Fix code style issues: remove trailing whitespace

Co-authored-by: tgrabiec <283695+tgrabiec@users.noreply.github.com>

.github/CODEOWNERS

@@ -1,5 +1,5 @@
 # AUTH
-auth/* @nuivall @ptrsmrn
+auth/* @nuivall
 
 # CACHE
 row_cache* @tgrabiec
@@ -25,11 +25,11 @@ compaction/* @raphaelsc
 transport/*
 
 # CQL QUERY LANGUAGE
-cql3/* @tgrabiec @nuivall @ptrsmrn
+cql3/* @tgrabiec @nuivall
 
 # COUNTERS
-counters* @nuivall @ptrsmrn
-tests/counter_test* @nuivall @ptrsmrn
+counters* @nuivall
+tests/counter_test* @nuivall
 
 # DOCS
 docs/* @annastuchlik @tzach

.github/scripts/auto-backport.py

@@ -62,7 +62,7 @@ def create_pull_request(repo, new_branch_name, base_branch_name, pr, backport_pr
         if is_draft:
             labels_to_add.append("conflicts")
             pr_comment = f"@{pr.user.login} - This PR was marked as draft because it has conflicts\n"
-            pr_comment += "Please resolve them and mark this PR as ready for review"
+            pr_comment += "Please resolve them and remove the 'conflicts' label. The PR will be made ready for review automatically."
             backport_pr.create_issue_comment(pr_comment)
         
         # Apply all labels at once if we have any

.github/workflows/backport-pr-fixes-validation.yaml

@@ -18,7 +18,7 @@ jobs:
             
             // Regular expression pattern to check for "Fixes" prefix
             // Adjusted to dynamically insert the repository full name
-            const pattern = `Fixes:? (?:#|${repo.replace('/', '\\/')}#|https://github\\.com/${repo.replace('/', '\\/')}/issues/)(\\d+)`;
+            const pattern = `Fixes:? ((?:#|${repo.replace('/', '\\/')}#|https://github\\.com/${repo.replace('/', '\\/')}/issues/)(\\d+)|([A-Z]+-\\d+))`;
             const regex = new RegExp(pattern);
             
             if (!regex.test(body)) {

.github/workflows/call_sync_milestone_to_jira.yml

@@ -0,0 +1,14 @@
+name: Call Jira release creation for new milestone
+
+on:
+  milestone:
+    types: [created]
+
+jobs:
+  sync-milestone-to-jira:
+    uses: scylladb/github-automation/.github/workflows/main_sync_milestone_to_jira_release.yml@main
+    with:
+      # Comma-separated list of Jira project keys
+      jira_project_keys: "SCYLLADB,CUSTOMER"
+    secrets:
+      caller_jira_auth: ${{ secrets.USER_AND_KEY_FOR_JIRA_AUTOMATION }}

.github/workflows/call_validate_pr_author_email.yml

@@ -0,0 +1,13 @@
+name: validate_pr_author_email
+
+on:
+  pull_request_target:
+    types:
+      - opened
+      - synchronize
+      - reopened
+
+jobs:
+  validate_pr_author_email:
+    uses: scylladb/github-automation/.github/workflows/validate_pr_author_email.yml@main
+

.github/workflows/docs-validate-metrics.yml

@@ -7,7 +7,7 @@ on:
       - enterprise
     paths:
       - '**/*.cc'
-      - 'scripts/metrics-config.yml' 
+      - 'scripts/metrics-config.yml'
       - 'scripts/get_description.py'
       - 'docs/_ext/scylladb_metrics.py'
 
@@ -15,20 +15,20 @@ jobs:
   validate-metrics:
     runs-on: ubuntu-latest
     name: Check metrics documentation coverage
-    
+
     steps:
     - name: Checkout code
       uses: actions/checkout@v4
       with:
         submodules: true
-      
+
     - name: Set up Python
       uses: actions/setup-python@v6
       with:
         python-version: '3.10'
-        
+
     - name: Install dependencies
       run: pip install PyYAML
-        
+
     - name: Validate metrics
       run: python3 scripts/get_description.py --validate -c scripts/metrics-config.yml

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

@@ -3,10 +3,13 @@ name: Trigger Scylla CI Route
 on:
   issue_comment:
     types: [created]
+  pull_request_target:
+    types:
+      - unlabeled
 
 jobs:
   trigger-jenkins:
-    if: github.event.comment.user.login != 'scylladbbot' && contains(github.event.comment.body, '@scylladbbot') && contains(github.event.comment.body, 'trigger-ci')
+    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: Trigger Scylla-CI-Route Jenkins Job

alternator/conditions.cc

@@ -42,7 +42,7 @@ comparison_operator_type get_comparison_operator(const rjson::value& comparison_
     if (!comparison_operator.IsString()) {
         throw api_error::validation(fmt::format("Invalid comparison operator definition {}", rjson::print(comparison_operator)));
     }
-    std::string op = comparison_operator.GetString();
+    std::string op = rjson::to_string(comparison_operator);
     auto it = ops.find(op);
     if (it == ops.end()) {
         throw api_error::validation(fmt::format("Unsupported comparison operator {}", op));
@@ -377,8 +377,8 @@ bool check_compare(const rjson::value* v1, const rjson::value& v2, const Compara
         return cmp(unwrap_number(*v1, cmp.diagnostic), unwrap_number(v2, cmp.diagnostic));
     }
     if (kv1.name == "S") {
-        return cmp(std::string_view(kv1.value.GetString(), kv1.value.GetStringLength()),
-                   std::string_view(kv2.value.GetString(), kv2.value.GetStringLength()));
+        return cmp(rjson::to_string_view(kv1.value),
+                   rjson::to_string_view(kv2.value));
     }
     if (kv1.name == "B") {
         auto d_kv1 = unwrap_bytes(kv1.value, v1_from_query);
@@ -470,9 +470,9 @@ static bool check_BETWEEN(const rjson::value* v, const rjson::value& lb, const r
         return check_BETWEEN(unwrap_number(*v, diag), unwrap_number(lb, diag), unwrap_number(ub, diag), bounds_from_query);
     }
     if (kv_v.name == "S") {
-        return check_BETWEEN(std::string_view(kv_v.value.GetString(), kv_v.value.GetStringLength()),
-                             std::string_view(kv_lb.value.GetString(), kv_lb.value.GetStringLength()),
-                             std::string_view(kv_ub.value.GetString(), kv_ub.value.GetStringLength()),
+        return check_BETWEEN(rjson::to_string_view(kv_v.value),
+                             rjson::to_string_view(kv_lb.value),
+                             rjson::to_string_view(kv_ub.value),
                              bounds_from_query);
     }
     if (kv_v.name == "B") {

alternator/consumed_capacity.cc

@@ -8,6 +8,8 @@
 
 #include "consumed_capacity.hh"
 #include "error.hh"
+#include "utils/rjson.hh"
+#include <fmt/format.h>
 
 namespace alternator {
 
@@ -32,12 +34,12 @@ bool consumed_capacity_counter::should_add_capacity(const rjson::value& request)
     if (!return_consumed->IsString()) {
         throw api_error::validation("Non-string ReturnConsumedCapacity field in request");
     }
-    std::string consumed = return_consumed->GetString();
+    std::string_view consumed = rjson::to_string_view(*return_consumed);
     if (consumed == "INDEXES") {
         throw api_error::validation("INDEXES consumed capacity is not supported");
     }
     if (consumed != "TOTAL") {
-        throw api_error::validation("Unknown consumed capacity "+ consumed);
+        throw api_error::validation(fmt::format("Unknown consumed capacity {}", consumed));
     }
     return true;
 }

alternator/controller.cc

@@ -169,7 +169,7 @@ future<> controller::request_stop_server() {
     });
 }
 
-future<utils::chunked_vector<client_data>> controller::get_client_data() {
+future<utils::chunked_vector<foreign_ptr<std::unique_ptr<client_data>>>> controller::get_client_data() {
     return _server.local().get_client_data();
 }
 

alternator/controller.hh

@@ -93,7 +93,7 @@ public:
     // This virtual function is called (on each shard separately) when the
     // virtual table "system.clients" is read. It is expected to generate a
     // list of clients connected to this server (on this shard).
-    virtual future<utils::chunked_vector<client_data>> get_client_data() override;
+    virtual future<utils::chunked_vector<foreign_ptr<std::unique_ptr<client_data>>>> get_client_data() override;
 };
 
 }

alternator/executor.cc

@@ -419,7 +419,7 @@ static std::optional<std::string> find_table_name(const rjson::value& request) {
     if (!table_name_value->IsString()) {
         throw api_error::validation("Non-string TableName field in request");
     }
-    std::string table_name = table_name_value->GetString();
+    std::string table_name = rjson::to_string(*table_name_value);
     return table_name;
 }
 
@@ -546,7 +546,7 @@ get_table_or_view(service::storage_proxy& proxy, const rjson::value& request) {
             // does exist but the index does not (ValidationException).
             if (proxy.data_dictionary().has_schema(keyspace_name, orig_table_name)) {
                 throw api_error::validation(
-                    fmt::format("Requested resource not found: Index '{}' for table '{}'", index_name->GetString(), orig_table_name));
+                    fmt::format("Requested resource not found: Index '{}' for table '{}'", rjson::to_string_view(*index_name), orig_table_name));
             } else {
                 throw api_error::resource_not_found(
                     fmt::format("Requested resource not found: Table: {} not found", orig_table_name));
@@ -587,7 +587,7 @@ static std::string get_string_attribute(const rjson::value& value, std::string_v
         throw api_error::validation(fmt::format("Expected string value for attribute {}, got: {}",
                 attribute_name, value));
     }
-    return std::string(attribute_value->GetString(), attribute_value->GetStringLength());
+    return rjson::to_string(*attribute_value);
 }
 
 // Convenience function for getting the value of a boolean attribute, or a
@@ -1080,8 +1080,8 @@ static void add_column(schema_builder& builder, const std::string& name, const r
     }
     for (auto it = attribute_definitions.Begin(); it != attribute_definitions.End(); ++it) {
         const rjson::value& attribute_info = *it;
-        if (attribute_info["AttributeName"].GetString() == name) {
-            auto type = attribute_info["AttributeType"].GetString();
+        if (rjson::to_string_view(attribute_info["AttributeName"]) == name) {
+            std::string_view type = rjson::to_string_view(attribute_info["AttributeType"]);
             data_type dt = parse_key_type(type);
             if (computed_column) {
                 // Computed column for GSI (doesn't choose a real column as-is
@@ -1116,7 +1116,7 @@ static std::pair<std::string, std::string> parse_key_schema(const rjson::value&
         throw api_error::validation("First element of KeySchema must be an object");
     }
     const rjson::value *v = rjson::find((*key_schema)[0], "KeyType");
-    if (!v || !v->IsString() || v->GetString() != std::string("HASH")) {
+    if (!v || !v->IsString() || rjson::to_string_view(*v) != "HASH") {
         throw api_error::validation("First key in KeySchema must be a HASH key");
     }
     v = rjson::find((*key_schema)[0], "AttributeName");
@@ -1124,14 +1124,14 @@ static std::pair<std::string, std::string> parse_key_schema(const rjson::value&
         throw api_error::validation("First key in KeySchema must have string AttributeName");
     }
     validate_attr_name_length(supplementary_context, v->GetStringLength(), true, "HASH key in KeySchema - ");
-    std::string hash_key = v->GetString();
+    std::string hash_key = rjson::to_string(*v);
     std::string range_key;
     if (key_schema->Size() == 2) {
         if (!(*key_schema)[1].IsObject()) {
             throw api_error::validation("Second element of KeySchema must be an object");
         }
         v = rjson::find((*key_schema)[1], "KeyType");
-        if (!v || !v->IsString() || v->GetString() != std::string("RANGE")) {
+        if (!v || !v->IsString() || rjson::to_string_view(*v) != "RANGE") {
             throw api_error::validation("Second key in KeySchema must be a RANGE key");
         }
         v = rjson::find((*key_schema)[1], "AttributeName");
@@ -1799,6 +1799,11 @@ static future<executor::request_return_type> create_table_on_shard0(service::cli
                 }
             }
         }
+        // Creating an index in tablets mode requires the rf_rack_valid_keyspaces option to be enabled.
+        // GSI and LSI indexes are based on materialized views which require this option to avoid consistency issues.
+        if (!view_builders.empty() && ksm->uses_tablets() && !sp.data_dictionary().get_config().rf_rack_valid_keyspaces()) {
+            co_return api_error::validation("GlobalSecondaryIndexes and LocalSecondaryIndexes with tablets require the rf_rack_valid_keyspaces option to be enabled.");
+        }
         try {
             schema_mutations = service::prepare_new_keyspace_announcement(sp.local_db(), ksm, ts);
         } catch (exceptions::already_exists_exception&) {
@@ -1887,8 +1892,8 @@ future<executor::request_return_type> executor::create_table(client_state& clien
         std::string def_type = type_to_string(def.type);
         for (auto it = attribute_definitions.Begin(); it != attribute_definitions.End(); ++it) {
             const rjson::value& attribute_info = *it;
-            if (attribute_info["AttributeName"].GetString() == def.name_as_text()) {
-                auto type = attribute_info["AttributeType"].GetString();
+            if (rjson::to_string_view(attribute_info["AttributeName"]) == def.name_as_text()) {
+                std::string_view type = rjson::to_string_view(attribute_info["AttributeType"]);
                 if (type != def_type) {
                     throw api_error::validation(fmt::format("AttributeDefinitions redefined {} to {} already a key attribute of type {} in this table", def.name_as_text(), type, def_type));
                 }
@@ -2019,6 +2024,10 @@ future<executor::request_return_type> executor::update_table(client_state& clien
                             co_return api_error::validation(fmt::format(
                                 "LSI {} already exists in table {}, can't use same name for GSI", index_name, table_name));
                         }
+                        if (p.local().local_db().find_keyspace(keyspace_name).get_replication_strategy().uses_tablets() &&
+                                !p.local().data_dictionary().get_config().rf_rack_valid_keyspaces()) {
+                            co_return api_error::validation("GlobalSecondaryIndexes with tablets require the rf_rack_valid_keyspaces option to be enabled.");
+                        }
 
                         elogger.trace("Adding GSI {}", index_name);
                         // FIXME: read and handle "Projection" parameter. This will
@@ -2362,7 +2371,7 @@ put_or_delete_item::put_or_delete_item(const rjson::value& item, schema_ptr sche
     _cells = std::vector<cell>();
     _cells->reserve(item.MemberCount());
     for (auto it = item.MemberBegin(); it != item.MemberEnd(); ++it) {
-        bytes column_name = to_bytes(it->name.GetString());
+        bytes column_name = to_bytes(rjson::to_string_view(it->name));
         validate_value(it->value, "PutItem");
         const column_definition* cdef = find_attribute(*schema, column_name);
         validate_attr_name_length("", column_name.size(), cdef && cdef->is_primary_key());
@@ -2783,10 +2792,10 @@ static void verify_all_are_used(const rjson::value* field,
         return;
     }
     for (auto it = field->MemberBegin(); it != field->MemberEnd(); ++it) {
-        if (!used.contains(it->name.GetString())) {
+        if (!used.contains(rjson::to_string(it->name))) {
             throw api_error::validation(
                 format("{} has spurious '{}', not used in {}",
-                    field_name, it->name.GetString(), operation));
+                    field_name, rjson::to_string_view(it->name), operation));
         }
     }
 }
@@ -3000,7 +3009,7 @@ future<executor::request_return_type> executor::delete_item(client_state& client
 }
 
 static schema_ptr get_table_from_batch_request(const service::storage_proxy& proxy, const rjson::value::ConstMemberIterator& batch_request) {
-    sstring table_name = batch_request->name.GetString(); // JSON keys are always strings
+    sstring table_name = rjson::to_sstring(batch_request->name); // JSON keys are always strings
     try {
         return proxy.data_dictionary().find_schema(sstring(executor::KEYSPACE_NAME_PREFIX) + table_name, table_name);
     } catch(data_dictionary::no_such_column_family&) {
@@ -3055,17 +3064,44 @@ public:
     }
 };
 
-static future<> cas_write(service::storage_proxy& proxy, schema_ptr schema, service::cas_shard cas_shard, const dht::decorated_key& dk, const std::vector<put_or_delete_item>& mutation_builders,
-        service::client_state& client_state, tracing::trace_state_ptr trace_state, service_permit permit) {
+future<> executor::cas_write(schema_ptr schema, service::cas_shard cas_shard, const dht::decorated_key& dk,
+        const std::vector<put_or_delete_item>& mutation_builders, service::client_state& client_state,
+        tracing::trace_state_ptr trace_state, service_permit permit)
+{
+    if (!cas_shard.this_shard()) {
+        _stats.shard_bounce_for_lwt++;
+        return container().invoke_on(cas_shard.shard(), _ssg,
+                    [cs = client_state.move_to_other_shard(),
+                    &mb = mutation_builders,
+                    &dk,
+                    ks = schema->ks_name(),
+                    cf = schema->cf_name(),
+                    gt = tracing::global_trace_state_ptr(trace_state),
+                    permit = std::move(permit)]
+                    (executor& self) mutable {
+            return do_with(cs.get(), [&mb, &dk, ks = std::move(ks), cf = std::move(cf),
+                                    trace_state = tracing::trace_state_ptr(gt), &self]
+                                    (service::client_state& client_state) mutable {
+                auto schema = self._proxy.data_dictionary().find_schema(ks, cf);
+                service::cas_shard cas_shard(*schema, dk.token());
+
+                //FIXME: Instead of passing empty_service_permit() to the background operation,
+                // the current permit's lifetime should be prolonged, so that it's destructed
+                // only after all background operations are finished as well.
+                return self.cas_write(schema, std::move(cas_shard), dk, mb, client_state, std::move(trace_state), empty_service_permit());
+            });
+        });
+    }
+
     auto timeout = executor::default_timeout();
     auto op = std::make_unique<put_or_delete_item_cas_request>(schema, mutation_builders);
     auto* op_ptr = op.get();
     auto cdc_opts = cdc::per_request_options{
         .alternator = true,
         .alternator_streams_increased_compatibility =
-                schema->cdc_options().enabled() && proxy.data_dictionary().get_config().alternator_streams_increased_compatibility(),
+                schema->cdc_options().enabled() && _proxy.data_dictionary().get_config().alternator_streams_increased_compatibility(),
     };
-    return proxy.cas(schema, std::move(cas_shard), *op_ptr, nullptr, to_partition_ranges(dk),
+    return _proxy.cas(schema, std::move(cas_shard), *op_ptr, nullptr, to_partition_ranges(dk),
             {timeout, std::move(permit), client_state, trace_state},
             db::consistency_level::LOCAL_SERIAL, db::consistency_level::LOCAL_QUORUM,
             timeout, timeout, true, std::move(cdc_opts)).finally([op = std::move(op)]{}).discard_result();
@@ -3091,13 +3127,11 @@ struct schema_decorated_key_equal {
 
 // FIXME: if we failed writing some of the mutations, need to return a list
 // of these failed mutations rather than fail the whole write (issue #5650).
-static future<> do_batch_write(service::storage_proxy& proxy,
-        smp_service_group ssg,
+future<> executor::do_batch_write(
         std::vector<std::pair<schema_ptr, put_or_delete_item>> mutation_builders,
         service::client_state& client_state,
         tracing::trace_state_ptr trace_state,
-        service_permit permit,
-        stats& stats) {
+        service_permit permit) {
     if (mutation_builders.empty()) {
         return make_ready_future<>();
     }
@@ -3119,7 +3153,7 @@ static future<> do_batch_write(service::storage_proxy& proxy,
             mutations.push_back(b.second.build(b.first, now));
             any_cdc_enabled |= b.first->cdc_options().enabled();
         }
-        return proxy.mutate(std::move(mutations),
+        return _proxy.mutate(std::move(mutations),
                 db::consistency_level::LOCAL_QUORUM,
                 executor::default_timeout(),
                 trace_state,
@@ -3128,7 +3162,7 @@ static future<> do_batch_write(service::storage_proxy& proxy,
                 false,
                 cdc::per_request_options{
                     .alternator = true,
-                    .alternator_streams_increased_compatibility = any_cdc_enabled && proxy.data_dictionary().get_config().alternator_streams_increased_compatibility(),
+                    .alternator_streams_increased_compatibility = any_cdc_enabled && _proxy.data_dictionary().get_config().alternator_streams_increased_compatibility(),
                 });
     } else {
         // Do the write via LWT:
@@ -3140,46 +3174,35 @@ static future<> do_batch_write(service::storage_proxy& proxy,
             schema_decorated_key_hash, 
             schema_decorated_key_equal>;
         auto key_builders = std::make_unique<map_type>(1, schema_decorated_key_hash{}, schema_decorated_key_equal{});
-        for (auto& b : mutation_builders) {
-            auto dk = dht::decorate_key(*b.first, b.second.pk());
-            auto [it, added] = key_builders->try_emplace(schema_decorated_key{b.first, dk});
+        for (auto&& b : std::move(mutation_builders)) {
+            auto [it, added] = key_builders->try_emplace(schema_decorated_key {
+                .schema = b.first,
+                .dk = dht::decorate_key(*b.first, b.second.pk())
+            });
             it->second.push_back(std::move(b.second));
         }
         auto* key_builders_ptr = key_builders.get();
-        return parallel_for_each(*key_builders_ptr, [&proxy, &client_state, &stats, trace_state, ssg, permit = std::move(permit)] (const auto& e) {
-            stats.write_using_lwt++;
+        return parallel_for_each(*key_builders_ptr, [this, &client_state, trace_state, permit = std::move(permit)] (const auto& e) {
+            _stats.write_using_lwt++;
             auto desired_shard = service::cas_shard(*e.first.schema, e.first.dk.token());
-            if (desired_shard.this_shard()) {
-                return cas_write(proxy, e.first.schema, std::move(desired_shard), e.first.dk, e.second, client_state, trace_state, permit);
-            } else {
-                stats.shard_bounce_for_lwt++;
-                return proxy.container().invoke_on(desired_shard.shard(), ssg,
-                            [cs = client_state.move_to_other_shard(),
-                             &mb = e.second,
-                             &dk = e.first.dk,
-                             ks = e.first.schema->ks_name(),
-                             cf = e.first.schema->cf_name(),
-                             gt =  tracing::global_trace_state_ptr(trace_state),
-                             permit = std::move(permit)]
-                            (service::storage_proxy& proxy) mutable {
-                    return do_with(cs.get(), [&proxy, &mb, &dk, ks = std::move(ks), cf = std::move(cf),
-                                              trace_state = tracing::trace_state_ptr(gt)]
-                                              (service::client_state& client_state) mutable {
-                        auto schema = proxy.data_dictionary().find_schema(ks, cf);
+            auto s = e.first.schema;
 
-                        // The desired_shard on the original shard remains alive for the duration
-                        // of cas_write on this shard and prevents any tablet operations.
-                        // However, we need a local instance of cas_shard on this shard
-                        // to pass it to sp::cas, so we just create a new one.
-                        service::cas_shard cas_shard(*schema, dk.token());
-
-                        //FIXME: Instead of passing empty_service_permit() to the background operation,
-                        // the current permit's lifetime should be prolonged, so that it's destructed
-                        // only after all background operations are finished as well.
-                        return cas_write(proxy, schema, std::move(cas_shard), dk, mb, client_state, std::move(trace_state), empty_service_permit());
-                    });
-                }).finally([desired_shard = std::move(desired_shard)]{});
-            }
+            static const auto* injection_name = "alternator_executor_batch_write_wait";
+            return utils::get_local_injector().inject(injection_name, [s = std::move(s)] (auto& handler) -> future<> {
+                const auto ks = handler.get("keyspace");
+                const auto cf = handler.get("table");
+                const auto shard = std::atoll(handler.get("shard")->data());
+                if (ks == s->ks_name() && cf == s->cf_name() && shard == this_shard_id()) {
+                    elogger.info("{}: hit", injection_name);
+                    co_await handler.wait_for_message(std::chrono::steady_clock::now() + std::chrono::minutes{5});
+                    elogger.info("{}: continue", injection_name);
+                }
+            }).then([&e, desired_shard = std::move(desired_shard),
+                 &client_state, trace_state = std::move(trace_state), permit = std::move(permit), this]() mutable
+            {
+                return cas_write(e.first.schema, std::move(desired_shard), e.first.dk,
+                    std::move(e.second), client_state, std::move(trace_state), std::move(permit));
+            });
         }).finally([key_builders = std::move(key_builders)]{});
     }
 }
@@ -3327,7 +3350,7 @@ future<executor::request_return_type> executor::batch_write_item(client_state& c
     _stats.wcu_total[stats::DELETE_ITEM] += wcu_delete_units;
     _stats.api_operations.batch_write_item_batch_total += total_items;
     _stats.api_operations.batch_write_item_histogram.add(total_items);
-    co_await do_batch_write(_proxy, _ssg, std::move(mutation_builders), client_state, trace_state, std::move(permit), _stats);
+    co_await do_batch_write(std::move(mutation_builders), client_state, trace_state, std::move(permit));
     // FIXME: Issue #5650: If we failed writing some of the updates,
     // need to return a list of these failed updates in UnprocessedItems
     // rather than fail the whole write (issue #5650).
@@ -3372,7 +3395,7 @@ static bool hierarchy_filter(rjson::value& val, const attribute_path_map_node<T>
         }
         rjson::value newv = rjson::empty_object();
         for (auto it = v.MemberBegin(); it != v.MemberEnd(); ++it) {
-            std::string attr = it->name.GetString();
+            std::string attr = rjson::to_string(it->name);
             auto x = members.find(attr);
             if (x != members.end()) {
                 if (x->second) {
@@ -3592,7 +3615,7 @@ static std::optional<attrs_to_get> calculate_attrs_to_get(const rjson::value& re
         const rjson::value& attributes_to_get = req["AttributesToGet"];
         attrs_to_get ret;
         for (auto it = attributes_to_get.Begin(); it != attributes_to_get.End(); ++it) {
-            attribute_path_map_add("AttributesToGet", ret, it->GetString());
+            attribute_path_map_add("AttributesToGet", ret, rjson::to_string(*it));
             validate_attr_name_length("AttributesToGet", it->GetStringLength(), false);
         }
         if (ret.empty()) {
@@ -4258,12 +4281,12 @@ inline void update_item_operation::apply_attribute_updates(const std::unique_ptr
         attribute_collector& modified_attrs, bool& any_updates, bool& any_deletes) const {
     for (auto it = _attribute_updates->MemberBegin(); it != _attribute_updates->MemberEnd(); ++it) {
         // Note that it.key() is the name of the column, *it is the operation
-        bytes column_name = to_bytes(it->name.GetString());
+        bytes column_name = to_bytes(rjson::to_string_view(it->name));
         const column_definition* cdef = _schema->get_column_definition(column_name);
         if (cdef && cdef->is_primary_key()) {
-            throw api_error::validation(format("UpdateItem cannot update key column {}", it->name.GetString()));
+            throw api_error::validation(format("UpdateItem cannot update key column {}", rjson::to_string_view(it->name)));
         }
-        std::string action = (it->value)["Action"].GetString();
+        std::string action = rjson::to_string((it->value)["Action"]);
         if (action == "DELETE") {
             // The DELETE operation can do two unrelated tasks. Without a
             // "Value" option, it is used to delete an attribute. With a
@@ -5460,7 +5483,7 @@ calculate_bounds_conditions(schema_ptr schema, const rjson::value& conditions) {
     std::vector<query::clustering_range> ck_bounds;
 
     for (auto it = conditions.MemberBegin(); it != conditions.MemberEnd(); ++it) {
-        std::string key = it->name.GetString();
+        sstring key = rjson::to_sstring(it->name);
         const rjson::value& condition = it->value;
 
         const rjson::value& comp_definition = rjson::get(condition, "ComparisonOperator");
@@ -5468,13 +5491,13 @@ calculate_bounds_conditions(schema_ptr schema, const rjson::value& conditions) {
 
         const column_definition& pk_cdef = schema->partition_key_columns().front();
         const column_definition* ck_cdef = schema->clustering_key_size() > 0 ? &schema->clustering_key_columns().front() : nullptr;
-        if (sstring(key) == pk_cdef.name_as_text()) {
+        if (key == pk_cdef.name_as_text()) {
             if (!partition_ranges.empty()) {
                 throw api_error::validation("Currently only a single restriction per key is allowed");
             }
             partition_ranges.push_back(calculate_pk_bound(schema, pk_cdef, comp_definition, attr_list));
         }
-        if (ck_cdef && sstring(key) == ck_cdef->name_as_text()) {
+        if (ck_cdef && key == ck_cdef->name_as_text()) {
             if (!ck_bounds.empty()) {
                 throw api_error::validation("Currently only a single restriction per key is allowed");
             }
@@ -5875,7 +5898,7 @@ future<executor::request_return_type> executor::list_tables(client_state& client
 
     rjson::value* exclusive_start_json = rjson::find(request, "ExclusiveStartTableName");
     rjson::value* limit_json = rjson::find(request, "Limit");
-    std::string exclusive_start = exclusive_start_json ? exclusive_start_json->GetString() : "";
+    std::string exclusive_start = exclusive_start_json ? rjson::to_string(*exclusive_start_json) : "";
     int limit = limit_json ? limit_json->GetInt() : 100;
     if (limit < 1 || limit > 100) {
         co_return api_error::validation("Limit must be greater than 0 and no greater than 100");

alternator/executor.hh

@@ -40,6 +40,7 @@ namespace cql3::selection {
 
 namespace service {
     class storage_proxy;
+    class cas_shard;
 }
 
 namespace cdc {
@@ -57,6 +58,7 @@ class schema_builder;
 namespace alternator {
 
 class rmw_operation;
+class put_or_delete_item;
 
 schema_ptr get_table(service::storage_proxy& proxy, const rjson::value& request);
 bool is_alternator_keyspace(const sstring& ks_name);
@@ -219,6 +221,16 @@ private:
 
     static void describe_key_schema(rjson::value& parent, const schema&, std::unordered_map<std::string,std::string> * = nullptr, const std::map<sstring, sstring> *tags = nullptr);
 
