Merge "mvcc: Fix incorrect schema version being used to copy the mutation when applying (#5099)" from Tomasz

"
Currently affects only counter tables.

Introduced in 27014a2.

mutation_partition(s, mp) is incorrect because it uses s to interpret
mp, while it should use mp_schema.

We may hit this if the current node has a newer schema than the
incoming mutation. This can happen during table schema altering when we receive the
mutation from a node which hasn't processed the schema change yet.

This is undefined behavior in general. If the alter was adding or
removing columns, this may result in corruption of the write where
values of one column are inserted into a different column.

Fixes #5095.
"

* 'fix-schema-alter-counter-tables' of https://github.com/tgrabiec/scylla:
  mvcc: Fix incorrect schema verison being used to copy the mutation when applying
  mutation_partition: Track and validate schema version in debug builds
  tests: Use the correct schema to access mutation_partition

(cherry picked from commit 83bc59a89f)

.gitmodules

@@ -1,6 +1,6 @@
 [submodule "seastar"]
 	path = seastar
-	url = ../seastar
+	url = ../scylla-seastar
 	ignore = dirty
 [submodule "swagger-ui"]
 	path = swagger-ui

SCYLLA-VERSION-GEN

@@ -1,6 +1,6 @@
 #!/bin/sh
 
-VERSION=666.development
+VERSION=2.3.6
 
 if test -f version
 then

api/api-doc/storage_service.json

@@ -2228,11 +2228,11 @@
                "description":"The column family"
             },
             "total":{
-               "type":"int",
+               "type":"long",
                "description":"The total snapshot size"
             },
             "live":{
-               "type":"int",
+               "type":"long",
                "description":"The live snapshot size"
             }
          }

atomic_cell.cc

@@ -187,7 +187,24 @@ size_t atomic_cell_or_collection::external_memory_usage(const abstract_type& t)
         return 0;
     }
     auto ctx = data::cell::context(_data.get(), t.imr_state().type_info());
-    return data::cell::structure::serialized_object_size(_data.get(), ctx);
+
+    auto view = data::cell::structure::make_view(_data.get(), ctx);
+    auto flags = view.get<data::cell::tags::flags>();
+
+    size_t external_value_size = 0;
+    if (flags.get<data::cell::tags::external_data>()) {
+        if (flags.get<data::cell::tags::collection>()) {
+            external_value_size = get_collection_mutation_view(_data.get()).data.size_bytes();
+        } else {
+            auto cell_view = data::cell::atomic_cell_view(t.imr_state().type_info(), view);
+            external_value_size = cell_view.value_size();
+        }
+        // Add overhead of chunk headers. The last one is a special case.
+        external_value_size += (external_value_size - 1) / data::cell::maximum_external_chunk_length * data::cell::external_chunk_overhead;
+        external_value_size += data::cell::external_last_chunk_overhead;
+    }
+    return data::cell::structure::serialized_object_size(_data.get(), ctx)
+        + imr_object_type::size_overhead + external_value_size;
 }
 
 std::ostream& operator<<(std::ostream& os, const atomic_cell_or_collection& c) {

auth/common.cc

@@ -28,6 +28,7 @@
 #include "database.hh"
 #include "schema_builder.hh"
 #include "service/migration_manager.hh"
+#include "timeout_config.hh"
 
 namespace auth {
 
@@ -94,4 +95,10 @@ future<> wait_for_schema_agreement(::service::migration_manager& mm, const datab
     });
 }
 
+const timeout_config& internal_distributed_timeout_config() noexcept {
+    static const auto t = 5s;
+    static const timeout_config tc{t, t, t, t, t, t, t};
+    return tc;
+}
+
 }

auth/common.hh

@@ -38,6 +38,7 @@
 using namespace std::chrono_literals;
 
 class database;
+class timeout_config;
 
 namespace service {
 class migration_manager;
@@ -82,4 +83,9 @@ future<> create_metadata_table_if_missing(
 
 future<> wait_for_schema_agreement(::service::migration_manager&, const database&);
 
+///
+/// Time-outs for internal, non-local CQL queries.
+///
+const timeout_config& internal_distributed_timeout_config() noexcept;
+
 }

auth/default_authorizer.cc

@@ -228,7 +228,7 @@ default_authorizer::modify(
         return _qp.process(
                 query,
                 db::consistency_level::ONE,
-                infinite_timeout_config,
+                internal_distributed_timeout_config(),
                 {permissions::to_strings(set), sstring(role_name), resource.name()}).discard_result();
     });
 }
@@ -254,7 +254,7 @@ future<std::vector<permission_details>> default_authorizer::list_all() const {
     return _qp.process(
             query,
             db::consistency_level::ONE,
-            infinite_timeout_config,
+            internal_distributed_timeout_config(),
             {},
             true).then([](::shared_ptr<cql3::untyped_result_set> results) {
         std::vector<permission_details> all_details;
@@ -282,7 +282,7 @@ future<> default_authorizer::revoke_all(stdx::string_view role_name) const {
     return _qp.process(
             query,
             db::consistency_level::ONE,
-            infinite_timeout_config,
+            internal_distributed_timeout_config(),
             {sstring(role_name)}).discard_result().handle_exception([role_name](auto ep) {
         try {
             std::rethrow_exception(ep);

auth/password_authenticator.cc

@@ -149,7 +149,9 @@ static sstring gensalt() {
     // blowfish 2011 fix, blowfish, sha512, sha256, md5
     for (sstring pfx : { "$2y$", "$2a$", "$6$", "$5$", "$1$" }) {
         salt = pfx + input;
-        if (crypt_r("fisk", salt.c_str(), &tlcrypt)) {
+        const char* e = crypt_r("fisk", salt.c_str(), &tlcrypt);
+
+        if (e && (e[0] != '*')) {
             prefix = pfx;
             return salt;
         }
@@ -184,7 +186,7 @@ future<> password_authenticator::migrate_legacy_metadata() const {
     return _qp.process(
             query,
             db::consistency_level::QUORUM,
-            infinite_timeout_config).then([this](::shared_ptr<cql3::untyped_result_set> results) {
+            internal_distributed_timeout_config()).then([this](::shared_ptr<cql3::untyped_result_set> results) {
         return do_for_each(*results, [this](const cql3::untyped_result_set_row& row) {
             auto username = row.get_as<sstring>("username");
             auto salted_hash = row.get_as<sstring>(SALTED_HASH);
@@ -192,7 +194,7 @@ future<> password_authenticator::migrate_legacy_metadata() const {
             return _qp.process(
                     update_row_query,
                     consistency_for_user(username),
-                    infinite_timeout_config,
+                    internal_distributed_timeout_config(),
                     {std::move(salted_hash), username}).discard_result();
         }).finally([results] {});
     }).then([] {
@@ -209,7 +211,7 @@ future<> password_authenticator::create_default_if_missing() const {
             return _qp.process(
                     update_row_query,
                     db::consistency_level::QUORUM,
-                    infinite_timeout_config,
+                    internal_distributed_timeout_config(),
                     {hashpw(DEFAULT_USER_PASSWORD), DEFAULT_USER_NAME}).then([](auto&&) {
                 plogger.info("Created default superuser authentication record.");
             });
@@ -309,13 +311,17 @@ future<authenticated_user> password_authenticator::authenticate(
         return _qp.process(
                 query,
                 consistency_for_user(username),
-                infinite_timeout_config,
+                internal_distributed_timeout_config(),
                 {username},
                 true);
     }).then_wrapped([=](future<::shared_ptr<cql3::untyped_result_set>> f) {
         try {
             auto res = f.get0();
-            if (res->empty() || !checkpw(password, res->one().get_as<sstring>(SALTED_HASH))) {
+            auto salted_hash = std::experimental::optional<sstring>();
+            if (!res->empty()) {
+                salted_hash = res->one().get_opt<sstring>(SALTED_HASH);
+            }
+            if (!salted_hash || !checkpw(password, *salted_hash)) {
                 throw exceptions::authentication_exception("Username and/or password are incorrect");
             }
             return make_ready_future<authenticated_user>(username);
@@ -337,7 +343,7 @@ future<> password_authenticator::create(stdx::string_view role_name, const authe
     return _qp.process(
             update_row_query,
             consistency_for_user(role_name),
-            infinite_timeout_config,
+            internal_distributed_timeout_config(),
             {hashpw(*options.password), sstring(role_name)}).discard_result();
 }
 
@@ -355,7 +361,7 @@ future<> password_authenticator::alter(stdx::string_view role_name, const authen
     return _qp.process(
             query,
             consistency_for_user(role_name),
-            infinite_timeout_config,
+            internal_distributed_timeout_config(),
             {hashpw(*options.password), sstring(role_name)}).discard_result();
 }
 
@@ -366,7 +372,10 @@ future<> password_authenticator::drop(stdx::string_view name) const {
             meta::roles_table::qualified_name(),
             meta::roles_table::role_col_name);
 
