monitor the bootstrap process from an existing node:
-
-   .. code-block:: shell
-
-      nodetool status
-
-   The new node will appear with ``UJ`` (Up, Joining) status while streaming
-   data from existing nodes. Wait until it transitions to ``UN`` (Up, Normal).
-
-   **Example output during bootstrap:**
-
-   .. code-block:: shell
-
-      Datacenter: dc1
-      Status=Up/Down
-      State=Normal/Leaving/Joining/Moving
-      --  Address        Load       Tokens  Owns   Host ID                               Rack
-      UN  192.168.1.10   500 MB     256     33.3%  8d5ed9f4-7764-4dbd-bad8-43fddce94b7c  RACK0
-      UN  192.168.1.11   500 MB     256     33.3%  125ed9f4-7777-1dbn-mac8-43fddce9123e  RACK1
-      UN  192.168.1.12   500 MB     256     33.3%  675ed9f4-6564-6dbd-can8-43fddce952gy  RACK2
-      UJ  192.168.2.10   250 MB     256     ?      a1b2c3d4-5678-90ab-cdef-112233445566  RACK0
-
-   **Example output after bootstrap completes:**
-
-   .. code-block:: shell
-
-      Datacenter: dc1
-      Status=Up/Down
-      State=Normal/Leaving/Joining/Moving
-      --  Address        Load       Tokens  Owns   Host ID                               Rack
-      UN  192.168.1.10   400 MB     256     25.0%  8d5ed9f4-7764-4dbd-bad8-43fddce94b7c  RACK0
-      UN  192.168.1.11   400 MB     256     25.0%  125ed9f4-7777-1dbn-mac8-43fddce9123e  RACK1
-      UN  192.168.1.12   400 MB     256     25.0%  675ed9f4-6564-6dbd-can8-43fddce952gy  RACK2
-      UN  192.168.2.10   400 MB     256     25.0%  a1b2c3d4-5678-90ab-cdef-112233445566  RACK0
-
-#. For tablets-enabled clusters, wait for tablet load balancing to complete.
-   After the node reaches ``UN`` status, verify no streaming is in progress:
-
-   .. code-block:: shell
-
-      nodetool netstats
-
-   Wait until output shows "Not sending any streams" and no active receiving streams.
-
-#. Repeat steps 1-6 for each new node to be added.
-
-.. note::
-
-   You can add multiple nodes in parallel if they are in different datacenters.
-   Within a single datacenter, add nodes one at a time for best results.
-
-
-Updating seed node configuration
-================================
-
-If any of your original nodes are configured as seed nodes, you must update
-the seed configuration before decommissioning them.
-
-#. Check the current seed configuration on any node:
-
-   .. code-block:: shell
-
-      grep -A 4 "seed_provider" /etc/scylla/scylla.yaml
-
-#. If the seeds include nodes you plan to decommission, update ``scylla.yaml``
-   on **all new nodes** to use the new node IPs as seeds:
-
-   .. code-block:: yaml
-
-      seed_provider:
-        - class_name: org.apache.cassandra.locator.SimpleSeedProvider
-          parameters:
-            - seeds: "192.168.2.10,192.168.2.11,192.168.2.12"
-
-   .. note::
-
-      Updating seed configuration on the **old nodes** (that will be
-      decommissioned) is optional. Seeds are only used during node startup
-      to discover the cluster. If you don't plan to restart the old nodes
-      before decommissioning them, their seed configuration doesn't matter.
-      However, updating all nodes is recommended for safety in case an old
-      node unexpectedly restarts during the migration.
-
-#. Restart ScyllaDB on each new node (one at a time) to apply the new seed
-   configuration:
-
-   .. code-block:: shell
-
-      sudo systemctl restart scylla-server
-
-   Wait for the node to fully start before restarting the next node.
-
-#. After restarting the new nodes, verify the cluster is healthy:
-
-   .. code-block:: shell
-
-      nodetool status
-      nodetool describecluster
-
-.. warning::
-
-   Complete this seed list update on **all new nodes** before decommissioning
-   any old nodes. This ensures the new nodes can reform the cluster after
-   the old nodes are removed.
-
-
-Decommissioning old nodes
-=========================
-
-After all new nodes are added and healthy, decommission the old nodes one
-at a time.
-
-#. Verify all nodes are healthy before starting decommission:
-
-   .. code-block:: shell
-
-      nodetool status
-
-   All nodes should show ``UN`` status.
-
-#. On the node to be decommissioned, run:
-
-   .. code-block:: shell
-
-      nodetool decommission
-
-   This command blocks until the decommission is complete. The node will
-   stream its data to the remaining nodes.
-
-#. Monitor the decommission progress from another node:
-
-   .. code-block:: shell
-
-      nodetool status
-
-   The decommissioning node will transition from ``UN`` → ``UL`` (Up, Leaving)
-   → removed from the cluster.
-
-   You can also monitor streaming progress:
-
-   .. code-block:: shell
-
-      nodetool netstats
-
-#. After decommission completes, verify the node is no longer in the cluster:
-
-   .. code-block:: shell
-
-      nodetool status
-
-   The decommissioned node should no longer appear in the output.
-
-#. Run ``nodetool cleanup`` on the remaining nodes to remove data that
-   no longer belongs to them after the topology change:
-
-   .. code-block:: shell
-
-      nodetool cleanup
-
-   .. note::
-
-      ``nodetool cleanup`` can be resource-intensive. Run it on one node at a
-      time during low-traffic periods.
-
-#. Wait for the cluster to stabilize before decommissioning the next node.
-   Ensure no streaming operations are in progress.
-
-#. Repeat steps 1-7 for each old node to be decommissioned.
-
-
-Post-migration verification
----------------------------
-
-After all old nodes are decommissioned, verify the migration was successful.
-
-Verify cluster topology
-=======================
-
-.. code-block:: shell
-
-   nodetool status
-
-Confirm:
-
-* All nodes show ``UN`` (Up, Normal) status
-* Only the new instance type nodes are present
-* Nodes are balanced across racks
-
-Verify schema agreement
-=======================
-
-.. code-block:: shell
-
-   nodetool describecluster
-
-All nodes should report the same schema version.
-
-Verify data connectivity
-========================
-
-Connect to the cluster and run a test query:
-
-.. code-block:: shell
-
-   cqlsh <node-ip> -e "SELECT count(*) FROM system_schema.keyspaces;"
-
-.. note::
-
-   If ScyllaDB is configured with ``listen_interface``, you must use the
-   node's interface IP address (not localhost) for cqlsh connections.
-
-Verify ScyllaDB version
-=======================
-
-Confirm all nodes are running the same ScyllaDB version:
-
-.. code-block:: shell
-
-   scylla --version
-
-Verify data integrity (optional)
-================================
-
-Run data validation on each keyspace to verify sstable integrity:
-
-.. code-block:: shell
-
-   nodetool scrub --mode=VALIDATE <keyspace_name>
-
-Rollback
---------
-
-If issues occur during the migration, you can roll back by reversing the
-procedure.
-
-During add phase
-================
-
-If a new node fails to bootstrap:
-
-#. Stop ScyllaDB on the new node:
-
-   .. code-block:: shell
-
-      sudo systemctl stop scylla-server
-
-#. From an existing node, remove the failed node:
-
-   .. code-block:: shell
-
-      nodetool removenode <host-id-of-failed-node>
-
-During decommission phase
-=========================
-
-If a decommission operation gets stuck:
-
-#. If the node is still reachable, try stopping and restarting ScyllaDB
-#. If the node is unresponsive, from another node:
-
-   .. code-block:: shell
-
-      nodetool removenode <host-id>
-
-   See :doc:`Remove a Node from a ScyllaDB Cluster </operating-scylla/procedures/cluster-management/remove-node>`
-   for more details.
-
-Full rollback
-=============
-
-To roll back after the migration is complete (all nodes on new instance type),
-apply the same add-and-replace procedure in reverse:
-
-#. Add new nodes on the original instance type
-#. Wait for data streaming to complete
-#. Decommission the nodes on the new instance type
-
-
-Troubleshooting
----------------
-
-Node stuck in Joining (UJ) state
-================================
-
-If a new node remains in ``UJ`` state for an extended period:
-
-* Check ScyllaDB logs for streaming errors: ``journalctl -u scylla-server``
-* Verify network connectivity between nodes
-* Ensure sufficient disk space on all nodes
-* Check for any ongoing operations that may be blocking
-
-Decommission taking too long
-============================
-
-Decommission duration depends on data size. If it appears stuck:
-
-* Check streaming progress: ``nodetool netstats``
-* Look for errors in ScyllaDB logs
-* Verify network bandwidth between nodes
-
-Schema disagreement
-===================
-
-If nodes report different schema versions:
-
-* Wait a few minutes for schema to propagate
-* If disagreement persists, restart the nodes one by one
-* Run ``nodetool describecluster`` to verify agreement
-
-
-Additional resources
---------------------
-
-* :doc:`Adding a New Node Into an Existing ScyllaDB Cluster </operating-scylla/procedures/cluster-management/add-node-to-cluster>`
-* :doc:`Remove a Node from a ScyllaDB Cluster </operating-scylla/procedures/cluster-management/remove-node>`
-* :doc:`Replace a Running Node in a ScyllaDB Cluster </operating-scylla/procedures/cluster-management/replace-running-node>`
-* :doc:`Upgrade </upgrade/index>`

docs/operating-scylla/procedures/cluster-management/decommissioning-data-center.rst

@@ -40,14 +40,12 @@ Prerequisites
 Procedure
 ---------
 
-#. If there are vnode keyspaces in this DC, run the ``nodetool repair -pr`` command on each node in the data-center that is going to be decommissioned. This will verify that all the data is in sync between the decommissioned data-center and the other data-centers in the cluster.
+#. Run the ``nodetool repair -pr`` command on each node in the data-center that is going to be decommissioned. This will verify that all the data is in sync between the decommissioned data-center and the other data-centers in the cluster.
 