+    future<> do_batch_write(
+        std::vector<std::pair<schema_ptr, put_or_delete_item>> mutation_builders,
+        service::client_state& client_state,
+        tracing::trace_state_ptr trace_state,
+        service_permit permit);
+
+    future<> cas_write(schema_ptr schema, service::cas_shard cas_shard, const dht::decorated_key& dk,
+        const std::vector<put_or_delete_item>& mutation_builders, service::client_state& client_state,
+        tracing::trace_state_ptr trace_state, service_permit permit);
+
 public:
     static void describe_key_schema(rjson::value& parent, const schema& schema, std::unordered_map<std::string,std::string>&, const std::map<sstring, sstring> *tags = nullptr);
 

alternator/serialization.cc

@@ -496,7 +496,7 @@ const std::pair<std::string, const rjson::value*> unwrap_set(const rjson::value&
         return {"", nullptr};
     }
     auto it = v.MemberBegin();
-    const std::string it_key = it->name.GetString();
+    const std::string it_key = rjson::to_string(it->name);
     if (it_key != "SS" && it_key != "BS" && it_key != "NS") {
         return {std::move(it_key), nullptr};
     }

alternator/server.cc

@@ -708,8 +708,12 @@ future<executor::request_return_type> server::handle_api_request(std::unique_ptr
     // As long as the system_clients_entry object is alive, this request will
     // be visible in the "system.clients" virtual table. When requested, this
     // entry will be formatted by server::ongoing_request::make_client_data().
+    auto user_agent_header = co_await _connection_options_keys_and_values.get_or_load(req->get_header("User-Agent"), [] (const client_options_cache_key_type&) {
+        return make_ready_future<options_cache_value_type>(options_cache_value_type{});
+    });
+
     auto system_clients_entry = _ongoing_requests.emplace(
-        req->get_client_address(), req->get_header("User-Agent"),
+        req->get_client_address(), std::move(user_agent_header),
         username, current_scheduling_group(),
         req->get_protocol_name() == "https");
 
@@ -985,10 +989,10 @@ client_data server::ongoing_request::make_client_data() const {
     return cd;
 }
 
-future<utils::chunked_vector<client_data>> server::get_client_data() {
-    utils::chunked_vector<client_data> ret;
+future<utils::chunked_vector<foreign_ptr<std::unique_ptr<client_data>>>> server::get_client_data() {
+    utils::chunked_vector<foreign_ptr<std::unique_ptr<client_data>>> ret;
     co_await _ongoing_requests.for_each_gently([&ret] (const ongoing_request& r) {
-        ret.emplace_back(r.make_client_data());
+        ret.emplace_back(make_foreign(std::make_unique<client_data>(r.make_client_data())));
     });
     co_return ret;
 }

alternator/server.hh

@@ -55,6 +55,7 @@ class server : public peering_sharded_service<server> {
     // though it isn't really relevant for Alternator which defines its own
     // timeouts separately. We can create this object only once.
     updateable_timeout_config _timeout_config;
+    client_options_cache_type _connection_options_keys_and_values;
 
     alternator_callbacks_map _callbacks;
 
@@ -88,7 +89,7 @@ class server : public peering_sharded_service<server> {
     // is called when reading the "system.clients" virtual table.
     struct ongoing_request {
         socket_address _client_address;
-        sstring _user_agent;
+        client_options_cache_entry_type _user_agent;
         sstring _username;
         scheduling_group _scheduling_group;
         bool _is_https;
@@ -107,7 +108,7 @@ public:
     // table "system.clients" is read. It is expected to generate a list of
     // clients connected to this server (on this shard). This function is
     // called by alternator::controller::get_client_data().
-    future<utils::chunked_vector<client_data>> get_client_data();
+    future<utils::chunked_vector<foreign_ptr<std::unique_ptr<client_data>>>> get_client_data();
 private:
     void set_routes(seastar::httpd::routes& r);
     // If verification succeeds, returns the authenticated user's username

alternator/ttl.cc

@@ -93,7 +93,7 @@ future<executor::request_return_type> executor::update_time_to_live(client_state
     if (v->GetStringLength() < 1 || v->GetStringLength() > 255) {
         co_return api_error::validation("The length of AttributeName must be between 1 and 255");
     }
-    sstring attribute_name(v->GetString(), v->GetStringLength());
+    sstring attribute_name = rjson::to_sstring(*v);
 
     co_await verify_permission(_enforce_authorization, _warn_authorization, client_state, schema, auth::permission::ALTER, _stats);
     co_await db::modify_tags(_mm, schema->ks_name(), schema->cf_name(), [&](std::map<sstring, sstring>& tags_map) {

api/CMakeLists.txt

@@ -31,6 +31,7 @@ set(swagger_files
   api-doc/column_family.json
   api-doc/commitlog.json
   api-doc/compaction_manager.json
+  api-doc/client_routes.json
   api-doc/config.json
   api-doc/cql_server_test.json
   api-doc/endpoint_snitch_info.json
@@ -68,6 +69,7 @@ target_sources(api
   PRIVATE
     api.cc
     cache_service.cc
+    client_routes.cc
     collectd.cc
     column_family.cc
     commitlog.cc

api/api-doc/client_routes.def.json

@@ -0,0 +1,23 @@
+    , "client_routes_entry": {
+        "id": "client_routes_entry",
+        "summary": "An entry storing client routes",
+        "properties": {
+            "connection_id": {"type": "string"},
+            "host_id": {"type": "string", "format": "uuid"},
+            "address": {"type": "string"},
+            "port": {"type": "integer"},
+            "tls_port": {"type": "integer"},
+            "alternator_port": {"type": "integer"},
+            "alternator_https_port": {"type": "integer"}
+        },
+        "required": ["connection_id", "host_id", "address"]
+    }
+    , "client_routes_key": {
+        "id": "client_routes_key",
+        "summary": "A key of client_routes_entry",
+        "properties": {
+            "connection_id": {"type": "string"},
+            "host_id": {"type": "string", "format": "uuid"}
+        }
+    }
+

api/api-doc/client_routes.json

@@ -0,0 +1,74 @@
+    , "/v2/client-routes":{
+        "get": {
+            "description":"List all client route entries",
+            "operationId":"get_client_routes",
+            "tags":["client_routes"],
+            "produces":[
+                "application/json"
+            ],
+            "parameters":[],
+            "responses":{
+                "200":{
+                    "schema":{
+                        "type":"array",
+                        "items":{ "$ref":"#/definitions/client_routes_entry" }
+                    }
+                },
+                "default":{
+                    "description":"unexpected error",
+                    "schema":{"$ref":"#/definitions/ErrorModel"}
+                }
+            }
+        },
+        "post": {
+            "description":"Upsert one or more client route entries",
+            "operationId":"set_client_routes",
+            "tags":["client_routes"],
+            "parameters":[
+                {
+                    "name":"body",
+                    "in":"body",
+                    "required":true,
+                    "schema":{
+                        "type":"array",
+                        "items":{ "$ref":"#/definitions/client_routes_entry" }
+                    }
+                }
+            ],
+            "responses":{
+                "200":{ "description": "OK" },
+                "default":{
+                    "description":"unexpected error",
+                    "schema":{ "$ref":"#/definitions/ErrorModel" }
+                }
+            }
+        },
+        "delete": {
+            "description":"Delete one or more client route entries",
+            "operationId":"delete_client_routes",
+            "tags":["client_routes"],
+            "parameters":[
+                {
+                    "name":"body",
+                    "in":"body",
+                    "required":true,
+                    "schema":{
+                        "type":"array",
+                        "items":{ "$ref":"#/definitions/client_routes_key" }
+                    }
+                }
+            ],
+            "responses":{
+                "200":{
+                    "description": "OK"
+                },
+                "default":{
+                    "description":"unexpected error",
+                    "schema":{
+                        "$ref":"#/definitions/ErrorModel"
+                    }
+                }
+            }
+        }
+    }
+

api/api-doc/storage_service.json

@@ -3051,7 +3051,7 @@
                   },
                   {
                      "name":"incremental_mode",
-                     "description":"Set the incremental repair mode. Can be 'disabled', 'incremental', or 'full'. 'incremental': The incremental repair logic is enabled. Unrepaired sstables will be included for repair. Repaired sstables will be skipped. The incremental repair states will be updated after repair. 'full': The incremental repair logic is enabled. Both repaired and unrepaired sstables will be included for repair. The incremental repair states will be updated after repair. 'disabled': The incremental repair logic is disabled completely. The incremental repair states, e.g., repaired_at in sstables and sstables_repaired_at in the system.tablets table, will not be updated after repair. When the option is not provided, it defaults to incremental mode.",
+                     "description":"Set the incremental repair mode. Can be 'disabled', 'incremental', or 'full'. 'incremental': The incremental repair logic is enabled. Unrepaired sstables will be included for repair. Repaired sstables will be skipped. The incremental repair states will be updated after repair. 'full': The incremental repair logic is enabled. Both repaired and unrepaired sstables will be included for repair. The incremental repair states will be updated after repair. 'disabled': The incremental repair logic is disabled completely. The incremental repair states, e.g., repaired_at in sstables and sstables_repaired_at in the system.tablets table, will not be updated after repair. When the option is not provided, it defaults to 'disabled' mode.",
                      "required":false,
                      "allowMultiple":false,
                      "type":"string",

api/api.cc

@@ -37,6 +37,7 @@
 #include "raft.hh"
 #include "gms/gossip_address_map.hh"
 #include "service_levels.hh"
+#include "client_routes.hh"
 
 logging::logger apilog("api");
 
@@ -67,9 +68,11 @@ future<> set_server_init(http_context& ctx) {
         rb02->set_api_doc(r);
         rb02->register_api_file(r, "swagger20_header");
         rb02->register_api_file(r, "metrics");
+        rb02->register_api_file(r, "client_routes");
         rb->register_function(r, "system",
                 "The system related API");
         rb02->add_definitions_file(r, "metrics");
+        rb02->add_definitions_file(r, "client_routes");
         set_system(ctx, r);
         rb->register_function(r, "error_injection",
             "The error injection API");
@@ -129,6 +132,16 @@ future<> unset_server_storage_service(http_context& ctx) {
     return ctx.http_server.set_routes([&ctx] (routes& r) { unset_storage_service(ctx, r); });
 }
 
+future<> set_server_client_routes(http_context& ctx, sharded<service::client_routes_service>& cr) {
+    return ctx.http_server.set_routes([&ctx, &cr] (routes& r) {
+        set_client_routes(ctx, r, cr);
+    });
+}
+
+future<> unset_server_client_routes(http_context& ctx) {
+    return ctx.http_server.set_routes([&ctx] (routes& r) { unset_client_routes(ctx, r); });
+}
+
 future<> set_load_meter(http_context& ctx, service::load_meter& lm) {
     return ctx.http_server.set_routes([&ctx, &lm] (routes& r) { set_load_meter(ctx, r, lm); });
 }

api/api_init.hh

@@ -29,6 +29,7 @@ class storage_proxy;
 class storage_service;
 class raft_group0_client;
 class raft_group_registry;
+class client_routes_service;
 
 } // namespace service
 
@@ -99,6 +100,8 @@ future<> set_server_snitch(http_context& ctx, sharded<locator::snitch_ptr>& snit
 future<> unset_server_snitch(http_context& ctx);
 future<> set_server_storage_service(http_context& ctx, sharded<service::storage_service>& ss, service::raft_group0_client&);
 future<> unset_server_storage_service(http_context& ctx);
+future<> set_server_client_routes(http_context& ctx, sharded<service::client_routes_service>& cr);
+future<> unset_server_client_routes(http_context& ctx);
 future<> set_server_sstables_loader(http_context& ctx, sharded<sstables_loader>& sst_loader);
 future<> unset_server_sstables_loader(http_context& ctx);
 future<> set_server_view_builder(http_context& ctx, sharded<db::view::view_builder>& vb, sharded<gms::gossiper>& g);

api/client_routes.cc

@@ -0,0 +1,176 @@
+/*
+ * Copyright (C) 2025-present ScyllaDB
+ *
+ */
+
+/*
+ * SPDX-License-Identifier: LicenseRef-ScyllaDB-Source-Available-1.0
+ */
+
+ #include <seastar/http/short_streams.hh>
+
+#include "client_routes.hh"
+#include "api/api.hh"
+#include "service/storage_service.hh"
+#include "service/client_routes.hh"
+#include "utils/rjson.hh"
+
+
+#include "api/api-doc/client_routes.json.hh"
+
+using namespace seastar::httpd;
+using namespace std::chrono_literals;
+using namespace json;
+
+extern logging::logger apilog;
+
+namespace api {
+
+static void validate_client_routes_endpoint(sharded<service::client_routes_service>& cr, sstring endpoint_name) {
+    if (!cr.local().get_feature_service().client_routes) {
+        apilog.warn("{}: called before the cluster feature was enabled", endpoint_name);
+        throw std::runtime_error(fmt::format("{} requires all nodes to support the CLIENT_ROUTES cluster feature", endpoint_name));
+    }
+}
+
+static sstring parse_string(const char* name, rapidjson::Value const& v) {
+    const auto it = v.FindMember(name);
+    if (it == v.MemberEnd()) {
+        throw bad_param_exception(fmt::format("Missing '{}'", name));
+    }
+    if (!it->value.IsString()) {
+        throw bad_param_exception(fmt::format("'{}' must be a string", name));
+    }
+    return {it->value.GetString(), it->value.GetStringLength()};
+}
+
+static std::optional<uint32_t> parse_port(const char* name, rapidjson::Value const& v) {
+    const auto it = v.FindMember(name);
+    if (it == v.MemberEnd()) {
+        return std::nullopt;
+    }
+    if (!it->value.IsInt()) {
+        throw bad_param_exception(fmt::format("'{}' must be an integer", name));
+    }
+    auto port = it->value.GetInt();
+    if (port < 1 || port > 65535) {
+        throw bad_param_exception(fmt::format("'{}' value={} is outside the allowed port range", name, port));
+    }
+    return port;
+}
+
+static std::vector<service::client_routes_service::client_route_entry> parse_set_client_array(const rapidjson::Document& root) {
+    if (!root.IsArray()) {
+        throw bad_param_exception("Body must be a JSON array");
+    }
+
+    std::vector<service::client_routes_service::client_route_entry> v;
+    v.reserve(root.GetArray().Size());
+    for (const auto& element : root.GetArray()) {
+        if (!element.IsObject()) { throw bad_param_exception("Each element must be object"); }
+
+        const auto port = parse_port("port", element);
+        const auto tls_port = parse_port("tls_port", element);
+        const auto alternator_port = parse_port("alternator_port", element);
+        const auto alternator_https_port = parse_port("alternator_https_port", element);
+
+        if (!port.has_value() && !tls_port.has_value() && !alternator_port.has_value() && !alternator_https_port.has_value()) {
+            throw bad_param_exception("At least one port field ('port', 'tls_port', 'alternator_port', 'alternator_https_port') must be specified");
+        }
+
+        v.emplace_back(
+            parse_string("connection_id", element),
+            utils::UUID{parse_string("host_id", element)},
+            parse_string("address", element),
+            port,
+            tls_port,
+            alternator_port,
+            alternator_https_port
+        );
+    }
+
+    return v;
+}
+
+static
+future<json::json_return_type>
+rest_set_client_routes(http_context& ctx, sharded<service::client_routes_service>& cr, std::unique_ptr<http::request> req) {
+    validate_client_routes_endpoint(cr, "rest_set_client_routes");
+
+    rapidjson::Document root;
+    auto content = co_await util::read_entire_stream_contiguous(*req->content_stream);
+    root.Parse(content.c_str());
+
+    co_await cr.local().set_client_routes(parse_set_client_array(root));
+    co_return seastar::json::json_void();
+}
+
+static std::vector<service::client_routes_service::client_route_key> parse_delete_client_array(const rapidjson::Document& root) {
+    if (!root.IsArray()) {
+        throw bad_param_exception("Body must be a JSON array");
+    }
+
+    std::vector<service::client_routes_service::client_route_key> v;
+    v.reserve(root.GetArray().Size());
+    for (const auto& element : root.GetArray()) {
+        v.emplace_back(
+            parse_string("connection_id", element),
+            utils::UUID{parse_string("host_id", element)}
+        );
+    }
+
+    return v;
+}
+
+static
+future<json::json_return_type>
+rest_delete_client_routes(http_context& ctx, sharded<service::client_routes_service>& cr, std::unique_ptr<http::request> req) {
+    validate_client_routes_endpoint(cr, "delete_client_routes");
+
+    rapidjson::Document root;
+    auto content = co_await util::read_entire_stream_contiguous(*req->content_stream);
+    root.Parse(content.c_str());
+
+    co_await cr.local().delete_client_routes(parse_delete_client_array(root));
+    co_return seastar::json::json_void();
+}
+
+static
+future<json::json_return_type>
+rest_get_client_routes(http_context& ctx, sharded<service::client_routes_service>& cr, std::unique_ptr<http::request> req) {
+    validate_client_routes_endpoint(cr, "get_client_routes");
+
+    co_return co_await cr.invoke_on(0, [] (service::client_routes_service& cr) -> future<json::json_return_type> {
+        co_return json::json_return_type(stream_range_as_array(co_await cr.get_client_routes(), [](const service::client_routes_service::client_route_entry & entry) {
+            seastar::httpd::client_routes_json::client_routes_entry obj;
+            obj.connection_id = entry.connection_id;
+            obj.host_id = fmt::to_string(entry.host_id);
+            obj.address = entry.address;
+            if (entry.port.has_value()) { obj.port = entry.port.value(); }
+            if (entry.tls_port.has_value()) { obj.tls_port = entry.tls_port.value(); }
+            if (entry.alternator_port.has_value()) { obj.alternator_port = entry.alternator_port.value(); }
+            if (entry.alternator_https_port.has_value()) { obj.alternator_https_port = entry.alternator_https_port.value(); }
+            return obj;
+        }));
+    });
+}
+
+void set_client_routes(http_context& ctx, routes& r, sharded<service::client_routes_service>& cr) {
+    seastar::httpd::client_routes_json::set_client_routes.set(r, [&ctx, &cr] (std::unique_ptr<seastar::http::request> req) {
+        return rest_set_client_routes(ctx, cr, std::move(req));
+    });
+    seastar::httpd::client_routes_json::delete_client_routes.set(r, [&ctx, &cr] (std::unique_ptr<seastar::http::request> req) {
+        return rest_delete_client_routes(ctx, cr, std::move(req));
+    });
+    seastar::httpd::client_routes_json::get_client_routes.set(r, [&ctx, &cr] (std::unique_ptr<seastar::http::request> req) {
+        return rest_get_client_routes(ctx, cr, std::move(req));
+    });
+}
+
+void unset_client_routes(http_context& ctx, routes& r) {
+    seastar::httpd::client_routes_json::set_client_routes.unset(r);
+    seastar::httpd::client_routes_json::delete_client_routes.unset(r);
+    seastar::httpd::client_routes_json::get_client_routes.unset(r);
+}
+
+}

api/client_routes.hh

@@ -0,0 +1,20 @@
+/*
+ * Copyright (C) 2025-present ScyllaDB
+ *
+ */
+
+/*
+ * SPDX-License-Identifier: LicenseRef-ScyllaDB-Source-Available-1.0
+ */
+#pragma once
+
+#include <seastar/core/sharded.hh>
+#include <seastar/json/json_elements.hh>
+#include "api/api_init.hh"
+
+namespace api {
+
+void set_client_routes(http_context& ctx, httpd::routes& r, sharded<service::client_routes_service>& cr);
+void unset_client_routes(http_context& ctx, httpd::routes& r);
+
+}

api/storage_service.cc

@@ -547,17 +547,13 @@ void set_view_builder(http_context& ctx, routes& r, sharded<db::view::view_build
                 vp.insert(b.second);
             }
         }
-        std::vector<sstring> res;
         replica::database& db = vb.local().get_db();
         auto uuid = validate_table(db, ks, cf_name);
         replica::column_family& cf = db.find_column_family(uuid);
-        res.reserve(cf.get_index_manager().list_indexes().size());
-        for (auto&& i : cf.get_index_manager().list_indexes()) {
-            if (vp.contains(secondary_index::index_table_name(i.metadata().name()))) {
-                res.emplace_back(i.metadata().name());
-            }
-        }
-        co_return res;
+        co_return cf.get_index_manager().list_indexes()
+                | std::views::transform([] (const auto& i) { return i.metadata().name(); })
+                | std::views::filter([&vp] (const auto& n) { return vp.contains(secondary_index::index_table_name(n)); })
+                | std::ranges::to<std::vector>();
     });
 
 }

auth/allow_all_authenticator.cc

@@ -9,7 +9,6 @@
 #include "auth/allow_all_authenticator.hh"
 
 #include "service/migration_manager.hh"
-#include "utils/alien_worker.hh"
 #include "utils/class_registrator.hh"
 
 namespace auth {
@@ -23,7 +22,6 @@ static const class_registrator<
         cql3::query_processor&,
         ::service::raft_group0_client&,
         ::service::migration_manager&,
-        cache&,
-        utils::alien_worker&> registration("org.apache.cassandra.auth.AllowAllAuthenticator");
+        cache&> registration("org.apache.cassandra.auth.AllowAllAuthenticator");
 
 }

auth/allow_all_authenticator.hh

@@ -14,7 +14,6 @@
 #include "auth/authenticator.hh"
 #include "auth/cache.hh"
 #include "auth/common.hh"
-#include "utils/alien_worker.hh"
 
 namespace cql3 {
 class query_processor;
@@ -30,7 +29,7 @@ extern const std::string_view allow_all_authenticator_name;
 
 class allow_all_authenticator final : public authenticator {
 public:
-    allow_all_authenticator(cql3::query_processor&, ::service::raft_group0_client&, ::service::migration_manager&, cache&, utils::alien_worker&) {
+    allow_all_authenticator(cql3::query_processor&, ::service::raft_group0_client&, ::service::migration_manager&, cache&) {
     }
 
     virtual future<> start() override {

auth/cache.cc

@@ -15,6 +15,7 @@
 #include "db/system_keyspace.hh"
 #include "schema/schema.hh"
 #include <iterator>
+#include <seastar/core/abort_source.hh>
 #include <seastar/coroutine/maybe_yield.hh>
 #include <seastar/core/format.hh>
 
@@ -22,9 +23,11 @@ namespace auth {
 
 logging::logger logger("auth-cache");
 
-cache::cache(cql3::query_processor& qp) noexcept
+cache::cache(cql3::query_processor& qp, abort_source& as) noexcept
     : _current_version(0)
-    , _qp(qp) {
+    , _qp(qp)
+    , _loading_sem(1)
+    , _as(as) {
 }
 
 lw_shared_ptr<const cache::role_record> cache::get(const role_name_t& role) const noexcept {
@@ -116,6 +119,8 @@ future<> cache::load_all() {
         co_return;
     }
     SCYLLA_ASSERT(this_shard_id() == 0);
+    auto units = co_await get_units(_loading_sem, 1, _as);
+
     ++_current_version;
 
     logger.info("Loading all roles");
@@ -146,6 +151,9 @@ future<> cache::load_roles(std::unordered_set<role_name_t> roles) {
     if (legacy_mode(_qp)) {
         co_return;
     }
+    SCYLLA_ASSERT(this_shard_id() == 0);
+    auto units = co_await get_units(_loading_sem, 1, _as);
+
     for (const auto& name : roles) {
         logger.info("Loading role {}", name);
         auto role = co_await fetch_role(name);

auth/cache.hh

@@ -8,6 +8,7 @@
 
 #pragma once
 
+#include <seastar/core/abort_source.hh>
 #include <unordered_set>
 #include <unordered_map>
 
@@ -15,6 +16,7 @@
 #include <seastar/core/future.hh>
 #include <seastar/core/sharded.hh>
 #include <seastar/core/shared_ptr.hh>
+#include <seastar/core/semaphore.hh>
 
 #include <absl/container/flat_hash_map.h>
 
@@ -41,7 +43,7 @@ public:
         version_tag_t version; // used for seamless cache reloads
     };
 
-    explicit cache(cql3::query_processor& qp) noexcept;
+    explicit cache(cql3::query_processor& qp, abort_source& as) noexcept;
     lw_shared_ptr<const role_record> get(const role_name_t& role) const noexcept;
     future<> load_all();
     future<> load_roles(std::unordered_set<role_name_t> roles);
@@ -52,6 +54,8 @@ private:
     roles_map _roles;
     version_tag_t _current_version;
     cql3::query_processor& _qp;
+    semaphore _loading_sem;
+    abort_source& _as;
 
     future<lw_shared_ptr<role_record>> fetch_role(const role_name_t& role) const;
     future<> prune_all() noexcept;

auth/certificate_authenticator.cc

@@ -35,14 +35,13 @@ static const class_registrator<auth::authenticator
     , cql3::query_processor&
     , ::service::raft_group0_client&
     , ::service::migration_manager&
-    , auth::cache&
-    , utils::alien_worker&> cert_auth_reg(CERT_AUTH_NAME);
+    , auth::cache&> cert_auth_reg(CERT_AUTH_NAME);
 
 enum class auth::certificate_authenticator::query_source {
     subject, altname
 };
 
-auth::certificate_authenticator::certificate_authenticator(cql3::query_processor& qp, ::service::raft_group0_client&, ::service::migration_manager&, auth::cache&, utils::alien_worker&)
+auth::certificate_authenticator::certificate_authenticator(cql3::query_processor& qp, ::service::raft_group0_client&, ::service::migration_manager&, auth::cache&)
     : _queries([&] {
         auto& conf = qp.db().get_config();
         auto queries = conf.auth_certificate_role_queries();
@@ -77,9 +76,9 @@ auth::certificate_authenticator::certificate_authenticator(cql3::query_processor
                         throw std::invalid_argument(fmt::format("Invalid source: {}", map.at(cfg_source_attr)));
                     }
                     continue;
-                } catch (std::out_of_range&) {
+                } catch (const std::out_of_range&) {
                     // just fallthrough
-                } catch (boost::regex_error&) {
+                } catch (const boost::regex_error&) {
                     std::throw_with_nested(std::invalid_argument(fmt::format("Invalid query expression: {}", map.at(cfg_query_attr))));
                 }
             }

auth/certificate_authenticator.hh

@@ -10,7 +10,6 @@
 #pragma once
 
 #include "auth/authenticator.hh"
-#include "utils/alien_worker.hh"
 #include <boost/regex_fwd.hpp>  // IWYU pragma: keep
 
 namespace cql3 {
@@ -34,7 +33,7 @@ class certificate_authenticator : public authenticator {
     enum class query_source;
     std::vector<std::pair<query_source, boost::regex>> _queries;
 public:
-    certificate_authenticator(cql3::query_processor&, ::service::raft_group0_client&, ::service::migration_manager&, cache&, utils::alien_worker&);
+    certificate_authenticator(cql3::query_processor&, ::service::raft_group0_client&, ::service::migration_manager&, cache&);
     ~certificate_authenticator();
 
     future<> start() override;

auth/common.cc

@@ -94,7 +94,7 @@ static future<> create_legacy_metadata_table_if_missing_impl(
         try {
             co_return co_await mm.announce(co_await ::service::prepare_new_column_family_announcement(qp.proxy(), table, ts),
                     std::move(group0_guard), format("auth: create {} metadata table", table->cf_name()));
-        } catch (exceptions::already_exists_exception&) {}
+        } catch (const exceptions::already_exists_exception&) {}
     }
 }
 

auth/default_authorizer.cc

@@ -256,7 +256,7 @@ future<> default_authorizer::revoke_all(std::string_view role_name, ::service::g
         } else {
             co_await collect_mutations(_qp, mc, query, {sstring(role_name)});
         }
-    } catch (exceptions::request_execution_exception& e) {
+    } catch (const exceptions::request_execution_exception& e) {
         alogger.warn("CassandraAuthorizer failed to revoke all permissions of {}: {}", role_name, e);
     }
 }
@@ -293,13 +293,13 @@ future<> default_authorizer::revoke_all_legacy(const resource& resource) {
                                 [resource](auto ep) {
                     try {
                         std::rethrow_exception(ep);
-                    } catch (exceptions::request_execution_exception& e) {
+                    } catch (const exceptions::request_execution_exception& e) {
                         alogger.warn("CassandraAuthorizer failed to revoke all permissions on {}: {}", resource, e);
                     }
 