-    return _qp.process(query, consistency_for_user(name), infinite_timeout_config, {sstring(name)}).discard_result();
+    return _qp.process(
+            query, consistency_for_user(name),
+            internal_distributed_timeout_config(),
+            {sstring(name)}).discard_result();
 }
 
 future<custom_options> password_authenticator::query_custom_options(stdx::string_view role_name) const {

auth/roles-metadata.cc

@@ -79,7 +79,7 @@ future<bool> default_role_row_satisfies(
                 return qp.process(
                         query,
                         db::consistency_level::QUORUM,
-                        infinite_timeout_config,
+                        internal_distributed_timeout_config(),
                         {meta::DEFAULT_SUPERUSER_NAME},
                         true).then([&p](::shared_ptr<cql3::untyped_result_set> results) {
                     if (results->empty()) {
@@ -104,7 +104,7 @@ future<bool> any_nondefault_role_row_satisfies(
         return qp.process(
                 query,
                 db::consistency_level::QUORUM,
-                infinite_timeout_config).then([&p](::shared_ptr<cql3::untyped_result_set> results) {
+                internal_distributed_timeout_config()).then([&p](::shared_ptr<cql3::untyped_result_set> results) {
             if (results->empty()) {
                 return false;
             }

auth/service.cc

@@ -196,6 +196,10 @@ future<> service::start() {
 }
 
 future<> service::stop() {
+    // Only one of the shards has the listener registered, but let's try to
+    // unregister on each one just to make sure.
+    _migration_manager.unregister_listener(_migration_listener.get());
+
     return _permissions_cache->stop().then([this] {
         return when_all_succeed(_role_manager->stop(), _authorizer->stop(), _authenticator->stop());
     });

auth/standard_role_manager.cc

@@ -89,7 +89,7 @@ static future<stdx::optional<record>> find_record(cql3::query_processor& qp, std
     return qp.process(
             query,
             consistency_for_role(role_name),
-            infinite_timeout_config,
+            internal_distributed_timeout_config(),
             {sstring(role_name)},
             true).then([](::shared_ptr<cql3::untyped_result_set> results) {
         if (results->empty()) {
@@ -174,7 +174,7 @@ future<> standard_role_manager::create_default_role_if_missing() const {
             return _qp.process(
                     query,
                     db::consistency_level::QUORUM,
-                    infinite_timeout_config,
+                    internal_distributed_timeout_config(),
                     {meta::DEFAULT_SUPERUSER_NAME}).then([](auto&&) {
                 log.info("Created default superuser role '{}'.", meta::DEFAULT_SUPERUSER_NAME);
                 return make_ready_future<>();
@@ -201,7 +201,7 @@ future<> standard_role_manager::migrate_legacy_metadata() const {
     return _qp.process(
             query,
             db::consistency_level::QUORUM,
-            infinite_timeout_config).then([this](::shared_ptr<cql3::untyped_result_set> results) {
+            internal_distributed_timeout_config()).then([this](::shared_ptr<cql3::untyped_result_set> results) {
         return do_for_each(*results, [this](const cql3::untyped_result_set_row& row) {
             role_config config;
             config.is_superuser = row.get_as<bool>("super");
@@ -263,7 +263,7 @@ future<> standard_role_manager::create_or_replace(stdx::string_view role_name, c
     return _qp.process(
             query,
             consistency_for_role(role_name),
-            infinite_timeout_config,
+            internal_distributed_timeout_config(),
             {sstring(role_name), c.is_superuser, c.can_login},
             true).discard_result();
 }
@@ -307,7 +307,7 @@ standard_role_manager::alter(stdx::string_view role_name, const role_config_upda
                         build_column_assignments(u),
                         meta::roles_table::role_col_name),
                 consistency_for_role(role_name),
-                infinite_timeout_config,
+                internal_distributed_timeout_config(),
                 {sstring(role_name)}).discard_result();
     });
 }
@@ -327,7 +327,7 @@ future<> standard_role_manager::drop(stdx::string_view role_name) const {
             return _qp.process(
                     query,
                     consistency_for_role(role_name),
-                    infinite_timeout_config,
+                    internal_distributed_timeout_config(),
                     {sstring(role_name)}).then([this, role_name](::shared_ptr<cql3::untyped_result_set> members) {
                 return parallel_for_each(
                         members->begin(),
@@ -367,7 +367,7 @@ future<> standard_role_manager::drop(stdx::string_view role_name) const {
             return _qp.process(
                     query,
                     consistency_for_role(role_name),
-                    infinite_timeout_config,
+                    internal_distributed_timeout_config(),
                     {sstring(role_name)}).discard_result();
         };
 
@@ -394,7 +394,7 @@ standard_role_manager::modify_membership(
         return _qp.process(
                 query,
                 consistency_for_role(grantee_name),
-                infinite_timeout_config,
+                internal_distributed_timeout_config(),
                 {role_set{sstring(role_name)}, sstring(grantee_name)}).discard_result();
     };
 
@@ -406,7 +406,7 @@ standard_role_manager::modify_membership(
                                 "INSERT INTO %s (role, member) VALUES (?, ?)",
                                 meta::role_members_table::qualified_name()),
                         consistency_for_role(role_name),
-                        infinite_timeout_config,
+                        internal_distributed_timeout_config(),
                         {sstring(role_name), sstring(grantee_name)}).discard_result();
 
             case membership_change::remove:
@@ -415,7 +415,7 @@ standard_role_manager::modify_membership(
                                 "DELETE FROM %s WHERE role = ? AND member = ?",
                                 meta::role_members_table::qualified_name()),
                         consistency_for_role(role_name),
-                        infinite_timeout_config,
+                        internal_distributed_timeout_config(),
                         {sstring(role_name), sstring(grantee_name)}).discard_result();
         }
 
@@ -516,7 +516,10 @@ future<role_set> standard_role_manager::query_all() const {
     // To avoid many copies of a view.
     static const auto role_col_name_string = sstring(meta::roles_table::role_col_name);
 
-    return _qp.process(query, db::consistency_level::QUORUM, infinite_timeout_config).then([](::shared_ptr<cql3::untyped_result_set> results) {
+    return _qp.process(
+            query,
+            db::consistency_level::QUORUM,
+            internal_distributed_timeout_config()).then([](::shared_ptr<cql3::untyped_result_set> results) {
         role_set roles;
 
         std::transform(

cache_flat_mutation_reader.hh

@@ -60,6 +60,7 @@ class cache_flat_mutation_reader final : public flat_mutation_reader::impl {
         // - _next_row_in_range = _next.position() < _upper_bound
         // - _last_row points at a direct predecessor of the next row which is going to be read.
         //   Used for populating continuity.
+        // - _population_range_starts_before_all_rows is set accordingly
         reading_from_underlying,
 
         end_of_stream
@@ -86,6 +87,13 @@ class cache_flat_mutation_reader final : public flat_mutation_reader::impl {
     partition_snapshot_row_cursor _next_row;
     bool _next_row_in_range = false;
 
+    // True iff current population interval, since the previous clustering row, starts before all clustered rows.
+    // We cannot just look at _lower_bound, because emission of range tombstones changes _lower_bound and
+    // because we mark clustering intervals as continuous when consuming a clustering_row, it would prevent
+    // us from marking the interval as continuous.
+    // Valid when _state == reading_from_underlying.
+    bool _population_range_starts_before_all_rows;
+
     // Whether _lower_bound was changed within current fill_buffer().
     // If it did not then we cannot break out of it (e.g. on preemption) because
     // forward progress is not guaranteed in case iterators are getting constantly invalidated.
@@ -231,6 +239,7 @@ inline
 future<> cache_flat_mutation_reader::do_fill_buffer(db::timeout_clock::time_point timeout) {
     if (_state == state::move_to_underlying) {
         _state = state::reading_from_underlying;
+        _population_range_starts_before_all_rows = _lower_bound.is_before_all_clustered_rows(*_schema);
         auto end = _next_row_in_range ? position_in_partition(_next_row.position())
                                       : position_in_partition(_upper_bound);
         return _read_context->fast_forward_to(position_range{_lower_bound, std::move(end)}, timeout).then([this, timeout] {
@@ -355,12 +364,12 @@ future<> cache_flat_mutation_reader::read_from_underlying(db::timeout_clock::tim
                 }
             });
             return make_ready_future<>();
-        });
+        }, timeout);
 }
 
 inline
 bool cache_flat_mutation_reader::ensure_population_lower_bound() {
-    if (!_ck_ranges_curr->start()) {
+    if (_population_range_starts_before_all_rows) {
         return true;
     }
     if (!_last_row.refresh(*_snp)) {
@@ -415,6 +424,7 @@ inline
 void cache_flat_mutation_reader::maybe_add_to_cache(const clustering_row& cr) {
     if (!can_populate()) {
         _last_row = nullptr;
+        _population_range_starts_before_all_rows = false;
         _read_context->cache().on_mispopulate();
         return;
     }
@@ -448,6 +458,7 @@ void cache_flat_mutation_reader::maybe_add_to_cache(const clustering_row& cr) {
         with_allocator(standard_allocator(), [&] {
             _last_row = partition_snapshot_row_weakref(*_snp, it, true);
         });
+        _population_range_starts_before_all_rows = false;
     });
 }
 

configure.py

@@ -303,6 +303,7 @@ scylla_tests = [
     'tests/imr_test',
     'tests/partition_data_test',
     'tests/reusable_buffer_test',
+    'tests/json_test'
 ]
 
 perf_tests = [
@@ -406,6 +407,7 @@ scylla_core = (['database.cc',
                  'mutation_reader.cc',
                  'flat_mutation_reader.cc',
                  'mutation_query.cc',
+                 'json.cc',
                  'keys.cc',
                  'counters.cc',                 
                  'compress.cc',
@@ -514,6 +516,7 @@ scylla_core = (['database.cc',
                  'db/consistency_level.cc',
                  'db/system_keyspace.cc',
                  'db/system_distributed_keyspace.cc',
+                 'db/size_estimates_virtual_reader.cc',
                  'db/schema_tables.cc',
                  'db/cql_type_parser.cc',
                  'db/legacy_schema_migrator.cc',
@@ -740,6 +743,7 @@ pure_boost_tests = set([
     'tests/imr_test',
     'tests/partition_data_test',
     'tests/reusable_buffer_test',
+    'tests/json_test',
 ])
 
 tests_not_using_seastar_test_framework = set([
@@ -791,7 +795,7 @@ deps['tests/log_heap_test'] = ['tests/log_heap_test.cc']
 deps['tests/anchorless_list_test'] = ['tests/anchorless_list_test.cc']
 deps['tests/perf/perf_fast_forward'] += ['release.cc']
 deps['tests/meta_test'] = ['tests/meta_test.cc']
-deps['tests/imr_test'] = ['tests/imr_test.cc']
+deps['tests/imr_test'] = ['tests/imr_test.cc', 'utils/logalloc.cc', 'utils/dynamic_bitset.cc']
 deps['tests/reusable_buffer_test'] = ['tests/reusable_buffer_test.cc']
 
 warnings = [

cql3/Cql.g

@@ -473,9 +473,9 @@ insertStatement returns [::shared_ptr<raw::modification_statement> expr]
         ::shared_ptr<cql3::term::raw> json_value;
     }
     : K_INSERT K_INTO cf=columnFamilyName
-          '(' c1=cident { column_names.push_back(c1); }  ( ',' cn=cident { column_names.push_back(cn); } )* ')'
-        ( K_VALUES
-              '(' v1=term { values.push_back(v1); } ( ',' vn=term { values.push_back(vn); } )* ')'
+        ('(' c1=cident { column_names.push_back(c1); }  ( ',' cn=cident { column_names.push_back(cn); } )* ')'
+            K_VALUES
+            '(' v1=term { values.push_back(v1); } ( ',' vn=term { values.push_back(vn); } )* ')'
             ( K_IF K_NOT K_EXISTS { if_not_exists = true; } )?
             ( usingClause[attrs] )?
               {

cql3/error_collector.hh

@@ -67,6 +67,12 @@ class error_collector : public error_listener<RecognizerType, ExceptionBaseType>
      */
     const sstring_view _query;
 