    For example:
 
    If the ASIA-DC cluster is to be removed, then, run the ``nodetool repair -pr`` command on all the nodes in the ASIA-DC
 
-#. If there are tablet keyspaces in this DC, run the ``nodetool cluster repair`` on an arbitrary node. The reason for running repair is to ensure that any updates stored only on the about-to-be-decommissioned replicas are propagated to the other replicas, before the replicas on the decommissioned datacenter are dropped.
-
 #. ALTER every cluster KEYSPACE, so that the keyspaces will no longer replicate data to the decommissioned data-center.
 
    For example:
@@ -75,33 +73,6 @@ Procedure
 
       cqlsh> ALTER KEYSPACE nba WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'US-DC' : 3, 'ASIA-DC' : 0, 'EUROPE-DC' : 3};
 
-   For tablet keyspaces, update the replication factor one by one:
-
-   .. code-block:: shell
-
-      cqlsh> DESCRIBE nba2
-      cqlsh> CREATE KEYSPACE nba2 WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'US-DC' : 3, 'ASIA-DC' : 2, 'EUROPE-DC' : 3} AND tablets = { 'enabled': true };
-
-   .. code-block:: shell
-
-      cqlsh> ALTER KEYSPACE nba2 WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'US-DC' : 3, 'ASIA-DC' : 1, 'EUROPE-DC' : 3} AND tablets = { 'enabled': true };
-      cqlsh> ALTER KEYSPACE nba2 WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'US-DC' : 3, 'ASIA-DC' : 0, 'EUROPE-DC' : 3} AND tablets = { 'enabled': true };
-
-   .. note::
-         If ``rf_rack_valid_keyspaces`` option is set, a tablet keyspace needs to use rack list replication factor, so that the DC can be removed. See :ref:`the conversion procedure <conversion-to-rack-list-rf>`. In this case, to remove a datacenter:
-
-         .. code-block:: shell
-
-            cqlsh> DESCRIBE nba3
-            cqlsh> CREATE KEYSPACE nba3 WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'US-DC' : ['RAC1', 'RAC2', 'RAC3'], 'ASIA-DC' : ['RAC4', 'RAC5'], 'EUROPE-DC' : ['RAC6', 'RAC7', 'RAC8']} AND tablets = { 'enabled': true };
-
-         .. code-block:: shell
-
-            cqlsh> ALTER KEYSPACE nba3 WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'US-DC' : ['RAC1', 'RAC2', 'RAC3'], 'ASIA-DC' : ['RAC4'], 'EUROPE-DC' : ['RAC6', 'RAC7', 'RAC8']} AND tablets = { 'enabled': true };
-            cqlsh> ALTER KEYSPACE nba3 WITH REPLICATION = {'class' : 'NetworkTopologyStrategy', 'US-DC' : ['RAC1', 'RAC2', 'RAC3'], 'ASIA-DC' : [], 'EUROPE-DC' : ['RAC6', 'RAC7', 'RAC8']} AND tablets = { 'enabled': true };
-
-         Consider :ref:`upgrading rf_rack_valid_keyspaces option to enforce_rack_list option <keyspace-rf-rack-valid-to-enforce-rack-list>` to ensure all tablet keyspaces use rack lists.
-
 #. Run :doc:`nodetool decommission </operating-scylla/nodetool-commands/decommission>` on every node in the data center that is to be removed.
    Refer to :doc:`Remove a Node from a ScyllaDB Cluster - Down Scale </operating-scylla/procedures/cluster-management/remove-node>` for further information.
 

docs/operating-scylla/procedures/cluster-management/index.rst

@@ -26,7 +26,6 @@ Cluster Management Procedures
    Safely Restart Your Cluster <safe-start>
    repair-based-node-operation
    Prevent Quorum Loss in Symmetrical Multi-DC Clusters <arbiter-dc>
-   Cluster Platform Migration <cluster-platform-migration>
 
 
 .. panel-box::
@@ -86,8 +85,6 @@ Cluster Management Procedures
 
   * :doc:`Preventing Quorum Loss in Symmetrical Multi-DC Clusters <arbiter-dc>`
 
-  * :doc:`Cluster Platform Migration Using Node Cycling </operating-scylla/procedures/cluster-management/cluster-platform-migration>`
-
 .. panel-box::
   :title: Topology Changes
   :id: "getting-started"

docs/operating-scylla/procedures/config-change/advanced-internode-compression.rst

@@ -57,11 +57,12 @@ To enable shared dictionaries:
     internode_compression_enable_advanced: true
     rpc_dict_training_when: when_leader
 
-.. note::
+.. warning:: Enabling shared dictionary training might leak unencrypted data to disk.
 
-   Some dictionary training data may be encrypted using storage-level encryption
-   (if enabled) instead of database-level encryption, meaning protection is
-   applied at the storage layer rather than within the database itself.
+             Trained dictionaries contain randomly chosen samples of data transferred between
+             nodes. The data samples are persisted in the Raft log, which is not encrypted.
+             As a result, some data from otherwise encrypted tables might be stored on disk
+             unencrypted.
 
 
 Reference

docs/operating-scylla/procedures/maintenance/repair.rst

@@ -61,9 +61,9 @@ See also
 Incremental Repair
 ------------------
 
-Built on top of :ref:`Row-level Repair <row-level-repair>` and :doc:`Tablets </architecture/tablets>`, Incremental Repair enables frequent and quick repairs. For more details, see :doc:`Incremental Repair </features/incremental-repair>`.
+Built on top of `Row-level Repair <row-level-repair_>`_ and `Tablets </architecture/tablets>`_, Incremental Repair enables frequent and quick repairs. For more details, see `Incremental Repair </features/incremental-repair>`_.
 
 Automatic Repair
 ----------------
 
-Built on top of :doc:`Incremental Repair </features/incremental-repair>`, :doc:`Automatic Repair </features/automatic-repair>` offers repair scheduling and execution directly in ScyllaDB, without external processes.
+Built on top of `Incremental Repair </features/incremental-repair>`_, `Automatic Repair </features/automatic-repair>`_ offers repair scheduling and execution directly in ScyllaDB, without external processes.

docs/upgrade/about-upgrade.rst

@@ -11,13 +11,9 @@ ScyllaDB. This means that:
 
 * You should follow the upgrade policy:
 
-   * Starting with version **2025.4**, upgrades can **skip minor versions** if:
-
-       * They remain within the same major version (for example, upgrading
-         directly from *2025.1 → 2025.4* is supported).
-       * You upgrade to the next major version (for example, upgrading
-         directly from *2025.3 → 2026.1* is supported).
-
+   * Starting with version **2025.4**, upgrades can skip minor versions as long
+     as they remain within the same major version (for example, upgrading directly
+     from 2025.1 → 2025.4 is supported).
    * For versions **prior to 2025.4**, upgrades must be performed consecutively—
      each successive X.Y version must be installed in order, **without skipping
      any major or minor version** (for example, upgrading directly from 2025.1 → 2025.3

docs/upgrade/upgrade-guides/index.rst

@@ -4,7 +4,8 @@ Upgrade ScyllaDB
 
 .. toctree::
    
-   ScyllaDB 2025.x to ScyllaDB 2026.1 <upgrade-guide-from-2025.x-to-2026.1/index>
+   ScyllaDB 2025.x to ScyllaDB 2025.4 <upgrade-guide-from-2025.x-to-2025.4/index>
+   ScyllaDB 2025.4 Patch Upgrades <upgrade-guide-from-2025.4.x-to-2025.4.y>
    ScyllaDB Image <ami-upgrade>
 