                 });
             });
-        } catch (exceptions::request_execution_exception& e) {
+        } catch (const exceptions::request_execution_exception& e) {
             alogger.warn("CassandraAuthorizer failed to revoke all permissions on {}: {}", resource, e);
             return make_ready_future();
         }

auth/password_authenticator.cc

@@ -49,8 +49,7 @@ static const class_registrator<
         cql3::query_processor&,
         ::service::raft_group0_client&,
         ::service::migration_manager&,
-        cache&,
-        utils::alien_worker&> password_auth_reg("org.apache.cassandra.auth.PasswordAuthenticator");
+        cache&> password_auth_reg("org.apache.cassandra.auth.PasswordAuthenticator");
 
 static thread_local auto rng_for_salt = std::default_random_engine(std::random_device{}());
 
@@ -64,14 +63,13 @@ std::string password_authenticator::default_superuser(const db::config& cfg) {
 password_authenticator::~password_authenticator() {
 }
 
-password_authenticator::password_authenticator(cql3::query_processor& qp, ::service::raft_group0_client& g0, ::service::migration_manager& mm, cache& cache, utils::alien_worker& hashing_worker)
+password_authenticator::password_authenticator(cql3::query_processor& qp, ::service::raft_group0_client& g0, ::service::migration_manager& mm, cache& cache)
     : _qp(qp)
     , _group0_client(g0)
     , _migration_manager(mm)
     , _cache(cache)
     , _stopped(make_ready_future<>()) 
     , _superuser(default_superuser(qp.db().get_config()))
-    , _hashing_worker(hashing_worker)
 {}
 
 static bool has_salted_hash(const cql3::untyped_result_set_row& row) {
@@ -330,20 +328,18 @@ future<authenticated_user> password_authenticator::authenticate(
             }
             salted_hash = role->salted_hash;
         }
-        const bool password_match = co_await _hashing_worker.submit<bool>([password = std::move(password), salted_hash] {
-            return passwords::check(password, *salted_hash);
-        });
+        const bool password_match = co_await passwords::check(password, *salted_hash);
         if (!password_match) {
             throw exceptions::authentication_exception("Username and/or password are incorrect");
         }
         co_return username;
-    } catch (std::system_error &) {
+    } catch (const std::system_error &) {
         std::throw_with_nested(exceptions::authentication_exception("Could not verify password"));
-    } catch (exceptions::request_execution_exception& e) {
+    } catch (const exceptions::request_execution_exception& e) {
         std::throw_with_nested(exceptions::authentication_exception(e.what()));
-    } catch (exceptions::authentication_exception& e) {
+    } catch (const exceptions::authentication_exception& e) {
         std::throw_with_nested(e);
-    } catch (exceptions::unavailable_exception& e) {
+    } catch (const exceptions::unavailable_exception& e) {
         std::throw_with_nested(exceptions::authentication_exception(e.get_message()));
     } catch (...) {
         std::throw_with_nested(exceptions::authentication_exception("authentication failed"));

auth/password_authenticator.hh

@@ -18,7 +18,6 @@
 #include "auth/passwords.hh"
 #include "auth/cache.hh"
 #include "service/raft/raft_group0_client.hh"
-#include "utils/alien_worker.hh"
 
 namespace db {
     class config;
@@ -49,13 +48,12 @@ class password_authenticator : public authenticator {
     shared_promise<> _superuser_created_promise;
     // We used to also support bcrypt, SHA-256, and MD5 (ref. scylladb#24524).
     constexpr static auth::passwords::scheme _scheme = passwords::scheme::sha_512;
-    utils::alien_worker& _hashing_worker;
 
 public:
     static db::consistency_level consistency_for_user(std::string_view role_name);
     static std::string default_superuser(const db::config&);
 
-    password_authenticator(cql3::query_processor&, ::service::raft_group0_client&, ::service::migration_manager&, cache&, utils::alien_worker&);
+    password_authenticator(cql3::query_processor&, ::service::raft_group0_client&, ::service::migration_manager&, cache&);
 
     ~password_authenticator();
 

auth/passwords.cc

@@ -7,6 +7,8 @@
  */
 
 #include "auth/passwords.hh"
+#include "utils/crypt_sha512.hh"
+#include <seastar/core/coroutine.hh>
 
 #include <cerrno>
 
@@ -21,27 +23,48 @@ static thread_local crypt_data tlcrypt = {};
 
 namespace detail {
 
+void verify_hashing_output(const char * res) {
+    if (!res || (res[0] == '*')) {
+        throw std::system_error(errno, std::system_category());
+    }
+}
+
 void verify_scheme(scheme scheme) {
     const sstring random_part_of_salt = "aaaabbbbccccdddd";
 
     const sstring salt = sstring(prefix_for_scheme(scheme)) + random_part_of_salt;
     const char* e = crypt_r("fisk", salt.c_str(), &tlcrypt);
-
-    if (e && (e[0] != '*')) {
-        return;
+    try {
+        verify_hashing_output(e);
+    } catch (const std::system_error& ex) {
+        throw no_supported_schemes();
     }
-
-    throw no_supported_schemes();
 }
 
 sstring hash_with_salt(const sstring& pass, const sstring& salt) {
     auto res = crypt_r(pass.c_str(), salt.c_str(), &tlcrypt);
-    if (!res || (res[0] == '*')) {
-        throw std::system_error(errno, std::system_category());
-    }
+    verify_hashing_output(res);
     return res;
 }
 
+seastar::future<sstring> hash_with_salt_async(const sstring& pass, const sstring& salt) {
+    sstring res;
+    // Only SHA-512 hashes for passphrases shorter than 256 bytes can be computed using
+    // the __crypt_sha512 method. For other computations, we fall back to the
+    // crypt_r implementation from `<crypt.h>`, which can stall.
+    if (salt.starts_with(prefix_for_scheme(scheme::sha_512)) && pass.size() <= 255) {
+        char buf[128];
+        const char * output_ptr = co_await __crypt_sha512(pass.c_str(), salt.c_str(), buf);
+        verify_hashing_output(output_ptr);
+        res = output_ptr;
+    } else {
+        const char * output_ptr = crypt_r(pass.c_str(), salt.c_str(), &tlcrypt);
+        verify_hashing_output(output_ptr);
+        res = output_ptr;
+    }
+    co_return res;
+}
+
 std::string_view prefix_for_scheme(scheme c) noexcept {
     switch (c) {
     case scheme::bcrypt_y: return "$2y$";
@@ -58,8 +81,9 @@ no_supported_schemes::no_supported_schemes()
         : std::runtime_error("No allowed hashing schemes are supported on this system") {
 }
 
-bool check(const sstring& pass, const sstring& salted_hash) {
-    return detail::hash_with_salt(pass, salted_hash) == salted_hash;
+seastar::future<bool> check(const sstring& pass, const sstring& salted_hash) {
+    const auto pwd_hash = co_await detail::hash_with_salt_async(pass, salted_hash);
+    co_return pwd_hash == salted_hash;
 }
 
 } // namespace auth::passwords

auth/passwords.hh

@@ -11,6 +11,7 @@
 #include <random>
 #include <stdexcept>
 
+#include <seastar/core/future.hh>
 #include <seastar/core/sstring.hh>
 
 #include "seastarx.hh"
@@ -75,10 +76,19 @@ sstring generate_salt(RandomNumberEngine& g, scheme scheme) {
 
 ///
 /// Hash a password combined with an implementation-specific salt string.
+/// Deprecated in favor of `hash_with_salt_async`.
 ///
 /// \throws \ref std::system_error when an unexpected implementation-specific error occurs.
 ///
-sstring hash_with_salt(const sstring& pass, const sstring& salt);
+[[deprecated("Use hash_with_salt_async instead")]] sstring hash_with_salt(const sstring& pass, const sstring& salt);
+
+///
+/// Async version of `hash_with_salt` that returns a future.
+/// If possible, hashing uses `coroutine::maybe_yield` to prevent reactor stalls.
+///
+/// \throws \ref std::system_error when an unexpected implementation-specific error occurs.
+///
+seastar::future<sstring> hash_with_salt_async(const sstring& pass, const sstring& salt);
 
 } // namespace detail
 
@@ -107,6 +117,6 @@ sstring hash(const sstring& pass, RandomNumberEngine& g, scheme scheme) {
 ///
 /// \throws \ref std::system_error when an unexpected implementation-specific error occurs.
 ///
-bool check(const sstring& pass, const sstring& salted_hash);
+seastar::future<bool> check(const sstring& pass, const sstring& salted_hash);
 
 } // namespace auth::passwords

auth/saslauthd_authenticator.cc

@@ -35,10 +35,9 @@ static const class_registrator<
         cql3::query_processor&,
         ::service::raft_group0_client&,
         ::service::migration_manager&,
-        cache&,
-        utils::alien_worker&> saslauthd_auth_reg("com.scylladb.auth.SaslauthdAuthenticator");
+        cache&> saslauthd_auth_reg("com.scylladb.auth.SaslauthdAuthenticator");
 
-saslauthd_authenticator::saslauthd_authenticator(cql3::query_processor& qp, ::service::raft_group0_client&, ::service::migration_manager&, cache&, utils::alien_worker&)
+saslauthd_authenticator::saslauthd_authenticator(cql3::query_processor& qp, ::service::raft_group0_client&, ::service::migration_manager&, cache&)
     : _socket_path(qp.db().get_config().saslauthd_socket_path())
 {}
 

auth/saslauthd_authenticator.hh

@@ -12,7 +12,6 @@
 
 #include "auth/authenticator.hh"
 #include "auth/cache.hh"
-#include "utils/alien_worker.hh"
 
 namespace cql3 {
 class query_processor;
@@ -30,7 +29,7 @@ namespace auth {
 class saslauthd_authenticator : public authenticator {
     sstring _socket_path; ///< Path to the domain socket on which saslauthd is listening.
 public:
-    saslauthd_authenticator(cql3::query_processor&, ::service::raft_group0_client&, ::service::migration_manager&, cache&,utils::alien_worker&);
+    saslauthd_authenticator(cql3::query_processor&, ::service::raft_group0_client&, ::service::migration_manager&, cache&);
 
     future<> start() override;
 

auth/service.cc

@@ -191,8 +191,7 @@ service::service(
         ::service::migration_manager& mm,
         const service_config& sc,
         maintenance_socket_enabled used_by_maintenance_socket,
-        cache& cache,
-        utils::alien_worker& hashing_worker)
+        cache& cache)
             : service(
                       std::move(c),
                       cache,
@@ -200,7 +199,7 @@ service::service(
                       g0,
                       mn,
                       create_object<authorizer>(sc.authorizer_java_name, qp, g0, mm),
-                      create_object<authenticator>(sc.authenticator_java_name, qp, g0, mm, cache, hashing_worker),
+                      create_object<authenticator>(sc.authenticator_java_name, qp, g0, mm, cache),
                       create_object<role_manager>(sc.role_manager_java_name, qp, g0, mm, cache),
                       used_by_maintenance_socket) {
 }
@@ -226,7 +225,7 @@ future<> service::create_legacy_keyspace_if_missing(::service::migration_manager
             try {
                 co_return co_await mm.announce(::service::prepare_new_keyspace_announcement(db.real_database(), ksm, ts),
                         std::move(group0_guard), seastar::format("auth_service: create {} keyspace", meta::legacy::AUTH_KS));
-            } catch (::service::group0_concurrent_modification&) {
+            } catch (const ::service::group0_concurrent_modification&) {
                 log.info("Concurrent operation is detected while creating {} keyspace, retrying.", meta::legacy::AUTH_KS);
             }
         }

auth/service.hh

@@ -27,7 +27,6 @@
 #include "cql3/description.hh"
 #include "seastarx.hh"
 #include "service/raft/raft_group0_client.hh"
-#include "utils/alien_worker.hh"
 #include "utils/observable.hh"
 #include "utils/serialized_action.hh"
 #include "service/maintenance_mode.hh"
@@ -131,8 +130,7 @@ public:
             ::service::migration_manager&,
             const service_config&,
             maintenance_socket_enabled,
-            cache&,
-            utils::alien_worker&);
+            cache&);
 
     future<> start(::service::migration_manager&, db::system_keyspace&);
 

auth/standard_role_manager.cc

@@ -192,7 +192,7 @@ future<> standard_role_manager::legacy_create_default_role_if_missing() {
                 {_superuser},
                 cql3::query_processor::cache_internal::no).discard_result();
         log.info("Created default superuser role '{}'.", _superuser);
-    } catch(const exceptions::unavailable_exception& e) {
+    } catch (const exceptions::unavailable_exception& e) {
         log.warn("Skipped default role setup: some nodes were not ready; will retry");
         throw e;
     }

auth/transitional.cc

@@ -38,8 +38,8 @@ class transitional_authenticator : public authenticator {
 public:
     static const sstring PASSWORD_AUTHENTICATOR_NAME;
 
-    transitional_authenticator(cql3::query_processor& qp, ::service::raft_group0_client& g0, ::service::migration_manager& mm, cache& cache, utils::alien_worker& hashing_worker)
-            : transitional_authenticator(std::make_unique<password_authenticator>(qp, g0, mm, cache, hashing_worker)) {
+    transitional_authenticator(cql3::query_processor& qp, ::service::raft_group0_client& g0, ::service::migration_manager& mm, cache& cache)
+            : transitional_authenticator(std::make_unique<password_authenticator>(qp, g0, mm, cache)) {
     }
     transitional_authenticator(std::unique_ptr<authenticator> a)
             : _authenticator(std::move(a)) {
@@ -81,7 +81,7 @@ public:
         }).handle_exception([](auto ep) {
             try {
                 std::rethrow_exception(ep);
-            } catch (exceptions::authentication_exception&) {
+            } catch (const exceptions::authentication_exception&) {
                 // return anon user
                 return make_ready_future<authenticated_user>(anonymous_user());
             }
@@ -126,7 +126,7 @@ public:
             virtual bytes evaluate_response(bytes_view client_response) override {
                 try {
                     return _sasl->evaluate_response(client_response);
-                } catch (exceptions::authentication_exception&) {
+                } catch (const exceptions::authentication_exception&) {
                     _complete = true;
                     return {};
                 }
@@ -141,7 +141,7 @@ public:
                     return _sasl->get_authenticated_user().handle_exception([](auto ep) {
                         try {
                             std::rethrow_exception(ep);
-                        } catch (exceptions::authentication_exception&) {
+                        } catch (const exceptions::authentication_exception&) {
                             // return anon user
                             return make_ready_future<authenticated_user>(anonymous_user());
                         }
@@ -241,8 +241,7 @@ static const class_registrator<
         cql3::query_processor&,
         ::service::raft_group0_client&,
         ::service::migration_manager&,
-        auth::cache&,
-        utils::alien_worker&> transitional_authenticator_reg(auth::PACKAGE_NAME + "TransitionalAuthenticator");
+        auth::cache&> transitional_authenticator_reg(auth::PACKAGE_NAME + "TransitionalAuthenticator");
 
 static const class_registrator<
         auth::authorizer,

client_data.hh

@@ -10,7 +10,9 @@
 #include <seastar/net/inet_address.hh>
 #include <seastar/core/sstring.hh>
 #include "seastarx.hh"
+#include "utils/loading_shared_values.hh"
 
+#include <list>
 #include <optional>
 
 enum class client_type {
@@ -27,6 +29,20 @@ enum class client_connection_stage {
     ready,
 };
 
+// We implement a keys cache using a map-like utils::loading_shared_values container by storing empty values.
+struct options_cache_value_type {};
+using client_options_cache_type = utils::loading_shared_values<sstring, options_cache_value_type>;
+using client_options_cache_entry_type = client_options_cache_type::entry_ptr;
+using client_options_cache_key_type = client_options_cache_type::key_type;
+
+// This struct represents a single OPTION key-value pair from the client's connection options.
+// Both key and value are represented by corresponding "references" to their cached values.
+// Each "reference" is effectively a lw_shared_ptr value.
+struct client_option_key_value_cached_entry {
+    client_options_cache_entry_type key;
+    client_options_cache_entry_type value;
+};
+
 sstring to_string(client_connection_stage ct);
 
 // Representation of a row in `system.clients'. std::optionals are for nullable cells.
@@ -37,8 +53,8 @@ struct client_data {
     client_connection_stage connection_stage = client_connection_stage::established;
     int32_t shard_id;  /// ID of server-side shard which is processing the connection.
 
-    std::optional<sstring> driver_name;
-    std::optional<sstring> driver_version;
+    std::optional<client_options_cache_entry_type> driver_name;
+    std::optional<client_options_cache_entry_type> driver_version;
     std::optional<sstring> hostname;
     std::optional<int32_t> protocol_version;
     std::optional<sstring> ssl_cipher_suite;
@@ -46,6 +62,7 @@ struct client_data {
     std::optional<sstring> ssl_protocol;
     std::optional<sstring> username;
     std::optional<sstring> scheduling_group_name;
+    std::list<client_option_key_value_cached_entry> client_options;
 
     sstring stage_str() const { return to_string(connection_stage); }
     sstring client_type_str() const { return to_string(ct); }

cmake/mode.common.cmake

@@ -125,10 +125,6 @@ if(target_arch)
   add_compile_options("-march=${target_arch}")
 endif()
 
-if(CMAKE_CXX_COMPILER_ID STREQUAL "Clang")
-  add_compile_options("SHELL:-Xclang -fexperimental-assignment-tracking=disabled")
-endif()
-
 function(maybe_limit_stack_usage_in_KB stack_usage_threshold_in_KB config)
   math(EXPR _stack_usage_threshold_in_bytes "${stack_usage_threshold_in_KB} * 1024")
   set(_stack_usage_threshold_flag "-Wstack-usage=${_stack_usage_threshold_in_bytes}")

compaction/compaction_group_view.hh

@@ -12,6 +12,7 @@
 #include <seastar/core/condition-variable.hh>
 
 #include "schema/schema_fwd.hh"
+#include "sstables/open_info.hh"
 #include "compaction_descriptor.hh"
 
 class reader_permit;
@@ -44,7 +45,7 @@ public:
     virtual compaction_strategy_state& get_compaction_strategy_state() noexcept = 0;
     virtual reader_permit make_compaction_reader_permit() const = 0;
     virtual sstables::sstables_manager& get_sstables_manager() noexcept = 0;
-    virtual sstables::shared_sstable make_sstable() const = 0;
+    virtual sstables::shared_sstable make_sstable(sstables::sstable_state) const = 0;
     virtual sstables::sstable_writer_config configure_writer(sstring origin) const = 0;
     virtual api::timestamp_type min_memtable_timestamp() const = 0;
     virtual api::timestamp_type min_memtable_live_timestamp() const = 0;

compaction/compaction_manager.cc

@@ -416,7 +416,9 @@ future<compaction_result> compaction_task_executor::compact_sstables(compaction_
         descriptor.enable_garbage_collection(co_await sstable_set_for_tombstone_gc(t));
     }
     descriptor.creator = [&t] (shard_id) {
-        return t.make_sstable();
+        // All compaction types going through this path will work on normal input sstables only.
+        // Off-strategy, for example, waits until the sstables move out of staging state.
+        return t.make_sstable(sstables::sstable_state::normal);
     };
     descriptor.replacer = [this, &t, &on_replace, offstrategy] (compaction_completion_desc desc) {
         t.get_compaction_strategy().notify_completion(t, desc.old_sstables, desc.new_sstables);
@@ -1847,6 +1849,10 @@ protected:
                 throw make_compaction_stopped_exception();
             }
         }, false);
+        if (utils::get_local_injector().is_enabled("split_sstable_force_stop_exception")) {
+            throw make_compaction_stopped_exception();
+        }
+
         co_return co_await do_rewrite_sstable(std::move(sst));
     }
 };
@@ -2284,12 +2290,16 @@ future<compaction_manager::compaction_stats_opt> compaction_manager::perform_spl
 }
 
 future<std::vector<sstables::shared_sstable>>
-compaction_manager::maybe_split_sstable(sstables::shared_sstable sst, compaction_group_view& t, compaction_type_options::split opt) {
+compaction_manager::maybe_split_new_sstable(sstables::shared_sstable sst, compaction_group_view& t, compaction_type_options::split opt) {
     if (!split_compaction_task_executor::sstable_needs_split(sst, opt)) {
         co_return std::vector<sstables::shared_sstable>{sst};
     }
-    if (!can_proceed(&t)) {
-        co_return std::vector<sstables::shared_sstable>{sst};
+    // Throw an error if split cannot be performed due to e.g. out of space prevention.
+    // We don't want to prevent split because compaction is temporarily disabled on a view only for synchronization,
+    // which is uneeded against new sstables that aren't part of any set yet, so never use can_proceed(&t) here.
+    if (is_disabled()) {
+        co_return coroutine::exception(std::make_exception_ptr(std::runtime_error(format("Cannot split {} because manager has compaction disabled, " \
+                                                                                         "reason might be out of space prevention", sst->get_filename()))));
     }
     std::vector<sstables::shared_sstable> ret;
 
@@ -2297,8 +2307,11 @@ compaction_manager::maybe_split_sstable(sstables::shared_sstable sst, compaction
     compaction_progress_monitor monitor;
     compaction_data info = create_compaction_data();
     compaction_descriptor desc = split_compaction_task_executor::make_descriptor(sst, opt);
-    desc.creator = [&t] (shard_id _) {
-        return t.make_sstable();
+    desc.creator = [&t, sst] (shard_id _) {
+        // NOTE: preserves the sstable state, since we want the output to be on the same state as the original.
+        // For example, if base table has views, it's important that sstable produced by repair will be
+        // in the staging state.
+        return t.make_sstable(sst->state());
     };
     desc.replacer = [&] (compaction_completion_desc d) {
         std::move(d.new_sstables.begin(), d.new_sstables.end(), std::back_inserter(ret));

compaction/compaction_manager.hh

@@ -376,7 +376,8 @@ public:
     // Splits a single SSTable by segregating all its data according to the classifier.
     // If SSTable doesn't need split, the same input SSTable is returned as output.
     // If SSTable needs split, then output SSTables are returned and the input SSTable is deleted.
-    future<std::vector<sstables::shared_sstable>> maybe_split_sstable(sstables::shared_sstable sst, compaction_group_view& t, compaction_type_options::split opt);
+    // Exception is thrown if the input sstable cannot be split due to e.g. out of space prevention.
+    future<std::vector<sstables::shared_sstable>> maybe_split_new_sstable(sstables::shared_sstable sst, compaction_group_view& t, compaction_type_options::split opt);
 
     // Run a custom job for a given table, defined by a function
     // it completes when future returned by job is ready or returns immediately

configure.py

@@ -368,6 +368,87 @@ def find_ninja():
     sys.exit(1)
 
 
+def find_compiler(name):
+    """
+    Find a compiler by name, skipping ccache wrapper directories.
+
+    This is useful when using sccache to avoid double-caching through ccache.
+
+    Args:
+        name: The compiler name (e.g., 'clang++', 'clang', 'gcc')
+
+    Returns:
+        Path to the compiler, skipping ccache directories, or None if not found.
+    """
+    ccache_dirs = {'/usr/lib/ccache', '/usr/lib64/ccache'}
+    for path_dir in os.environ.get('PATH', '').split(os.pathsep):
+        # Skip ccache wrapper directories
+        if os.path.realpath(path_dir) in ccache_dirs or path_dir in ccache_dirs:
+            continue
+        candidate = os.path.join(path_dir, name)
+        if os.path.isfile(candidate) and os.access(candidate, os.X_OK):
+            return candidate
+    return None
+
+
+def resolve_compilers_for_sccache(args, compiler_cache):
+    """
+    When using sccache, resolve compiler paths to avoid ccache directories.
+
+    This prevents double-caching when ccache symlinks are in PATH.
+
+    Args:
+        args: The argument namespace with cc and cxx attributes.
+        compiler_cache: Path to the compiler cache binary, or None.
+    """
+    if not compiler_cache or 'sccache' not in compiler_cache:
+        return
+    if not os.path.isabs(args.cxx):
+        real_cxx = find_compiler(args.cxx)
+        if real_cxx:
+            args.cxx = real_cxx
+    if not os.path.isabs(args.cc):
+        real_cc = find_compiler(args.cc)
+        if real_cc:
+            args.cc = real_cc
+
+
+def find_compiler_cache(preference):
+    """
+    Find a compiler cache based on the preference.
+
+    Args:
+        preference: One of 'auto', 'sccache', 'ccache', 'none', or a path to a binary.
+
+    Returns:
+        Path to the compiler cache binary, or None if not found/disabled.
+    """
+    if preference == 'none':
+        return None
+
+    if preference == 'auto':
+        # Prefer sccache over ccache
+        for cache in ['sccache', 'ccache']:
+            path = which(cache)
+            if path:
+                return path
+        return None
+
+    if preference in ('sccache', 'ccache'):
+        path = which(preference)
+        if path:
+            return path
+        print(f"Warning: {preference} not found on PATH, disabling compiler cache")
+        return None
+
+    # Assume it's a path to a binary
+    if os.path.isfile(preference) and os.access(preference, os.X_OK):
+        return preference
+
+    print(f"Warning: compiler cache '{preference}' not found or not executable, disabling compiler cache")
+    return None
+
+
 modes = {
     'debug': {
         'cxxflags': '-DDEBUG -DSANITIZE -DDEBUG_LSA_SANITIZER -DSCYLLA_ENABLE_ERROR_INJECTION',
@@ -732,6 +813,8 @@ arg_parser.add_argument('--compiler', action='store', dest='cxx', default='clang
                         help='C++ compiler path')
 arg_parser.add_argument('--c-compiler', action='store', dest='cc', default='clang',
                         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')
 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='',
@@ -859,6 +942,7 @@ scylla_core = (['message/messaging_service.cc',
                 'utils/alien_worker.cc',
                 'utils/array-search.cc',
                 'utils/base64.cc',
+                'utils/crypt_sha512.cc',
                 'utils/logalloc.cc',
                 'utils/large_bitset.cc',
                 'utils/buffer_input_stream.cc',
@@ -1157,6 +1241,7 @@ scylla_core = (['message/messaging_service.cc',
                 'locator/topology.cc',
                 'locator/util.cc',
                 'service/client_state.cc',
+                'service/client_routes.cc',
                 'service/storage_service.cc',
                 'service/session.cc',
                 'service/task_manager_module.cc',
@@ -1317,6 +1402,8 @@ api = ['api/api.cc',
        'api/storage_proxy.cc',
        Json2Code('api/api-doc/cache_service.json'),
        'api/cache_service.cc',
+       Json2Code('api/api-doc/client_routes.json'),
+       'api/client_routes.cc',
        Json2Code('api/api-doc/collectd.json'),
        'api/collectd.cc',
        Json2Code('api/api-doc/endpoint_snitch_info.json'),
@@ -1479,7 +1566,6 @@ deps = {
 
 pure_boost_tests = set([
     'test/boost/anchorless_list_test',
-    'test/boost/auth_passwords_test',
     'test/boost/auth_resource_test',
     'test/boost/big_decimal_test',
     'test/boost/caching_options_test',
@@ -1695,6 +1781,18 @@ deps['test/vector_search/vector_store_client_test'] =  ['test/vector_search/vect
 deps['test/vector_search/load_balancer_test'] = ['test/vector_search/load_balancer_test.cc'] + scylla_tests_dependencies
 deps['test/vector_search/client_test'] = ['test/vector_search/client_test.cc'] + scylla_tests_dependencies
 
+boost_tests_prefixes = ["test/boost/", "test/vector_search/", "test/raft/", "test/manual/", "test/ldap/"]
+
+# We need to link these files to all Boost tests to make sure that
+# we can execute `--list_json_content` on them. That will produce
+# a similar result as calling `--list_content={HRF,DOT}`.
+# Unfortunately, to be able to do that, we're forced to link the
+# relevant code by hand.
+for key in deps.keys():
+    for prefix in boost_tests_prefixes:
+        if key.startswith(prefix):
+            deps[key] += ["test/lib/boost_tree_lister_injector.cc", "test/lib/boost_test_tree_lister.cc"]
+
 wasm_deps = {}
 
 wasm_deps['wasm/return_input.wat'] = 'test/resource/wasm/rust/return_input.rs'
@@ -1999,7 +2097,7 @@ def semicolon_separated(*flags):
 def real_relpath(path, start):
     return os.path.relpath(os.path.realpath(path), os.path.realpath(start))
 