+    /**
+     * An empty bitset to be used as a workaround for AntLR null dereference
+     * bug.
+     */
+    static typename ExceptionBaseType::BitsetListType _empty_bit_list;
+
 public:
 
     /**
@@ -144,6 +150,14 @@ private:
             break;
         }
         default:
+            // AntLR Exception class has a bug of dereferencing a null
+            // pointer in the displayRecognitionError. The following
+            // if statement makes sure it will not be null before the
+            // call to that function (displayRecognitionError).
+            // bug reference: https://github.com/antlr/antlr3/issues/191
+            if (!ex->get_expectingSet()) {
+                ex->set_expectingSet(&_empty_bit_list);
+            }
             ex->displayRecognitionError(token_names, msg);
         }
         return msg.str();
@@ -345,4 +359,8 @@ private:
 #endif
 };
 
+template<typename RecognizerType, typename TokenType, typename ExceptionBaseType>
+typename ExceptionBaseType::BitsetListType
+error_collector<RecognizerType,TokenType,ExceptionBaseType>::_empty_bit_list = typename ExceptionBaseType::BitsetListType();
+
 }

cql3/functions/functions.cc

@@ -177,7 +177,7 @@ shared_ptr<function>
 make_to_json_function(data_type t) {
     return make_native_scalar_function<true>("tojson", utf8_type, {t},
             [t](cql_serialization_format sf, const std::vector<bytes_opt>& parameters) -> bytes_opt {
-        return utf8_type->decompose(t->to_json_string(parameters[0].value()));
+        return utf8_type->decompose(t->to_json_string(parameters[0]));
     });
 }
 

cql3/query_options.cc

@@ -217,19 +217,18 @@ void query_options::prepare(const std::vector<::shared_ptr<column_specification>
     }
 
     auto& names = *_names;
-    std::vector<cql3::raw_value> ordered_values;
+    std::vector<cql3::raw_value_view> ordered_values;
     ordered_values.reserve(specs.size());
     for (auto&& spec : specs) {
         auto& spec_name = spec->name->text();
         for (size_t j = 0; j < names.size(); j++) {
             if (names[j] == spec_name) {
-                ordered_values.emplace_back(_values[j]);
+                ordered_values.emplace_back(_value_views[j]);
                 break;
             }
         }
     }
-    _values = std::move(ordered_values);
-    fill_value_views();
+    _value_views = std::move(ordered_values);
 }
 
 void query_options::fill_value_views()

cql3/query_processor.cc

@@ -239,11 +239,11 @@ query_processor::process(const sstring_view& query_string, service::query_state&
     log.trace("process: \"{}\"", query_string);
     tracing::trace(query_state.get_trace_state(), "Parsing a statement");
     auto p = get_statement(query_string, query_state.get_client_state());
-    options.prepare(p->bound_names);
     auto cql_statement = p->statement;
     if (cql_statement->get_bound_terms() != options.get_values_count()) {
         throw exceptions::invalid_request_exception("Invalid amount of bind variables");
     }
+    options.prepare(p->bound_names);
 
     warn(unimplemented::cause::METRICS);
 #if 0

cql3/restrictions/single_column_restriction.hh

@@ -202,6 +202,14 @@ public:
                                  const query_options& options,
                                  gc_clock::time_point now) const override;
 
+    virtual std::vector<bytes_opt> values_raw(const query_options& options) const = 0;
+
+    virtual std::vector<bytes_opt> values(const query_options& options) const override {
+        std::vector<bytes_opt> ret = values_raw(options);
+        std::sort(ret.begin(),ret.end());
+        ret.erase(std::unique(ret.begin(),ret.end()),ret.end());
+        return ret;
+    }
 #if 0
     @Override
     protected final boolean isSupportedBy(SecondaryIndex index)
@@ -224,7 +232,7 @@ public:
         return abstract_restriction::term_uses_function(_values, ks_name, function_name);
     }
 
-    virtual std::vector<bytes_opt> values(const query_options& options) const override {
+    virtual std::vector<bytes_opt> values_raw(const query_options& options) const override {
         std::vector<bytes_opt> ret;
         for (auto&& v : _values) {
             ret.emplace_back(to_bytes_opt(v->bind_and_get(options)));
@@ -249,7 +257,7 @@ public:
         return false;
     }
 
-    virtual std::vector<bytes_opt> values(const query_options& options) const override {
+    virtual std::vector<bytes_opt> values_raw(const query_options& options) const override {
         auto&& lval = dynamic_pointer_cast<multi_item_terminal>(_marker->bind(options));
         if (!lval) {
             throw exceptions::invalid_request_exception("Invalid null value for IN restriction");

cql3/selection/selector.hh

@@ -105,9 +105,11 @@ public:
     virtual void reset() = 0;
 
     virtual assignment_testable::test_result test_assignment(database& db, const sstring& keyspace, ::shared_ptr<column_specification> receiver) override {
-        if (receiver->type == get_type()) {
+        auto t1 = receiver->type->underlying_type();
+        auto t2 = get_type()->underlying_type();
+        if (t1 == t2) {
             return assignment_testable::test_result::EXACT_MATCH;
-        } else if (receiver->type->is_value_compatible_with(*get_type())) {
+        } else if (t1->is_value_compatible_with(*t2)) {
             return assignment_testable::test_result::WEAKLY_ASSIGNABLE;
         } else {
             return assignment_testable::test_result::NOT_ASSIGNABLE;

cql3/statements/select_statement.cc

@@ -96,12 +96,8 @@ public:
                 encoded_row.write("\\\"", 2);
             }
             encoded_row.write("\": ", 3);
-            if (parameters[i]) {
-                sstring row_sstring = _selector_types[i]->to_json_string(parameters[i].value());
-                encoded_row.write(row_sstring.c_str(), row_sstring.size());
-            } else {
-                encoded_row.write("null", 4);
-            }
+            sstring row_sstring = _selector_types[i]->to_json_string(parameters[i]);
+            encoded_row.write(row_sstring.c_str(), row_sstring.size());
         }
         encoded_row.write("}", 1);
         return encoded_row.linearize().to_string();
@@ -974,6 +970,10 @@ std::unique_ptr<prepared_statement> select_statement::prepare(database& db, cql_
     }
 
     check_needs_filtering(restrictions);
+    size_t restrictions_size = restrictions->get_partition_key_restrictions()->size() + restrictions->get_clustering_columns_restrictions()->size() + restrictions->get_non_pk_restriction().size();
+    if (restrictions->uses_secondary_indexing() && restrictions_size > 1) {
+        throw exceptions::invalid_request_exception("Indexed query may not contain multiple restrictions in 2.3");
+    }
 
     ::shared_ptr<cql3::statements::select_statement> stmt;
     if (restrictions->uses_secondary_indexing()) {

cql3/statements/update_statement.cc

@@ -179,7 +179,21 @@ modification_statement::json_cache_opt insert_prepared_json_statement::maybe_pre
 void
 insert_prepared_json_statement::execute_set_value(mutation& m, const clustering_key_prefix& prefix, const update_parameters& params, const column_definition& column, const bytes_opt& value) {
     if (!value) {
+        if (column.type->is_collection()) {
+            auto& k = static_pointer_cast<const collection_type_impl>(column.type)->_kind;
+            if (&k == &collection_type_impl::kind::list) {
+                lists::setter::execute(m, prefix, params, column, make_shared<lists::value>(lists::value(std::vector<bytes_opt>())));
+            } else if (&k == &collection_type_impl::kind::set) {
+                sets::setter::execute(m, prefix, params, column, make_shared<sets::value>(sets::value(std::set<bytes, serialized_compare>(serialized_compare(empty_type)))));
+            } else if (&k == &collection_type_impl::kind::map) {
+                maps::setter::execute(m, prefix, params, column, make_shared<maps::value>(maps::value(std::map<bytes, bytes, serialized_compare>(serialized_compare(empty_type)))));
+            } else {
+                throw exceptions::invalid_request_exception("Incorrect value kind in JSON INSERT statement");
+            }
+            return;
+        }
         m.set_cell(prefix, column, std::move(operation::make_dead_cell(params)));
+        return;
     } else if (!column.type->is_collection()) {
         constants::setter::execute(m, prefix, params, column, raw_value_view::make_value(bytes_view(*value)));
         return;
@@ -204,15 +218,17 @@ insert_prepared_json_statement::execute_set_value(mutation& m, const clustering_
 dht::partition_range_vector
 insert_prepared_json_statement::build_partition_keys(const query_options& options, const json_cache_opt& json_cache) {
     dht::partition_range_vector ranges;
+    std::vector<bytes_opt> exploded;
     for (const auto& def : s->partition_key_columns()) {
         auto json_value = json_cache->at(def.name_as_text());
-        auto k = query::range<partition_key>::make_singular(partition_key::from_single_value(*s, json_value.value()));
-        ranges.emplace_back(std::move(k).transform(
-                    [this] (partition_key&& k) -> query::ring_position {
-                        auto token = dht::global_partitioner().get_token(*s, k);
-                        return { std::move(token), std::move(k) };
-                    }));
+        if (!json_value) {
+            throw exceptions::invalid_request_exception(sprint("Missing mandatory PRIMARY KEY part %s", def.name_as_text()));
+        }
+        exploded.emplace_back(*json_value);
     }
+    auto pkey = partition_key::from_optional_exploded(*s, std::move(exploded));
+    auto k = query::range<query::ring_position>::make_singular(dht::global_partitioner().decorate_key(*s, std::move(pkey)));
+    ranges.emplace_back(std::move(k));
     return ranges;
 }
 