 

docs/upgrade/upgrade-guides/upgrade-guide-from-2025.4.x-to-2025.4.y.rst

@@ -0,0 +1,266 @@
+.. |SCYLLA_NAME| replace:: ScyllaDB
+
+.. |SRC_VERSION| replace:: 2025.4.x
+.. |NEW_VERSION| replace:: 2025.4.y
+
+==========================================================================
+Upgrade - |SCYLLA_NAME| |SRC_VERSION| to |NEW_VERSION| (Patch Upgrades)
+==========================================================================
+
+This document describes a step-by-step procedure for upgrading from
+|SCYLLA_NAME| |SRC_VERSION|  to |SCYLLA_NAME| |NEW_VERSION| (where "y" is
+the latest available version), and rolling back to version |SRC_VERSION|
+if necessary.
+
+This guide covers upgrading ScyllaDB on Red Hat Enterprise Linux (RHEL),
+CentOS, Debian, and Ubuntu.
+See :doc:`OS Support by Platform and Version </getting-started/os-support>`
+for information about supported versions.
+
+It also applies to the ScyllaDB official image on EC2, GCP, or Azure.
+
+Upgrade Procedure
+=================
+
+.. note::
+   Apply the following procedure **serially** on each node. Do not move to the next
+   node before validating that the node is up and running the new version.
+
+A ScyllaDB upgrade is a rolling procedure that does **not** require a full cluster
+shutdown. For each of the nodes in the cluster, you will:
+
+#. Drain the node and back up the data.
+#. Backup configuration file.
+#. Stop ScyllaDB.
+#. Download and install new ScyllaDB packages.
+#. Start ScyllaDB.
+#. Validate that the upgrade was successful.
+
+**Before** upgrading, check which version you are running now using
+``scylla --version``. Note the current version in case you want to roll back
+the upgrade.
+
+**During** the rolling upgrade it is highly recommended:
+
+* Not to use new |NEW_VERSION| features.
+* Not to run administration functions, like repairs, refresh, rebuild or add
+  or remove nodes. See
+  `sctool <https://manager.docs.scylladb.com/stable/sctool/>`_ for suspending
+  ScyllaDB Manager's scheduled or running repairs.
+* Not to apply schema changes.
+
+Upgrade Steps
+=============
+
+Back up the data
+------------------------------
+
+Back up all the data to an external device. We recommend using
+`ScyllaDB Manager <https://manager.docs.scylladb.com/stable/backup/index.html>`_
+to create backups.
+
+Alternatively, you can use the ``nodetool snapshot`` command.
+For **each** node in the cluster, run the following:
+
+.. code:: sh
+
+   nodetool drain
+   nodetool snapshot
+
+Take note of the directory name that nodetool gives you, and copy all
+the directories with this name under ``/var/lib/scylla`` to a backup device.
+
+When the upgrade is completed on all nodes, remove the snapshot with the 
+``nodetool clearsnapshot -t <snapshot>`` command to prevent running out of
+space.
+
+Back up the configuration file
+------------------------------
+
+Back up the ``scylla.yaml`` configuration file and the ScyllaDB packages
+in case you need to roll back the upgrade.
+
+.. tabs::
+
+   .. group-tab:: Debian/Ubuntu
+
+      .. code:: sh
+
+         sudo cp -a /etc/scylla/scylla.yaml /etc/scylla/scylla.yaml.backup
+         sudo cp /etc/apt/sources.list.d/scylla.list ~/scylla.list-backup
+
+   .. group-tab:: RHEL/CentOS
+
+      .. code:: sh 
+         
+         sudo cp -a /etc/scylla/scylla.yaml /etc/scylla/scylla.yaml.backup
+         sudo cp /etc/yum.repos.d/scylla.repo ~/scylla.repo-backup
+
+Gracefully stop the node
+------------------------
+
+.. code:: sh
+
+   sudo service scylla-server stop
+
+Download and install the new release
+------------------------------------
+
+You don’t need to update the ScyllaDB DEB or RPM repo when you upgrade to
+a patch release.
+
+.. tabs::
+
+   .. group-tab:: Debian/Ubuntu
+
+        To install a patch version on Debian or Ubuntu, run:
+
+        .. code:: sh
+            
+            sudo apt-get clean all
+            sudo apt-get update
+            sudo apt-get dist-upgrade scylla
+
+        Answer ‘y’ to the first two questions.
+
+   .. group-tab:: RHEL/CentOS
+
+        To install a patch version on RHEL or CentOS, run:
+
+        .. code:: sh
+            
+            sudo yum clean all
+            sudo yum update scylla\* -y
+        
+   .. group-tab:: EC2/GCP/Azure Ubuntu Image
+
+        If you're using the ScyllaDB official image (recommended), see 
+        the **Debian/Ubuntu** tab for upgrade instructions.
+
+        If you're using your own image and have installed ScyllaDB packages for 
+        Ubuntu or Debian, you need to apply an extended upgrade procedure:
+
+        #. Install the new ScyllaDB version with the additional
+           ``scylla-machine-image`` package:
+
+            .. code-block:: console
+
+               sudo apt-get clean all
+               sudo apt-get update
+               sudo apt-get dist-upgrade scylla
+               sudo apt-get dist-upgrade scylla-machine-image
+        #. Run ``scylla_setup`` without ``running io_setup``.
+        #. Run ``sudo /opt/scylladb/scylla-machine-image/scylla_cloud_io_setup``.
+
+Start the node
+--------------
+
+.. code:: sh
+
+   sudo service start scylla-server
+
+Validate
+--------
+#. Check cluster status with ``nodetool status`` and make sure **all** nodes,
+   including the one you just upgraded, are in UN status.
+#. Use ``curl -X GET "http://localhost:10000/storage_service/scylla_release_version"``
+   to check the ScyllaDB version.
+#. Use ``journalctl _COMM=scylla`` to check there are no new errors in the log.
+#. Check again after 2 minutes to validate that no new issues are introduced.
+
+Once you are sure the node upgrade is successful, move to the next node in
+the cluster.
+
+Rollback Procedure
+==================
+
+The following procedure describes a rollback from ScyllaDB release
+|NEW_VERSION| to |SRC_VERSION|. Apply this procedure if an upgrade from
+|SRC_VERSION| to |NEW_VERSION| failed before completing on all nodes. 
+
+* Use this procedure only on nodes you upgraded to |NEW_VERSION|.
+* Execute the following commands one node at a time, moving to the next node only
+  after the rollback procedure is completed successfully.
+
+ScyllaDB rollback is a rolling procedure that does **not** require a full
+cluster shutdown. For each of the nodes to roll back to |SRC_VERSION|, you will:
+
+#. Drain the node and stop ScyllaDB.
+#. Downgrade to the previous release.
+#. Restore the configuration file.
+#. Restart ScyllaDB.
+#. Validate the rollback success.
+
+Rollback Steps
+==============
+
+Gracefully shutdown ScyllaDB
+-----------------------------
+
+.. code:: sh
+    
+   nodetool drain
+   sudo service stop scylla-server
+
+Downgrade to the previous release
+----------------------------------
+
+.. tabs::
+
+   .. group-tab:: Debian/Ubuntu
+
+        To downgrade to |SRC_VERSION| on Debian or Ubuntu, run:
+
+        .. code-block:: console
+            :substitutions:
+
+            sudo apt-get install scylla=|SRC_VERSION|\* scylla-server=|SRC_VERSION|\* scylla-tools=|SRC_VERSION|\* scylla-tools-core=|SRC_VERSION|\* scylla-kernel-conf=|SRC_VERSION|\* scylla-conf=|SRC_VERSION|\*
+        
+        Answer ‘y’ to the first two questions.
+
+   .. group-tab:: RHEL/CentOS
+
+        To downgrade to |SRC_VERSION| on RHEL or CentOS, run:
+
+        .. code-block:: console
+            :substitutions:
+
+            sudo yum downgrade scylla\*-|SRC_VERSION|-\* -y
+
+   .. group-tab:: EC2/GCP/Azure Ubuntu Image
+
+        If you’re using the ScyllaDB official image (recommended), see
+        the **Debian/Ubuntu** tab for upgrade instructions.
+
+        If you’re using your own image and have installed ScyllaDB packages for
+        Ubuntu or Debian, you need to additionally downgrade
+        the ``scylla-machine-image`` package.
+
+        .. code-block:: console
+            :substitutions:
+
+            sudo apt-get install scylla=|SRC_VERSION|\* scylla-server=|SRC_VERSION|\* scylla-tools=|SRC_VERSION|\* scylla-tools-core=|SRC_VERSION|\* scylla-kernel-conf=|SRC_VERSION|\* scylla-conf=|SRC_VERSION|\*
+            sudo apt-get install scylla-machine-image=|SRC_VERSION|\*
+        
+        Answer ‘y’ to the first two questions.
+
+
+Restore the configuration file
+------------------------------
+
+.. code:: sh
+   
+   sudo rm -rf /etc/scylla/scylla.yaml
+   sudo cp -a /etc/scylla/scylla.yaml.backup /etc/scylla/scylla.yaml
+
+Start the node
+--------------
+
+.. code:: sh
+
+   sudo service scylla-server start
+
+Validate
+--------
+Check upgrade instruction above for validation. Once you are sure the node
+rollback is successful, move to the next node in the cluster.