-def configure_seastar(build_dir, mode, mode_config):
+def configure_seastar(build_dir, mode, mode_config, compiler_cache=None):
     seastar_cxx_ld_flags = mode_config['cxx_ld_flags']
     # We want to "undo" coverage for seastar if we have it enabled.
     if args.coverage:
@@ -2046,6 +2144,10 @@ def configure_seastar(build_dir, mode, mode_config):
         '-DSeastar_IO_URING=ON',
     ]
 
+    if compiler_cache:
+        seastar_cmake_args += [f'-DCMAKE_CXX_COMPILER_LAUNCHER={compiler_cache}',
+                               f'-DCMAKE_C_COMPILER_LAUNCHER={compiler_cache}']
+
     if args.stack_guards is not None:
         stack_guards = 'ON' if args.stack_guards else 'OFF'
         seastar_cmake_args += ['-DSeastar_STACK_GUARDS={}'.format(stack_guards)]
@@ -2077,7 +2179,7 @@ def configure_seastar(build_dir, mode, mode_config):
     subprocess.check_call(seastar_cmd, shell=False, cwd=cmake_dir)
 
 
-def configure_abseil(build_dir, mode, mode_config):
+def configure_abseil(build_dir, mode, mode_config, compiler_cache=None):
     abseil_cflags = mode_config['lib_cflags']
     cxx_flags = mode_config['cxxflags']
     if '-DSANITIZE' in cxx_flags:
@@ -2103,6 +2205,10 @@ def configure_abseil(build_dir, mode, mode_config):
         '-DABSL_PROPAGATE_CXX_STD=ON',
     ]
 
+    if compiler_cache:
+        abseil_cmake_args += [f'-DCMAKE_CXX_COMPILER_LAUNCHER={compiler_cache}',
+                              f'-DCMAKE_C_COMPILER_LAUNCHER={compiler_cache}']
+
     cmake_args = abseil_cmake_args[:]
     abseil_build_dir = os.path.join(build_dir, mode, 'abseil')
     abseil_cmd = ['cmake', '-G', 'Ninja', real_relpath('abseil', abseil_build_dir)] + cmake_args
@@ -2248,15 +2354,6 @@ def get_extra_cxxflags(mode, mode_config, cxx, debuginfo):
     if debuginfo and mode_config['can_have_debug_info']:
         cxxflags += ['-g', '-gz']
 
-    if 'clang' in cxx:
-        # Since AssignmentTracking was enabled by default in clang
-        # (llvm/llvm-project@de6da6ad55d3ca945195d1cb109cb8efdf40a52a)
-        # coroutine frame debugging info (`coro_frame_ty`) is broken.
-        #
-        # It seems that we aren't losing much by disabling AssigmentTracking,
-        # so for now we choose to disable it to get `coro_frame_ty` back.
-        cxxflags.append('-Xclang -fexperimental-assignment-tracking=disabled')
-
     return cxxflags
 
 
@@ -2284,10 +2381,15 @@ def write_build_file(f,
                      scylla_product,
                      scylla_version,
                      scylla_release,
+                     compiler_cache,
                      args):
     use_precompiled_header = not args.disable_precompiled_header
     warnings = get_warning_options(args.cxx)
     rustc_target = pick_rustc_target('wasm32-wasi', 'wasm32-wasip1')
+    # 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 ''
     f.write(textwrap.dedent('''\
         configure_args = {configure_args}
         builddir = {outdir}
@@ -2350,7 +2452,7 @@ def write_build_file(f,
             command = clang --target=wasm32 --no-standard-libraries -Wl,--export-all -Wl,--no-entry $in -o $out
             description = C2WASM $out
         rule rust2wasm
-            command = cargo build --target={rustc_target} --example=$example --locked --manifest-path=test/resource/wasm/rust/Cargo.toml --target-dir=$builddir/wasm/ $
+            command = {rustc_wrapper}cargo build --target={rustc_target} --example=$example --locked --manifest-path=test/resource/wasm/rust/Cargo.toml --target-dir=$builddir/wasm/ $
                 && wasm-opt -Oz $builddir/wasm/{rustc_target}/debug/examples/$example.wasm -o $builddir/wasm/$example.wasm $
                 && wasm-strip $builddir/wasm/$example.wasm
             description = RUST2WASM $out
@@ -2366,7 +2468,7 @@ def write_build_file(f,
           command = llvm-profdata merge $in -output=$out
         ''').format(configure_args=configure_args,
                     outdir=outdir,
-                    cxx=args.cxx,
+                    cxx=cxx_with_cache,
                     user_cflags=user_cflags,
                     warnings=warnings,
                     defines=defines,
@@ -2374,6 +2476,7 @@ def write_build_file(f,
                     user_ldflags=user_ldflags,
                     libs=libs,
                     rustc_target=rustc_target,
+                    rustc_wrapper=rustc_wrapper,
                     link_pool_depth=link_pool_depth,
                     seastar_path=args.seastar_path,
                     ninja=ninja,
@@ -2458,10 +2561,10 @@ def write_build_file(f,
               description = TEST {mode}
             # This rule is unused for PGO stages. They use the rust lib from the parent mode.
             rule rust_lib.{mode}
-              command = CARGO_BUILD_DEP_INFO_BASEDIR='.' cargo build --locked --manifest-path=rust/Cargo.toml --target-dir=$builddir/{mode} --profile=rust-{mode} $
+              command = CARGO_BUILD_DEP_INFO_BASEDIR='.' {rustc_wrapper}cargo build --locked --manifest-path=rust/Cargo.toml --target-dir=$builddir/{mode} --profile=rust-{mode} $
                         && touch $out
               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, **modeval))
+            ''').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} {vector_search_validator_bins}\n'.format(
                 mode=mode,
@@ -2525,7 +2628,7 @@ def write_build_file(f,
                 # In debug/sanitize modes, we compile with fsanitizers,
                 # so must use the same options during the link:
                 if '-DSANITIZE' in modes[mode]['cxxflags']:
-                    f.write('   libs = -fsanitize=address -fsanitize=undefined\n')
+                    f.write('   libs = -fsanitize=address -fsanitize=undefined -lubsan\n')
                 else:
                     f.write('   libs =\n')
                 f.write(f'build $builddir/{mode}/{binary}.stripped: strip $builddir/{mode}/{binary}\n')
@@ -2921,6 +3024,9 @@ def create_build_system(args):
 
     os.makedirs(outdir, exist_ok=True)
 
+    compiler_cache = find_compiler_cache(args.compiler_cache)
+    resolve_compilers_for_sccache(args, compiler_cache)
+
     scylla_product, scylla_version, scylla_release = generate_version(args.date_stamp)
 
     for mode, mode_config in build_modes.items():
@@ -2937,8 +3043,8 @@ def create_build_system(args):
         # {outdir}/{mode}/seastar/build.ninja, and
         # {outdir}/{mode}/seastar/seastar.pc is queried for building flags
         for mode, mode_config in build_modes.items():
-            configure_seastar(outdir, mode, mode_config)
-            configure_abseil(outdir, mode, mode_config)
+            configure_seastar(outdir, mode, mode_config, compiler_cache)
+            configure_abseil(outdir, mode, mode_config, compiler_cache)
         user_cflags += ' -isystem abseil'
 
     for mode, mode_config in build_modes.items():
@@ -2961,6 +3067,7 @@ def create_build_system(args):
                          scylla_product,
                          scylla_version,
                          scylla_release,
+                         compiler_cache,
                          args)
     generate_compdb('compile_commands.json', ninja, args.buildfile, selected_modes)
 
@@ -3003,6 +3110,10 @@ def configure_using_cmake(args):
     selected_modes = args.selected_modes or default_modes
     selected_configs = ';'.join(build_modes[mode].cmake_build_type for mode
                                 in selected_modes)
+
+    compiler_cache = find_compiler_cache(args.compiler_cache)
+    resolve_compilers_for_sccache(args, compiler_cache)
+
     settings = {
         'CMAKE_CONFIGURATION_TYPES': selected_configs,
         'CMAKE_CROSS_CONFIGS': selected_configs,
@@ -3020,6 +3131,14 @@ def configure_using_cmake(args):
         'Scylla_WITH_DEBUG_INFO' : 'ON' if args.debuginfo else 'OFF',
         'Scylla_USE_PRECOMPILED_HEADER': 'OFF' if args.disable_precompiled_header else 'ON',
     }
+
+    if compiler_cache:
+        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:
+            settings['Scylla_RUSTC_WRAPPER'] = compiler_cache
+
     if args.date_stamp:
         settings['Scylla_DATE_STAMP'] = args.date_stamp
     if args.staticboost:
@@ -3051,7 +3170,7 @@ def configure_using_cmake(args):
 
     if not args.dist_only:
         for mode in selected_modes:
-            configure_seastar(build_dir, build_modes[mode].cmake_build_type, modes[mode])
+            configure_seastar(build_dir, build_modes[mode].cmake_build_type, modes[mode], compiler_cache)
 
     cmake_command = ['cmake']
     cmake_command += [f'-D{var}={value}' for var, value in settings.items()]

cql3/query_processor.cc

@@ -64,6 +64,10 @@ bool query_processor::topology_global_queue_empty() {
     return remote().first.get().ss.topology_global_queue_empty();
 }
 
+future<bool> query_processor::ongoing_rf_change(const service::group0_guard& guard, sstring ks) {
+    return remote().first.get().ss.ongoing_rf_change(guard, std::move(ks));
+}
+
 static service::query_state query_state_for_internal_call() {
     return {service::client_state::for_internal_calls(), empty_service_permit()};
 }

cql3/query_processor.hh

@@ -474,6 +474,7 @@ public:
     void reset_cache();
 
     bool topology_global_queue_empty();
+    future<bool> ongoing_rf_change(const service::group0_guard& guard, sstring ks);
 
     query_options make_internal_options(
             const statements::prepared_statement::checked_weak_ptr& p,

cql3/restrictions/statement_restrictions.cc

@@ -1322,6 +1322,10 @@ const std::vector<expr::expression>& statement_restrictions::index_restrictions(
     return _index_restrictions;
 }
 
+bool statement_restrictions::is_empty() const {
+    return !_where.has_value();
+}
+
 // Current score table:
 // local and restrictions include full partition key: 2
 // global: 1

cql3/restrictions/statement_restrictions.hh

@@ -408,6 +408,8 @@ public:
 
     /// Checks that the primary key restrictions don't contain null values, throws invalid_request_exception otherwise.
     void validate_primary_key(const query_options& options) const;
+
+    bool is_empty() const;
 };
 
 statement_restrictions analyze_statement_restrictions(

cql3/statements/alter_keyspace_statement.cc

@@ -19,6 +19,7 @@
 #include "locator/abstract_replication_strategy.hh"
 #include "mutation/canonical_mutation.hh"
 #include "prepared_statement.hh"
+#include "seastar/coroutine/exception.hh"
 #include "service/migration_manager.hh"
 #include "service/storage_proxy.hh"
 #include "service/topology_mutation.hh"
@@ -138,6 +139,7 @@ bool cql3::statements::alter_keyspace_statement::changes_tablets(query_processor
 future<std::tuple<::shared_ptr<cql_transport::event::schema_change>, cql3::cql_warnings_vec>>
 cql3::statements::alter_keyspace_statement::prepare_schema_mutations(query_processor& qp, service::query_state& state, const query_options& options, service::group0_batch& mc) const {
     using namespace cql_transport;
+    bool unknown_keyspace = false;
     try {
         event::schema_change::target_type target_type = event::schema_change::target_type::KEYSPACE;
         auto ks = qp.db().find_keyspace(_name);
@@ -158,8 +160,12 @@ cql3::statements::alter_keyspace_statement::prepare_schema_mutations(query_proce
         //       when in reality nothing or only schema is being changed
         if (changes_tablets(qp)) {
             if (!qp.proxy().features().topology_global_request_queue && !qp.topology_global_queue_empty()) {
-                return make_exception_future<std::tuple<::shared_ptr<::cql_transport::event::schema_change>, cql3::cql_warnings_vec>>(
-                        exceptions::invalid_request_exception("Another global topology request is ongoing, please retry."));
+                co_await coroutine::return_exception(
+                    exceptions::invalid_request_exception("Another global topology request is ongoing, please retry."));
+            }
+            if (qp.proxy().features().rack_list_rf && co_await qp.ongoing_rf_change(mc.guard(),_name)) {
+                co_await coroutine::return_exception(
+                        exceptions::invalid_request_exception(format("Another RF change for this keyspace {} ongoing, please retry.", _name)));
             }
             qp.db().real_database().validate_keyspace_update(*ks_md_update);
 
@@ -242,10 +248,15 @@ cql3::statements::alter_keyspace_statement::prepare_schema_mutations(query_proce
                 target_type,
                 keyspace());
         mc.add_mutations(std::move(muts), "CQL alter keyspace");
-        return make_ready_future<std::tuple<::shared_ptr<cql_transport::event::schema_change>, cql3::cql_warnings_vec>>(std::make_tuple(std::move(ret), warnings));
+        co_return std::make_tuple(std::move(ret), warnings);
     } catch (data_dictionary::no_such_keyspace& e) {
-        return make_exception_future<std::tuple<::shared_ptr<cql_transport::event::schema_change>, cql3::cql_warnings_vec>>(exceptions::invalid_request_exception("Unknown keyspace " + _name));
+        unknown_keyspace = true;
     }
+    if (unknown_keyspace) {
+        co_await coroutine::return_exception(
+                exceptions::invalid_request_exception("Unknown keyspace " + _name));
+    }
+    std::unreachable();
 }
 
 std::unique_ptr<cql3::statements::prepared_statement>

cql3/statements/batch_statement.cc

@@ -190,7 +190,7 @@ future<utils::chunked_vector<mutation>> batch_statement::get_mutations(query_pro
     co_return vresult;
 }
 
-void batch_statement::verify_batch_size(query_processor& qp, const utils::chunked_vector<mutation>& mutations) {
+void batch_statement::verify_batch_size(query_processor& qp, const utils::chunked_vector<mutation>& mutations) const {
     if (mutations.size() <= 1) {
         return;     // We only warn for batch spanning multiple mutations
     }
@@ -209,8 +209,9 @@ void batch_statement::verify_batch_size(query_processor& qp, const utils::chunke
             for (auto&& m : mutations) {
                 ks_cf_pairs.insert(m.schema()->ks_name() + "." + m.schema()->cf_name());
             }
-            return seastar::format("Batch modifying {:d} partitions in {} is of size {:d} bytes, exceeding specified {} threshold of {:d} by {:d}.",
-                    mutations.size(), fmt::join(ks_cf_pairs, ", "), size, type, threshold, size - threshold);
+            const auto batch_type = _type == type::LOGGED ? "Logged" : "Unlogged";
+            return seastar::format("{} batch modifying {:d} partitions in {} is of size {:d} bytes, exceeding specified {} threshold of {:d} by {:d}.",
+                    batch_type, mutations.size(), fmt::join(ks_cf_pairs, ", "), size, type, threshold, size - threshold);
         };
         if (size > fail_threshold) {
             _logger.error("{}", error("FAIL", fail_threshold).c_str());

cql3/statements/batch_statement.hh

@@ -116,7 +116,7 @@ public:
      * Checks batch size to ensure threshold is met. If not, a warning is logged.
      * @param cfs ColumnFamilies that will store the batch's mutations.
      */
-    static void verify_batch_size(query_processor& qp, const utils::chunked_vector<mutation>& mutations);
+    void verify_batch_size(query_processor& qp, const utils::chunked_vector<mutation>& mutations) const;
 
     virtual future<shared_ptr<cql_transport::messages::result_message>> execute(
             query_processor& qp, service::query_state& state, const query_options& options, std::optional<service::group0_guard> guard) const override;

cql3/statements/ks_prop_defs.cc

@@ -61,7 +61,7 @@ expand_to_racks(const locator::token_metadata& tm,
 
     // Handle ALTER:
     // ([]|0) -> numeric is allowed, there are no existing replicas
-    // numeric -> numeric' is not supported. User should convert RF to rack list of equal count first.
+    // numeric -> numeric' is not supported unless numeric == numeric'. User should convert RF to rack list of equal count first.
     // rack_list -> len(rack_list) is allowed (no-op)
     // rack_list -> numeric is not allowed
     if (old_options.contains(dc)) {
@@ -75,6 +75,8 @@ expand_to_racks(const locator::token_metadata& tm,
                         "Cannot change replication factor for '{}' from {} to numeric {}, use rack list instead",
                         dc, old_rf_val, data.count()));
             }
+        } else if (old_rf.count() == data.count()) {
+            return rf;
         } else if (old_rf.count() > 0) {
             throw exceptions::configuration_exception(fmt::format(
                     "Cannot change replication factor for '{}' from {} to {}, only rack list is allowed",
@@ -153,6 +155,8 @@ static locator::replication_strategy_config_options prepare_options(
     }
 
     // Validate options.
+    bool numeric_to_rack_list_transition = false;
+    bool rf_change = false;
     for (auto&& [dc, opt] : options) {
         locator::replication_factor_data rf(opt);
 
@@ -162,6 +166,7 @@ static locator::replication_strategy_config_options prepare_options(
             old_rf = locator::replication_factor_data(i->second);
         }
 
+        rf_change = rf_change || (old_rf && old_rf->count() != rf.count()) || (!old_rf && rf.count() != 0);
         if (!rf.is_rack_based()) {
             if (old_rf && old_rf->is_rack_based() && rf.count() != 0) {
                 if (old_rf->count() != rf.count()) {
@@ -187,12 +192,11 @@ static locator::replication_strategy_config_options prepare_options(
             throw exceptions::configuration_exception(fmt::format(
                     "Rack list for '{}' contains duplicate entries", dc));
         }
-        if (old_rf && !old_rf->is_rack_based() && old_rf->count() != 0) {
-            // FIXME: Allow this if replicas already conform to the given rack list.
-            // FIXME: Implement automatic colocation to allow transition to rack list.
-            throw exceptions::configuration_exception(fmt::format(
-                    "Cannot change replication factor from numeric to rack list for '{}'", dc));
-        }
+        numeric_to_rack_list_transition = numeric_to_rack_list_transition || (old_rf && !old_rf->is_rack_based() && old_rf->count() != 0);
+    }
+
+    if (numeric_to_rack_list_transition && rf_change) {
+        throw exceptions::configuration_exception("Cannot change replication factor from numeric to rack list and rf value at the same time");
     }
 
     if (!rf && options.empty() && old_options.empty()) {
@@ -412,7 +416,7 @@ lw_shared_ptr<data_dictionary::keyspace_metadata> ks_prop_defs::as_ks_metadata(s
             ? std::optional<unsigned>(0) : std::nullopt;
     auto initial_tablets = get_initial_tablets(default_initial_tablets, cfg.enforce_tablets());
     bool uses_tablets = initial_tablets.has_value();
-    bool rack_list_enabled = feat.rack_list_rf;
+    bool rack_list_enabled = utils::get_local_injector().enter("create_with_numeric") ? false : feat.rack_list_rf;
     auto options = prepare_options(sc, tm, cfg.rf_rack_valid_keyspaces(), get_replication_options(), {}, rack_list_enabled, uses_tablets);
     return data_dictionary::keyspace_metadata::new_keyspace(ks_name, sc,
             std::move(options), initial_tablets, get_consistency_option(), get_boolean(KW_DURABLE_WRITES, true), get_storage_options());
@@ -428,7 +432,7 @@ lw_shared_ptr<data_dictionary::keyspace_metadata> ks_prop_defs::as_ks_metadata_u
         throw exceptions::invalid_request_exception("Cannot alter replication strategy vnode/tablets flavor");
     }
     auto sc = get_replication_strategy_class();
-    bool rack_list_enabled = feat.rack_list_rf;
+    bool rack_list_enabled = utils::get_local_injector().enter("create_with_numeric") ? false : feat.rack_list_rf;
     if (sc) {
         options = prepare_options(*sc, tm, cfg.rf_rack_valid_keyspaces(), get_replication_options(), old_options, rack_list_enabled, uses_tablets);
     } else {

cql3/statements/select_statement.cc

@@ -1976,7 +1976,7 @@ mutation_fragments_select_statement::do_execute(query_processor& qp, service::qu
     if (it == indexes.end()) {
         throw exceptions::invalid_request_exception("ANN ordering by vector requires the column to be indexed using 'vector_index'");
     }
-    if (index_opt || parameters->allow_filtering() || restrictions->need_filtering() || check_needs_allow_filtering_anyway(*restrictions)) {
+    if (index_opt || parameters->allow_filtering() || !(restrictions->is_empty()) || check_needs_allow_filtering_anyway(*restrictions)) {
         throw exceptions::invalid_request_exception("ANN ordering by vector does not support filtering");
     }
     index_opt = *it;

data_dictionary/data_dictionary.cc

@@ -42,6 +42,11 @@ table::get_index_manager() const {
     return _ops->get_index_manager(*this);
 }
 
+db_clock::time_point
+table::get_truncation_time() const {
+    return _ops->get_truncation_time(*this);
+}
+
 lw_shared_ptr<keyspace_metadata>
 keyspace::metadata() const {
     return _ops->get_keyspace_metadata(*this);

data_dictionary/data_dictionary.hh

@@ -77,6 +77,7 @@ public:
     schema_ptr schema() const;
     const std::vector<view_ptr>& views() const;
     const secondary_index::secondary_index_manager& get_index_manager() const;
+    db_clock::time_point get_truncation_time() const;
 };
 
 class keyspace {

data_dictionary/impl.hh

@@ -27,6 +27,7 @@ public:
     virtual std::optional<table> try_find_table(database db, table_id id) const = 0;
     virtual const secondary_index::secondary_index_manager& get_index_manager(table t) const = 0;
     virtual schema_ptr get_table_schema(table t) const = 0;
+    virtual db_clock::time_point get_truncation_time(table t) const = 0;
     virtual lw_shared_ptr<keyspace_metadata> get_keyspace_metadata(keyspace ks) const = 0;
     virtual bool is_internal(keyspace ks) const = 0;
     virtual const locator::abstract_replication_strategy& get_replication_strategy(keyspace ks) const = 0;

db/batchlog.hh

@@ -0,0 +1,20 @@
+/*
+ * Copyright (C) 2025-present ScyllaDB
+ */
+
+/*
+ * SPDX-License-Identifier: (LicenseRef-ScyllaDB-Source-Available-1.0 and Apache-2.0)
+ */
+
+#pragma once
+
+#include "mutation/mutation.hh"
+#include "utils/UUID.hh"
+
+namespace db {
+
+mutation get_batchlog_mutation_for(schema_ptr schema, const utils::chunked_vector<mutation>& mutations, int32_t version, db_clock::time_point now, const utils::UUID& id);
+
+mutation get_batchlog_delete_mutation(schema_ptr schema, int32_t version, db_clock::time_point now, const utils::UUID& id);
+
+}

db/batchlog_manager.cc

@@ -10,20 +10,24 @@
 
 #include <chrono>
 #include <exception>
+#include <ranges>
 #include <seastar/core/future-util.hh>
 #include <seastar/core/do_with.hh>
 #include <seastar/core/semaphore.hh>
 #include <seastar/core/metrics.hh>
 #include <seastar/core/coroutine.hh>
 #include <seastar/core/sleep.hh>
+#include <seastar/coroutine/parallel_for_each.hh>
 
 #include "batchlog_manager.hh"
+#include "batchlog.hh"
 #include "data_dictionary/data_dictionary.hh"
 #include "mutation/canonical_mutation.hh"
 #include "service/storage_proxy.hh"
 #include "system_keyspace.hh"
 #include "utils/rate_limiter.hh"
 #include "utils/log.hh"
+#include "utils/murmur_hash.hh"
 #include "db_clock.hh"
 #include "unimplemented.hh"
 #include "idl/frozen_schema.dist.hh"
@@ -33,17 +37,94 @@
 #include "cql3/untyped_result_set.hh"
 #include "service_permit.hh"
 #include "cql3/query_processor.hh"
-#include "replica/database.hh"
 
 static logging::logger blogger("batchlog_manager");
 
+namespace db {
+
+// Yields 256 batchlog shards. Even on the largest nodes we currently run on,
+// this should be enough to give every core a batchlog partition.
+static constexpr unsigned batchlog_shard_bits = 8;
+
+int32_t batchlog_shard_of(db_clock::time_point written_at) {
+    const int64_t count = written_at.time_since_epoch().count();
+    std::array<uint64_t, 2> result;
+    utils::murmur_hash::hash3_x64_128(bytes_view(reinterpret_cast<const signed char*>(&count), sizeof(count)), 0, result);
+    uint64_t hash = result[0] ^ result[1];
+    return hash & ((1ULL << batchlog_shard_bits) - 1);
+}
+
+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) {
+    auto pkey = partition_key::from_exploded(schema, {serialized(version), serialized(int8_t(stage)), serialized(batchlog_shard)});
+
+    std::vector<bytes> ckey_components;
+    ckey_components.reserve(2);
+    ckey_components.push_back(serialized(written_at));
+    if (id) {
+        ckey_components.push_back(serialized(*id));
+    }
+    auto ckey = clustering_key::from_exploded(schema, ckey_components);
+
+    return {std::move(pkey), std::move(ckey)};
+}
+
+std::pair<partition_key, clustering_key>
+get_batchlog_key(const schema& schema, int32_t version, db::batchlog_stage stage, db_clock::time_point written_at, std::optional<utils::UUID> id) {
+    return get_batchlog_key(schema, version, stage, batchlog_shard_of(written_at), written_at, id);
+}
+
+mutation get_batchlog_mutation_for(schema_ptr schema, managed_bytes data, int32_t version, db::batchlog_stage stage, db_clock::time_point now, const utils::UUID& id) {
+    auto [key, ckey] = get_batchlog_key(*schema, version, stage, now, id);
+
+    auto timestamp = api::new_timestamp();
+
+    mutation m(schema, key);
+    // Avoid going through data_value and therefore `bytes`, as it can be large (#24809).
+    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)));
+
+    return m;
+}
+
+mutation get_batchlog_mutation_for(schema_ptr schema, const utils::chunked_vector<mutation>& mutations, int32_t version, db::batchlog_stage stage, db_clock::time_point now, const utils::UUID& id) {
+    auto data = [&mutations] {
+        utils::chunked_vector<canonical_mutation> fm(mutations.begin(), mutations.end());
+        bytes_ostream out;
+        for (auto& m : fm) {
+            ser::serialize(out, m);
+        }
+        return std::move(out).to_managed_bytes();
+    }();
+
+    return get_batchlog_mutation_for(std::move(schema), std::move(data), version, stage, now, id);
+}
+
+mutation get_batchlog_mutation_for(schema_ptr schema, const utils::chunked_vector<mutation>& mutations, int32_t version, db_clock::time_point now, const utils::UUID& id) {
+    return get_batchlog_mutation_for(std::move(schema), mutations, version, batchlog_stage::initial, now, id);
+}
+
+mutation get_batchlog_delete_mutation(schema_ptr schema, int32_t version, db::batchlog_stage stage, db_clock::time_point now, const utils::UUID& id) {
+    auto [key, ckey] = get_batchlog_key(*schema, version, stage, now, id);
+    mutation m(schema, key);
+    auto timestamp = api::new_timestamp();
+    m.partition().apply_delete(*schema, ckey, tombstone(timestamp, gc_clock::now()));
+    return m;
+}
+
+mutation get_batchlog_delete_mutation(schema_ptr schema, int32_t version, db_clock::time_point now, const utils::UUID& id) {
+    return get_batchlog_delete_mutation(std::move(schema), version, batchlog_stage::initial, now, id);
+}
+
+} // namespace db
+
 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, batchlog_manager_config config)
         : _qp(qp)
         , _sys_ks(sys_ks)
-        , _write_request_timeout(std::chrono::duration_cast<db_clock::duration>(config.write_request_timeout))
+        , _replay_timeout(config.replay_timeout)
         , _replay_rate(config.replay_rate)
         , _delay(config.delay)
         , _replay_cleanup_after_replays(config.replay_cleanup_after_replays)
@@ -152,18 +233,75 @@ future<> db::batchlog_manager::stop() {
 }
 
 future<size_t> db::batchlog_manager::count_all_batches() const {
-    sstring query = format("SELECT count(*) FROM {}.{} BYPASS CACHE", system_keyspace::NAME, system_keyspace::BATCHLOG);
+    sstring query = format("SELECT count(*) FROM {}.{} BYPASS CACHE", system_keyspace::NAME, system_keyspace::BATCHLOG_V2);
     return _qp.execute_internal(query, cql3::query_processor::cache_internal::yes).then([](::shared_ptr<cql3::untyped_result_set> rs) {
        return size_t(rs->one().get_as<int64_t>("count"));
     });
 }
 