@@ -221,7 +237,10 @@ query::clustering_row_ranges insert_prepared_json_statement::create_clustering_r
     std::vector<bytes_opt> exploded;
     for (const auto& def : s->clustering_key_columns()) {
         auto json_value = json_cache->at(def.name_as_text());
-        exploded.emplace_back(json_value.value());
+        if (!json_value) {
+            throw exceptions::invalid_request_exception(sprint("Missing mandatory PRIMARY KEY part %s", def.name_as_text()));
+        }
+        exploded.emplace_back(*json_value);
     }
     auto k = query::range<clustering_key_prefix>::make_singular(clustering_key_prefix::from_optional_exploded(*s, std::move(exploded)));
     ranges.emplace_back(query::clustering_range(std::move(k)));

cql3/tuples.hh

@@ -405,7 +405,7 @@ public:
         in_marker(int32_t bind_index, ::shared_ptr<column_specification> receiver)
             : abstract_marker(bind_index, std::move(receiver))
         {
-            assert(dynamic_pointer_cast<const list_type_impl>(receiver->type));
+            assert(dynamic_pointer_cast<const list_type_impl>(_receiver->type));
         }
 
         virtual shared_ptr<terminal> bind(const query_options& options) override {

cql3/update_parameters.cc

@@ -53,6 +53,9 @@ update_parameters::get_prefetched_list(
         return {};
     }
 
+    if (column.is_static()) {
+        ckey = clustering_key_view::make_empty();
+    }
     auto i = _prefetched->rows.find(std::make_pair(std::move(pkey), std::move(ckey)));
     if (i == _prefetched->rows.end()) {
         return {};

data/cell.hh

@@ -211,6 +211,7 @@ struct cell {
         imr::member<tags::chunk_next, imr::pod<uint8_t*>>,
         imr::member<tags::chunk_data, imr::buffer<tags::chunk_data>>
     >;
+    static constexpr size_t external_chunk_overhead = sizeof(uint8_t*) * 2;
 
     using external_last_chunk_size = imr::pod<uint16_t>;
     /// The last fragment of an externally stored value
@@ -224,6 +225,7 @@ struct cell {
         imr::member<tags::last_chunk_size, external_last_chunk_size>,
         imr::member<tags::chunk_data, imr::buffer<tags::chunk_data>>
     >;
+    static constexpr size_t external_last_chunk_overhead = sizeof(uint8_t*) + sizeof(uint16_t);
 
     class context;
     class minimal_context;

database.cc

@@ -383,9 +383,13 @@ filter_sstable_for_reader(std::vector<sstables::shared_sstable>&& sstables, colu
     };
     sstables.erase(boost::remove_if(sstables, sstable_has_not_key), sstables.end());
 
+    // FIXME: Workaround for https://github.com/scylladb/scylla/issues/3552
+    // and https://github.com/scylladb/scylla/issues/3553
+    const bool filtering_broken = true;
+
     // no clustering filtering is applied if schema defines no clustering key or
     // compaction strategy thinks it will not benefit from such an optimization.
-    if (!schema->clustering_key_size() || !cf.get_compaction_strategy().use_clustering_key_filter()) {
+    if (filtering_broken || !schema->clustering_key_size() || !cf.get_compaction_strategy().use_clustering_key_filter()) {
          return sstables;
     }
     ::cf_stats* stats = cf.cf_stats();
@@ -957,6 +961,11 @@ table::seal_active_memtable(flush_permit&& permit) {
     }
     _memtables->add_memtable();
     _stats.memtable_switch_count++;
+    // This will set evictable occupancy of the old memtable region to zero, so that
+    // this region is considered last for flushing by dirty_memory_manager::flush_when_needed().
+    // If we don't do that, the flusher may keep picking up this memtable list for flushing after
+    // the permit is released even though there is not much to flush in the active memtable of this list.
+    old->region().ground_evictable_occupancy();
     auto previous_flush = _flush_barrier.advance_and_await();
     auto op = _flush_barrier.start();
 
@@ -1325,6 +1334,7 @@ table::on_compaction_completion(const std::vector<sstables::shared_sstable>& new
 
     // This is done in the background, so we can consider this compaction completed.
     seastar::with_gate(_sstable_deletion_gate, [this, sstables_to_remove] {
+       return with_semaphore(_sstable_deletion_sem, 1, [this, sstables_to_remove = std::move(sstables_to_remove)] {
         return sstables::delete_atomically(sstables_to_remove, *get_large_partition_handler()).then_wrapped([this, sstables_to_remove] (future<> f) {
             std::exception_ptr eptr;
             try {
@@ -1348,6 +1358,7 @@ table::on_compaction_completion(const std::vector<sstables::shared_sstable>& new
                 return make_exception_future<>(eptr);
             }
             return make_ready_future<>();
+         });
         }).then([this] {
             // refresh underlying data source in row cache to prevent it from holding reference
             // to sstables files which were previously deleted.
@@ -1469,7 +1480,10 @@ future<> table::cleanup_sstables(sstables::compaction_descriptor descriptor) {
             static thread_local semaphore sem(1);
 
             return with_semaphore(sem, 1, [this, &sst] {
-                return this->compact_sstables(sstables::compaction_descriptor({ sst }, sst->get_sstable_level()), true);
+                // release reference to sstables cleaned up, otherwise space usage from their data and index
+                // components cannot be reclaimed until all of them are cleaned.
+                auto sstable_level = sst->get_sstable_level();
+                return this->compact_sstables(sstables::compaction_descriptor({ std::move(sst) }, sstable_level), true);
             });
         });
     });
@@ -1651,9 +1665,9 @@ future<> distributed_loader::open_sstable(distributed<database>& db, sstables::e
     // to distribute evenly the resource usage among all shards.
 
     return db.invoke_on(column_family::calculate_shard_from_sstable_generation(comps.generation),
-            [&db, comps = std::move(comps), func = std::move(func), pc] (database& local) {
+            [&db, comps = std::move(comps), func = std::move(func), &pc] (database& local) {
 
-        return with_semaphore(local.sstable_load_concurrency_sem(), 1, [&db, &local, comps = std::move(comps), func = std::move(func), pc] {
+        return with_semaphore(local.sstable_load_concurrency_sem(), 1, [&db, &local, comps = std::move(comps), func = std::move(func), &pc] {
             auto& cf = local.find_column_family(comps.ks, comps.cf);
 
             auto f = sstables::sstable::load_shared_components(cf.schema(), cf._config.datadir, comps.generation, comps.version, comps.format, pc);
@@ -2699,7 +2713,7 @@ future<> database::drop_column_family(const sstring& ks_name, const sstring& cf_
     remove(*cf);
     cf->clear_views();
     auto& ks = find_keyspace(ks_name);
-    return cf->await_pending_writes().then([this, &ks, cf, tsf = std::move(tsf), snapshot] {
+    return when_all_succeed(cf->await_pending_writes(), cf->await_pending_reads()).then([this, &ks, cf, tsf = std::move(tsf), snapshot] {
         return truncate(ks, *cf, std::move(tsf), snapshot).finally([this, cf] {
             return cf->stop();
         });
@@ -3139,7 +3153,7 @@ database::query(schema_ptr s, const query::read_command& cmd, query::result_opti
             seastar::ref(get_result_memory_limiter()),
             max_result_size,
             timeout,
-            std::move(cache_ctx)).then_wrapped([this, s = _stats, hit_rate = cf.get_global_cache_hit_rate()] (auto f) {
+            std::move(cache_ctx)).then_wrapped([this, s = _stats, hit_rate = cf.get_global_cache_hit_rate(), op = cf.read_in_progress()] (auto f) {
         if (f.failed()) {
             ++s->total_reads_failed;
             return make_exception_future<lw_shared_ptr<query::result>, cache_temperature>(f.get_exception());
@@ -3167,7 +3181,7 @@ database::query_mutations(schema_ptr s, const query::read_command& cmd, const dh
             std::move(accounter),
             std::move(trace_state),
             timeout,
-            std::move(cache_ctx)).then_wrapped([this, s = _stats, hit_rate = cf.get_global_cache_hit_rate()] (auto f) {
+            std::move(cache_ctx)).then_wrapped([this, s = _stats, hit_rate = cf.get_global_cache_hit_rate(), op = cf.read_in_progress()] (auto f) {
         if (f.failed()) {
             ++s->total_reads_failed;
             return make_exception_future<reconcilable_result, cache_temperature>(f.get_exception());
@@ -3433,6 +3447,13 @@ future<> dirty_memory_manager::flush_when_needed() {
                 // release the biggest amount of memory and is less likely to be generating tiny
                 // SSTables.
                 memtable& candidate_memtable = memtable::from_region(*(this->_virtual_region_group.get_largest_region()));
+
+                if (candidate_memtable.empty()) {
+                    // Soft pressure, but nothing to flush. It could be due to fsync or memtable_to_cache lagging.
+                    // Back off to avoid OOMing with flush continuations.
+                    return sleep(1ms);
+                }
+
                 // Do not wait. The semaphore will protect us against a concurrent flush. But we
                 // want to start a new one as soon as the permits are destroyed and the semaphore is
                 // made ready again, not when we are done with the current one.
@@ -3980,6 +4001,7 @@ seal_snapshot(sstring jsondir) {
 
 future<> table::snapshot(sstring name) {
     return flush().then([this, name = std::move(name)]() {
+       return with_semaphore(_sstable_deletion_sem, 1, [this, name = std::move(name)]() {
         auto tables = boost::copy_range<std::vector<sstables::shared_sstable>>(*_sstables->all());
         return do_with(std::move(tables), [this, name](std::vector<sstables::shared_sstable> & tables) {
             auto jsondir = _config.datadir + "/snapshots/" + name;
@@ -4044,6 +4066,7 @@ future<> table::snapshot(sstring name) {
                 });
             });
         });
+       });
     });
 }
 