docs/upgrade/upgrade-guides/upgrade-guide-from-2025.x-to-2025.4/index.rst

@@ -0,0 +1,13 @@
+==========================================================
+Upgrade - ScyllaDB 2025.x to ScyllaDB 2025.4
+==========================================================
+
+.. toctree::
+   :maxdepth: 2
+   :hidden:
+
+   Upgrade ScyllaDB <upgrade-guide-from-2025.x-to-2025.4>
+   Metrics Update <metric-update-2025.x-to-2025.4>
+
+* :doc:`Upgrade from ScyllaDB 2025.x to ScyllaDB 2025.4 <upgrade-guide-from-2025.x-to-2025.4>`
+* :doc:`Metrics Update Between 2025.x and 2025.4 <metric-update-2025.x-to-2025.4>`

docs/upgrade/upgrade-guides/upgrade-guide-from-2025.x-to-2025.4/metric-update-2025.x-to-2025.4.rst

@@ -0,0 +1,68 @@
+.. |SRC_VERSION| replace:: 2025.x
+.. |NEW_VERSION| replace:: 2025.4
+.. |PRECEDING_VERSION| replace:: 2025.3
+
+================================================================
+Metrics Update Between |SRC_VERSION| and |NEW_VERSION|
+================================================================
+
+.. toctree::
+   :maxdepth: 2
+   :hidden:
+
+ScyllaDB |NEW_VERSION| Dashboards are available as part of the latest |mon_root|.
+
+
+New Metrics in |NEW_VERSION|
+--------------------------------------
+
+The following metrics are new in ScyllaDB |NEW_VERSION| compared to |PRECEDING_VERSION|.
+
+.. list-table::
+   :widths: 25 150
+   :header-rows: 1
+
+   * - Metric
+     - Description
+   * - scylla_database_total_view_updates_due_to_replica_count_mismatch	
+     - The total number of view updates for which there were more view replicas
+       than base replicas and we had to generate an extra view update because
+       the additional view replica wouldn't get paired with any base replica.
+       It should only increase during the Replication Factor (RF) change. It
+       should stop increasing shortly after finishing the RF change.
+   * - scylla_database_total_writes_rejected_due_to_out_of_space_prevention
+     - Counts write operations that were rejected due to disabled user tables
+       writes.
+   * - scylla_index_query_latencies
+     - Index query latencies.
+   * - scylla_reactor_aio_retries
+     - The total number of IOCB-s re-submitted via thread-pool.
+   * - scylla_reactor_io_threaded_fallbacks
+     - The total number of io-threaded-fallbacks operations.
+   * - scylla_repair_inc_sst_read_bytes
+     - The total number of bytes read from SStables for incremental repair
+       on this shard.
+   * - scylla_repair_inc_sst_skipped_bytes
+     - The total number of bytes skipped from SStables for incremental repair
+       on this shard.
+   * - scylla_repair_tablet_time_ms
+     - The time spent on tablet repair on this shard (in milliseconds).
+   * - scylla_s3_downloads_blocked_on_memory
+     - Counts the number of times the S3 client downloads were delayed due to
+       insufficient memory availability.
+   * - scylla_s3_memory_usage
+     - The total number of bytes consumed by the S3 client.
+   * - scylla_s3_total_read_prefetch_bytes
+     - The total number of bytes requested from object.
+   * - scylla_storage_proxy_replica_fenced_out_requests
+     - The number of requests that resulted in a stale_topology_exception.
+   * - scylla_vector_store_dns_refreshes	
+     - The number of DNS refreshes.
+
+New and Updated Metrics in Previous 2025.x Releases
+-------------------------------------------------------
+
+* `Metrics Update Between 2025.2 and 2025.3 <https://docs.scylladb.com/manual/branch-2025.3/upgrade/upgrade-guides/upgrade-guide-from-2025.2-to-2025.3/metric-update-2025.2-to-2025.3.html>`_
+* `Metrics Update Between 2025.1 and 2025.2 <https://docs.scylladb.com/manual/branch-2025.2/upgrade/upgrade-guides/upgrade-guide-from-2025.1-to-2025.2/metric-update-2025.1-to-2025.2.html>`_
+
+

docs/upgrade/upgrade-guides/upgrade-guide-from-2025.x-to-2025.4/upgrade-guide-from-2025.x-to-2025.4.rst

@@ -1,13 +1,13 @@
 .. |SCYLLA_NAME| replace:: ScyllaDB
 
 .. |SRC_VERSION| replace:: 2025.x
-.. |NEW_VERSION| replace:: 2026.1
+.. |NEW_VERSION| replace:: 2025.4
 
 .. |ROLLBACK| replace:: rollback
 .. _ROLLBACK: ./#rollback-procedure
 
-.. |SCYLLA_METRICS| replace:: ScyllaDB Metrics Update - ScyllaDB 2025.x to 2026.1
-.. _SCYLLA_METRICS: ../metric-update-2025.x-to-2026.1
+.. |SCYLLA_METRICS| replace:: ScyllaDB Metrics Update - ScyllaDB 2025.x to 2025.4
+.. _SCYLLA_METRICS: ../metric-update-2025.x-to-2025.4
 
 =======================================================================================
 Upgrade from |SCYLLA_NAME| |SRC_VERSION| to |SCYLLA_NAME| |NEW_VERSION|
@@ -17,12 +17,10 @@ This document describes a step-by-step procedure for upgrading from |SCYLLA_NAME
 to |SCYLLA_NAME| |NEW_VERSION| and rollback to version |SRC_VERSION| if necessary.
 
 This guide covers upgrading ScyllaDB on Red Hat Enterprise Linux (RHEL), CentOS, Debian, 
-and Ubuntu.
-See `OS Support by Platform and Version <https://docs.scylladb.com/stable/versioning/os-support-per-version.html>`_ 
-for information about supported versions. It also applies when using
-the ScyllaDB official image on EC2, GCP, or Azure.
+and Ubuntu. See :doc:`OS Support by Platform and Version </getting-started/os-support>` 
+for information about supported versions.
 
-See :doc:`About Upgrade </upgrade/about-upgrade/>` for the ScyllaDB upgrade policy.
+It also applies when using the ScyllaDB official image on EC2, GCP, or Azure.
 
 Before You Upgrade ScyllaDB
 ==============================
@@ -151,9 +149,8 @@ You should take note of the current version in case you want to |ROLLBACK|_ the
         #. Update the ScyllaDB deb repo to |NEW_VERSION|.
 
             .. code-block:: console
-               :substitutions:
 
-               sudo wget -O /etc/apt/sources.list.d/scylla.list https://downloads.scylladb.com/deb/debian/|UBUNTU_SCYLLADB_LIST|
+               sudo wget -O /etc/apt/sources.list.d/scylla.list https://downloads.scylladb.com/deb/debian/scylla-2025.4.list
 
         #. Install the new ScyllaDB version:
 
@@ -170,9 +167,8 @@ You should take note of the current version in case you want to |ROLLBACK|_ the
         #. Update the ScyllaDB rpm repo to |NEW_VERSION|.
 
             .. code-block:: console
-               :substitutions:
 
-               sudo curl -o /etc/yum.repos.d/scylla.repo -L https://downloads.scylladb.com/rpm/centos/|CENTOS_SCYLLADB_REPO|
+               sudo curl -o /etc/yum.repos.d/scylla.repo -L https://downloads.scylladb.com/rpm/centos/scylla-2025.4.repo
 
         #. Install the new ScyllaDB version:
 
@@ -202,6 +198,11 @@ You should take note of the current version in case you want to |ROLLBACK|_ the
       #. Run ``scylla_setup`` without ``running io_setup``.
       #. Run ``sudo /opt/scylladb/scylla-machine-image/scylla_cloud_io_setup``.
 