-db_clock::duration db::batchlog_manager::get_batch_log_timeout() const {
-    // enough time for the actual write + BM removal mutation
-    return _write_request_timeout * 2;
+future<> db::batchlog_manager::maybe_migrate_v1_to_v2() {
+    if (_migration_done) {
+        return make_ready_future<>();
+    }
+    return with_gate(_gate, [this] () mutable -> future<> {
+        blogger.info("Migrating batchlog entries from v1 -> v2");
+
+        auto schema_v1 = _qp.db().find_schema(system_keyspace::NAME, system_keyspace::BATCHLOG);
+        auto schema_v2 = _qp.db().find_schema(system_keyspace::NAME, system_keyspace::BATCHLOG_V2);
+
+        auto batch = [this, schema_v1, schema_v2] (const cql3::untyped_result_set::row& row) -> future<stop_iteration> {
+            // check version of serialization format
+            if (!row.has("version")) {
+                blogger.warn("Not migrating logged batch because of unknown version");
+                co_return stop_iteration::no;
+            }
+
+            auto version = row.get_as<int32_t>("version");
+            if (version != netw::messaging_service::current_version) {
+                blogger.warn("Not migrating logged batch because of incorrect version");
+                co_return stop_iteration::no;
+            }
+
+            auto id = row.get_as<utils::UUID>("id");
+            auto written_at = row.get_as<db_clock::time_point>("written_at");
+            auto data = row.get_blob_fragmented("data");
+
+            auto& sp = _qp.proxy();
+
+            utils::get_local_injector().inject("batchlog_manager_fail_migration", [] { throw std::runtime_error("Error injection: failing batchlog migration"); });
+
+            auto migrate_mut = get_batchlog_mutation_for(schema_v2, std::move(data), version, batchlog_stage::failed_replay, written_at, id);
+            co_await sp.mutate_locally(migrate_mut, tracing::trace_state_ptr(), db::commitlog::force_sync::no);
+
+            mutation delete_mut(schema_v1, partition_key::from_single_value(*schema_v1, serialized(id)));
+            delete_mut.partition().apply_delete(*schema_v1, clustering_key_prefix::make_empty(), tombstone(api::new_timestamp(), gc_clock::now()));
+            co_await sp.mutate_locally(delete_mut, tracing::trace_state_ptr(), db::commitlog::force_sync::no);
+
+            co_return stop_iteration::no;
+        };
+        try {
+            co_await _qp.query_internal(
+                    format("SELECT * FROM {}.{} BYPASS CACHE", system_keyspace::NAME, system_keyspace::BATCHLOG),
+                    db::consistency_level::ONE,
+                    {},
+                    page_size,
+                    std::move(batch));
+        } catch (...) {
+            blogger.warn("Batchlog v1 to v2 migration failed: {}; will retry", std::current_exception());
+            co_return;
+        }
+
+        co_await container().invoke_on_all([] (auto& bm) {
+            bm._migration_done = true;
+        });
+
+        blogger.info("Done migrating batchlog entries from v1 -> v2");
+    });
 }
 
 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;
@@ -172,21 +310,26 @@ future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches
     auto throttle = _replay_rate / _qp.proxy().get_token_metadata_ptr()->count_normal_token_owners();
     auto limiter = make_lw_shared<utils::rate_limiter>(throttle);
 
-    auto schema = _qp.db().find_schema(system_keyspace::NAME, system_keyspace::BATCHLOG);
-    auto delete_batch = [this, schema = std::move(schema)] (utils::UUID id) {
-        auto key = partition_key::from_singular(*schema, id);
-        mutation m(schema, key);
-        auto now = service::client_state(service::client_state::internal_tag()).get_timestamp();
-        m.partition().apply_delete(*schema, clustering_key_prefix::make_empty(), tombstone(now, gc_clock::now()));
-        return _qp.proxy().mutate_locally(m, tracing::trace_state_ptr(), db::commitlog::force_sync::no);
+    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;
     };
 
-    auto batch = [this, limiter, delete_batch = std::move(delete_batch), &all_replayed](const cql3::untyped_result_set::row& row) -> future<stop_iteration> {
+    std::unordered_map<int32_t, replay_stats> replay_stats_per_shard;
+
+    // Use a stable `now` accross all batches, so skip/replay decisions are the
+    // same accross a while prefix of written_at (accross 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) -> 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 now = db_clock::now();
-        auto timeout = get_batch_log_timeout();
+        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");
@@ -194,52 +337,48 @@ future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches
             co_return stop_iteration::no;
         }
 
-        // check version of serialization format
-        if (!row.has("version")) {
-            blogger.warn("Skipping logged batch because of unknown version");
-            co_await delete_batch(id);
-            co_return stop_iteration::no;
-        }
-
-        auto version = row.get_as<int32_t>("version");
-        if (version != netw::messaging_service::current_version) {
-            blogger.warn("Skipping logged batch because of incorrect version {}; current version = {}", version, netw::messaging_service::current_version);
-            co_await delete_batch(id);
-            co_return stop_iteration::no;
-        }
-
         auto data = row.get_blob_unfragmented("data");
 
-        blogger.debug("Replaying batch {}", id);
+        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 {
-            auto fms = make_lw_shared<std::deque<canonical_mutation>>();
+            utils::chunked_vector<std::pair<canonical_mutation, schema_ptr>> fms;
             auto in = ser::as_input_stream(data);
             while (in.size()) {
-                fms->emplace_back(ser::deserialize(in, std::type_identity<canonical_mutation>()));
-                schema_ptr s = _qp.db().find_schema(fms->back().column_family_id());
-                timeout = std::min(timeout, std::chrono::duration_cast<db_clock::duration>(s->tombstone_gc_options().propagation_delay_in_seconds()));
+                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();
 
-            auto mutations = co_await map_reduce(*fms, [this, written_at] (canonical_mutation& fm) {
-                const auto& cf = _qp.proxy().local_db().find_column_family(fm.column_family_id());
-                return make_ready_future<canonical_mutation*>(written_at > cf.get_truncation_time() ? &fm : nullptr);
-            },
-            utils::chunked_vector<mutation>(),
-            [this] (utils::chunked_vector<mutation> mutations, canonical_mutation* fm) {
-                if (fm) {
-                    schema_ptr s = _qp.db().find_schema(fm->column_family_id());
-                    mutations.emplace_back(fm->to_mutation(s));
-                }
-                return mutations;
-            });
+            for (const auto& [fm, s] : fms) {
+                mutations.emplace_back(fm.to_mutation(s));
+                co_await maybe_yield();
+            }
 
             if (!mutations.empty()) {
                 const auto ttl = [written_at]() -> clock_type {
@@ -265,7 +404,11 @@ future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches
                     co_await limiter->reserve(size);
                     _stats.write_attempts += mutations.size();
                     auto timeout = db::timeout_clock::now() + write_timeout;
-                    co_await _qp.proxy().send_batchlog_replay_to_all_replicas(std::move(mutations), 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) {
@@ -279,31 +422,80 @@ future<db::all_batches_replayed> db::batchlog_manager::replay_all_failed_batches
             // 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.
-            co_return stop_iteration::no;
+            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
-        co_await delete_batch(id);
+        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;
     };
 
-    co_await with_gate(_gate, [this, cleanup, batch = std::move(batch)] () mutable -> future<> {
-        blogger.debug("Started replayAllFailedBatches (cpu {})", this_shard_id());
+    co_await with_gate(_gate, [this, cleanup, &all_replayed, batch = std::move(batch), now, &replay_stats_per_shard] () mutable -> future<> {
+        blogger.debug("Started replayAllFailedBatches with cleanup: {}", cleanup);
         co_await utils::get_local_injector().inject("add_delay_to_batch_replay", std::chrono::milliseconds(1000));
-        co_await _qp.query_internal(
-                format("SELECT id, data, written_at, version FROM {}.{} BYPASS CACHE", system_keyspace::NAME, system_keyspace::BATCHLOG),
-                db::consistency_level::ONE,
-                {},
-                page_size,
-                std::move(batch)).then([this, cleanup] {
-            if (cleanup == post_replay_cleanup::no) {
-                return make_ready_future<>();
+
+        auto schema = _qp.db().find_schema(system_keyspace::NAME, system_keyspace::BATCHLOG_V2);
+
+        co_await coroutine::parallel_for_each(std::views::iota(0, 16), [&] (int32_t chunk) -> future<> {
+            const int32_t batchlog_chunk_base = chunk * 16;
+            for (int32_t i = 0; i < 16; ++i) {
+                int32_t batchlog_shard = batchlog_chunk_base + i;
+
+                co_await _qp.query_internal(
+                        format("SELECT * FROM {}.{} WHERE version = ? AND stage = ? AND shard = ? BYPASS CACHE", system_keyspace::NAME, system_keyspace::BATCHLOG_V2),
+                        db::consistency_level::ONE,
+                        {data_value(netw::messaging_service::current_version), data_value(int8_t(batchlog_stage::failed_replay)), data_value(batchlog_shard)},
+                        page_size,
+                        batch);
+
+                co_await _qp.query_internal(
+                        format("SELECT * FROM {}.{} WHERE version = ? AND stage = ? AND shard = ? BYPASS CACHE", system_keyspace::NAME, system_keyspace::BATCHLOG_V2),
+                        db::consistency_level::ONE,
+                        {data_value(netw::messaging_service::current_version), data_value(int8_t(batchlog_stage::initial)), data_value(batchlog_shard)},
+                        page_size,
+                        batch);
+
+                if (cleanup != post_replay_cleanup::yes) {
+                    continue;
+                }
+
+                auto it = replay_stats_per_shard.find(batchlog_shard);
+                if (it == replay_stats_per_shard.end() || !it->second.need_cleanup) {
+                    // Nothing was replayed on this batchlog shard, nothing to cleanup.
+                    continue;
+                }
+
+                const auto write_time = it->second.min_too_fresh.value_or(now - _replay_timeout);
+                const auto end_weight  = it->second.min_too_fresh ? bound_weight::before_all_prefixed : bound_weight::after_all_prefixed;
+                auto [key, ckey] = get_batchlog_key(*schema, netw::messaging_service::current_version, batchlog_stage::initial, batchlog_shard, write_time, {});
+                auto end_pos = position_in_partition(partition_region::clustered, end_weight, std::move(ckey));
+
+                range_tombstone rt(position_in_partition::before_all_clustered_rows(), std::move(end_pos), tombstone(api::new_timestamp(), gc_clock::now()));
+
+                blogger.trace("Clean up batchlog shard {} with range tombstone {}", batchlog_shard, rt);
+
+                mutation m(schema, key);
+                m.partition().apply_row_tombstone(*schema, std::move(rt));
+                co_await _qp.proxy().mutate_locally(m, tracing::trace_state_ptr(), db::commitlog::force_sync::no);
             }
-            // Replaying batches could have generated tombstones, flush to disk,
-            // where they can be compacted away.
-            return replica::database::flush_table_on_all_shards(_qp.proxy().get_db(), system_keyspace::NAME, system_keyspace::BATCHLOG);
-        }).then([] {
-            blogger.debug("Finished replayAllFailedBatches");
         });
+
+        blogger.debug("Finished replayAllFailedBatches with all_replayed: {}", all_replayed);
     });
 
     co_return all_replayed;

db/batchlog_manager.hh

@@ -34,12 +34,17 @@ class system_keyspace;
 using all_batches_replayed = bool_class<struct all_batches_replayed_tag>;
 
 struct batchlog_manager_config {
-    std::chrono::duration<double> write_request_timeout;
+    db_clock::duration replay_timeout;
     uint64_t replay_rate = std::numeric_limits<uint64_t>::max();
     std::chrono::milliseconds delay = std::chrono::milliseconds(0);
     unsigned replay_cleanup_after_replays;
 };
 
+enum class batchlog_stage : int8_t {
+    initial,
+    failed_replay
+};
+
 class batchlog_manager : public peering_sharded_service<batchlog_manager> {
 public:
     using post_replay_cleanup = bool_class<class post_replay_cleanup_tag>;
@@ -59,7 +64,7 @@ private:
 
     cql3::query_processor& _qp;
     db::system_keyspace& _sys_ks;
-    db_clock::duration _write_request_timeout;
+    db_clock::duration _replay_timeout;
     uint64_t _replay_rate;
     std::chrono::milliseconds _delay;
     unsigned _replay_cleanup_after_replays = 100;
@@ -71,6 +76,14 @@ private:
 
     gc_clock::time_point _last_replay;
 
+    // Was the v1 -> v2 migration already done since last restart?
+    // The migration is attempted once after each restart. This is redundant but
+    // keeps thing simple. Once no upgrade path exists from a ScyllaDB version
+    // which can still produce v1 entries, this migration code can be removed.
+    bool _migration_done = false;
+
+    future<> maybe_migrate_v1_to_v2();
+
     future<all_batches_replayed> replay_all_failed_batches(post_replay_cleanup cleanup);
 public:
     // Takes a QP, not a distributes. Because this object is supposed
@@ -85,10 +98,13 @@ public:
     future<all_batches_replayed> do_batch_log_replay(post_replay_cleanup cleanup);
 
     future<size_t> count_all_batches() const;
-    db_clock::duration get_batch_log_timeout() const;
     gc_clock::time_point get_last_replay() const {
         return _last_replay;
     }
+
+    const stats& stats() const {
+        return _stats;
+    }
 private:
     future<> batchlog_replay_loop();
 };

db/commitlog/commitlog_replayer.cc

@@ -54,12 +54,14 @@ public:
         uint64_t applied_mutations = 0;
         uint64_t corrupt_bytes = 0;
         uint64_t truncated_at = 0;
+        uint64_t broken_files = 0;
 
         stats& operator+=(const stats& s) {
             invalid_mutations += s.invalid_mutations;
             skipped_mutations += s.skipped_mutations;
             applied_mutations += s.applied_mutations;
             corrupt_bytes += s.corrupt_bytes;
+            broken_files += s.broken_files;
             return *this;
         }
         stats operator+(const stats& s) const {
@@ -192,6 +194,8 @@ db::commitlog_replayer::impl::recover(const commitlog::descriptor& d, const comm
             s->corrupt_bytes += e.bytes();
         } catch (commitlog::segment_truncation& e) {
             s->truncated_at = e.position();
+        } catch (commitlog::header_checksum_error&) {
+            ++s->broken_files;
         } catch (...) {
             throw;
         }
@@ -370,6 +374,9 @@ future<> db::commitlog_replayer::recover(std::vector<sstring> files, sstring fna
                     if (stats.truncated_at != 0) {
                         rlogger.warn("Truncated file: {} at position {}.", f, stats.truncated_at);
                     }
+                    if (stats.broken_files != 0) {
+                        rlogger.warn("Corrupted file header: {}. Skipped.", f);
+                    }
                     rlogger.debug("Log replay of {} complete, {} replayed mutations ({} invalid, {} skipped)"
                                     , f
                                     , stats.applied_mutations

db/config.cc

@@ -1105,6 +1105,14 @@ db::config::config(std::shared_ptr<db::extensions> exts)
         "Like native_transport_port, but clients-side port number (modulo smp) is used to route the connection to the specific shard.")
     , native_shard_aware_transport_port_ssl(this, "native_shard_aware_transport_port_ssl", value_status::Used, 19142,
         "Like native_transport_port_ssl, but clients-side port number (modulo smp) is used to route the connection to the specific shard.")
+    , native_transport_port_proxy_protocol(this, "native_transport_port_proxy_protocol", value_status::Used, 0,
+        "Port on which the CQL native transport listens for clients using proxy protocol v2. Disabled (0) by default.")
+    , native_transport_port_ssl_proxy_protocol(this, "native_transport_port_ssl_proxy_protocol", value_status::Used, 0,
+        "Port on which the CQL TLS native transport listens for clients using proxy protocol v2. Disabled (0) by default.")
+    , native_shard_aware_transport_port_proxy_protocol(this, "native_shard_aware_transport_port_proxy_protocol", value_status::Used, 0,
+        "Like native_transport_port_proxy_protocol, but clients-side port number (modulo smp) is used to route the connection to the specific shard.")
+    , native_shard_aware_transport_port_ssl_proxy_protocol(this, "native_shard_aware_transport_port_ssl_proxy_protocol", value_status::Used, 0,
+        "Like native_transport_port_ssl_proxy_protocol, but clients-side port number (modulo smp) is used to route the connection to the specific shard.")
     , native_transport_max_threads(this, "native_transport_max_threads", value_status::Invalid, 128,
         "The maximum number of thread handling requests. The meaning is the same as rpc_max_threads.\n"
         "Default is different (128 versus unlimited).\n"
@@ -1152,7 +1160,7 @@ db::config::config(std::shared_ptr<db::extensions> exts)
         "Number of threads with which to deliver hints. In multiple data-center deployments, consider increasing this number because cross data-center handoff is generally slower.")
     , batchlog_replay_throttle_in_kb(this, "batchlog_replay_throttle_in_kb", value_status::Unused, 1024,
         "Total maximum throttle. Throttling is reduced proportionally to the number of nodes in the cluster.")
-    , batchlog_replay_cleanup_after_replays(this, "batchlog_replay_cleanup_after_replays", liveness::LiveUpdate, value_status::Used, 60,
+    , batchlog_replay_cleanup_after_replays(this, "batchlog_replay_cleanup_after_replays", liveness::LiveUpdate, value_status::Used, 1,
         "Clean up batchlog memtable after every N replays. Replays are issued on a timer, every 60 seconds. So if batchlog_replay_cleanup_after_replays is set to 60, the batchlog memtable is flushed every 60 * 60 seconds.")
     /**
     * @Group Request scheduler properties

db/config.hh

@@ -324,6 +324,10 @@ public:
     named_value<uint16_t> native_transport_port_ssl;
     named_value<uint16_t> native_shard_aware_transport_port;
     named_value<uint16_t> native_shard_aware_transport_port_ssl;
+    named_value<uint16_t> native_transport_port_proxy_protocol;
+    named_value<uint16_t> native_transport_port_ssl_proxy_protocol;
+    named_value<uint16_t> native_shard_aware_transport_port_proxy_protocol;
+    named_value<uint16_t> native_shard_aware_transport_port_ssl_proxy_protocol;
     named_value<uint32_t> native_transport_max_threads;
     named_value<uint32_t> native_transport_max_frame_size_in_mb;
     named_value<sstring> broadcast_rpc_address;

db/hints/internal/hint_endpoint_manager.cc

@@ -248,7 +248,7 @@ future<db::commitlog> hint_endpoint_manager::add_store() noexcept {
             // which is larger than the segment ID of the RP of the last written hint.
             cfg.base_segment_id = _last_written_rp.base_id();
 
-            return commitlog::create_commitlog(std::move(cfg)).then([this] (commitlog l) -> future<commitlog> {
+            return commitlog::create_commitlog(std::move(cfg)).then([this] (this auto, commitlog l) -> future<commitlog> {
                 // add_store() is triggered every time hint files are forcefully flushed to I/O (every hints_flush_period).
                 // When this happens we want to refill _sender's segments only if it has finished with the segments he had before.
                 if (_sender.have_segments()) {

db/hints/manager.cc

@@ -26,6 +26,7 @@
 #include <seastar/core/smp.hh>
 #include <seastar/coroutine/exception.hh>
 #include <seastar/coroutine/parallel_for_each.hh>
+#include <seastar/util/file.hh>
 
 // Boost features.
 
@@ -899,7 +900,7 @@ future<> manager::migrate_ip_directories() {
     co_await coroutine::parallel_for_each(dirs_to_remove, [] (auto& directory) -> future<> {
         try {
             manager_logger.warn("Removing hint directory {}", directory.native());
-            co_await lister::rmdir(directory);
+            co_await seastar::recursive_remove_directory(directory);
         } catch (...) {
             on_internal_error(manager_logger,
                     seastar::format("Removing a hint directory has failed. Reason: {}", std::current_exception()));

db/object_storage_endpoint_param.cc

@@ -12,8 +12,6 @@
 #include <yaml-cpp/yaml.h>
 
 #include <boost/lexical_cast.hpp>
-
-#include "utils/s3/creds.hh"
 #include "object_storage_endpoint_param.hh"
 
 using namespace std::string_literals;
@@ -21,9 +19,6 @@ using namespace std::string_literals;
 db::object_storage_endpoint_param::object_storage_endpoint_param(s3_storage s)
     : _data(std::move(s))
 {}
-db::object_storage_endpoint_param::object_storage_endpoint_param(std::string endpoint, s3::endpoint_config config)
-    : object_storage_endpoint_param(s3_storage{std::move(endpoint), std::move(config)})
-{}
 db::object_storage_endpoint_param::object_storage_endpoint_param(gs_storage s)
     : _data(std::move(s))
 {}
@@ -32,8 +27,8 @@ db::object_storage_endpoint_param::object_storage_endpoint_param() = default;
 db::object_storage_endpoint_param::object_storage_endpoint_param(const object_storage_endpoint_param&) = default;
 
 std::string db::object_storage_endpoint_param::s3_storage::to_json_string() const {
-    return fmt::format("{{ \"port\": {}, \"use_https\": {}, \"aws_region\": \"{}\", \"iam_role_arn\": \"{}\" }}",
-        config.port, config.use_https, config.region, config.role_arn
+    return fmt::format("{{ \"type\": \"s3\", \"aws_region\": \"{}\", \"iam_role_arn\": \"{}\" }}",
+        region, iam_role_arn
     );
 }
 
@@ -99,8 +94,6 @@ const std::string& db::object_storage_endpoint_param::type() const {
 
 db::object_storage_endpoint_param db::object_storage_endpoint_param::decode(const YAML::Node& node) {
     auto name = node["name"];
-    auto aws_region = node["aws_region"];
-    auto iam_role_arn = node["iam_role_arn"];
     auto type = node["type"];
 
     auto get_opt = [](auto& node, const std::string& key, auto def) {
@@ -108,13 +101,19 @@ db::object_storage_endpoint_param db::object_storage_endpoint_param::decode(cons
         return tmp ? tmp.template as<std::decay_t<decltype(def)>>() : def;
     };
     // aws s3 endpoint. 
-    if (!type || type.as<std::string>() == s3_type || aws_region || iam_role_arn) {
+    if (!type || type.as<std::string>() == s3_type ) {
         s3_storage ep;
         ep.endpoint = name.as<std::string>();
-        ep.config.port = node["port"].as<unsigned>();
-        ep.config.use_https = node["https"].as<bool>(false);
-        ep.config.region = aws_region ? aws_region.as<std::string>() : std::getenv("AWS_DEFAULT_REGION");
-        ep.config.role_arn = iam_role_arn ? iam_role_arn.as<std::string>() : "";
+        auto aws_region = node["aws_region"];
+        ep.region = aws_region ? aws_region.as<std::string>() : std::getenv("AWS_DEFAULT_REGION");
+        ep.iam_role_arn = get_opt(node, "iam_role_arn", ""s);
+
+        if (maybe_legacy_endpoint_name(ep.endpoint)) {
+            // Support legacy config for a while
+            auto port = node["port"].as<unsigned>();
+            auto use_https = node["https"].as<bool>(false);
+            ep.endpoint = fmt::format("http{}://{}:{}", use_https ? "s" : "", ep.endpoint, port);
+        }
 
         return object_storage_endpoint_param{std::move(ep)};
     }
@@ -135,5 +134,5 @@ const std::string db::object_storage_endpoint_param::gs_type = "gs";
 
 auto fmt::formatter<db::object_storage_endpoint_param>::format(const db::object_storage_endpoint_param& e, fmt::format_context& ctx) const
     -> decltype(ctx.out()) {
-    return fmt::format_to(ctx.out(), "object_storage_endpoint_param{{}}", e.to_json_string());
+    return fmt::format_to(ctx.out(), "object_storage_endpoint_param{}", e.to_json_string());
 }

db/object_storage_endpoint_param.hh

@@ -13,7 +13,6 @@
 #include <variant>
 #include <compare>
 #include <fmt/core.h>
-#include "utils/s3/creds.hh"
 
 namespace YAML {
     class Node;
@@ -25,7 +24,8 @@ class object_storage_endpoint_param {
 public:
     struct s3_storage {
         std::string endpoint;
-        s3::endpoint_config config;
+        std::string region;
+        std::string iam_role_arn;
 
         std::strong_ordering operator<=>(const s3_storage&) const = default;
         std::string to_json_string() const;
@@ -43,7 +43,6 @@ public:
     object_storage_endpoint_param();
     object_storage_endpoint_param(const object_storage_endpoint_param&);
     object_storage_endpoint_param(s3_storage);
-    object_storage_endpoint_param(std::string endpoint, s3::endpoint_config config);
     object_storage_endpoint_param(gs_storage);
 
     std::strong_ordering operator<=>(const object_storage_endpoint_param&) const;
@@ -77,3 +76,7 @@ template <>
 struct fmt::formatter<db::object_storage_endpoint_param> : fmt::formatter<std::string_view> {
     auto format(const db::object_storage_endpoint_param&, fmt::format_context& ctx) const -> decltype(ctx.out());
 };
+
+inline bool maybe_legacy_endpoint_name(std::string_view ep) noexcept {
+    return !(ep.starts_with("http://") || ep.starts_with("https://"));
+}

db/schema_applier.cc

@@ -1262,16 +1262,9 @@ static future<> do_merge_schema(sharded<service::storage_proxy>& proxy,  sharded
 {
     slogger.trace("do_merge_schema: {}", mutations);
     schema_applier ap(proxy, ss, sys_ks, reload);
-    std::exception_ptr ex;
-    try {
-        co_await execute_do_merge_schema(proxy, ap, std::move(mutations));
-    } catch (...) {
-        ex = std::current_exception();
-    }
-    co_await ap.destroy();
-    if (ex) {
-        throw ex;
-    }
+    co_await execute_do_merge_schema(proxy, ap, std::move(mutations)).finally([&ap]() {
+        return ap.destroy();
+    });
 }
 
 /**

db/system_keyspace.cc

@@ -110,6 +110,7 @@ namespace {
             system_keyspace::v3::CDC_LOCAL,
             system_keyspace::DICTS,
             system_keyspace::VIEW_BUILDING_TASKS,
+            system_keyspace::CLIENT_ROUTES,
         };
         if (ks_name == system_keyspace::NAME && tables.contains(cf_name)) {
             props.enable_schema_commitlog();
@@ -137,6 +138,7 @@ namespace {
                 system_keyspace::ROLE_PERMISSIONS,
                 system_keyspace::DICTS,
                 system_keyspace::VIEW_BUILDING_TASKS,
+                system_keyspace::CLIENT_ROUTES,
             };
             if (ks_name == system_keyspace::NAME && tables.contains(cf_name)) {
                 props.is_group0_table = true;
@@ -213,6 +215,30 @@ schema_ptr system_keyspace::batchlog() {
     return batchlog;
 }
 
+schema_ptr system_keyspace::batchlog_v2() {
+    static thread_local auto batchlog_v2 = [] {
+        schema_builder builder(generate_legacy_id(NAME, BATCHLOG_V2), NAME, BATCHLOG_V2,
+        // partition key
+        {{"version", int32_type}, {"stage", byte_type}, {"shard", int32_type}},
+        // clustering key
+        {{"written_at", timestamp_type}, {"id", uuid_type}},
+        // regular columns
+        {{"data", bytes_type}},
+        // static columns
+        {},
+        // regular column name type
+        utf8_type,
+        // comment
+        "batches awaiting replay"
+       );
+       builder.set_gc_grace_seconds(0);
+       builder.set_caching_options(caching_options::get_disabled_caching_options());
+       builder.with_hash_version();
+       return builder.build(schema_builder::compact_storage::no);
+    }();
+    return batchlog_v2;
+}
+
 /*static*/ schema_ptr system_keyspace::paxos() {
     static thread_local auto paxos = [] {
         // FIXME: switch to the new schema_builder interface (with_column(...), etc)
@@ -285,6 +311,7 @@ schema_ptr system_keyspace::topology() {
             .with_column("tablet_balancing_enabled", boolean_type, column_kind::static_column)
             .with_column("upgrade_state", utf8_type, column_kind::static_column)
             .with_column("global_requests", set_type_impl::get_instance(timeuuid_type, true), column_kind::static_column)
+            .with_column("paused_rf_change_requests", set_type_impl::get_instance(timeuuid_type, true), column_kind::static_column)
             .set_comment("Current state of topology change machine")
             .with_hash_version()
             .build();
@@ -1391,6 +1418,23 @@ schema_ptr system_keyspace::view_building_tasks() {
     return schema;
 }
 
+schema_ptr system_keyspace::client_routes() {
+    static thread_local auto schema = [] {
+        auto id = generate_legacy_id(NAME, CLIENT_ROUTES);
+        return schema_builder(NAME, CLIENT_ROUTES, std::make_optional(id))
+                .with_column("connection_id", utf8_type, column_kind::partition_key)
+                .with_column("host_id", uuid_type, column_kind::clustering_key)
+                .with_column("address", utf8_type)
+                .with_column("port", int32_type)
+                .with_column("tls_port", int32_type)
+                .with_column("alternator_port", int32_type)
+                .with_column("alternator_https_port", int32_type)
+                .with_hash_version()
+                .build();
+    }();
+    return schema;
+}
+
 future<system_keyspace::local_info> system_keyspace::load_local_info() {
     auto msg = co_await execute_cql(format("SELECT host_id, cluster_name, data_center, rack FROM system.{} WHERE key=?", LOCAL), sstring(LOCAL));
 