@@ -4175,6 +4198,7 @@ future<> table::fail_streaming_mutations(utils::UUID plan_id) {
     _streaming_memtables_big.erase(it);
     return entry->flush_in_progress.close().then([this, entry] {
         for (auto&& sst : entry->sstables) {
+            sst.monitor->write_failed();
             sst.sstable->mark_for_deletion();
         }
     });

database.hh

@@ -294,6 +294,8 @@ public:
 class table;
 using column_family = table;
 
+class database_sstable_write_monitor;
+
 class table : public enable_lw_shared_from_this<table> {
 public:
     struct config {
@@ -389,7 +391,7 @@ private:
     // plan memtables and the resulting sstables are not made visible until
     // the streaming is complete.
     struct monitored_sstable {
-        std::unique_ptr<sstables::write_monitor> monitor;
+        std::unique_ptr<database_sstable_write_monitor> monitor;
         sstables::shared_sstable sstable;
     };
 
@@ -428,6 +430,10 @@ private:
     std::unordered_map<uint64_t, sstables::shared_sstable> _sstables_need_rewrite;
     // Control background fibers waiting for sstables to be deleted
     seastar::gate _sstable_deletion_gate;
+    // This semaphore ensures that an operation like snapshot won't have its selected
+    // sstables deleted by compaction in parallel, a race condition which could
+    // easily result in failure.
+    seastar::semaphore _sstable_deletion_sem = {1};
     // There are situations in which we need to stop writing sstables. Flushers will take
     // the read lock, and the ones that wish to stop that process will take the write lock.
     rwlock _sstables_lock;
@@ -475,6 +481,8 @@ private:
     // after some modification, needs to ensure that news writes will see it before
     // it can proceed, such as the view building code.
     utils::phased_barrier _pending_writes_phaser;
+    // Corresponding phaser for in-progress reads.
+    utils::phased_barrier _pending_reads_phaser;
 private:
     void update_stats_for_new_sstable(uint64_t disk_space_used_by_sstable, const std::vector<unsigned>& shards_for_the_sstable) noexcept;
     // Adds new sstable to the set of sstables
@@ -817,6 +825,14 @@ public:
         return _pending_writes_phaser.advance_and_await();
     }
 
+    utils::phased_barrier::operation read_in_progress() {
+        return _pending_reads_phaser.start();
+    }
+
+    future<> await_pending_reads() {
+        return _pending_reads_phaser.advance_and_await();
+    }
+
     void add_or_update_view(view_ptr v);
     void remove_view(view_ptr v);
     void clear_views();

db/commitlog/commitlog_replayer.cc

@@ -163,7 +163,7 @@ future<> db::commitlog_replayer::impl::init() {
                 // Get all truncation records for the CF and initialize max rps if
                 // present. Cannot do this on demand, as there may be no sstables to
                 // mark the CF as "needed".
-                return db::system_keyspace::get_truncated_position(uuid).then([&map, &uuid](std::vector<db::replay_position> tpps) {
+                return db::system_keyspace::get_truncated_position(uuid).then([&map, uuid](std::vector<db::replay_position> tpps) {
                     for (auto& p : tpps) {
                         rlogger.trace("CF {} truncated at {}", uuid, p);
                         auto& pp = map[p.shard_id()][uuid];

db/schema_tables.cc

@@ -686,33 +686,7 @@ read_keyspace_mutation(distributed<service::storage_proxy>& proxy, const sstring
 static semaphore the_merge_lock {1};
 
 future<> merge_lock() {
-    // ref:  #1088
-    // to avoid deadlocks, we don't want long-standing calls to the shard 0
-    // as they can cause a deadlock:
-    //
-    //   fiber1                fiber2
-    //   merge_lock()                         (succeeds)
-    //                         merge_lock()   (waits)
-    //   invoke_on_all()                      (waits on merge_lock to relinquish smp::submit_to slot)
-    //
-    // so we issue the lock calls with a timeout; the slot will be relinquished, and invoke_on_all()
-    // can complete
-    return repeat([] () mutable {
-        return smp::submit_to(0, [] {
-            return the_merge_lock.try_wait();
-        }).then([] (bool result) {
-            if (result) {
-                return make_ready_future<stop_iteration>(stop_iteration::yes);
-            } else {
-                static thread_local auto rand_engine = std::default_random_engine();
-                auto dist = std::uniform_int_distribution<int>(0, 100);
-                auto to = std::chrono::microseconds(dist(rand_engine));
-                return sleep(to).then([] {
-                    return make_ready_future<stop_iteration>(stop_iteration::no);
-                });
-            }
-        });
-    });
+    return smp::submit_to(0, [] { return the_merge_lock.wait(); });
 }
 