+
+If you need JMX server, see
+:doc:`Install scylla-jmx Package </getting-started/installation-common/install-jmx>`
+and get new version. 
+
 Start the node
 --------------
 

docs/upgrade/upgrade-guides/upgrade-guide-from-2025.x-to-2026.1/index.rst

@@ -1,13 +0,0 @@
-==========================================================
-Upgrade - ScyllaDB 2025.x to ScyllaDB 2026.1
-==========================================================
-
-.. toctree::
-   :maxdepth: 2
-   :hidden:
-
-   Upgrade ScyllaDB <upgrade-guide-from-2025.x-to-2026.1>
-   Metrics Update <metric-update-2025.x-to-2026.1>
-
-* :doc:`Upgrade from ScyllaDB 2025.x to ScyllaDB 2026.1 <upgrade-guide-from-2025.x-to-2026.1>`
-* :doc:`Metrics Update Between 2025.x and 2026.1 <metric-update-2025.x-to-2026.1>`

docs/upgrade/upgrade-guides/upgrade-guide-from-2025.x-to-2026.1/metric-update-2025.x-to-2026.1.rst

@@ -1,82 +0,0 @@
-.. |SRC_VERSION| replace:: 2025.x
-.. |NEW_VERSION| replace:: 2026.1
-.. |PRECEDING_VERSION| replace:: 2025.4
-
-================================================================
-Metrics Update Between |SRC_VERSION| and |NEW_VERSION|
-================================================================
-
-.. toctree::
-   :maxdepth: 2
-   :hidden:
-
-ScyllaDB |NEW_VERSION| Dashboards are available as part of the latest |mon_root|.
-
-
-New Metrics in |NEW_VERSION|
---------------------------------------
-
-The following metrics are new in ScyllaDB |NEW_VERSION| compared to |PRECEDING_VERSION|.
-
-.. list-table::
-   :widths: 25 150
-   :header-rows: 1
-
-   * - Metric
-     - Description
-   * - scylla_alternator_operation_size_kb
-     - Histogram of item sizes involved in a request.
-   * - scylla_column_family_total_disk_space_before_compression
-     - Hypothetical total disk space used if data files weren't compressed
-   * - scylla_group_name_auto_repair_enabled_nr
-     - Number of tablets with auto repair enabled.
-   * - scylla_group_name_auto_repair_needs_repair_nr
-     - Number of tablets with auto repair enabled that currently need repair.
-   * - scylla_lsa_compact_time_ms
-     - Total time spent on segment compaction that was not accounted under ``reclaim_time_ms``.
-   * - scylla_lsa_evict_time_ms
-     - Total time spent on evicting objects that was not accounted under ``reclaim_time_ms``,
-   * - scylla_lsa_reclaim_time_ms
-     - Total time spent in reclaiming LSA memory back to std allocator.
-   * - scylla_object_storage_memory_usage
-     - Total number of bytes consumed by the object storage client.
-   * - scylla_tablet_ops_failed
-     - Number of failed tablet auto repair attempts.
-   * - scylla_tablet_ops_succeeded
-     - Number of successful tablet auto repair attempts.
-   
-Renamed Metrics in |NEW_VERSION|
---------------------------------------
-
-The following metrics are renamed in ScyllaDB |NEW_VERSION| compared to |PRECEDING_VERSION|.
-
-.. list-table::
-   :widths: 25 150
-   :header-rows: 1
-
-   * - Metric Name in |PRECEDING_VERSION|
-     - Metric Name in |NEW_VERSION|
-   * - scylla_s3_memory_usage
-     - scylla_object_storage_memory_usage
-
-Removed Metrics in |NEW_VERSION|
---------------------------------------
-
-The following metrics are removed in ScyllaDB |NEW_VERSION|.
-
-* scylla_redis_current_connections
-* scylla_redis_op_latency
-* scylla_redis_operation
-* scylla_redis_operation
-* scylla_redis_requests_latency
-* scylla_redis_requests_served
-* scylla_redis_requests_serving
-
-New and Updated Metrics in Previous Releases
--------------------------------------------------------
-
-* `Metrics Update Between 2025.3 and 2025.4 <https://docs.scylladb.com/manual/branch-2025.4/upgrade/upgrade-guides/upgrade-guide-from-2025.x-to-2025.4/metric-update-2025.x-to-2025.4.html>`_
-* `Metrics Update Between 2025.2 and 2025.3 <https://docs.scylladb.com/manual/branch-2025.3/upgrade/upgrade-guides/upgrade-guide-from-2025.2-to-2025.3/metric-update-2025.2-to-2025.3.html>`_
-* `Metrics Update Between 2025.1 and 2025.2 <https://docs.scylladb.com/manual/branch-2025.2/upgrade/upgrade-guides/upgrade-guide-from-2025.1-to-2025.2/metric-update-2025.1-to-2025.2.html>`_
-
-

ent/encryption/gcp_host.cc

@@ -284,7 +284,6 @@ future<rjson::value> encryption::gcp_host::impl::gcp_auth_post_with_retry(std::s
                 }
                 [[fallthrough]];
             case httpclient::reply_status::request_timeout:
-            case httpclient::reply_status::too_many_requests:
                 if (retry < max_retries) {
                     // service unavailable etc -> backoff + retry
                     do_backoff = true;

gms/feature_service.hh

@@ -182,7 +182,7 @@ public:
     gms::feature removenode_with_left_token_ring { *this, "REMOVENODE_WITH_LEFT_TOKEN_RING"sv };
     gms::feature size_based_load_balancing { *this, "SIZE_BASED_LOAD_BALANCING"sv };
     gms::feature topology_noop_request { *this, "TOPOLOGY_NOOP_REQUEST"sv };
-    gms::feature batchlog_v2 { *this, "BATCHLOG_V2"sv };
+    gms::feature tablets_intermediate_fallback_cleanup { *this, "TABLETS_INTERMEDIATE_FALLBACK_CLEANUP"sv };
 public:
 
     const std::unordered_map<sstring, std::reference_wrapper<feature>>& registered_features() const;

index/vector_index.cc

@@ -17,11 +17,11 @@
 #include "index/secondary_index.hh"
 #include "index/secondary_index_manager.hh"
 #include "types/concrete_types.hh"
-#include "types/types.hh"
 #include "utils/managed_string.hh"
 #include <seastar/core/sstring.hh>
 #include <boost/algorithm/string.hpp>
 
+
 namespace secondary_index {
 
 static void validate_positive_option(int max, const sstring& value_name, const sstring& value) {
@@ -147,88 +147,17 @@ std::optional<cql3::description> vector_index::describe(const index_metadata& im
 }
 
 void vector_index::check_target(const schema& schema, const std::vector<::shared_ptr<cql3::statements::index_target>>& targets) const {
-
-    struct validate_visitor {
-        const class schema& schema;
-        bool& is_vector;
-
-        /// Vector indexes support filtering on native types that can be used as primary key columns.
-        /// There is no counter (it cannot be used with vector columns)
-        /// and no duration (it cannot be used as a primary key or in secondary indexes).
-        static bool is_supported_filtering_column(abstract_type const & kind_type) {
-            switch (kind_type.get_kind()) {
-                case abstract_type::kind::ascii:
-                case abstract_type::kind::boolean:
-                case abstract_type::kind::byte:
-                case abstract_type::kind::bytes:
-                case abstract_type::kind::date:
-                case abstract_type::kind::decimal:
-                case abstract_type::kind::double_kind:
-                case abstract_type::kind::float_kind:
-                case abstract_type::kind::inet:
-                case abstract_type::kind::int32:
-                case abstract_type::kind::long_kind:
-                case abstract_type::kind::short_kind:
-                case abstract_type::kind::simple_date:
-                case abstract_type::kind::time:
-                case abstract_type::kind::timestamp:
-                case abstract_type::kind::timeuuid:
-                case abstract_type::kind::utf8:
-                case abstract_type::kind::uuid:
-                case abstract_type::kind::varint:
-                    return true;
-                default:
-                    break;
-            }
-            return false;
-        }
-
-        void validate(cql3::column_identifier const& column, bool is_vector) const {
-            auto const& c_name = column.to_string();
-            auto const* c_def = schema.get_column_definition(column.name());
-            if (c_def == nullptr) {
-                throw exceptions::invalid_request_exception(format("Column {} not found in schema", c_name));
-            }
-
-            auto type = c_def->type;
-
-            if (is_vector) {
-                auto const* vector_type = dynamic_cast<const vector_type_impl*>(type.get());
-                if (vector_type == nullptr) {
-                    throw exceptions::invalid_request_exception("Vector indexes are only supported on columns of vectors of floats");
-                }
-
-                auto elements_type = vector_type->get_elements_type();
-                if (elements_type->get_kind() != abstract_type::kind::float_kind) {
-                    throw exceptions::invalid_request_exception("Vector indexes are only supported on columns of vectors of floats");
-                }
-                return;
-            }
-
-            if (!is_supported_filtering_column(*type)) {
-                throw exceptions::invalid_request_exception(format("Unsupported vector index filtering column {} type", c_name));
-            }
-        }
-
-        void operator()(const std::vector<::shared_ptr<cql3::column_identifier>>& columns) const {
-            for (const auto& column : columns) {
-                // CQL restricts the secondary local index to have multiple columns with partition key only.
-                // Vectors shouldn't be partition key columns and they aren't supported as a filtering column,
-                // so we can assume here that these are non-vectors filtering columns.
-                validate(*column, false);
-            }
-        }
-
-        void operator()(const ::shared_ptr<cql3::column_identifier>& column) {
-            validate(*column, is_vector);
-            // The first column is the vector column, the rest mustn't be vectors.
-            is_vector = false;
-        }
-    };
-
-    bool is_vector = true;
-    for (const auto& target : targets) {
-        std::visit(validate_visitor{.schema = schema, .is_vector = is_vector}, target->value);
+    if (targets.size() != 1) {
+        throw exceptions::invalid_request_exception("Vector index can only be created on a single column");
+    }
+    auto target = targets[0];
+    auto c_def = schema.get_column_definition(to_bytes(target->column_name()));
+    if (!c_def) {
+        throw exceptions::invalid_request_exception(format("Column {} not found in schema", target->column_name()));
+    }
+    auto type = c_def->type;
+    if (!type->is_vector() || static_cast<const vector_type_impl*>(type.get())->get_elements_type()->get_kind() != abstract_type::kind::float_kind) {
+        throw exceptions::invalid_request_exception(format("Vector indexes are only supported on columns of vectors of floats", target->column_name()));
     }
 }
 