@@ -2304,7 +2348,7 @@ std::vector<schema_ptr> system_keyspace::all_tables(const db::config& cfg) {
     std::copy(schema_tables.begin(), schema_tables.end(), std::back_inserter(r));
     auto auth_tables = system_keyspace::auth_tables();
     std::copy(auth_tables.begin(), auth_tables.end(), std::back_inserter(r));
-    r.insert(r.end(), { built_indexes(), hints(), batchlog(), paxos(), local(),
+    r.insert(r.end(), { built_indexes(), hints(), batchlog(), batchlog_v2(), paxos(), local(),
                     peers(), peer_events(), range_xfers(),
                     compactions_in_progress(), compaction_history(),
                     sstable_activity(), size_estimates(), large_partitions(), large_rows(), large_cells(),
@@ -2318,7 +2362,7 @@ std::vector<schema_ptr> system_keyspace::all_tables(const db::config& cfg) {
                     v3::cdc_local(),
                     raft(), raft_snapshots(), raft_snapshot_config(), group0_history(), discovery(),
                     topology(), cdc_generations_v3(), topology_requests(), service_levels_v2(), view_build_status_v2(),
-                    dicts(), view_building_tasks(), cdc_streams_state(), cdc_streams_history()
+                    dicts(), view_building_tasks(), client_routes(), cdc_streams_state(), cdc_streams_history()
     });
 
     if (cfg.check_experimental(db::experimental_features_t::feature::BROADCAST_TABLES)) {
@@ -2335,7 +2379,9 @@ std::vector<schema_ptr> system_keyspace::all_tables(const db::config& cfg) {
 }
 
 static bool maybe_write_in_user_memory(schema_ptr s) {
-    return (s.get() == system_keyspace::batchlog().get()) || (s.get() == system_keyspace::paxos().get())
+    return (s.get() == system_keyspace::batchlog().get())
+            || (s.get() == system_keyspace::batchlog_v2().get())
+            || (s.get() == system_keyspace::paxos().get())
             || s == system_keyspace::v3::scylla_views_builds_in_progress();
 }
 
@@ -3111,7 +3157,10 @@ static bool must_have_tokens(service::node_state nst) {
     // A decommissioning node doesn't have tokens at the end, they are
     // removed during transition to the left_token_ring state.
     case service::node_state::decommissioning: return false;
-    case service::node_state::removing: return true;
+    // A removing node might or might not have tokens depending on whether
+    // REMOVENODE_WITH_LEFT_TOKEN_RING feature is enabled. To support both
+    // cases, we allow removing nodes to not have tokens.
+    case service::node_state::removing: return false;
     case service::node_state::rebuilding: return true;
     case service::node_state::normal: return true;
     case service::node_state::left: return false;
@@ -3351,6 +3400,12 @@ future<service::topology> system_keyspace::load_topology_state(const std::unorde
             }
         }
 
+        if (some_row.has("paused_rf_change_requests")) {
+            for (auto&& v : deserialize_set_column(*topology(), some_row, "paused_rf_change_requests")) {
+                ret.paused_rf_change_requests.insert(value_cast<utils::UUID>(v));
+            }
+        }
+
         if (some_row.has("enabled_features")) {
             ret.enabled_features = decode_features(deserialize_set_column(*topology(), some_row, "enabled_features"));
         }
@@ -3562,35 +3617,43 @@ system_keyspace::topology_requests_entry system_keyspace::topology_request_row_t
     return entry;
 }
 
-future<system_keyspace::topology_requests_entry> system_keyspace::get_topology_request_entry(utils::UUID id, bool require_entry) {
+future<system_keyspace::topology_requests_entry> system_keyspace::get_topology_request_entry(utils::UUID id) {
+    auto r = co_await get_topology_request_entry_opt(id);
+    if (!r) {
+        on_internal_error(slogger, format("no entry for request id {}", id));
+    }
+    co_return std::move(*r);
+}
+
+future<std::optional<system_keyspace::topology_requests_entry>> system_keyspace::get_topology_request_entry_opt(utils::UUID id) {
     auto rs = co_await execute_cql(
         format("SELECT * FROM system.{} WHERE id = {}", TOPOLOGY_REQUESTS, id));
 
     if (!rs || rs->empty()) {
-        if (require_entry) {
-            on_internal_error(slogger, format("no entry for request id {}", id));
-        } else {
-            co_return topology_requests_entry{
-                .id = utils::null_uuid()
-            };
-        }
+        co_return std::nullopt;
     }
 
     const auto& row = rs->one();
     co_return topology_request_row_to_entry(id, row);
 }
 
-future<system_keyspace::topology_requests_entries> system_keyspace::get_node_ops_request_entries(db_clock::time_point end_time_limit) {
+future<system_keyspace::topology_requests_entries> system_keyspace::get_topology_request_entries(std::vector<std::variant<service::topology_request, service::global_topology_request>> request_types, db_clock::time_point end_time_limit) {
+    sstring request_types_str = "";
+    bool first = true;
+    for (const auto& rt : request_types) {
+        if (!std::exchange(first, false)) {
+            request_types_str += ", ";
+        }
+        request_types_str += std::visit([] (auto&& arg) { return fmt::format("'{}'", arg); }, rt);
+    }
+
     // Running requests.
     auto rs_running = co_await execute_cql(
-        format("SELECT * FROM system.{} WHERE done = false AND request_type IN ('{}', '{}', '{}', '{}', '{}') ALLOW FILTERING", TOPOLOGY_REQUESTS,
-            service::topology_request::join, service::topology_request::replace, service::topology_request::rebuild, service::topology_request::leave, service::topology_request::remove));
-
+        format("SELECT * FROM system.{} WHERE done = false AND request_type IN ({}) ALLOW FILTERING", TOPOLOGY_REQUESTS, request_types_str));
 
     // Requests which finished after end_time_limit.
     auto rs_done = co_await execute_cql(
-        format("SELECT * FROM system.{} WHERE end_time > {} AND request_type IN ('{}', '{}', '{}', '{}', '{}') ALLOW FILTERING", TOPOLOGY_REQUESTS, end_time_limit.time_since_epoch().count(),
-            service::topology_request::join, service::topology_request::replace, service::topology_request::rebuild, service::topology_request::leave, service::topology_request::remove));
+        format("SELECT * FROM system.{} WHERE end_time > {} AND request_type IN ({}) ALLOW FILTERING", TOPOLOGY_REQUESTS, end_time_limit.time_since_epoch().count(), request_types_str));
 
     topology_requests_entries m;
     for (const auto& row: *rs_done) {
@@ -3608,6 +3671,16 @@ future<system_keyspace::topology_requests_entries> system_keyspace::get_node_ops
     co_return m;
 }
 
+future<system_keyspace::topology_requests_entries> system_keyspace::get_node_ops_request_entries(db_clock::time_point end_time_limit) {
+    return get_topology_request_entries({
+        service::topology_request::join,
+        service::topology_request::replace,
+        service::topology_request::rebuild,
+        service::topology_request::leave,
+        service::topology_request::remove
+    }, end_time_limit);
+}
+
 future<mutation> system_keyspace::get_insert_dict_mutation(
     std::string_view name,
     bytes data,

db/system_keyspace.hh

@@ -163,6 +163,7 @@ public:
     static constexpr auto NAME = "system";
     static constexpr auto HINTS = "hints";
     static constexpr auto BATCHLOG = "batchlog";
+    static constexpr auto BATCHLOG_V2 = "batchlog_v2";
     static constexpr auto PAXOS = "paxos";
     static constexpr auto BUILT_INDEXES = "IndexInfo";
     static constexpr auto LOCAL = "local";
@@ -198,6 +199,8 @@ public:
     static constexpr auto VIEW_BUILD_STATUS_V2 = "view_build_status_v2";
     static constexpr auto DICTS = "dicts";
     static constexpr auto VIEW_BUILDING_TASKS = "view_building_tasks";
+    static constexpr auto CLIENT_ROUTES = "client_routes";
+    static constexpr auto VERSIONS = "versions";
 
     // auth
     static constexpr auto ROLES = "roles";
@@ -255,6 +258,7 @@ public:
 
     static schema_ptr hints();
     static schema_ptr batchlog();
+    static schema_ptr batchlog_v2();
     static schema_ptr paxos();
     static schema_ptr built_indexes(); // TODO (from Cassandra): make private
     static schema_ptr raft();
@@ -274,6 +278,7 @@ public:
     static schema_ptr view_build_status_v2();
     static schema_ptr dicts();
     static schema_ptr view_building_tasks();
+    static schema_ptr client_routes();
 
     // auth
     static schema_ptr roles();
@@ -665,7 +670,9 @@ public:
 
     future<service::topology_request_state> get_topology_request_state(utils::UUID id, bool require_entry);
     topology_requests_entry topology_request_row_to_entry(utils::UUID id, const cql3::untyped_result_set_row& row);
-    future<topology_requests_entry> get_topology_request_entry(utils::UUID id, bool require_entry);
+    future<topology_requests_entry> get_topology_request_entry(utils::UUID id);
+    future<std::optional<topology_requests_entry>> get_topology_request_entry_opt(utils::UUID id);
+    future<system_keyspace::topology_requests_entries> get_topology_request_entries(std::vector<std::variant<service::topology_request, service::global_topology_request>> request_types, db_clock::time_point end_time_limit);
     future<topology_requests_entries> get_node_ops_request_entries(db_clock::time_point end_time_limit);
 
 public:

db/view/view.cc

@@ -1744,6 +1744,115 @@ bool should_generate_view_updates_on_this_shard(const schema_ptr& base, const lo
         && std::ranges::contains(shards, this_shard_id());
 }
 
+static endpoints_to_update get_view_natural_endpoint_vnodes(
+        locator::host_id me,
+        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 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;
+    auto& my_datacenter = my_location.dc;
+
+    auto process_candidate = [&] (node_vector& nodes, std::reference_wrapper<const locator::node> node) {
+        if (!network_topology || node.get().dc() == my_datacenter) {
+            nodes.emplace_back(node);
+        }
+    };
+
+    for (auto&& base_node : base_nodes) {
+        process_candidate(base_endpoints, base_node);
+    }
+
+    for (auto&& view_node : view_nodes) {
+        auto it = std::ranges::find(base_endpoints, view_node.get().host_id(), std::mem_fn(&locator::node::host_id));
+        // If this base replica is also one of the view replicas, we use
+        // ourselves as the view replica.
+        // We don't return an extra endpoint, as it's only needed when
+        // using tablets (so !use_legacy_self_pairing)
+        if (view_node.get().host_id() == me && it != base_endpoints.end()) {
+            return {.natural_endpoint = me};
+        }
+
+        // We have to remove any endpoint which is shared between the base
+        // and the view, as it will select itself and throw off the counts
+        // otherwise.
+        if (it != base_endpoints.end()) {
+            base_endpoints.erase(it);
+        } else if (!network_topology || view_node.get().dc() == my_datacenter) {
+            view_endpoints.push_back(view_node);
+        }
+    }
+
+    auto base_it = std::ranges::find(base_endpoints, me, std::mem_fn(&locator::node::host_id));
+    if (base_it == base_endpoints.end()) {
+        // This node is not a base replica of this key, so we return empty
+        // FIXME: This case shouldn't happen, and if it happens, a view update
+        // would be lost.
+        ++cf_stats.total_view_updates_on_wrong_node;
+        vlogger.warn("Could not find {} in base_endpoints={}", me,
+                base_endpoints | std::views::transform(std::mem_fn(&locator::node::host_id)));
+        return {};
+    }
+    size_t idx = base_it - base_endpoints.begin();
+    return {.natural_endpoint = view_endpoints[idx].get().host_id()};
+}
+
+static std::optional<locator::host_id> get_unpaired_view_endpoint(
+        std::vector<std::reference_wrapper<const locator::node>> base_nodes,
+        std::vector<std::reference_wrapper<const locator::node>> view_nodes,
+        replica::cf_stats& cf_stats) {
+    std::unordered_set<locator::endpoint_dc_rack> base_dc_racks;
+    for (auto&& base_node : base_nodes) {
+        if (base_dc_racks.contains(base_node.get().dc_rack())) {
+            // We can't do rack-aware pairing if there are multiple replicas in the same rack.
+            ++cf_stats.total_view_updates_failed_pairing;
+            vlogger.warn("Can't perform base-view pairing in this topology. There are multiple base table replicas in the same dc/rack({}/{}):",
+                    base_node.get().dc(), base_node.get().rack());
+            return std::nullopt;
+        }
+        base_dc_racks.insert(base_node.get().dc_rack());
+    }
+
+    std::unordered_set<locator::endpoint_dc_rack> paired_view_dc_racks;
+    std::unordered_map<locator::endpoint_dc_rack, locator::host_id> unpaired_view_dc_rack_replicas;
+    for (auto&& view_node : view_nodes) {
+        if (paired_view_dc_racks.contains(view_node.get().dc_rack()) || unpaired_view_dc_rack_replicas.contains(view_node.get().dc_rack())) {
+            // We can't do rack-aware pairing if there are multiple replicas in the same rack.
+            ++cf_stats.total_view_updates_failed_pairing;
+            vlogger.warn("Can't perform base-view pairing in this topology. There are multiple view table replicas in the same dc/rack({}/{}):",
+                    view_node.get().dc(), view_node.get().rack());
+            return std::nullopt;
+        }
+        // Track unpaired replicas in both sets
+        if (base_dc_racks.contains(view_node.get().dc_rack())) {
+            paired_view_dc_racks.insert(view_node.get().dc_rack());
+        } else {
+            unpaired_view_dc_rack_replicas.insert({view_node.get().dc_rack(), view_node.get().host_id()});
+        }
+    }
+
+    if (unpaired_view_dc_rack_replicas.size() > 0) {
+        // There are view replicas that can't be paired with any base replica
+        // This can happen as a result of an RF change when the view replica finishes streaming
+        // before the base replica.
+        // Because of this, a view replica might not get paired with any base replica, so we need
+        // to send an additional update to it.
+        ++cf_stats.total_view_updates_due_to_replica_count_mismatch;
+        auto extra_replica = unpaired_view_dc_rack_replicas.begin()->second;
+        unpaired_view_dc_rack_replicas.erase(unpaired_view_dc_rack_replicas.begin());
+        if (unpaired_view_dc_rack_replicas.size() > 0) {
+            // We only expect one extra replica to appear due to an RF change. If there's more, that's an error,
+            // but we'll still perform updates to the paired and last replicas to minimize degradation.
+            vlogger.warn("There are too many view endpoints for base-view pairing. View updates may get lost on view_endpoints={}",
+                    unpaired_view_dc_rack_replicas | std::views::values);
+        }
+        return extra_replica;
+    }
+    return std::nullopt;
+}
+
 // Calculate the node ("natural endpoint") to which this node should send
 // a view update.
 //
@@ -1756,29 +1865,19 @@ bool should_generate_view_updates_on_this_shard(const schema_ptr& base, const lo
 // of this function is to find, assuming that this node is one of the base
 // replicas for a given partition, the paired view replica.
 //
-// In the past, we used an optimization called "self-pairing" that if a single
-// node was both a base replica and a view replica for a write, the pairing is
-// modified so that this node would send the update to itself. This self-
-// pairing optimization could cause the pairing to change after view ranges
-// are moved between nodes, so currently we only use it if
-// use_legacy_self_pairing is set to true. When using tablets - where range
-// movements are common - it is strongly recommended to set it to false.
+// When using vnodes, we have an optimization called "self-pairing" - if a single
+// node is both a base replica and a view replica for a write, the pairing is
+// modified so that this node sends the update to itself and this node is removed
+// from the lists of nodes paired by index. This self-pairing optimization can
+// cause the pairing to change after view ranges are moved between nodes.
 //
 // If the keyspace's replication strategy is a NetworkTopologyStrategy,
 // we pair only nodes in the same datacenter.
 //
-// When use_legacy_self_pairing is enabled, if one of the base replicas
-// also happens to be a view replica, it is paired with itself
-// (with the other nodes paired by order in the list
-// after taking this node out).
-//
-// If the table uses tablets and the replication strategy is NetworkTopologyStrategy
-// and the replication factor in the node's datacenter is a multiple of the number
-// of racks in the datacenter, then pairing is rack-aware.  In this case,
-// all racks have the same number of replicas, and those are never migrated
-// outside their racks. Therefore, the base replicas are naturally paired with the
-// view replicas that are in the same rack, based on the ordinal position.
-// Note that typically, there is a single replica per rack and pairing is trivial.
+// If the table uses tablets, then pairing is rack-aware. In this case, in each
+// rack where we have a base replica there is also one replica of each view tablet.
+// Therefore, the base replicas are naturally paired with the view replicas that
+// are in the same rack.
 //
 // If the assumption that the given base token belongs to this replica
 // does not hold, we return an empty optional.
@@ -1806,19 +1905,12 @@ endpoints_to_update get_view_natural_endpoint(
         const locator::abstract_replication_strategy& replication_strategy,
         const dht::token& base_token,
         const dht::token& view_token,
-        bool use_legacy_self_pairing,
-        bool use_tablets_rack_aware_view_pairing,
+        bool use_tablets,
         replica::cf_stats& cf_stats) {
     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& my_datacenter = my_location.dc;
     auto* network_topology = dynamic_cast<const locator::network_topology_strategy*>(&replication_strategy);
-    auto rack_aware_pairing = use_tablets_rack_aware_view_pairing && network_topology;
-    bool simple_rack_aware_pairing = false;
-    using node_vector = std::vector<std::reference_wrapper<const locator::node>>;
-    node_vector orig_base_endpoints, orig_view_endpoints;
-    node_vector base_endpoints, view_endpoints;
 
     auto resolve = [&] (const locator::topology& topology, const locator::host_id& ep, bool is_view) -> const locator::node& {
         if (auto* np = topology.find_node(ep)) {
@@ -1829,6 +1921,7 @@ endpoints_to_update get_view_natural_endpoint(
 
     // We need to use get_replicas() for pairing to be stable in case base or view tablet
     // is rebuilding a replica which has left the ring. get_natural_endpoints() filters such replicas.
+    using node_vector = std::vector<std::reference_wrapper<const locator::node>>;
     auto base_nodes = base_erm->get_replicas(base_token) | std::views::transform([&] (const locator::host_id& ep) -> const locator::node& {
         return resolve(topology, ep, false);
     }) | std::ranges::to<node_vector>();
@@ -1852,231 +1945,43 @@ endpoints_to_update get_view_natural_endpoint(
                 // 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, replication_strategy, base_token,
-                        view_token, use_legacy_self_pairing, use_tablets_rack_aware_view_pairing, cf_stats);
+                        view_token, use_tablets, cf_stats);
             }
         }
     }
 
-    std::function<bool(const locator::node&)> is_candidate;
-    if (network_topology) {
-        is_candidate = [&] (const locator::node& node) { return node.dc() == my_datacenter; };
-    } else {
-        is_candidate = [&] (const locator::node&) { return true; };
-    }
-    auto process_candidate = [&] (node_vector& nodes, std::reference_wrapper<const locator::node> node) {
-        if (is_candidate(node)) {
-            nodes.emplace_back(node);
-        }
-    };
-
-    for (auto&& base_node : base_nodes) {
-        process_candidate(base_endpoints, base_node);
+    if (!use_tablets) {
+        return get_view_natural_endpoint_vnodes(
+                me,
+                base_nodes,
+                view_nodes,
+                my_location,
+                network_topology,
+                cf_stats);
     }
 
-    if (use_legacy_self_pairing) {
-        for (auto&& view_node : view_nodes) {
-            auto it = std::ranges::find(base_endpoints, view_node.get().host_id(), std::mem_fn(&locator::node::host_id));
-            // If this base replica is also one of the view replicas, we use
-            // ourselves as the view replica.
-            // We don't return an extra endpoint, as it's only needed when
-            // using tablets (so !use_legacy_self_pairing)
-            if (view_node.get().host_id() == me && it != base_endpoints.end()) {
-                return {.natural_endpoint = me};
-            }
-
-            // We have to remove any endpoint which is shared between the base
-            // and the view, as it will select itself and throw off the counts
-            // otherwise.
-            if (it != base_endpoints.end()) {
-                base_endpoints.erase(it);
-            } else if (is_candidate(view_node)) {
-                view_endpoints.push_back(view_node);
-            }
-        }
-    } else {
-        for (auto&& view_node : view_nodes) {
-            process_candidate(view_endpoints, view_node);
+    std::optional<locator::host_id> paired_replica;
+    for (auto&& view_node : view_nodes) {
+        if (view_node.get().dc_rack() == my_location) {
+            paired_replica = view_node.get().host_id();
+            break;
         }
     }
-
-    // Try optimizing for simple rack-aware pairing
-    // If the numbers of base and view replica differ, that means an RF change is taking place
-    // and we can't use simple rack-aware pairing.
-    if (rack_aware_pairing && base_endpoints.size() == view_endpoints.size()) {
-        auto dc_rf = network_topology->get_replication_factor(my_datacenter);
-        const auto& racks = topology.get_datacenter_rack_nodes().at(my_datacenter);
-        // Simple rack-aware pairing is possible when the datacenter replication factor
-        // is a multiple of the number of racks in the datacenter.
-        if (dc_rf % racks.size() == 0) {
-            simple_rack_aware_pairing = true;
-            size_t rack_rf = dc_rf / racks.size();
-            // If any rack doesn't have enough nodes to satisfy the per-rack rf
-            // simple rack-aware pairing is disabled.
-            for (const auto& [rack, nodes] : racks) {
-                if (nodes.size() < rack_rf) {
-                    simple_rack_aware_pairing = false;
-                    break;
-                }
-            }
-        }
-        if (dc_rf != base_endpoints.size()) {
-            // If the datacenter replication factor is not equal to the number of base replicas,
-            // we're in progress of a RF change and we can't use simple rack-aware pairing.
-            simple_rack_aware_pairing = false;
-        }
-        if (simple_rack_aware_pairing) {
-            std::erase_if(base_endpoints, [&] (const locator::node& node) { return node.dc_rack() != my_location; });
-            std::erase_if(view_endpoints, [&] (const locator::node& node) { return node.dc_rack() != my_location; });
-        }
+    if (paired_replica && base_nodes.size() == view_nodes.size()) {
+        // We don't need to find any extra replicas, so we can return early
+        return {.natural_endpoint = paired_replica};
     }
-
-    orig_base_endpoints = base_endpoints;
-    orig_view_endpoints = view_endpoints;
-
-    // For the complex rack_aware_pairing case, nodes are already filtered by datacenter
-    // Use best-match, for the minimum number of base and view replicas in each rack,
-    // and ordinal match for the rest.
-    std::optional<std::reference_wrapper<const locator::node>> paired_replica;
-    if (rack_aware_pairing && !simple_rack_aware_pairing) {
-        struct indexed_replica {
-            size_t idx;
-            std::reference_wrapper<const locator::node> node;
-        };
-        std::unordered_map<sstring, std::vector<indexed_replica>> base_racks, view_racks;
-
-        // First, index all replicas by rack
-        auto index_replica_set = [] (std::unordered_map<sstring, std::vector<indexed_replica>>& racks, const node_vector& replicas) {
-            size_t idx = 0;
-            for (const auto& r: replicas) {
-                racks[r.get().rack()].emplace_back(idx++, r);
-            }
-        };
-        index_replica_set(base_racks, base_endpoints);
-        index_replica_set(view_racks, view_endpoints);
-
-        // Try optimistically pairing `me` first
-        const auto& my_base_replicas = base_racks[my_location.rack];
-        auto base_it = std::ranges::find(my_base_replicas, me, [] (const indexed_replica& ir) { return ir.node.get().host_id(); });
-        if (base_it == my_base_replicas.end()) {
-            return {};
-        }
-        const auto& my_view_replicas = view_racks[my_location.rack];
-        size_t idx = base_it - my_base_replicas.begin();
-        if (idx < my_view_replicas.size()) {
-            if (orig_view_endpoints.size() <= orig_base_endpoints.size()) {
-                return {.natural_endpoint = my_view_replicas[idx].node.get().host_id()};
-            } else {
-                // If the number of view replicas is larger than the number of base replicas,
-                // we need to find the unpaired view replica, so we can't return yet.
-                paired_replica = my_view_replicas[idx].node;
-            }
-        }
-
-        // Collect all unpaired base and view replicas,
-        // where the number of replicas in the base rack is different than the respective view rack
-        std::vector<indexed_replica> unpaired_base_replicas, unpaired_view_replicas;
-        for (const auto& [rack, base_replicas] : base_racks) {
-            const auto& view_replicas = view_racks[rack];
-            for (auto i = view_replicas.size(); i < base_replicas.size(); ++i) {
-                unpaired_base_replicas.emplace_back(base_replicas[i]);
-            }
-        }
-        for (const auto& [rack, view_replicas] : view_racks) {
-            const auto& base_replicas = base_racks[rack];
-            for (auto i = base_replicas.size(); i < view_replicas.size(); ++i) {
-                unpaired_view_replicas.emplace_back(view_replicas[i]);
-            }
-        }
-
-        // Sort by the original ordinality, and copy the sorted results
-        // back into {base,view}_endpoints, for backward compatible processing below.
-        std::ranges::sort(unpaired_base_replicas, std::less(), std::mem_fn(&indexed_replica::idx));
-        base_endpoints.clear();
-        std::ranges::transform(unpaired_base_replicas, std::back_inserter(base_endpoints), std::mem_fn(&indexed_replica::node));
-
-        std::ranges::sort(unpaired_view_replicas, std::less(), std::mem_fn(&indexed_replica::idx));
-        view_endpoints.clear();
-        std::ranges::transform(unpaired_view_replicas, std::back_inserter(view_endpoints), std::mem_fn(&indexed_replica::node));
-    }
-
-    auto base_it = std::ranges::find(base_endpoints, me, std::mem_fn(&locator::node::host_id));
-    if (!paired_replica && base_it == base_endpoints.end()) {
-        // This node is not a base replica of this key, so we return empty
-        // FIXME: This case shouldn't happen, and if it happens, a view update
-        // would be lost.
-        ++cf_stats.total_view_updates_on_wrong_node;
-        vlogger.warn("Could not find {} in base_endpoints={}", me,
-                orig_base_endpoints | std::views::transform(std::mem_fn(&locator::node::host_id)));
-        return {};
-    }
-    size_t idx = base_it - base_endpoints.begin();
-    std::optional<std::reference_wrapper<const locator::node>> no_pairing_replica;
-    if (!paired_replica && idx >= view_endpoints.size()) {
-        // There are fewer view replicas than base replicas
-        // FIXME: This might still happen when reducing replication factor with tablets,
-        // see https://github.com/scylladb/scylladb/issues/21492
-        ++cf_stats.total_view_updates_failed_pairing;
-        vlogger.warn("Could not pair {}: rack_aware={} base_endpoints={} view_endpoints={}", me,
-                rack_aware_pairing ? (simple_rack_aware_pairing ? "simple" : "complex") : "none",
-                orig_base_endpoints | std::views::transform(std::mem_fn(&locator::node::host_id)),
-                orig_view_endpoints | std::views::transform(std::mem_fn(&locator::node::host_id)));
-        return {};
-    } else if (base_endpoints.size() < view_endpoints.size()) {
-        // There are fewer base replicas than view replicas.
-        // This can happen as a result of an RF change when the view replica finishes streaming
-        // before the base replica.
-        // Because of this, a view replica might not get paired with any base replica, so we need
-        // to send an additional update to it.
-        ++cf_stats.total_view_updates_due_to_replica_count_mismatch;
-        no_pairing_replica = view_endpoints.back();
-        if (base_endpoints.size() < view_endpoints.size() - 1) {
-            // We only expect one extra replica to appear due to an RF change. If there's more, that's an error,
-            // but we'll still perform updates to the paired and last replicas to minimize degradation.
-            vlogger.warn("There are too many view endpoints for base-view pairing. View updates may get lost on view_endpoints={}",
-                    std::span(view_endpoints.begin() + base_endpoints.size(), view_endpoints.end() - 1) | std::views::transform(std::mem_fn(&locator::node::host_id)));
-        }
-    }
-
     if (!paired_replica) {
-        paired_replica = view_endpoints[idx];
+        // We couldn't find any view replica in our rack
+        ++cf_stats.total_view_updates_failed_pairing;
+        vlogger.warn("Could not find a view replica in the same rack as base replica {} for base_endpoints={} view_endpoints={}", 
+                me,
+                base_nodes | std::views::transform(std::mem_fn(&locator::node::host_id)),
+                view_nodes | std::views::transform(std::mem_fn(&locator::node::host_id)));
     }
-    if (!no_pairing_replica && base_nodes.size() < view_nodes.size()) {
-        // This can happen when the view replica with no pairing is in another DC.
-        // We need to send an update to it if there are no base replicas in that DC yet,
-        // as it won't receive updates otherwise.
-        std::unordered_set<sstring> dcs_with_base_replicas;
-        for (const auto& base_node : base_nodes) {
-            dcs_with_base_replicas.insert(base_node.get().dc());
-        }
-        for (const auto& view_node : view_nodes) {
-            if (!dcs_with_base_replicas.contains(view_node.get().dc())) {
-                ++cf_stats.total_view_updates_due_to_replica_count_mismatch;
-                no_pairing_replica = view_node;
-                break;
-            }
-        }
-    }
-    // https://github.com/scylladb/scylladb/issues/19439
-    // With tablets, a node being replaced might transition to "left" state
-    // but still be kept as a replica.
-    // As of writing this hints are not prepared to handle nodes that are left
-    // but are still replicas. Therefore, there is no other sensible option
-    // right now but to give up attempt to send the update or write a hint
-    // to the paired, permanently down replica.
-    // We use the same workaround for the extra replica.
-    auto return_host_id_if_not_left = [] (const auto& replica) -> std::optional<locator::host_id> {
-        if (!replica) {
-            return std::nullopt;
-        }
-        const auto& node = replica->get();
-        if (!node.left()) {
-            return node.host_id();
-        } else {
-            return std::nullopt;
-        }
-    };
-    return {.natural_endpoint = return_host_id_if_not_left(paired_replica),
-            .endpoint_with_no_pairing = return_host_id_if_not_left(no_pairing_replica)};
+    std::optional<locator::host_id> no_pairing_replica = get_unpaired_view_endpoint(base_nodes, view_nodes, cf_stats);
+    return {.natural_endpoint = paired_replica,
+            .endpoint_with_no_pairing = no_pairing_replica};
 }
 
 static future<> apply_to_remote_endpoints(service::storage_proxy& proxy, locator::effective_replication_map_ptr ermp,
@@ -2136,12 +2041,6 @@ future<> view_update_generator::mutate_MV(
 {
     auto& ks = _db.find_keyspace(base->ks_name());
     auto& replication = ks.get_replication_strategy();
-    // We set legacy self-pairing for old vnode-based tables (for backward
-    // compatibility), and unset it for tablets - where range movements
-    // are more frequent and backward compatibility is less important.
-    // TODO: Maybe allow users to set use_legacy_self_pairing explicitly
-    // on a view, like we have the synchronous_updates_flag.
-    bool use_legacy_self_pairing = !ks.uses_tablets();
     std::unordered_map<table_id, locator::effective_replication_map_ptr> erms;
     auto get_erm = [&] (table_id id) {
         auto it = erms.find(id);
@@ -2154,10 +2053,6 @@ future<> view_update_generator::mutate_MV(
     for (const auto& mut : view_updates) {
         (void)get_erm(mut.s->id());
     }
-    // Enable rack-aware view updates pairing for tablets
-    // when the cluster feature is enabled so that all replicas agree
-    // on the pairing algorithm.
-    bool use_tablets_rack_aware_view_pairing = _db.features().tablet_rack_aware_view_pairing && ks.uses_tablets();
     auto me = base_ermp->get_topology().my_host_id();
     static constexpr size_t max_concurrent_updates = 128;
     co_await utils::get_local_injector().inject("delay_before_get_view_natural_endpoint", 8000ms);
@@ -2165,7 +2060,7 @@ future<> view_update_generator::mutate_MV(
         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, replication, base_token, view_token,
-                use_legacy_self_pairing, use_tablets_rack_aware_view_pairing, cf_stats);
+                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) {

db/view/view.hh

@@ -305,8 +305,7 @@ endpoints_to_update get_view_natural_endpoint(
     const locator::abstract_replication_strategy& replication_strategy,
     const dht::token& base_token,
     const dht::token& view_token,
-    bool use_legacy_self_pairing,
-    bool use_tablets_basic_rack_aware_view_pairing,
+    bool use_tablets,
     replica::cf_stats& cf_stats);
 