 future<> merge_unlock() {

db/size_estimates_virtual_reader.cc

@@ -0,0 +1,329 @@
+/*
+ * Copyright (C) 2019 ScyllaDB
+ *
+ * Modified by ScyllaDB
+ */
+
+/*
+ * This file is part of Scylla.
+ *
+ * Scylla is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * Scylla is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include <boost/range/adaptor/indirected.hpp>
+#include <boost/range/adaptor/map.hpp>
+#include <boost/range/adaptor/transformed.hpp>
+#include <boost/range/algorithm/find_if.hpp>
+
+#include "clustering_bounds_comparator.hh"
+#include "database.hh"
+#include "db/system_keyspace.hh"
+#include "dht/i_partitioner.hh"
+#include "partition_range_compat.hh"
+#include "range.hh"
+#include "service/storage_service.hh"
+#include "stdx.hh"
+#include "mutation_fragment.hh"
+#include "sstables/sstables.hh"
+#include "db/timeout_clock.hh"
+#include "database.hh"
+
+#include "db/size_estimates_virtual_reader.hh"
+
+namespace db {
+
+namespace size_estimates {
+
+struct virtual_row {
+    const bytes& cf_name;
+    const token_range& tokens;
+    clustering_key_prefix as_key() const {
+        return clustering_key_prefix::from_exploded(std::vector<bytes_view>{cf_name, tokens.start, tokens.end});
+    }
+};
+
+struct virtual_row_comparator {
+    schema_ptr _schema;
+    virtual_row_comparator(schema_ptr schema) : _schema(schema) { }
+    bool operator()(const clustering_key_prefix& key1, const clustering_key_prefix& key2) {
+        return clustering_key_prefix::prefix_equality_less_compare(*_schema)(key1, key2);
+    }
+    bool operator()(const virtual_row& row, const clustering_key_prefix& key) {
+        return operator()(row.as_key(), key);
+    }
+    bool operator()(const clustering_key_prefix& key, const virtual_row& row) {
+        return operator()(key, row.as_key());
+    }
+};
+
+// Iterating over the cartesian product of cf_names and token_ranges.
+class virtual_row_iterator : public std::iterator<std::input_iterator_tag, const virtual_row> {
+    std::reference_wrapper<const std::vector<bytes>> _cf_names;
+    std::reference_wrapper<const std::vector<token_range>> _ranges;
+    size_t _cf_names_idx = 0;
+    size_t _ranges_idx = 0;
+public:
+    struct end_iterator_tag {};
+    virtual_row_iterator(const std::vector<bytes>& cf_names, const std::vector<token_range>& ranges)
+            : _cf_names(std::ref(cf_names))
+            , _ranges(std::ref(ranges))
+    { }
+    virtual_row_iterator(const std::vector<bytes>& cf_names, const std::vector<token_range>& ranges, end_iterator_tag)
+            : _cf_names(std::ref(cf_names))
+            , _ranges(std::ref(ranges))
+            , _cf_names_idx(cf_names.size())
+            , _ranges_idx(ranges.size())
+    {
+        if (cf_names.empty() || ranges.empty()) {
+            // The product of an empty range with any range is an empty range.
+            // In this case we want the end iterator to be equal to the begin iterator,
+            // which has_ranges_idx = _cf_names_idx = 0.
+            _ranges_idx = _cf_names_idx = 0;
+        }
+    }
+    virtual_row_iterator& operator++() {
+        if (++_ranges_idx == _ranges.get().size() && ++_cf_names_idx < _cf_names.get().size()) {
+            _ranges_idx = 0;
+        }
+        return *this;
+    }
+    virtual_row_iterator operator++(int) {
+        virtual_row_iterator i(*this);
+        ++(*this);
+        return i;
+    }
+    const value_type operator*() const {
+        return { _cf_names.get()[_cf_names_idx], _ranges.get()[_ranges_idx] };
+    }
+    bool operator==(const virtual_row_iterator& i) const {
+        return _cf_names_idx == i._cf_names_idx
+            && _ranges_idx == i._ranges_idx;
+    }
+    bool operator!=(const virtual_row_iterator& i) const {
+        return !(*this == i);
+    }
+};
+
+/**
+ * Returns the keyspaces, ordered by name, as selected by the partition_range.
+ */
+static std::vector<sstring> get_keyspaces(const schema& s, const database& db, dht::partition_range range) {
+    struct keyspace_less_comparator {
+        const schema& _s;
+        keyspace_less_comparator(const schema& s) : _s(s) { }
+        dht::ring_position as_ring_position(const sstring& ks) {
+            auto pkey = partition_key::from_single_value(_s, utf8_type->decompose(ks));
+            return dht::global_partitioner().decorate_key(_s, std::move(pkey));
+        }
+        bool operator()(const sstring& ks1, const sstring& ks2) {
+            return as_ring_position(ks1).less_compare(_s, as_ring_position(ks2));
+        }
+        bool operator()(const sstring& ks, const dht::ring_position& rp) {
+            return as_ring_position(ks).less_compare(_s, rp);
+        }
+        bool operator()(const dht::ring_position& rp, const sstring& ks) {
+            return rp.less_compare(_s, as_ring_position(ks));
+        }
+    };
+    auto keyspaces = db.get_non_system_keyspaces();
+    auto cmp = keyspace_less_comparator(s);
+    boost::sort(keyspaces, cmp);
+    return boost::copy_range<std::vector<sstring>>(
+        range.slice(keyspaces, std::move(cmp)) | boost::adaptors::filtered([&s] (const auto& ks) {
+            // If this is a range query, results are divided between shards by the partition key (keyspace_name).
+            return shard_of(dht::global_partitioner().get_token(s,
+                        partition_key::from_single_value(s, utf8_type->decompose(ks))))
+                == engine().cpu_id();
+        })
+    );
+}
+
+/**
+ * Makes a wrapping range of ring_position from a nonwrapping range of token, used to select sstables.
+ */
+static dht::partition_range as_ring_position_range(dht::token_range& r) {
+    stdx::optional<range<dht::ring_position>::bound> start_bound, end_bound;
+    if (r.start()) {
+        start_bound = {{ dht::ring_position(r.start()->value(), dht::ring_position::token_bound::start), r.start()->is_inclusive() }};
+    }
+    if (r.end()) {
+        end_bound = {{ dht::ring_position(r.end()->value(), dht::ring_position::token_bound::end), r.end()->is_inclusive() }};
+    }
+    return dht::partition_range(std::move(start_bound), std::move(end_bound), r.is_singular());
+}
+
+/**
+ * Add a new range_estimates for the specified range, considering the sstables associated with `cf`.
+ */
+static system_keyspace::range_estimates estimate(const column_family& cf, const token_range& r) {
+    int64_t count{0};
+    utils::estimated_histogram hist{0};
+    auto from_bytes = [] (auto& b) {
+        return dht::global_partitioner().from_sstring(utf8_type->to_string(b));
+    };
+    dht::token_range_vector ranges;
+    compat::unwrap_into(
+        wrapping_range<dht::token>({{ from_bytes(r.start) }}, {{ from_bytes(r.end) }}),
+        dht::token_comparator(),
+        [&] (auto&& rng) { ranges.push_back(std::move(rng)); });
+    for (auto&& r : ranges) {
+        auto rp_range = as_ring_position_range(r);
+        for (auto&& sstable : cf.select_sstables(rp_range)) {
+            count += sstable->estimated_keys_for_range(r);
+            hist.merge(sstable->get_stats_metadata().estimated_row_size);
+        }
+    }
+    return {cf.schema(), r.start, r.end, count, count > 0 ? hist.mean() : 0};
+}
+
+future<std::vector<token_range>> get_local_ranges() {
+    auto& ss = service::get_local_storage_service();
+    return ss.get_local_tokens().then([&ss] (auto&& tokens) {
+        auto ranges = ss.get_token_metadata().get_primary_ranges_for(std::move(tokens));
+        std::vector<token_range> local_ranges;
+        auto to_bytes = [](const stdx::optional<dht::token_range::bound>& b) {
+            assert(b);
+            return utf8_type->decompose(dht::global_partitioner().to_sstring(b->value()));
+        };
+        // We merge the ranges to be compatible with how Cassandra shows it's size estimates table.
+        // All queries will be on that table, where all entries are text and there's no notion of
+        // token ranges form the CQL point of view.
+        auto left_inf = boost::find_if(ranges, [] (auto&& r) {
+            return !r.start() || r.start()->value() == dht::minimum_token();
+        });
+        auto right_inf = boost::find_if(ranges, [] (auto&& r) {
+            return !r.end() || r.start()->value() == dht::maximum_token();
+        });
+        if (left_inf != right_inf && left_inf != ranges.end() && right_inf != ranges.end()) {
+            local_ranges.push_back(token_range{to_bytes(right_inf->start()), to_bytes(left_inf->end())});
+            ranges.erase(left_inf);
+            ranges.erase(right_inf);
+        }
+        for (auto&& r : ranges) {
+            local_ranges.push_back(token_range{to_bytes(r.start()), to_bytes(r.end())});
+        }
+        boost::sort(local_ranges, [] (auto&& tr1, auto&& tr2) {
+            return utf8_type->less(tr1.start, tr2.start);
+        });
+        return local_ranges;
+    });
+}
+
+size_estimates_mutation_reader::size_estimates_mutation_reader(schema_ptr schema, const dht::partition_range& prange, const query::partition_slice& slice, streamed_mutation::forwarding fwd)
+            : impl(schema)
+            , _schema(std::move(schema))
+            , _prange(&prange)
+            , _slice(slice)
+            , _fwd(fwd)
+    { }
+
+future<> size_estimates_mutation_reader::get_next_partition() {
+    auto& db = service::get_local_storage_proxy().get_db().local();
+    if (!_keyspaces) {
+        _keyspaces = get_keyspaces(*_schema, db, *_prange);
+        _current_partition = _keyspaces->begin();
+    }
+    if (_current_partition == _keyspaces->end()) {
+        _end_of_stream = true;
+        return make_ready_future<>();
+    }
+    return get_local_ranges().then([&db, this] (auto&& ranges) {
+        auto estimates = this->estimates_for_current_keyspace(db, std::move(ranges));
+        auto mutations = db::system_keyspace::make_size_estimates_mutation(*_current_partition, std::move(estimates));
+        ++_current_partition;
+        std::vector<mutation> ms;
+        ms.emplace_back(std::move(mutations));
+        _partition_reader = flat_mutation_reader_from_mutations(std::move(ms), _fwd);
+    });
+}
+
+future<> size_estimates_mutation_reader::fill_buffer(db::timeout_clock::time_point timeout) {
+    return do_until([this, timeout] { return is_end_of_stream() || is_buffer_full(); }, [this, timeout] {
+        if (!_partition_reader) {
+            return get_next_partition();
+        }
+        return _partition_reader->consume_pausable([this] (mutation_fragment mf) {
+            push_mutation_fragment(std::move(mf));
+            return stop_iteration(is_buffer_full());
+        }, timeout).then([this] {
+            if (_partition_reader->is_end_of_stream() && _partition_reader->is_buffer_empty()) {
+                _partition_reader = stdx::nullopt;
+            }
+        });
+    });
+}
+
+void size_estimates_mutation_reader::next_partition() {
+    clear_buffer_to_next_partition();
+    if (is_buffer_empty()) {
+        _partition_reader = stdx::nullopt;
+    }
+}
+
+future<> size_estimates_mutation_reader::fast_forward_to(const dht::partition_range& pr, db::timeout_clock::time_point timeout) {
+    clear_buffer();
+    _prange = &pr;
+    _keyspaces = stdx::nullopt;
+    _partition_reader = stdx::nullopt;
+    _end_of_stream = false;
+    return make_ready_future<>();
+}
+
+future<> size_estimates_mutation_reader::fast_forward_to(position_range pr, db::timeout_clock::time_point timeout) {
+    forward_buffer_to(pr.start());
+    _end_of_stream = false;
+    if (_partition_reader) {
+        return _partition_reader->fast_forward_to(std::move(pr), timeout);
+    }
+    return make_ready_future<>();
+}
+
+size_t size_estimates_mutation_reader::buffer_size() const {
+    if (_partition_reader) {
+        return flat_mutation_reader::impl::buffer_size() + _partition_reader->buffer_size();
+    }
+    return flat_mutation_reader::impl::buffer_size();
+}
+
+std::vector<db::system_keyspace::range_estimates>
+size_estimates_mutation_reader::estimates_for_current_keyspace(const database& db, std::vector<token_range> local_ranges) const {
+    // For each specified range, estimate (crudely) mean partition size and partitions count.
+    auto pkey = partition_key::from_single_value(*_schema, utf8_type->decompose(*_current_partition));
+    auto cfs = db.find_keyspace(*_current_partition).metadata()->cf_meta_data();
+    auto cf_names = boost::copy_range<std::vector<bytes>>(cfs | boost::adaptors::transformed([] (auto&& cf) {
+        return utf8_type->decompose(cf.first);
+    }));
+    boost::sort(cf_names, [] (auto&& n1, auto&& n2) {
+        return utf8_type->less(n1, n2);
+    });
+    std::vector<db::system_keyspace::range_estimates> estimates;
+    for (auto& range : _slice.row_ranges(*_schema, pkey)) {
+        auto rows = boost::make_iterator_range(
+                virtual_row_iterator(cf_names, local_ranges),
+                virtual_row_iterator(cf_names, local_ranges, virtual_row_iterator::end_iterator_tag()));
+        auto rows_to_estimate = range.slice(rows, virtual_row_comparator(_schema));
+        for (auto&& r : rows_to_estimate) {
+            auto& cf = db.find_column_family(*_current_partition, utf8_type->to_string(r.cf_name));
+            estimates.push_back(estimate(cf, r.tokens));
+            if (estimates.size() >= _slice.partition_row_limit()) {
+                return estimates;
+            }
+        }
+    }
+    return estimates;
+}
+
+} // namespace size_estimates
+
+} // namespace db

db/size_estimates_virtual_reader.hh

@@ -21,33 +21,19 @@
  * along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
  */
 