install-dependencies.sh

@@ -31,6 +31,7 @@ fi
 
 debian_base_packages=(
     clang
+    clang-tools
     gdb
     cargo
     wabt
@@ -72,6 +73,7 @@ debian_base_packages=(
 
 fedora_packages=(
     clang
+    clang-tools-extra
     compiler-rt
     libasan
     libubsan

install.sh

@@ -347,8 +347,8 @@ install -d -m755 "$retc"/scylla.d
 scylla_yaml_dir=$(mktemp -d)
 scylla_yaml=$scylla_yaml_dir/scylla.yaml
 grep -v api_ui_dir conf/scylla.yaml | grep -v api_doc_dir > $scylla_yaml
-echo "api_ui_dir: $prefix/swagger-ui/dist/" >> $scylla_yaml
-echo "api_doc_dir: $prefix/api/api-doc/" >> $scylla_yaml
+echo "api_ui_dir: /opt/scylladb/swagger-ui/dist/" >> $scylla_yaml
+echo "api_doc_dir: /opt/scylladb/api/api-doc/" >> $scylla_yaml
 installconfig 644 $scylla_yaml "$retc"/scylla
 rm -rf $scylla_yaml_dir
 

locator/tablets.cc

@@ -50,6 +50,8 @@ write_replica_set_selector get_selector_for_writes(tablet_transition_stage stage
             return write_replica_set_selector::previous;
         case tablet_transition_stage::write_both_read_old:
             return write_replica_set_selector::both;
+        case tablet_transition_stage::write_both_read_old_fallback_cleanup:
+            return write_replica_set_selector::both;
         case tablet_transition_stage::streaming:
             return write_replica_set_selector::both;
         case tablet_transition_stage::rebuild_repair:
@@ -81,6 +83,8 @@ read_replica_set_selector get_selector_for_reads(tablet_transition_stage stage)
             return read_replica_set_selector::previous;
         case tablet_transition_stage::write_both_read_old:
             return read_replica_set_selector::previous;
+        case tablet_transition_stage::write_both_read_old_fallback_cleanup:
+            return read_replica_set_selector::previous;
         case tablet_transition_stage::streaming:
             return read_replica_set_selector::previous;
         case tablet_transition_stage::rebuild_repair:
@@ -612,16 +616,12 @@ tablet_replica tablet_map::get_primary_replica(tablet_id id, const locator::topo
     return maybe_get_primary_replica(id, replicas, topo, [&] (const auto& _) { return true; }).value();
 }
 
-tablet_replica tablet_map::get_secondary_replica(tablet_id id, const locator::topology& topo) const {
-    const auto& orig_replicas = get_tablet_info(id).replicas;
-    if (orig_replicas.size() < 2) {
+tablet_replica tablet_map::get_secondary_replica(tablet_id id) const {
+    if (get_tablet_info(id).replicas.size() < 2) {
         throw std::runtime_error(format("No secondary replica for tablet id {}", id));
     }
-    tablet_replica_set replicas = orig_replicas;
-    std::ranges::sort(replicas, tablet_replica_comparator(topo));
-    // This formula must match the one in get_primary_replica(),
-    // just with + 1.
-    return replicas.at((size_t(id) + size_t(id) / replicas.size() + 1) % replicas.size());
+    const auto& replicas = get_tablet_info(id).replicas;
+    return replicas.at((size_t(id)+1) % replicas.size());
 }
 
 std::optional<tablet_replica> tablet_map::maybe_get_selected_replica(tablet_id id, const topology& topo, const tablet_task_info& tablet_task_info) const {
@@ -745,6 +745,7 @@ void tablet_map::set_tablet_raft_info(tablet_id id, tablet_raft_info raft_info)
 static const std::unordered_map<tablet_transition_stage, sstring> tablet_transition_stage_to_name = {
     {tablet_transition_stage::allow_write_both_read_old, "allow_write_both_read_old"},
     {tablet_transition_stage::write_both_read_old, "write_both_read_old"},
+    {tablet_transition_stage::write_both_read_old_fallback_cleanup, "write_both_read_old_fallback_cleanup"},
     {tablet_transition_stage::write_both_read_new, "write_both_read_new"},
     {tablet_transition_stage::streaming, "streaming"},
     {tablet_transition_stage::rebuild_repair, "rebuild_repair"},

locator/tablets.hh

@@ -277,6 +277,7 @@ std::optional<tablet_info> merge_tablet_info(tablet_info a, tablet_info b);
 enum class tablet_transition_stage {
     allow_write_both_read_old,
     write_both_read_old,
+    write_both_read_old_fallback_cleanup,
     streaming,
     rebuild_repair,
     write_both_read_new,
@@ -647,10 +648,9 @@ public:
     /// Returns the primary replica for the tablet
     tablet_replica get_primary_replica(tablet_id id, const locator::topology& topo) const;
 
-    /// Returns the secondary replica for the tablet: the replica that immediately follows the primary
-    /// replica in the topology-sorted replica list.
+    /// Returns the secondary replica for the tablet, which is assumed to be directly following the primary replica in the replicas vector
     /// \throws std::runtime_error if the tablet has less than 2 replicas.
-    tablet_replica get_secondary_replica(tablet_id id, const locator::topology& topo) const;
+    tablet_replica get_secondary_replica(tablet_id id) const;
 
     // Returns the replica that matches hosts and dcs filters for tablet_task_info.
     std::optional<tablet_replica> maybe_get_selected_replica(tablet_id id, const topology& topo, const tablet_task_info& tablet_task_info) const;

main.cc

@@ -571,7 +571,7 @@ sharded<service::storage_proxy> *the_storage_proxy;
 // This is used by perf-alternator to allow running scylla together with the tool
 // in a single process. So that it's easier to measure internals. It's not added
 // to main_func_type to not complicate common flow as no other tool needs such logic.
-std::function<void(lw_shared_ptr<db::config>)> after_init_func;
+std::function<future<>(lw_shared_ptr<db::config>, sharded<abort_source>&)> after_init_func;
 
 static locator::host_id initialize_local_info_thread(sharded<db::system_keyspace>& sys_ks,
         sharded<locator::snitch_ptr>& snitch,
@@ -906,6 +906,7 @@ To start the scylla server proper, simply invoke as: scylla server (or just scyl
 
             auto background_reclaim_scheduling_group = create_scheduling_group("background_reclaim", "bgre", 50).get();
             auto maintenance_scheduling_group = create_scheduling_group("streaming", "strm", 200).get();
+            debug::streaming_scheduling_group = maintenance_scheduling_group;
 
             smp::invoke_on_all([&cfg, background_reclaim_scheduling_group] {
                 logalloc::tracker::config st_cfg;
@@ -1306,6 +1307,7 @@ To start the scylla server proper, simply invoke as: scylla server (or just scyl
             checkpoint(stop_signal, "starting storage proxy");
             service::storage_proxy::config spcfg {
                 .hints_directory_initializer = hints_dir_initializer,
+                .hints_sched_group = maintenance_scheduling_group,
             };
             spcfg.hinted_handoff_enabled = hinted_handoff_enabled;
             spcfg.available_memory = memory::stats().total_memory();
@@ -1677,7 +1679,9 @@ To start the scylla server proper, simply invoke as: scylla server (or just scyl
                     gossiper.local(), feature_service.local(), sys_ks.local(), group0_client, dbcfg.gossip_scheduling_group};
 
             checkpoint(stop_signal, "starting tablet allocator");
-            service::tablet_allocator::config tacfg;
+            service::tablet_allocator::config tacfg {
+                .background_sg = maintenance_scheduling_group,
+            };
             sharded<service::tablet_allocator> tablet_allocator;
             tablet_allocator.start(tacfg, std::ref(mm_notifier), std::ref(db)).get();
             auto stop_tablet_allocator = defer_verbose_shutdown("tablet allocator", [&tablet_allocator] {
@@ -2417,7 +2421,7 @@ To start the scylla server proper, simply invoke as: scylla server (or just scyl
             bm_cfg.delay = std::chrono::milliseconds(cfg->ring_delay_ms());
             bm_cfg.replay_cleanup_after_replays = cfg->batchlog_replay_cleanup_after_replays();
 