 /// Verify that the provided keyspace is eligible for storing materialized views.

db/view/view_building_worker.cc

@@ -198,6 +198,7 @@ future<> view_building_worker::register_staging_sstable_tasks(std::vector<sstabl
 
 future<> view_building_worker::run_staging_sstables_registrator() {
     while (!_as.abort_requested()) {
+        bool sleep = false;
         try {
             auto lock = co_await get_units(_staging_sstables_mutex, 1, _as);
             co_await create_staging_sstable_tasks();
@@ -214,6 +215,14 @@ future<> view_building_worker::run_staging_sstables_registrator() {
             vbw_logger.warn("Got group0_concurrent_modification while creating staging sstable tasks");
         } catch (raft::request_aborted&) {
             vbw_logger.warn("Got raft::request_aborted while creating staging sstable tasks");
+        } catch (...) {
+            vbw_logger.error("Exception while creating staging sstable tasks: {}", std::current_exception());
+            sleep = true;
+        }
+
+        if (sleep) {
+            vbw_logger.debug("Sleeping after exception.");
+            co_await seastar::sleep_abortable(1s, _as).handle_exception([] (auto x) { return make_ready_future<>(); });
         }
     }
 }
@@ -417,9 +426,12 @@ future<> view_building_worker::check_for_aborted_tasks() {
 
         auto my_host_id = vbw._db.get_token_metadata().get_topology().my_host_id();
         auto my_replica = locator::tablet_replica{my_host_id, this_shard_id()};
-        auto tasks_map = vbw._state._batch->tasks; // Potentially, we'll remove elements from the map, so we need a copy to iterate over it
-        for (auto& [id, t]: tasks_map) {
-            auto task_opt = building_state.get_task(t.base_id, my_replica, id);
+        auto it = vbw._state._batch->tasks.begin();
+        while (it != vbw._state._batch->tasks.end()) {
+            auto id = it->first;
+            auto task_opt = building_state.get_task(it->second.base_id, my_replica, id);
+
+            ++it; // Advance the iterator before potentially removing the entry from the map.
             if (!task_opt || task_opt->get().aborted) {
                 co_await vbw._state._batch->abort_task(id);
             }
@@ -449,7 +461,7 @@ static std::unordered_set<table_id> get_ids_of_all_views(replica::database& db,
     }) | std::ranges::to<std::unordered_set>();;
 }
 
-// If `state::processing_base_table` is diffrent that the `view_building_state::currently_processed_base_table`,
+// If `state::processing_base_table` is different that the `view_building_state::currently_processed_base_table`,
 // clear the state, save and flush new base table
 future<> view_building_worker::state::update_processing_base_table(replica::database& db, const view_building_state& building_state, abort_source& as) {
     if (processing_base_table != building_state.currently_processed_base_table) {
@@ -571,8 +583,6 @@ future<> view_building_worker::batch::do_work() {
             break;
         }
     }
-
-    _vbw.local()._vb_state_machine.event.broadcast();
 }
 
 future<> view_building_worker::do_build_range(table_id base_id, std::vector<table_id> views_ids, dht::token last_token, abort_source& as) {
@@ -774,13 +784,15 @@ future<std::vector<utils::UUID>> view_building_worker::work_on_tasks(raft::term_
             tasks.insert({id, *task_opt});
         }
 #ifdef SEASTAR_DEBUG
-        auto& some_task = tasks.begin()->second;
-        for (auto& [_, t]: tasks) {
-            SCYLLA_ASSERT(t.base_id == some_task.base_id);
-            SCYLLA_ASSERT(t.last_token == some_task.last_token);
-            SCYLLA_ASSERT(t.replica == some_task.replica);
-            SCYLLA_ASSERT(t.type == some_task.type);
-            SCYLLA_ASSERT(t.replica.shard == this_shard_id());
+        {
+            auto& some_task = tasks.begin()->second;
+            for (auto& [_, t]: tasks) {
+                SCYLLA_ASSERT(t.base_id == some_task.base_id);
+                SCYLLA_ASSERT(t.last_token == some_task.last_token);
+                SCYLLA_ASSERT(t.replica == some_task.replica);
+                SCYLLA_ASSERT(t.type == some_task.type);
+                SCYLLA_ASSERT(t.replica.shard == this_shard_id());
+            }
         }
 #endif
 
@@ -811,25 +823,6 @@ future<std::vector<utils::UUID>> view_building_worker::work_on_tasks(raft::term_
     co_return collect_completed_tasks();
 }
 
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
-
 }
 
 }

db/virtual_tables.cc

@@ -605,8 +605,8 @@ public:
     }
 
     static schema_ptr build_schema() {
-        auto id = generate_legacy_id(system_keyspace::NAME, "versions");
-        return schema_builder(system_keyspace::NAME, "versions", std::make_optional(id))
+        auto id = generate_legacy_id(system_keyspace::NAME, system_keyspace::VERSIONS);
+        return schema_builder(system_keyspace::NAME, system_keyspace::VERSIONS, std::make_optional(id))
             .with_column("key", utf8_type, column_kind::partition_key)
             .with_column("version", utf8_type)
             .with_column("build_mode", utf8_type)
@@ -749,6 +749,7 @@ class clients_table : public streaming_virtual_table {
             .with_column("ssl_protocol", utf8_type)
             .with_column("username", utf8_type)
             .with_column("scheduling_group", utf8_type)
+            .with_column("client_options", map_type_impl::get_instance(utf8_type, utf8_type, false))
             .with_hash_version()
             .build();
     }
@@ -766,7 +767,7 @@ class clients_table : public streaming_virtual_table {
 
     future<> execute(reader_permit permit, result_collector& result, const query_restrictions& qr) override {
         // Collect
-        using client_data_vec = utils::chunked_vector<client_data>;
+        using client_data_vec = utils::chunked_vector<foreign_ptr<std::unique_ptr<client_data>>>;
         using shard_client_data = std::vector<client_data_vec>;
         std::vector<foreign_ptr<std::unique_ptr<shard_client_data>>> cd_vec;
         cd_vec.resize(smp::count);
@@ -806,13 +807,13 @@ class clients_table : public streaming_virtual_table {
         for (unsigned i = 0; i < smp::count; i++) {
             for (auto&& ps_cdc : *cd_vec[i]) {
                 for (auto&& cd : ps_cdc) {
-                    if (cd_map.contains(cd.ip)) {
-                        cd_map[cd.ip].emplace_back(std::move(cd));
+                    if (cd_map.contains(cd->ip)) {
+                        cd_map[cd->ip].emplace_back(std::move(cd));
                     } else {
-                        dht::decorated_key key = make_partition_key(cd.ip);
+                        dht::decorated_key key = make_partition_key(cd->ip);
                         if (this_shard_owns(key) && contains_key(qr.partition_range(), key)) {
-                            ips.insert(decorated_ip{std::move(key), cd.ip});
-                            cd_map[cd.ip].emplace_back(std::move(cd));
+                            ips.insert(decorated_ip{std::move(key), cd->ip});
+                            cd_map[cd->ip].emplace_back(std::move(cd));
                         }
                     }
                     co_await coroutine::maybe_yield();
@@ -825,39 +826,58 @@ class clients_table : public streaming_virtual_table {
             co_await result.emit_partition_start(dip.key);
             auto& clients = cd_map[dip.ip];
 
-            std::ranges::sort(clients, [] (const client_data& a, const client_data& b) {
-                return a.port < b.port || a.client_type_str() < b.client_type_str();
+            std::ranges::sort(clients, [] (const foreign_ptr<std::unique_ptr<client_data>>& a, const foreign_ptr<std::unique_ptr<client_data>>& b) {
+                return a->port < b->port || a->client_type_str() < b->client_type_str();
             });
 
             for (const auto& cd : clients) {
-                clustering_row cr(make_clustering_key(cd.port, cd.client_type_str()));
-                set_cell(cr.cells(), "shard_id", cd.shard_id);
-                set_cell(cr.cells(), "connection_stage", cd.stage_str());
-                if (cd.driver_name) {
-                    set_cell(cr.cells(), "driver_name", *cd.driver_name);
+                clustering_row cr(make_clustering_key(cd->port, cd->client_type_str()));
+                set_cell(cr.cells(), "shard_id", cd->shard_id);
+                set_cell(cr.cells(), "connection_stage", cd->stage_str());
+                if (cd->driver_name) {
+                    set_cell(cr.cells(), "driver_name", cd->driver_name->key());
                 }
-                if (cd.driver_version) {
-                    set_cell(cr.cells(), "driver_version", *cd.driver_version);
+                if (cd->driver_version) {
+                    set_cell(cr.cells(), "driver_version", cd->driver_version->key());
                 }
-                if (cd.hostname) {
-                    set_cell(cr.cells(), "hostname", *cd.hostname);
+                if (cd->hostname) {
+                    set_cell(cr.cells(), "hostname", *cd->hostname);
                 }
-                if (cd.protocol_version) {
-                    set_cell(cr.cells(), "protocol_version", *cd.protocol_version);
+                if (cd->protocol_version) {
+                    set_cell(cr.cells(), "protocol_version", *cd->protocol_version);
                 }
-                if (cd.ssl_cipher_suite) {
-                    set_cell(cr.cells(), "ssl_cipher_suite", *cd.ssl_cipher_suite);
+                if (cd->ssl_cipher_suite) {
+                    set_cell(cr.cells(), "ssl_cipher_suite", *cd->ssl_cipher_suite);
                 }
-                if (cd.ssl_enabled) {
-                    set_cell(cr.cells(), "ssl_enabled", *cd.ssl_enabled);
+                if (cd->ssl_enabled) {
+                    set_cell(cr.cells(), "ssl_enabled", *cd->ssl_enabled);
                 }
-                if (cd.ssl_protocol) {
-                    set_cell(cr.cells(), "ssl_protocol", *cd.ssl_protocol);
+                if (cd->ssl_protocol) {
+                    set_cell(cr.cells(), "ssl_protocol", *cd->ssl_protocol);
                 }
-                set_cell(cr.cells(), "username", cd.username ? *cd.username : sstring("anonymous"));
-                if (cd.scheduling_group_name) {
-                    set_cell(cr.cells(), "scheduling_group", *cd.scheduling_group_name);
+                set_cell(cr.cells(), "username", cd->username ? *cd->username : sstring("anonymous"));
+                if (cd->scheduling_group_name) {
+                    set_cell(cr.cells(), "scheduling_group", *cd->scheduling_group_name);
                 }
+
+                auto map_type = map_type_impl::get_instance(
+                    utf8_type,
+                    utf8_type,
+                    false
+                );
+
+                auto prepare_client_options = [] (const auto& client_options) {
+                    map_type_impl::native_type tmp;
+                    for (auto& co: client_options) {
+                        auto map_element = std::make_pair(data_value(co.key.key()), data_value(co.value.key()));
+                        tmp.push_back(std::move(map_element));
+                    }
+                    return tmp;
+                };
+
+                set_cell(cr.cells(), "client_options",
+                    make_map_value(map_type, prepare_client_options(cd->client_options)));
+
                 co_await result.emit_row(std::move(cr));
             }
             co_await result.emit_partition_end();

dist/common/sysconfig/scylla-node-exporter

@@ -1 +1 @@
-SCYLLA_NODE_EXPORTER_ARGS="--collector.interrupts --no-collector.hwmon --no-collector.bcache --no-collector.btrfs --no-collector.fibrechannel --no-collector.infiniband --no-collector.ipvs --no-collector.nfs --no-collector.nfsd --no-collector.powersupplyclass --no-collector.rapl --no-collector.tapestats --no-collector.thermal_zone --no-collector.udp_queues --no-collector.zfs"
+SCYLLA_NODE_EXPORTER_ARGS="--collector.interrupts --collector.ethtool.metrics-include='(bw_in_allowance_exceeded|bw_out_allowance_exceeded|conntrack_allowance_exceeded|conntrack_allowance_available|linklocal_allowance_exceeded)' --collector.ethtool --no-collector.hwmon --no-collector.bcache --no-collector.btrfs --no-collector.fibrechannel --no-collector.infiniband --no-collector.ipvs --no-collector.nfs --no-collector.nfsd --no-collector.powersupplyclass --no-collector.rapl --no-collector.tapestats --no-collector.thermal_zone --no-collector.udp_queues --no-collector.zfs"

docs/README.md

@@ -71,7 +71,7 @@ Use "Bash on Ubuntu on Windows" for the same tools and capabilities as on Linux
 
 ### Building the Docs 
 
-1. Run `make preview` to build the documentation.
+1. Run `make preview` in the `docs/` directory to build the documentation.
 1. Preview the built documentation locally at http://127.0.0.1:5500/.
 
 ### Cleanup

docs/_ext/scylladb_metrics.py

@@ -41,6 +41,8 @@ class MetricsProcessor:
                 # Get metrics from the file
                 try:
                     metrics_file = metrics.get_metrics_from_file(relative_path, "scylla_", metrics_info, strict=strict)
+                except SystemExit:
+                    pass
                 finally:
                     os.chdir(old_cwd)
                 if metrics_file:

docs/alternator/alternator.md

@@ -1,17 +1,17 @@
-# Alternator: DynamoDB API in Scylla
+# Alternator: DynamoDB API in ScyllaDB
 
 ## Introduction
-Alternator is a Scylla feature adding compatibility with Amazon DynamoDB(TM).
+Alternator is a ScyllaDB feature adding compatibility with Amazon DynamoDB(TM).
 DynamoDB's API uses JSON-encoded requests and responses which are sent over
 an HTTP or HTTPS transport. It is described in detail in Amazon's [DynamoDB
 API Reference](https://docs.aws.amazon.com/amazondynamodb/latest/APIReference/).
 
 Our goal is that any application written to use Amazon DynamoDB could
-be run, unmodified, against Scylla with Alternator enabled. Alternator's
+be run, unmodified, against ScyllaDB with Alternator enabled. Alternator's
 compatibility with DynamoDB is fairly complete, but users should be aware
 of some differences and some unimplemented features. The extent of
 Alternator's compatibility with DynamoDB is described in the
-[Scylla Alternator for DynamoDB users](compatibility.md) document,
+[ScyllaDB Alternator for DynamoDB users](compatibility.md) document,
 which is updated as the work on Alternator progresses and compatibility
 continues to improve.
 
@@ -19,8 +19,8 @@ Alternator also adds several features and APIs that are not available in
 DynamoDB. These are described in [Alternator-specific APIs](new-apis.md).
 
 ## Running Alternator
-By default, Scylla does not listen for DynamoDB API requests. To enable
-this API in Scylla you must set at least two configuration options,
+By default, ScyllaDB does not listen for DynamoDB API requests. To enable
+this API in ScyllaDB you must set at least two configuration options,
 **alternator_port** and **alternator_write_isolation**. For example in the
 YAML configuration file:
 ```yaml
@@ -30,7 +30,7 @@ alternator_write_isolation: only_rmw_uses_lwt # or always, forbid or unsafe
 or, equivalently, via command-line arguments: `--alternator-port=8000
 --alternator-write-isolation=only_rmw_uses_lwt.
 
-the **alternator_port** option determines on which port Scylla listens for
+the **alternator_port** option determines on which port ScyllaDB listens for
 DynamoDB API requests. By default, it listens on this port on all network
 interfaces. To listen only on a specific interface, configure also the
 **alternator_address** option.
@@ -41,12 +41,12 @@ Alternator has four different choices
 for the implementation of writes, each with different advantages. You should
 carefully consider which of the options makes more sense for your intended
 use case and configure alternator_write_isolation accordingly. There is
-currently no default for this option: Trying to run Scylla with an Alternator
+currently no default for this option: Trying to run ScyllaDB with an Alternator
 port selected but without configuring write isolation will result in an error message,
 asking you to set it.
 
 In addition to (or instead of) serving HTTP requests on alternator_port,
-Scylla can accept DynamoDB API requests over HTTPS (encrypted), on the port
+ScyllaDB can accept DynamoDB API requests over HTTPS (encrypted), on the port
 specified by **alternator_https_port**. As usual for HTTPS servers, the
 operator must specify certificate and key files. By default these should
 be placed in `/etc/scylla/scylla.crt` and `/etc/scylla/scylla.key`, but
@@ -54,7 +54,7 @@ these default locations can overridden by specifying
 `--alternator-encryption-options keyfile="..."` and
 `--alternator-encryption-options certificate="..."`.
 
-By default, Scylla saves a snapshot of deleted tables. But Alternator does
+By default, ScyllaDB saves a snapshot of deleted tables. But Alternator does
 not offer an API to restore these snapshots, so these snapshots are not useful
 and waste disk space - deleting a table does not recover any disk space.
 It is therefore recommended to disable this automatic-snapshotting feature
@@ -73,11 +73,11 @@ itself. Instructions, code and examples for doing this can be found in the
 
 This section provides only a very brief introduction to Alternator's
 design. A much more detailed document about the features of the DynamoDB
-API and how they are, or could be, implemented in Scylla can be found in:
+API and how they are, or could be, implemented in ScyllaDB can be found in:
 <https://docs.google.com/document/d/1i4yjF5OSAazAY_-T8CBce9-2ykW4twx_E_Nt2zDoOVs>
 
 Almost all of Alternator's source code (except some initialization code)
-can be found in the alternator/ subdirectory of Scylla's source code.
+can be found in the alternator/ subdirectory of ScyllaDB's source code.
 Extensive functional tests can be found in the test/alternator
 subdirectory. These tests are written in Python, and can be run against
 both Alternator and Amazon's DynamoDB; This allows verifying that
@@ -85,15 +85,15 @@ Alternator's behavior matches the one observed on DynamoDB.
 See test/alternator/README.md for more information about the tests and
 how to run them.
 
-With Alternator enabled on port 8000 (for example), every Scylla node
+With Alternator enabled on port 8000 (for example), every ScyllaDB node
 listens for DynamoDB API requests on this port. These requests, in
 JSON format over HTTP, are parsed and result in calls to internal Scylla
 C++ functions - there is no CQL generation or parsing involved.
-In Scylla terminology, the node receiving the request acts as the
+In ScyllaDB terminology, the node receiving the request acts as the
 *coordinator*, and often passes the request on to one or more other nodes -
 *replicas* which hold copies of the requested data.
 
-Alternator tables are stored as Scylla tables, each in a separate keyspace.
+Alternator tables are stored as ScyllaDB tables, each in a separate keyspace.
 Each keyspace is initialized when the corresponding Alternator table is
 created (with a CreateTable request). The replication factor (RF) for this
 keyspace is chosen at that point, depending on the size of the cluster:
@@ -101,19 +101,19 @@ RF=3 is used on clusters with three or more nodes, and RF=1 is used for
 smaller clusters. Such smaller clusters are, of course, only recommended
 for tests because of the risk of data loss.
 
-Each table in Alternator is stored as a Scylla table in a separate
+Each table in Alternator is stored as a ScyllaDB table in a separate
 keyspace. The DynamoDB key columns (hash and sort key) have known types,
-and become partition and clustering key columns of the Scylla table.
+and become partition and clustering key columns of the ScyllaDB table.
 All other attributes may be different for each row, so are stored in one
-map column in Scylla, and not as separate columns.
+map column in ScyllaDB, and not as separate columns.
 
 DynamoDB supports two consistency levels for reads, "eventual consistency"
-and "strong consistency". These two modes are implemented using Scylla's CL
+and "strong consistency". These two modes are implemented using ScyllaDB's CL
 (consistency level) feature: All writes are done using the `LOCAL_QUORUM`
 consistency level, then strongly-consistent reads are done with
 `LOCAL_QUORUM`, while eventually-consistent reads are with just `LOCAL_ONE`.
 
-In Scylla (and its inspiration, Cassandra), high write performance is
+In ScyllaDB (and its inspiration, Cassandra), high write performance is
 achieved by ensuring that writes do not require reads from disk.
 The DynamoDB API, however, provides many types of requests that need a read
 before the write (a.k.a. RMW requests - read-modify-write). For example,
@@ -121,7 +121,7 @@ a request may copy an existing attribute, increment an attribute,
 be conditional on some expression involving existing values of attribute,
 or request that the previous values of attributes be returned. These
 read-modify-write transactions should be _isolated_ from each other, so
-by default Alternator implements every write operation using Scylla's
+by default Alternator implements every write operation using ScyllaDB's
 LWT (lightweight transactions). This default can be overridden on a per-table
 basis, by tagging the table as explained above in the "write isolation
 policies" section.

docs/alternator/compatibility.md

@@ -1,6 +1,6 @@
 # ScyllaDB Alternator for DynamoDB users
 
-Scylla supports the DynamoDB API (this feature is codenamed "Alternator").
+ScyllaDB supports the DynamoDB API (this feature is codenamed "Alternator").
 Our goal is to support any application written for Amazon DynamoDB.
 Nevertheless, there are a few differences between DynamoDB and Scylla, and
 and a few DynamoDB features that have not yet been implemented in Scylla.
@@ -8,16 +8,16 @@ The purpose of this document is to inform users of these differences.
 
 ## Provisioning
 
-The most obvious difference between DynamoDB and Scylla is that while
-DynamoDB is a shared cloud service, Scylla is a dedicated service running
+The most obvious difference between DynamoDB and ScyllaDB is that while
+DynamoDB is a shared cloud service, ScyllaDB is a dedicated service running
 on your private cluster. Whereas DynamoDB allows you to "provision" the
 number of requests per second you'll need - or at an extra cost not even
-provision that - Scylla requires you to provision your cluster. You need
+provision that - ScyllaDB requires you to provision your cluster. You need
 to reason about the number and size of your nodes - not the throughput.
 
 Moreover, DynamoDB's per-table provisioning (`BillingMode=PROVISIONED`) is
 not yet supported by Scylla. The BillingMode and ProvisionedThroughput options
-on a table need to be valid but are ignored, and Scylla behaves like DynamoDB's
+on a table need to be valid but are ignored, and ScyllaDB behaves like DynamoDB's
 `BillingMode=PAY_PER_REQUEST`: All requests are accepted without a per-table
 throughput cap.
 
@@ -33,7 +33,7 @@ Instructions for doing this can be found in:
 
 ## Write isolation policies
 
-Scylla was designed to optimize the performance of pure write operations -
+ScyllaDB was designed to optimize the performance of pure write operations -
 writes which do not need to read the previous value of the item.
 In CQL, writes which do need the previous value of the item must explicitly
 use the slower LWT ("LightWeight Transaction") feature to be correctly
@@ -79,11 +79,11 @@ a _higher_ timestamp - and this will be the "last write" that wins.
 To avoid or mitigate this write reordering issue, users may consider
 one or more of the following:
 
-1. Use NTP to keep the clocks on the different Scylla nodes synchronized.
+1. Use NTP to keep the clocks on the different ScyllaDB nodes synchronized.
    If the delay between the two writes is longer than NTP's accuracy,
    they will not be reordered.
 2. If an application wants to ensure that two specific writes are not
-   reordered, it should send both requests to the same Scylla node.
+   reordered, it should send both requests to the same ScyllaDB node.
    Care should be taken when using a load balancer - which might redirect
    two requests to two different nodes.
 3. Consider using the `always_use_lwt` write isolation policy.
@@ -210,7 +210,7 @@ CREATE SERVICE_LEVEL IF NOT EXISTS oltp WITH SHARES = 1000;
 ATTACH SERVICE_LEVEL olap TO alice;
 ATTACH SERVICE_LEVEL oltp TO bob;
 ```
-Note that `alternator_enforce_authorization` has to be enabled in Scylla configuration.
+Note that `alternator_enforce_authorization` has to be enabled in ScyllaDB configuration.
 
 See [Authorization](##Authorization) section to learn more about roles and authorization.
 See [Workload Prioritization](../features/workload-prioritization)
@@ -218,11 +218,11 @@ to read about Workload Prioritization in detail.
 
 ## Metrics
 
-Scylla has an advanced and extensive monitoring framework for inspecting
-and graphing hundreds of different metrics of Scylla's usage and performance.
-Scylla's monitoring stack, based on Grafana and Prometheus, is described in
+ScyllaDB has an advanced and extensive monitoring framework for inspecting
+and graphing hundreds of different metrics of ScyllaDB's usage and performance.
+ScyllaDB's monitoring stack, based on Grafana and Prometheus, is described in
 <https://docs.scylladb.com/operating-scylla/monitoring/>.
-This monitoring stack is different from DynamoDB's offering - but Scylla's
+This monitoring stack is different from DynamoDB's offering - but ScyllaDB's
 is significantly more powerful and gives the user better insights on
 the internals of the database and its performance.
 
@@ -248,7 +248,7 @@ data in different partition order. Applications mustn't rely on that
 undocumented order.
 
 Note that inside each partition, the individual items will be sorted the same
-in DynamoDB and Scylla - determined by the _sort key_ defined for that table.
+in DynamoDB and ScyllaDB - determined by the _sort key_ defined for that table.
 