-#include <boost/range/adaptor/indirected.hpp>
-#include <boost/range/adaptor/map.hpp>
-#include <boost/range/adaptor/transformed.hpp>
-#include <boost/range/algorithm/find_if.hpp>
-
-#include "clustering_bounds_comparator.hh"
-#include "database.hh"
 #include "db/system_keyspace.hh"
-#include "dht/i_partitioner.hh"
 #include "mutation_reader.hh"
-#include "partition_range_compat.hh"
-#include "range.hh"
-#include "service/storage_service.hh"
-#include "stdx.hh"
-#include "mutation_fragment.hh"
-#include "sstables/sstables.hh"
-#include "db/timeout_clock.hh"
 
 namespace db {
 
 namespace size_estimates {
 
+struct token_range {
+    bytes start;
+    bytes end;
+};
+
 class size_estimates_mutation_reader final : public flat_mutation_reader::impl {
-    struct token_range {
-        bytes start;
-        bytes end;
-    };
     schema_ptr _schema;
     const dht::partition_range* _prange;
     const query::partition_slice& _slice;
@@ -57,267 +43,18 @@ class size_estimates_mutation_reader final : public flat_mutation_reader::impl {
     streamed_mutation::forwarding _fwd;
     flat_mutation_reader_opt _partition_reader;
 public:
-    size_estimates_mutation_reader(schema_ptr schema, const dht::partition_range& prange, const query::partition_slice& slice, streamed_mutation::forwarding fwd)
-            : impl(schema)
-            , _schema(std::move(schema))
-            , _prange(&prange)
-            , _slice(slice)
-            , _fwd(fwd)
-    { }
+    size_estimates_mutation_reader(schema_ptr, const dht::partition_range&, const query::partition_slice&, streamed_mutation::forwarding);
 
+    virtual future<> fill_buffer(db::timeout_clock::time_point) override;
+    virtual void next_partition() override;
+    virtual future<> fast_forward_to(const dht::partition_range&, db::timeout_clock::time_point) override;
+    virtual future<> fast_forward_to(position_range, db::timeout_clock::time_point) override;
+    virtual size_t buffer_size() const override;
 private:
-    future<> get_next_partition() {
-        // For each specified range, estimate (crudely) mean partition size and partitions count.
-        auto& db = service::get_local_storage_proxy().get_db().local();
-        if (!_keyspaces) {
-            _keyspaces = get_keyspaces(*_schema, db, *_prange);
-            _current_partition = _keyspaces->begin();
-        }
-        if (_current_partition == _keyspaces->end()) {
-            _end_of_stream = true;
-            return make_ready_future<>();
-        }
-        return get_local_ranges().then([&db, this] (auto&& ranges) {
-            auto estimates = this->estimates_for_current_keyspace(db, std::move(ranges));
-            auto mutations = db::system_keyspace::make_size_estimates_mutation(*_current_partition, std::move(estimates));
-            ++_current_partition;
-            std::vector<mutation> ms;
-            ms.emplace_back(std::move(mutations));
-            _partition_reader = flat_mutation_reader_from_mutations(std::move(ms), _fwd);
-        });
-    }
-public:
-    virtual future<> fill_buffer(db::timeout_clock::time_point timeout) override {
-        return do_until([this, timeout] { return is_end_of_stream() || is_buffer_full(); }, [this, timeout] {
-            if (!_partition_reader) {
-                return get_next_partition();
-            }
-            return _partition_reader->consume_pausable([this] (mutation_fragment mf) {
-                push_mutation_fragment(std::move(mf));
-                return stop_iteration(is_buffer_full());
-            }, timeout).then([this] {
-                if (_partition_reader->is_end_of_stream() && _partition_reader->is_buffer_empty()) {
-                    _partition_reader = stdx::nullopt;
-                }
-            });
-        });
-    }
-    virtual void next_partition() override {
-        clear_buffer_to_next_partition();
-        if (is_buffer_empty()) {
-            _partition_reader = stdx::nullopt;
-        }
-    }
-    virtual future<> fast_forward_to(const dht::partition_range& pr, db::timeout_clock::time_point timeout) override {
-        clear_buffer();
-        _prange = &pr;
-        _keyspaces = stdx::nullopt;
-        _partition_reader = stdx::nullopt;
-        _end_of_stream = false;
-        return make_ready_future<>();
-    }
-    virtual future<> fast_forward_to(position_range pr, db::timeout_clock::time_point timeout) override {
-        forward_buffer_to(pr.start());
-        _end_of_stream = false;
-        if (_partition_reader) {
-            return _partition_reader->fast_forward_to(std::move(pr), timeout);
-        }
-        return make_ready_future<>();
-    }
-    virtual size_t buffer_size() const override {
-        if (_partition_reader) {
-            return flat_mutation_reader::impl::buffer_size() + _partition_reader->buffer_size();
-        }
-        return flat_mutation_reader::impl::buffer_size();
-    }
-    /**
-     * Returns the primary ranges for the local node.
-     * Used for testing as well.
-     */
-    static future<std::vector<token_range>> get_local_ranges() {
-        auto& ss = service::get_local_storage_service();
-        return ss.get_local_tokens().then([&ss] (auto&& tokens) {
-            auto ranges = ss.get_token_metadata().get_primary_ranges_for(std::move(tokens));
-            std::vector<token_range> local_ranges;
-            auto to_bytes = [](const stdx::optional<dht::token_range::bound>& b) {
-                assert(b);
-                return utf8_type->decompose(dht::global_partitioner().to_sstring(b->value()));
-            };
-            // We merge the ranges to be compatible with how Cassandra shows it's size estimates table.
-            // All queries will be on that table, where all entries are text and there's no notion of
-            // token ranges form the CQL point of view.
-            auto left_inf = boost::find_if(ranges, [] (auto&& r) {
-                return !r.start() || r.start()->value() == dht::minimum_token();
-            });
-            auto right_inf = boost::find_if(ranges, [] (auto&& r) {
-                return !r.end() || r.start()->value() == dht::maximum_token();
-            });
-            if (left_inf != right_inf && left_inf != ranges.end() && right_inf != ranges.end()) {
-                local_ranges.push_back(token_range{to_bytes(right_inf->start()), to_bytes(left_inf->end())});
-                ranges.erase(left_inf);
-                ranges.erase(right_inf);
-            }
-            for (auto&& r : ranges) {
-                local_ranges.push_back(token_range{to_bytes(r.start()), to_bytes(r.end())});
-            }
-            boost::sort(local_ranges, [] (auto&& tr1, auto&& tr2) {
-                return utf8_type->less(tr1.start, tr2.start);
-            });
-            return local_ranges;
-        });
-    }
-private:
-    struct virtual_row {
-        const bytes& cf_name;
-        const token_range& tokens;
-        clustering_key_prefix as_key() const {
-            return clustering_key_prefix::from_exploded(std::vector<bytes_view>{cf_name, tokens.start, tokens.end});
-        }
-    };
-    struct virtual_row_comparator {
-        schema_ptr _schema;
-        virtual_row_comparator(schema_ptr schema) : _schema(schema) { }
-        bool operator()(const clustering_key_prefix& key1, const clustering_key_prefix& key2) {
-            return clustering_key_prefix::prefix_equality_less_compare(*_schema)(key1, key2);
-        }
-        bool operator()(const virtual_row& row, const clustering_key_prefix& key) {
-            return operator()(row.as_key(), key);
-        }
-        bool operator()(const clustering_key_prefix& key, const virtual_row& row) {
-            return operator()(key, row.as_key());
-        }
-    };
-    class virtual_row_iterator : public std::iterator<std::input_iterator_tag, const virtual_row> {
-        std::reference_wrapper<const std::vector<bytes>> _cf_names;
-        std::reference_wrapper<const std::vector<token_range>> _ranges;
-        size_t _cf_names_idx = 0;
-        size_t _ranges_idx = 0;
-    public:
-        struct end_iterator_tag {};
-        virtual_row_iterator(const std::vector<bytes>& cf_names, const std::vector<token_range>& ranges)
-                : _cf_names(std::ref(cf_names))
-                , _ranges(std::ref(ranges))
-        { }
-        virtual_row_iterator(const std::vector<bytes>& cf_names, const std::vector<token_range>& ranges, end_iterator_tag)
-                : _cf_names(std::ref(cf_names))
-                , _ranges(std::ref(ranges))
-                , _cf_names_idx(cf_names.size())
-                , _ranges_idx(ranges.size())
-        { }
-        virtual_row_iterator& operator++() {
-            if (++_ranges_idx == _ranges.get().size() && ++_cf_names_idx < _cf_names.get().size()) {
-                _ranges_idx = 0;
-            }
-            return *this;
-        }
-        virtual_row_iterator operator++(int) {
-            virtual_row_iterator i(*this);
-            ++(*this);
-            return i;
-        }
-        const value_type operator*() const {
-            return { _cf_names.get()[_cf_names_idx], _ranges.get()[_ranges_idx] };
-        }
-        bool operator==(const virtual_row_iterator& i) const {
-            return _cf_names_idx == i._cf_names_idx
-                && _ranges_idx == i._ranges_idx;
-        }
-        bool operator!=(const virtual_row_iterator& i) const {
-            return !(*this == i);
-        }
-    };
+    future<> get_next_partition();
 