-            bm.start(std::ref(qp), std::ref(sys_ks), std::ref(feature_service), bm_cfg).get();
+            bm.start(std::ref(qp), std::ref(sys_ks), bm_cfg).get();
             auto stop_batchlog_manager = defer_verbose_shutdown("batchlog manager", [&bm] {
                 bm.stop().get();
             });
@@ -2490,7 +2494,9 @@ To start the scylla server proper, simply invoke as: scylla server (or just scyl
 
             if (cfg->view_building()) {
                 checkpoint(stop_signal, "starting view builders");
-                view_builder.invoke_on_all(&db::view::view_builder::start, std::ref(mm), utils::cross_shard_barrier()).get();
+                with_scheduling_group(maintenance_scheduling_group, [&mm] {
+                    return view_builder.invoke_on_all(&db::view::view_builder::start, std::ref(mm), utils::cross_shard_barrier());
+                }).get();
             }
             auto drain_view_builder = defer_verbose_shutdown("draining view builders", [&] {
                 view_builder.invoke_on_all(&db::view::view_builder::drain).get();
@@ -2576,11 +2582,13 @@ To start the scylla server proper, simply invoke as: scylla server (or just scyl
             supervisor::notify("serving");
 
             startlog.info("Scylla version {} initialization completed.", scylla_version());
+            future<> after_init_fut = make_ready_future<>();
             if (after_init_func) {
-                after_init_func(cfg);
+                after_init_fut = after_init_func(cfg, stop_signal.as_sharded_abort_source());
             }
             stop_signal.wait().get();
             startlog.info("Signal received; shutting down");
+            std::move(after_init_fut).get();
 	    // At this point, all objects destructors and all shutdown hooks registered with defer() are executed
           } catch (const sleep_aborted&) {
             startlog.info("Startup interrupted");
@@ -2650,7 +2658,8 @@ int main(int ac, char** av) {
         {"perf-load-balancing", perf::scylla_tablet_load_balancing_main, "run tablet load balancer tests"},
         {"perf-simple-query", perf::scylla_simple_query_main, "run performance tests by sending simple queries to this server"},
         {"perf-sstable", perf::scylla_sstable_main, "run performance tests by exercising sstable related operations on this server"},
-        {"perf-alternator", perf::alternator(scylla_main, &after_init_func), "run performance tests on full alternator stack"}
+        {"perf-alternator", perf::alternator(scylla_main, &after_init_func), "run performance tests on full alternator stack"},
+        {"perf-cql-raw", perf::perf_cql_raw(scylla_main, &after_init_func), "run performance tests using raw CQL protocol frames"}
     };
 
     main_func_type main_func;

mutation/collection_mutation.cc

@@ -261,7 +261,7 @@ static collection_mutation serialize_collection_mutation(
 
         writev(v.serialize());
     }
-    return collection_mutation(type, std::move(ret));
+    return collection_mutation(type, ret);
 }
 
 collection_mutation collection_mutation_description::serialize(const abstract_type& type) const {

node_ops/task_manager_module.cc

@@ -103,7 +103,7 @@ future<std::optional<tasks::task_status>> node_ops_virtual_task::get_status(task
         .entity = stats.entity,
         .progress_units = "",
         .progress = tasks::task_manager::task::progress{},
-        .children = co_await get_children(get_module(), id, _ss.get_token_metadata_ptr())
+        .children = co_await get_children(get_module(), id, std::bind_front(&gms::gossiper::is_alive, &_ss.gossiper()))
     };
 }
 

pgo/profiles/aarch64/profile.profdata.xz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:088a9d7e165d33436eb3029ab092582cbae61f0e17486c226d8947ff44658c78
-size 6535832
+oid sha256:a4710f1f0b0bb329721c21d133618e811e820f2e70553b0aca28fb278bff89c9
+size 6492280

pgo/profiles/x86_64/profile.profdata.xz

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d1a869ebfe4e90d9681499246eb86bb032ae402c350357e19d97b989037a5bd3
-size 6528308
+oid sha256:2433f7a1fc5cda0dd990ab59587eb6046dca0fe1ae48d599953d1936fe014ed9
+size 6492176

reader_concurrency_semaphore.cc

@@ -148,6 +148,7 @@ public:
     };
 
 private:
+    const db::timeout_clock::time_point _created;
     reader_concurrency_semaphore& _semaphore;
     schema_ptr _schema;
 
@@ -237,17 +238,25 @@ private:
                 break;
             case state::inactive:
                 _semaphore.evict(*this, reader_concurrency_semaphore::evict_reason::time);
-                break;
+                // Return here on purpose. The evicted permit is destroyed when closing a reader.
+                // As a consequence, any member access beyond this point is invalid.
+                return;
             case state::evicted:
+            case state::preemptive_aborted:
                 break;
         }
+
+        // The function call not only sets state to reader_permit::state::preemptive_aborted
+        // but also correctly decreases the statistics i.e. need_cpu_permits and awaits_permits.
+        on_permit_inactive(reader_permit::state::preemptive_aborted);
     }
 
 public:
     struct value_tag {};
 
     impl(reader_concurrency_semaphore& semaphore, schema_ptr schema, const std::string_view& op_name, reader_resources base_resources, db::timeout_clock::time_point timeout, tracing::trace_state_ptr trace_ptr)
-        : _semaphore(semaphore)
+        : _created(db::timeout_clock::now())
+        , _semaphore(semaphore)
         , _schema(std::move(schema))
         , _op_name_view(op_name)
         , _base_resources(base_resources)
@@ -258,7 +267,8 @@ public:
         _semaphore.on_permit_created(*this);
     }
     impl(reader_concurrency_semaphore& semaphore, schema_ptr schema, sstring&& op_name, reader_resources base_resources, db::timeout_clock::time_point timeout, tracing::trace_state_ptr trace_ptr)
-        : _semaphore(semaphore)
+        : _created(db::timeout_clock::now())
+        , _semaphore(semaphore)
         , _schema(std::move(schema))
         , _op_name(std::move(op_name))
         , _op_name_view(_op_name)
@@ -360,6 +370,17 @@ public:
         on_permit_active();
     }
 
+    void on_preemptive_aborted() {
+        if (_state != reader_permit::state::waiting_for_admission && _state != reader_permit::state::waiting_for_memory) {
+            on_internal_error(rcslog, format("on_preemptive_aborted(): permit in invalid state {}", _state));
+        }
+
+        _ttl_timer.cancel();
+        _state = reader_permit::state::preemptive_aborted;
+        _aux_data.pr.set_exception(named_semaphore_aborted(_semaphore._name));
+        _semaphore.on_permit_preemptive_aborted();
+    }
+
     void on_register_as_inactive() {
         SCYLLA_ASSERT(_state == reader_permit::state::active || _state == reader_permit::state::active_need_cpu || _state == reader_permit::state::waiting_for_memory);
         on_permit_inactive(reader_permit::state::inactive);
@@ -467,6 +488,10 @@ public:
         return _semaphore.do_wait_admission(*this);
     }
 
+    db::timeout_clock::time_point created() const noexcept {
+        return _created;
+    }
+
     db::timeout_clock::time_point timeout() const noexcept {
         return _ttl_timer.armed() ? _ttl_timer.get_timeout() : db::no_timeout;
     }
@@ -689,6 +714,9 @@ auto fmt::formatter<reader_permit::state>::format(reader_permit::state s, fmt::f
         case reader_permit::state::evicted:
             name = "evicted";
             break;
+        case reader_permit::state::preemptive_aborted:
+            name = "preemptive_aborted";
+            break;
     }
     return formatter<string_view>::format(name, ctx);
 }
@@ -1021,8 +1049,8 @@ void reader_concurrency_semaphore::signal(const resources& r) noexcept {
         on_internal_error_noexcept(rcslog,
                 format("reader_concurrency_semaphore::signal(): semaphore {} detected resource leak, available {} exceeds initial {}", _name,
                         _resources, _initial_resources));
-        _resources.count = std::min(_resources.count, _initial_resources.count);
-        _resources.memory = std::min(_resources.memory, _initial_resources.memory);
+        _resources.count = std::max(_resources.count, _initial_resources.count);
+        _resources.memory = std::max(_resources.memory, _initial_resources.memory);
     }
     maybe_wake_execution_loop();
 }
@@ -1038,6 +1066,7 @@ reader_concurrency_semaphore::reader_concurrency_semaphore(
         utils::updateable_value<uint32_t> serialize_limit_multiplier,
         utils::updateable_value<uint32_t> kill_limit_multiplier,
         utils::updateable_value<uint32_t> cpu_concurrency,
+        utils::updateable_value<float> preemptive_abort_factor,
         register_metrics metrics)
     : _initial_resources(count, memory)
     , _resources(count, memory)
@@ -1047,6 +1076,7 @@ reader_concurrency_semaphore::reader_concurrency_semaphore(
     , _serialize_limit_multiplier(std::move(serialize_limit_multiplier))
     , _kill_limit_multiplier(std::move(kill_limit_multiplier))
     , _cpu_concurrency(cpu_concurrency)
+    , _preemptive_abort_factor(preemptive_abort_factor)
     , _close_readers_gate(format("[reader_concurrency_semaphore {}] close_readers", _name))
     , _permit_gate(format("[reader_concurrency_semaphore {}] permit", _name))
 {
@@ -1114,6 +1144,7 @@ reader_concurrency_semaphore::reader_concurrency_semaphore(no_limits, sstring na
             utils::updateable_value(std::numeric_limits<uint32_t>::max()),
             utils::updateable_value(std::numeric_limits<uint32_t>::max()),
             utils::updateable_value(uint32_t(1)),
+            utils::updateable_value(float(0.0)),
             metrics) {}
 