 ---
 
@@ -274,7 +274,7 @@ is different, or can be configured in Alternator:
 ## Experimental API features
 
 Some DynamoDB API features are supported by Alternator, but considered
-**experimental** in this release. An experimental feature in Scylla is a
+**experimental** in this release. An experimental feature in ScyllaDB is a
 feature whose functionality is complete, or mostly complete, but it is not
 as thoroughly tested or optimized as regular features. Also, an experimental
 feature's implementation is still subject to change and upgrades may not be
@@ -351,8 +351,8 @@ they should be easy to detect. Here is a list of these unimplemented features:
 
 * The on-demand backup APIs are not supported: CreateBackup, DescribeBackup,
   DeleteBackup, ListBackups, RestoreTableFromBackup.
-  For now, users can use Scylla's existing backup solutions such as snapshots
-  or Scylla Manager.
+  For now, users can use ScyllaDB's existing backup solutions such as snapshots
+  or ScyllaDB Manager.
   <https://github.com/scylladb/scylla/issues/5063>
 
 * Continuous backup (the ability to restore any point in time) is also not
@@ -370,7 +370,7 @@ they should be easy to detect. Here is a list of these unimplemented features:
   <https://github.com/scylladb/scylla/issues/5068>
 
 * DAX (DynamoDB Accelerator), an in-memory cache for DynamoDB, is not
-  available in for Alternator. Anyway, it should not be necessary - Scylla's
+  available in for Alternator. Anyway, it should not be necessary - ScyllaDB's
   internal cache is already rather advanced and there is no need to place
   another cache in front of the it. We wrote more about this here:
   <https://www.scylladb.com/2017/07/31/database-caches-not-good/>
@@ -384,7 +384,7 @@ they should be easy to detect. Here is a list of these unimplemented features:
 * The PartiQL syntax (SQL-like SELECT/UPDATE/INSERT/DELETE expressions)
   and the operations ExecuteStatement, BatchExecuteStatement and
   ExecuteTransaction are not yet supported.
-  A user that is interested in an SQL-like syntax can consider using Scylla's
+  A user that is interested in an SQL-like syntax can consider using ScyllaDB's
   CQL protocol instead.
   This feature was added to DynamoDB in November 2020.
   <https://github.com/scylladb/scylla/issues/8787>
@@ -393,7 +393,7 @@ they should be easy to detect. Here is a list of these unimplemented features:
   which is different from AWS's. In particular, the operations
   DescribeContributorInsights, ListContributorInsights and
   UpdateContributorInsights that configure Amazon's "CloudWatch Contributor
-  Insights" are not yet supported. Scylla has different ways to retrieve the
+  Insights" are not yet supported. ScyllaDB has different ways to retrieve the
   same information, such as which items were accessed most often.
   <https://github.com/scylladb/scylla/issues/8788>
 

docs/alternator/getting-started.md

@@ -11,7 +11,7 @@ This section will guide you through the steps for setting up the cluster:
    <https://hub.docker.com/r/scylladb/scylla/>, but add to every `docker run`
    command a `-p 8000:8000` before the image name and
    `--alternator-port=8000 --alternator-write-isolation=always` at the end.
-   The "alternator-port" option specifies on which port Scylla will listen for
+   The "alternator-port" option specifies on which port ScyllaDB will listen for
    the (unencrypted) DynamoDB API, and the "alternator-write-isolation" chooses
    whether or not Alternator will use LWT for every write.
    For example,
@@ -24,10 +24,10 @@ This section will guide you through the steps for setting up the cluster:
 By default, ScyllaDB run in this way will not have authentication or
 authorization enabled, and any DynamoDB API request will be honored without
 requiring them to be signed appropriately. See the
-[Scylla Alternator for DynamoDB users](compatibility.md#authentication-and-authorization)
+[ScyllaDB Alternator for DynamoDB users](compatibility.md#authentication-and-authorization)
 document on how to configure authentication and authorization.
 
-## Testing Scylla's DynamoDB API support:
+## Testing ScyllaDB's DynamoDB API support:
 ### Running AWS Tic Tac Toe demo app to test the cluster:
 1. Follow the instructions on the [AWS github page](https://github.com/awsdocs/amazon-dynamodb-developer-guide/blob/master/doc_source/TicTacToe.Phase1.md)
 2. Enjoy your tic-tac-toe game :-)

docs/alternator/new-apis.md

@@ -2,9 +2,9 @@
 
 Alternator's primary goal is to be compatible with Amazon DynamoDB(TM)
 and its APIs, so that any application written to use Amazon DynamoDB could
-be run, unmodified, against Scylla with Alternator enabled. The extent of
+be run, unmodified, against ScyllaDB with Alternator enabled. The extent of
 Alternator's compatibility with DynamoDB is described in the
-[Scylla Alternator for DynamoDB users](compatibility.md) document.
+[ScyllaDB Alternator for DynamoDB users](compatibility.md) document.
 
 But Alternator also adds several features and APIs that are not available in
 DynamoDB. These Alternator-specific APIs are documented here.
@@ -15,7 +15,7 @@ _conditional_ update or an update based on the old value of an attribute.
 The read and the write should be treated as a single transaction - protected
 (_isolated_) from other parallel writes to the same item.
 
-Alternator could do this isolation by using Scylla's LWT (lightweight
+Alternator could do this isolation by using ScyllaDB's LWT (lightweight
 transactions) for every write operation, but this significantly slows
 down writes, and not necessary for workloads which don't use read-modify-write
 (RMW) updates.
@@ -41,7 +41,7 @@ isolation policy for a specific table can be overridden by tagging the table
     which need a read before the write. An attempt to use such statements
     (e.g.,  UpdateItem with a ConditionExpression) will result in an error.
     In this mode, the remaining write requests which are allowed - pure writes
-    without a read - are performed using standard Scylla writes, not LWT,
+    without a read - are performed using standard ScyllaDB writes, not LWT,
     so they are significantly faster than they would have been in the
     `always_use_lwt`, but their isolation is still correct.
 
@@ -65,19 +65,19 @@ isolation policy for a specific table can be overridden by tagging the table
     read-modify-write updates. This mode is not recommended for any use case,
     and will likely be removed in the future.
 
-## Accessing system tables from Scylla
-Scylla exposes lots of useful information via its internal system tables,
+## Accessing system tables from ScyllaDB
+ScyllaDB exposes lots of useful information via its internal system tables,
 which can be found in system keyspaces: 'system', 'system\_auth', etc.
 In order to access to these tables via alternator interface,
 Scan and Query requests can use a special table name:
 `.scylla.alternator.KEYSPACE_NAME.TABLE_NAME`
-which will return results fetched from corresponding Scylla table.
+which will return results fetched from corresponding ScyllaDB table.
 
 This interface can be used only to fetch data from system tables.
 Attempts to read regular tables via the virtual interface will result
 in an error.
 
-Example: in order to query the contents of Scylla's `system.large_rows`,
+Example: in order to query the contents of ScyllaDB's `system.large_rows`,
 pass `TableName='.scylla.alternator.system.large_rows'` to a Query/Scan
 request.
 
@@ -113,14 +113,14 @@ connection (either active or idle), not necessarily an active request as
 in Alternator.
 
 ## Service discovery
-As explained in [Scylla Alternator for DynamoDB users](compatibility.md),
+As explained in [ScyllaDB Alternator for DynamoDB users](compatibility.md),
 Alternator requires a load-balancer or a client-side load-balancing library
-to distribute requests between all Scylla nodes. This load-balancer needs
-to be able to _discover_ the Scylla nodes. Alternator provides two special
+to distribute requests between all ScyllaDB nodes. This load-balancer needs
+to be able to _discover_ the ScyllaDB nodes. Alternator provides two special
 requests, `/` and `/localnodes`, to help with this service discovery, which
 we will now explain.
 
-Some setups know exactly which Scylla nodes were brought up, so all that
+Some setups know exactly which ScyllaDB nodes were brought up, so all that
 remains is to periodically verify that each node is still functional. The
 easiest way to do this is to make an HTTP (or HTTPS) GET request to the node,
 with URL `/`. This is a trivial GET request and does **not** need to be
@@ -133,10 +133,10 @@ $ curl http://localhost:8000/
 healthy: localhost:8000
 ```
 
-In other setups, the load balancer might not know which Scylla nodes exist.
-For example, it may be possible to add or remove Scylla nodes without a
+In other setups, the load balancer might not know which ScyllaDB nodes exist.
+For example, it may be possible to add or remove ScyllaDB nodes without a
 client-side load balancer knowing. For these setups we have the `/localnodes`
-request that can be used to discover which Scylla nodes exist: A load balancer
+request that can be used to discover which ScyllaDB nodes exist: A load balancer
 that already knows at least one live node can discover the rest by sending
 a `/localnodes` request to the known node. It's again an unauthenticated
 HTTP (or HTTPS) GET request:
@@ -160,7 +160,7 @@ list the nodes in a specific _data center_ or _rack_. These options are
 useful for certain use cases:
 
 * A `dc` option (e.g., `/localnodes?dc=dc1`) can be passed to list the
-  nodes in a specific Scylla data center, not the data center of the node
+  nodes in a specific ScyllaDB data center, not the data center of the node
   being contacted. This is useful when a client knowns of _some_ Scylla
   node belonging to an unknown DC, but wants to list the nodes in _its_
   DC, which it knows by name.
@@ -191,7 +191,7 @@ tells them to.
 
 If you want to influence whether a specific Alternator table is created with tablets or vnodes,
 you can do this by specifying the `system:initial_tablets` tag
-(in earlier versions of Scylla the tag was `experimental:initial_tablets`)
+(in earlier versions of ScyllaDB the tag was `experimental:initial_tablets`)
 in the CreateTable operation. The value of this tag can be:
 
 * Any valid integer as the value of this tag enables tablets.

docs/cql/ddl.rst

@@ -365,7 +365,7 @@ Modifying a keyspace with tablets enabled is possible and doesn't require any sp
 
 - The replication factor (RF) can be increased or decreased by at most 1 at a time. To reach the desired RF value, modify the RF repeatedly.
 - The ``ALTER`` statement rejects the ``replication_factor`` tag. List the DCs explicitly when altering a keyspace. See :ref:`NetworkTopologyStrategy <replication-strategy>`.
-- If there's any other ongoing global topology operation, executing the ``ALTER`` statement will fail (with an explicit and specific error) and needs to be repeated.
+- An RF change cannot be requested while another RF change is pending for the same keyspace. Attempting to execute an ``ALTER`` statement in this scenario will fail with an explicit error. Wait for the ongoing RF change to complete before issuing another ``ALTER`` statement.
 - The ``ALTER`` statement may take longer than the regular query timeout, and even if it times out, it will continue to execute in the background.
 - The replication strategy cannot be modified, as keyspaces with tablets only support ``NetworkTopologyStrategy``.
 - The ``ALTER`` statement will fail if it would make the keyspace :term:`RF-rack-invalid <RF-rack-valid keyspace>`.
@@ -1043,6 +1043,8 @@ The following modes are available:
    * - ``immediate``
      - Tombstone GC is immediately performed. There is no wait time or repair requirement. This mode is useful for a table that uses the TWCS compaction strategy with no user deletes. After data is expired after TTL, ScyllaDB can perform compaction to drop the expired data immediately.
 
+.. warning:: The ``repair`` mode is not supported for :term:`Colocated Tables <Colocated Table>` in this version.
+
 .. _cql-per-table-tablet-options:
 
 Per-table tablet options

docs/cql/time-to-live.rst

@@ -102,6 +102,7 @@ Additional Information
 
 To learn more about TTL, and see a hands-on example, check out `this lesson <https://university.scylladb.com/courses/data-modeling/lessons/advanced-data-modeling/topic/expiring-data-with-ttl-time-to-live/>`_ on ScyllaDB University.
 
+* `Video: Managing data expiration with Time-To-Live <https://www.youtube.com/watch?v=SXkbu7mFHeA>`_
 * :doc:`Apache Cassandra Query Language (CQL) Reference </cql/index>`
 * :doc:`KB Article:How to Change gc_grace_seconds for a Table </kb/gc-grace-seconds/>`
 * :doc:`KB Article:Time to Live (TTL) and Compaction </kb/ttl-facts/>`

docs/dev/audit.md

@@ -1,6 +1,6 @@
 # Introduction
 
-Similar to the approach described in CASSANDRA-14471, we add the
+Similar to the approach described in CASSANDRA-12151, we add the
 concept of an audit specification.  An audit has a target (syslog or a
 table) and a set of events/actions that it wants recorded.  We
 introduce new CQL syntax for Scylla users to describe and manipulate

docs/dev/docker-hub.md

@@ -2,8 +2,11 @@
 
 ## What is ScyllaDB?
 
-ScyllaDB is a high-performance NoSQL database system, fully compatible with Apache Cassandra.
-ScyllaDB is released under the GNU Affero General Public License version 3 and the Apache License, ScyllaDB is free and open-source software.
+ScyllaDB is a high-performance NoSQL database optimized for speed and scalability.
+It is designed to efficiently handle large volumes of data with minimal latency,
+making it ideal for data-intensive applications.
+
+ScyllaDB is distributed under the [ScyllaDB Source Available License](https://github.com/scylladb/scylladb/blob/master/LICENSE-ScyllaDB-Source-Available.md).
 
 > [ScyllaDB](http://www.scylladb.com/)
 

docs/dev/object_storage.md

@@ -20,9 +20,7 @@ command line option when launchgin scylla.
 You can define endpoint details in the `scylla.yaml` file. For example:
 ```yaml
 object_storage_endpoints:
-  - name: s3.us-east-1.amazonaws.com
-    port: 443
-    https: true
+  - name: https://s3.us-east-1.amazonaws.com:443
     aws_region: us-east-1
 ```
 
@@ -78,9 +76,7 @@ The examples above are intended for development or local environments. You shoul
 For the EC2 Instance Metadata Service to function correctly, no additional configuration is required. However, STS requires the IAM Role ARN to be defined in the `scylla.yaml` file, as shown below:
 ```yaml
 object_storage_endpoints:
-  - name: s3.us-east-1.amazonaws.com
-    port: 443
-    https: true
+  - name: https://s3.us-east-1.amazonaws.com:443
     aws_region: us-east-1
     iam_role_arn: arn:aws:iam::123456789012:instance-profile/my-instance-instance-profile
 ```
@@ -100,9 +96,7 @@ in `scylla.yaml`:
 
 ```yaml
 object_storage_endpoints:
-  - name: s3.us-east-2.amazonaws.com
-    port: 443
-    https: true
+  - name: https://s3.us-east-2.amazonaws.com:443
     aws_region: us-east-2
 ```
 

docs/dev/protocol-extensions.md

@@ -74,6 +74,8 @@ The keys and values are:
     as an indicator to which shard client wants to connect. The desired shard number
     is calculated as: `desired_shard_no = client_port % SCYLLA_NR_SHARDS`.
     Its value is a decimal representation of type `uint16_t`, by default `19142`.
+  - `CLIENT_OPTIONS` is a string containing a JSON object representation that
+    contains CQL Driver configuration, e.g. load balancing policy, retry policy, timeouts, etc.
 
 Currently, one `SCYLLA_SHARDING_ALGORITHM` is defined,
 `biased-token-round-robin`. To apply the algorithm,
@@ -236,3 +238,26 @@ the same mechanism for other protocol versions, such as CQLv4.
 
 The feature is identified by the `SCYLLA_USE_METADATA_ID` key, which is meant to be sent
 in the SUPPORTED message.
+
+## Sending the CLIENT_ROUTES_CHANGE event
+
+This extension allows a driver to update its connections when the
+`system.client_routes` table is modified.
+
+In some network topologies a specific mapping of addresses and ports is required (e.g.
+to support Private Link). This mapping can change dynamically even when no nodes are
+added or removed. The driver must adapt to those changes; otherwise connectivity can be
+lost.
+
+The extension is implemented as a new `EVENT` type: `CLIENT_ROUTES_CHANGE`. The event
+body consists of:
+- [string] change
+- [string list] connection_ids
+- [string list] host_ids
+
+There is only one change value: `UPDATE_NODES`, which means at least one client route
+was inserted, updated, or deleted.
+
+Events already have a subscription mechanism similar to protocol extensions (that is,
+the driver only receives the events it explicitly subscribed to), so no additional
+`cql_protocol_extension` key is introduced for this feature.

docs/dev/topology-over-raft.md

@@ -86,6 +86,7 @@ stateDiagram-v2
         de_left_token_ring --> [*]
     }
     state removing {
+        re_left_token_ring : left_token_ring
         re_tablet_draining : tablet_draining
         re_tablet_migration : tablet_migration
         re_write_both_read_old : write_both_read_old
@@ -98,7 +99,8 @@ stateDiagram-v2
         re_tablet_draining --> re_write_both_read_old
         re_write_both_read_old --> re_write_both_read_new: streaming completed
         re_write_both_read_old --> re_rollback_to_normal: rollback
-        re_write_both_read_new --> [*]
+        re_write_both_read_new --> re_left_token_ring
+        re_left_token_ring --> [*]
     }
     rebuilding --> normal: streaming completed
     decommissioning --> left: operation succeeded
@@ -122,9 +124,10 @@ Note that these are not all states, as there are other states specific to tablet
     Writes to vnodes-based tables are going to both new and old replicas (new replicas means calculated according
     to modified token ring), reads are using old replicas.
 - `write_both_read_new` - as above, but reads are using new replicas.
-- `left_token_ring` - the decommissioning node left the token ring, but we still need to wait until other
-    nodes observe it and stop sending writes to this node. Then, we tell the node to shut down and remove
-    it from group 0. We also use this state to rollback a failed bootstrap or decommission.
+- `left_token_ring` - the decommissioning or removing node left the token ring, but we still need to wait until other
+    nodes observe it and stop sending writes to this node. For decommission, we tell the node to shut down,
+    then remove it from group 0. For removenode, the node is already down, so we skip the shutdown step.
+    We also use this state to rollback a failed bootstrap or decommission.
 - `rollback_to_normal` - the decommission or removenode operation failed. Rollback the operation by
     moving the node we tried to decommission/remove back to the normal state.
 - `lock` - the topology stays in this state until externally changed (to null state), preventing topology
@@ -141,7 +144,9 @@ reads that started before this point exist in the system. Finally we remove the
 transitioning state.
 
 Decommission, removenode and replace work similarly, except they don't go through
-`commit_cdc_generation`.
+`commit_cdc_generation`. Both decommission and removenode go through the
+`left_token_ring` state to run a global barrier ensuring all nodes are aware
+of the topology change before the operation completes.
 
 The state machine may also go only through the `commit_cdc_generation` state
 after getting a request from the user to create a new CDC generation if the

docs/dev/view-building-coordinator.md

@@ -41,12 +41,12 @@ Unless the task was aborted, the worker will eventually reply that the task was
 it temporarily saves list of ids of finished tasks and removes those tasks from group0 state (pernamently marking them as finished) in 200ms intervals. (*)
 This batching of removing finished tasks is done in order to reduce number of generated group0 operations.
 
-On the other hand, view buildind tasks can can also be aborted due to 2 main reasons:
+On the other hand, view building tasks can can also be aborted due to 2 main reasons:
 - a keyspace/view was dropped
 - tablet operations (see [tablet operations section](#tablet-operations))
 In the first case we simply delete relevant view building tasks as they are no longer needed.
-But if a task needs to be aborted due to tablet operation, we're firstly setting the `aborted` flag to true. We need to do this because we need the task informations
-to created a new adjusted tasks (if the operation succeeded) or rollback them (if the operation failed).
+But if a task needs to be aborted due to tablet operation, we're firstly setting the `aborted` flag to true. We need to do this because we need the task information
+to create new adjusted tasks (if the operation succeeded) or rollback them (if the operation failed).
 Once a task is aborted by setting the flag, this cannot be revoked, so rolling back a task means creating its duplicate and removing the original task.
 
 (*) - Because there is a time gap between when the coordinator learns that a task is finished (from the RPC response) and when the task is marked as completed,

docs/features/cdc/cdc-stream-changes.rst

@@ -29,9 +29,6 @@ A CDC generation consists of:
 
 This is the mapping used to decide on which stream IDs to use when making writes, as explained in the :doc:`./cdc-streams` document. It is a global property of the cluster: it doesn't depend on the table you're making writes to.
 
-.. caution::
-   The tables mentioned in the following sections: ``system_distributed.cdc_generation_timestamps`` and ``system_distributed.cdc_streams_descriptions_v2`` have been introduced in ScyllaDB 4.4. It is highly recommended to upgrade to 4.4 for efficient CDC usage. The last section explains how to run the below examples in ScyllaDB 4.3.
-
 When CDC generations change
 ---------------------------
 

docs/features/incremental-repair.rst

@@ -28,7 +28,8 @@ Incremental Repair is only supported for tables that use the tablets architectur
 Incremental Repair Modes
 ------------------------
 
-While incremental repair is the default and recommended mode, you can control its behavior for a given repair operation using the ``incremental_mode`` parameter. This is useful for situations where you might need to force a full data validation.
+Incremental is currently disabled by default. You can control its behavior for a given repair operation using the ``incremental_mode`` parameter.
+This is useful for enabling incremental repair, or in situations where you might need to force a full data validation.
 
 The available modes are:
 

docs/features/index.rst

@@ -17,6 +17,7 @@ This document highlights ScyllaDB's key data modeling features.
    Workload Prioritization </features/workload-prioritization>
    Backup and Restore </features/backup-and-restore>
    Incremental Repair </features/incremental-repair/>
+   Vector Search </features/vector-search/>
 
 .. panel-box::
   :title: ScyllaDB Features
@@ -43,3 +44,5 @@ This document highlights ScyllaDB's key data modeling features.
   * :doc:`Incremental Repair </features/incremental-repair/>` provides a much more
     efficient and lightweight approach to maintaining data consistency by
     repairing only the data that has changed since the last repair.
+  * :doc:`Vector Search in ScyllaDB </features/vector-search/>` enables
+    similarity-based queries on vector embeddings.

docs/features/vector-search.rst

@@ -0,0 +1,55 @@
+=================================
+Vector Search in ScyllaDB
+=================================
+
+.. note::
+
+    This feature is currently available only in `ScyllaDB Cloud <https://cloud.docs.scylladb.com/>`_.
+
+What Is Vector Search
+-------------------------
+
+Vector Search enables similarity-based queries over high-dimensional data,
+such as text, images, audio, or user behavior. Instead of searching for exact
+matches, it allows applications to find items that are semantically similar to
+a given input.
+
+To do this, Vector Search works on vector embeddings, which are numerical
+representations of data that capture semantic meaning. This enables queries
+such as:
+
+* “Find documents similar to this paragraph”
+* “Find products similar to what the user just viewed”
+* “Find previous tickets related to this support request”
+
+Rather than relying on exact values or keywords, Vector Search returns results
+based on distance or similarity between vectors. This capability is
+increasingly used in modern workloads such as AI-powered search, recommendation
+systems, and retrieval-augmented generation (RAG).
+
+Why Vector Search Matters
+------------------------------------
+
+Many applications already rely on ScyllaDB for high throughput, low and
+predictable latency, and large-scale data storage.
+
+Vector Search complements these strengths by enabling new classes of workloads,
+including:
+
+* Semantic search over text or documents
+* Recommendations based on user or item similarity
+* AI and ML applications, including RAG pipelines
+* Anomaly and pattern detection
+
+With Vector Search, ScyllaDB can serve as the similarity search backend for
+AI-driven applications.
+
+Availability
+--------------
+
+Vector Search is currently available only in ScyllaDB Cloud, the fully managed
+ScyllaDB service.
+
+
+👉 For details on using Vector Search, refer to the
+`ScyllaDB Cloud documentation <https://cloud.docs.scylladb.com/stable/vector-search/index.html>`_.

docs/getting-started/cloud-instance-recommendations.rst

@@ -20,7 +20,10 @@ You can run your ScyllaDB workloads on AWS, GCE, and Azure using a ScyllaDB imag
 Amazon Web Services (AWS)
 -----------------------------
 
-The recommended instance types are :ref:`i3en <system-requirements-i3en-instances>`, :ref:`i4i <system-requirements-i4i-instances>`, :ref:`i7i <system-requirements-i7i-instances>`, and :ref:`i7ie <system-requirements-i7ie-instances>`.
+The recommended instance types are :ref:`i3en <system-requirements-i3en-instances>`,
+:ref:`i4i <system-requirements-i4i-instances>`, :ref:`i7i <system-requirements-i7i-instances>`,
+:ref:`i7ie <system-requirements-i7ie-instances>`, :ref:`i8g<system-requirements-i8g-instances>`,
+and :ref:`i8ge <system-requirements-i8ge-instances>`.
 
 .. note::
 
@@ -195,6 +198,118 @@ All i7i instances have the following specs:
 
 See `Amazon EC2 I7i Instances <https://aws.amazon.com/ec2/instance-types/i7i/>`_ for details.
 
+
+.. _system-requirements-i8g-instances:
+
+i8g instances
+^^^^^^^^^^^^^^
+
+The following i8g instances are supported.
+
+.. list-table::
+   :widths: 30 20 20 30
+   :header-rows: 1
+
+   * - Model
+     - vCPU
+     - Mem (GiB)
+     - Storage (GB)
+   * - i8g.large
+     - 2
+     - 16
+     - 1 x 468 GB
+   * - i8g.xlarge
+     - 4
+     - 32
+     - 1 x 937 GB
+   * - i8g.2xlarge
+     - 8
+     - 64
+     - 1 x 1,875 GB
+   * - i8g.4xlarge
+     - 16
+     - 128
+     - 1 x 3,750 GB
+   * - i8g.8xlarge
+     - 32
+     - 256
+     - 2 x 3,750 GB
+   * - i8g.12xlarge
+     - 48
+     - 384
+     - 3 x 3,750 GB
+   * - i8g.16xlarge
+     - 64
+     - 512
+     - 4 x 3,750 GB
+
+All i8g instances have the following specs:
+
+* Powered by AWS Graviton4 processors
+* 3rd generation AWS Nitro SSD storage
+* DDR5-5600 memory for improved throughput
+* Up to 100 Gbps of networking bandwidth and up to 60 Gbps of bandwidth to
+  Amazon Elastic Block Store (EBS)
+* Instance sizes offer up to 45 TB of total local NVMe instance storage
+
+See `Amazon EC2 I8g Instances <https://aws.amazon.com/ec2/instance-types/i8g/>`_ for details.
+
+.. _system-requirements-i8ge-instances:
+
+i8ge instances
+^^^^^^^^^^^^^^
+
+The following i8ge instances are supported.
+
+.. list-table::
+   :widths: 30 20 20 30
+   :header-rows: 1
+
+   * - Model
+     - vCPU
+     - Mem (GiB)
+     - Storage (GB)
+   * - i8ge.large
+     - 2
+     - 16
+     - 1 x 1,250 GB
+   * - i8ge.xlarge
+     - 4
+     - 32
+     - 1 x 2,500 GB
+   * - i8ge.2xlarge
+     - 8
+     - 64
+     - 2 x 2,500 GB
+   * - i8ge.3xlarge
+     - 12
+     - 96
+     - 1 x 7,500 GB
+   * - i8ge.6xlarge
+     - 24
+     - 192
+     - 2 x 7,500 GB
+   * - i8ge.12xlarge
+     - 48
+     - 384
+     - 4 x 7,500 GB
+   * - i8ge.18xlarge
+     - 72
+     - 576
+     - 6 x 7,500 GB
+
+All i8ge instances have the following specs:
+
+* Powered by AWS Graviton4 processors
+* 3rd generation AWS Nitro SSD storage
+* DDR5-5600 memory for improved throughput
+* Up to 300 Gbps of networking bandwidth and up to 60 Gbps of bandwidth to
+  Amazon Elastic Block Store (EBS)
+* Instance sizes offer up to 120 TB of total local NVMe instance storage
+
+See `Amazon EC2 I8g Instances <https://aws.amazon.com/ec2/instance-types/i8g/>`_ for details.
+
+
 Im4gn and Is4gen instances
 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^
 ScyllaDB supports Arm-based Im4gn and Is4gen instances. See  `Amazon EC2 Im4gn and Is4gen instances <https://aws.amazon.com/ec2/instance-types/i4g/>`_ for specification details. 

docs/getting-started/index.rst

@@ -25,8 +25,7 @@ Getting Started
   :id: "getting-started"
   :class: my-panel
 
-  * `Install ScyllaDB (Binary Packages, Docker, or EC2) <https://www.scylladb.com/download/#core>`_ - Links to the ScyllaDB Download Center
-  
+  * :doc:`Install ScyllaDB </getting-started/install-scylla/index/>`
   * :doc:`Configure ScyllaDB </getting-started/system-configuration/>`
   * :doc:`Run ScyllaDB in a Shared Environment </getting-started/scylla-in-a-shared-environment>`
   * :doc:`Create a ScyllaDB Cluster - Single Data Center (DC) </operating-scylla/procedures/cluster-management/create-cluster/>`