 reader_concurrency_semaphore::~reader_concurrency_semaphore() {
@@ -1489,6 +1520,25 @@ void reader_concurrency_semaphore::maybe_admit_waiters() noexcept {
         auto& permit = _wait_list.front();
         dequeue_permit(permit);
         try {
+            // Do not admit the read as it is unlikely to finish before its timeout. The condition is:
+            // permit's remaining time <= preemptive_abort_factor * permit's time budget
+            //
+            // The additional check for remaining_time > 0 is to avoid preemptive aborting reads
+            // that already timed out but are still in the wait list due to scheduling delays.
+            // It also effectively disables preemptive aborting when the factor is set to 0.
+            const auto time_budget = permit.timeout() - permit.created();
+            const auto remaining_time = permit.timeout() - db::timeout_clock::now();
+            if (remaining_time > db::timeout_clock::duration::zero() && remaining_time <= _preemptive_abort_factor() * time_budget) {
+                permit.on_preemptive_aborted();
+                using ms = std::chrono::milliseconds;
+                tracing::trace(permit.trace_state(), "[reader concurrency semaphore {}] read shed as unlikely to finish (elapsed: {}, timeout: {}, preemptive_factor: {})",
+                               _name,
+                               std::chrono::duration_cast<ms>(time_budget - remaining_time),
+                               std::chrono::duration_cast<ms>(time_budget),
+                               _preemptive_abort_factor());
+                continue;
+            }
+
             if (permit.get_state() == reader_permit::state::waiting_for_memory) {
                 _blessed_permit = &permit;
                 permit.on_granted_memory();
@@ -1549,7 +1599,11 @@ void reader_concurrency_semaphore::dequeue_permit(reader_permit::impl& permit) {
         case reader_permit::state::waiting_for_admission:
         case reader_permit::state::waiting_for_memory:
         case reader_permit::state::waiting_for_execution:
-            --_stats.waiters;
+            if (_stats.waiters > 0) {
+                --_stats.waiters;
+            } else {
+                on_internal_error_noexcept(rcslog, "reader_concurrency_semaphore::dequeue_permit(): invalid state: no waiters yet dequeueing a waiting permit");
+            }
             break;
         case reader_permit::state::inactive:
         case reader_permit::state::evicted:
@@ -1558,12 +1612,17 @@ void reader_concurrency_semaphore::dequeue_permit(reader_permit::impl& permit) {
         case reader_permit::state::active:
         case reader_permit::state::active_need_cpu:
         case reader_permit::state::active_await:
+        case reader_permit::state::preemptive_aborted:
             on_internal_error_noexcept(rcslog, format("reader_concurrency_semaphore::dequeue_permit(): unrecognized queued state: {}", permit.get_state()));
     }
     permit.unlink();
     _permit_list.push_back(permit);
 }
 
+void reader_concurrency_semaphore::on_permit_preemptive_aborted() noexcept {
+    ++_stats.total_reads_shed_due_to_overload;
+}
+
 void reader_concurrency_semaphore::on_permit_created(reader_permit::impl& permit) {
     _permit_gate.enter();
     _permit_list.push_back(permit);

reader_concurrency_semaphore.hh

@@ -42,7 +42,7 @@ using mutation_reader_opt = optimized_optional<mutation_reader>;
 /// number of waiting readers becomes equal or greater than
 /// `max_queue_length` (upon calling `obtain_permit()`) an exception of
 /// type `std::runtime_error` is thrown. Optionally, some additional
-/// code can be executed just before throwing (`prethrow_action` 
+/// code can be executed just before throwing (`prethrow_action`
 /// constructor parameter).
 ///
 /// The semaphore has 3 layers of defense against consuming more memory
@@ -89,6 +89,7 @@ public:
         // Total number of failed reads executed through this semaphore.
         uint64_t total_failed_reads = 0;
         // Total number of reads rejected because the admission queue reached its max capacity
+        // or rejected due to a high probability of not getting finalized on time.
         uint64_t total_reads_shed_due_to_overload = 0;
         // Total number of reads killed due to the memory consumption reaching the kill limit.
         uint64_t total_reads_killed_due_to_kill_limit = 0;
@@ -192,6 +193,8 @@ private:
     utils::updateable_value<uint32_t> _serialize_limit_multiplier;
     utils::updateable_value<uint32_t> _kill_limit_multiplier;
     utils::updateable_value<uint32_t> _cpu_concurrency;
+    utils::updateable_value<float> _preemptive_abort_factor;
+
     stats _stats;
     std::optional<seastar::metrics::metric_groups> _metrics;
     bool _stopped = false;
@@ -250,6 +253,8 @@ private:
     void on_permit_created(reader_permit::impl&);
     void on_permit_destroyed(reader_permit::impl&) noexcept;
 
+    void on_permit_preemptive_aborted() noexcept;
+
     void on_permit_need_cpu() noexcept;
     void on_permit_not_need_cpu() noexcept;
 
@@ -287,6 +292,7 @@ public:
             utils::updateable_value<uint32_t> serialize_limit_multiplier,
             utils::updateable_value<uint32_t> kill_limit_multiplier,
             utils::updateable_value<uint32_t> cpu_concurrency,
+            utils::updateable_value<float> preemptive_abort_factor,
             register_metrics metrics);
 
     reader_concurrency_semaphore(
@@ -296,9 +302,12 @@ public:
             size_t max_queue_length,
             utils::updateable_value<uint32_t> serialize_limit_multiplier,
             utils::updateable_value<uint32_t> kill_limit_multiplier,
+            utils::updateable_value<uint32_t> cpu_concurrency,
+            utils::updateable_value<float> preemptive_abort_factor,
             register_metrics metrics)
         : reader_concurrency_semaphore(utils::updateable_value(count), memory, std::move(name), max_queue_length,
-                std::move(serialize_limit_multiplier), std::move(kill_limit_multiplier), utils::updateable_value<uint32_t>(1), metrics)
+                std::move(serialize_limit_multiplier), std::move(kill_limit_multiplier), std::move(cpu_concurrency),
+                std::move(preemptive_abort_factor), metrics)
     { }
 
     /// Create a semaphore with practically unlimited count and memory.
@@ -318,9 +327,10 @@ public:
             utils::updateable_value<uint32_t> serialize_limit_multipler = utils::updateable_value(std::numeric_limits<uint32_t>::max()),
             utils::updateable_value<uint32_t> kill_limit_multipler = utils::updateable_value(std::numeric_limits<uint32_t>::max()),
             utils::updateable_value<uint32_t> cpu_concurrency = utils::updateable_value<uint32_t>(1),
+            utils::updateable_value<float> preemptive_abort_factor = utils::updateable_value<float>(0.0f),
             register_metrics metrics = register_metrics::no)
         : reader_concurrency_semaphore(utils::updateable_value(count), memory, std::move(name), max_queue_length, std::move(serialize_limit_multipler),
-                std::move(kill_limit_multipler), std::move(cpu_concurrency), metrics)
+                std::move(kill_limit_multipler), std::move(cpu_concurrency), std::move(preemptive_abort_factor), metrics)
     {}
 
     virtual ~reader_concurrency_semaphore();

reader_concurrency_semaphore_group.cc

@@ -70,7 +70,8 @@ reader_concurrency_semaphore& reader_concurrency_semaphore_group::add_or_update(
             _max_queue_length,
             _serialize_limit_multiplier,
             _kill_limit_multiplier,
-            _cpu_concurrency
+            _cpu_concurrency,
+            _preemptive_abort_factor
         );
     auto&& it = result.first;
     // since we serialize all group changes this change wait will be queues and no further operations