     std::vector<db::system_keyspace::range_estimates>
-    estimates_for_current_keyspace(const database& db, std::vector<token_range> local_ranges) const {
-        auto pkey = partition_key::from_single_value(*_schema, utf8_type->decompose(*_current_partition));
-        auto cfs = db.find_keyspace(*_current_partition).metadata()->cf_meta_data();
-        auto cf_names = boost::copy_range<std::vector<bytes>>(cfs | boost::adaptors::transformed([] (auto&& cf) {
-            return utf8_type->decompose(cf.first);
-        }));
-        boost::sort(cf_names, [] (auto&& n1, auto&& n2) {
-            return utf8_type->less(n1, n2);
-        });
-        std::vector<db::system_keyspace::range_estimates> estimates;
-        for (auto& range : _slice.row_ranges(*_schema, pkey)) {
-            auto rows = boost::make_iterator_range(
-                    virtual_row_iterator(cf_names, local_ranges),
-                    virtual_row_iterator(cf_names, local_ranges, virtual_row_iterator::end_iterator_tag()));
-            auto rows_to_estimate = range.slice(rows, virtual_row_comparator(_schema));
-            for (auto&& r : rows_to_estimate) {
-                auto& cf = db.find_column_family(*_current_partition, utf8_type->to_string(r.cf_name));
-                estimates.push_back(estimate(cf, r.tokens));
-                if (estimates.size() >= _slice.partition_row_limit()) {
-                    return estimates;
-                }
-            }
-        }
-        return estimates;
-    }
-
-    /**
-     * Returns the keyspaces, ordered by name, as selected by the partition_range.
-     */
-    static ks_range get_keyspaces(const schema& s, const database& db, dht::partition_range range) {
-        struct keyspace_less_comparator {
-            const schema& _s;
-            keyspace_less_comparator(const schema& s) : _s(s) { }
-            dht::ring_position as_ring_position(const sstring& ks) {
-                auto pkey = partition_key::from_single_value(_s, utf8_type->decompose(ks));
-                return dht::global_partitioner().decorate_key(_s, std::move(pkey));
-            }
-            bool operator()(const sstring& ks1, const sstring& ks2) {
-                return as_ring_position(ks1).less_compare(_s, as_ring_position(ks2));
-            }
-            bool operator()(const sstring& ks, const dht::ring_position& rp) {
-                return as_ring_position(ks).less_compare(_s, rp);
-            }
-            bool operator()(const dht::ring_position& rp, const sstring& ks) {
-                return rp.less_compare(_s, as_ring_position(ks));
-            }
-        };
-        auto keyspaces = db.get_non_system_keyspaces();
-        auto cmp = keyspace_less_comparator(s);
-        boost::sort(keyspaces, cmp);
-        return boost::copy_range<ks_range>(range.slice(keyspaces, std::move(cmp)));
-    }
-
-    /**
-     * Makes a wrapping range of ring_position from a nonwrapping range of token, used to select sstables.
-     */
-    static dht::partition_range as_ring_position_range(dht::token_range& r) {
-        stdx::optional<range<dht::ring_position>::bound> start_bound, end_bound;
-        if (r.start()) {
-            start_bound = {{ dht::ring_position(r.start()->value(), dht::ring_position::token_bound::start), r.start()->is_inclusive() }};
-        }
-        if (r.end()) {
-            end_bound = {{ dht::ring_position(r.end()->value(), dht::ring_position::token_bound::end), r.end()->is_inclusive() }};
-        }
-        return dht::partition_range(std::move(start_bound), std::move(end_bound), r.is_singular());
-    }
-
-    /**
-     * Add a new range_estimates for the specified range, considering the sstables associated with `cf`.
-     */
-    static system_keyspace::range_estimates estimate(const column_family& cf, const token_range& r) {
-        int64_t count{0};
-        utils::estimated_histogram hist{0};
-        auto from_bytes = [] (auto& b) {
-            return dht::global_partitioner().from_sstring(utf8_type->to_string(b));
-        };
-        dht::token_range_vector ranges;
-        compat::unwrap_into(
-            wrapping_range<dht::token>({{ from_bytes(r.start) }}, {{ from_bytes(r.end) }}),
-            dht::token_comparator(),
-            [&] (auto&& rng) { ranges.push_back(std::move(rng)); });
-        for (auto&& r : ranges) {
-            auto rp_range = as_ring_position_range(r);
-            for (auto&& sstable : cf.select_sstables(rp_range)) {
-                count += sstable->estimated_keys_for_range(r);
-                hist.merge(sstable->get_stats_metadata().estimated_row_size);
-            }
-        }
-        return {cf.schema(), r.start, r.end, count, count > 0 ? hist.mean() : 0};
-    }
+    estimates_for_current_keyspace(const database&, std::vector<token_range> local_ranges) const;
 };
 
 struct virtual_reader {
@@ -332,6 +69,12 @@ struct virtual_reader {
     }
 };
 
+/**
+ * Returns the primary ranges for the local node.
+ * Used for testing as well.
+ */
+future<std::vector<token_range>> get_local_ranges();
+
 } // namespace size_estimates
 
 } // namespace db

db/system_distributed_keyspace.cc

@@ -26,6 +26,7 @@
 #include "db/consistency_level_type.hh"
 #include "db/system_keyspace.hh"
 #include "schema_builder.hh"
+#include "timeout_config.hh"
 #include "types.hh"
 
 #include <seastar/core/reactor.hh>
@@ -97,11 +98,17 @@ future<> system_distributed_keyspace::stop() {
     return make_ready_future<>();
 }
 
+static const timeout_config internal_distributed_timeout_config = [] {
+    using namespace std::chrono_literals;
+    const auto t = 10s;
+    return timeout_config{ t, t, t, t, t, t, t };
+}();
+
 future<std::unordered_map<utils::UUID, sstring>> system_distributed_keyspace::view_status(sstring ks_name, sstring view_name) const {
     return _qp.process(
             sprint("SELECT host_id, status FROM %s.%s WHERE keyspace_name = ? AND view_name = ?", NAME, VIEW_BUILD_STATUS),
             db::consistency_level::ONE,
-            infinite_timeout_config,
+            internal_distributed_timeout_config,
             { std::move(ks_name), std::move(view_name) },
             false).then([this] (::shared_ptr<cql3::untyped_result_set> cql_result) {
         return boost::copy_range<std::unordered_map<utils::UUID, sstring>>(*cql_result
@@ -118,7 +125,7 @@ future<> system_distributed_keyspace::start_view_build(sstring ks_name, sstring
         return _qp.process(
                 sprint("INSERT INTO %s.%s (keyspace_name, view_name, host_id, status) VALUES (?, ?, ?, ?)", NAME, VIEW_BUILD_STATUS),
                 db::consistency_level::ONE,
-                infinite_timeout_config,
+                internal_distributed_timeout_config,
                 { std::move(ks_name), std::move(view_name), std::move(host_id), "STARTED" },
                 false).discard_result();
     });
@@ -129,7 +136,7 @@ future<> system_distributed_keyspace::finish_view_build(sstring ks_name, sstring
         return _qp.process(
                 sprint("UPDATE %s.%s SET status = ? WHERE keyspace_name = ? AND view_name = ? AND host_id = ?", NAME, VIEW_BUILD_STATUS),
                 db::consistency_level::ONE,
-                infinite_timeout_config,
+                internal_distributed_timeout_config,
                 { "SUCCESS", std::move(ks_name), std::move(view_name), std::move(host_id) },
                 false).discard_result();
     });
@@ -139,7 +146,7 @@ future<> system_distributed_keyspace::remove_view(sstring ks_name, sstring view_
     return _qp.process(
             sprint("DELETE FROM %s.%s WHERE keyspace_name = ? AND view_name = ?", NAME, VIEW_BUILD_STATUS),
             db::consistency_level::ONE,
-            infinite_timeout_config,
+            internal_distributed_timeout_config,
             { std::move(ks_name), std::move(view_name) },
             false).discard_result();
 }

db/system_keyspace.cc

@@ -1635,6 +1635,9 @@ void make(database& db, bool durable, bool volatile_testing_only) {
         auto cfg = ks.make_column_family_config(*table, db.get_config(), db.get_large_partition_handler());
         if (maybe_write_in_user_memory(table, db)) {
             cfg.dirty_memory_manager = &db._dirty_memory_manager;
+        } else {
+            cfg.memtable_scheduling_group = default_scheduling_group();
+            cfg.memtable_to_cache_scheduling_group = default_scheduling_group();
         }
         db.add_column_family(ks, table, std::move(cfg));
         maybe_add_virtual_reader(table, db);

dht/i_partitioner.hh

@@ -384,6 +384,10 @@ public:
         return "biased-token-round-robin";
     }
 
+    virtual unsigned sharding_ignore_msb() const {
+        return 0;
+    }
+
     friend bool operator==(token_view t1, token_view t2);
     friend bool operator<(token_view t1, token_view t2);
     friend int tri_compare(token_view t1, token_view t2);

dht/murmur3_partitioner.cc

@@ -290,6 +290,11 @@ murmur3_partitioner::token_for_next_shard(const token& t, shard_id shard, unsign
     return bias(n);
 }
 
+unsigned
+murmur3_partitioner::sharding_ignore_msb() const {
+    return _sharding_ignore_msb_bits;
+}
+
 
 using registry = class_registrator<i_partitioner, murmur3_partitioner, const unsigned&, const unsigned&>;
 static registry registrator("org.apache.cassandra.dht.Murmur3Partitioner");

dht/murmur3_partitioner.hh

@@ -52,6 +52,7 @@ public:
 
     virtual unsigned shard_of(const token& t) const override;
     virtual token token_for_next_shard(const token& t, shard_id shard, unsigned spans) const override;
+    virtual unsigned sharding_ignore_msb() const override;
 private:
     using uint128_t = unsigned __int128;
     static int64_t normalize(int64_t in);

dist/ami/build_ami.sh

@@ -11,11 +11,9 @@ print_usage() {
     echo "  --repo  repository for both install and update, specify .repo/.list file URL"
     echo "  --repo-for-install  repository for install, specify .repo/.list file URL"
     echo "  --repo-for-update  repository for update, specify .repo/.list file URL"
-    echo "  --target    specify target distribution"
     exit 1
 }
 LOCALRPM=0
-TARGET=centos
 while [ $# -gt 0 ]; do
     case "$1" in
         "--localrpm")
@@ -34,10 +32,6 @@ while [ $# -gt 0 ]; do
             INSTALL_ARGS="$INSTALL_ARGS --repo-for-update $2"
             shift 2
             ;;
-        "--target")
-            TARGET="$2"
-            shift 2
-            ;;
         *)
             print_usage
             ;;
@@ -62,91 +56,42 @@ pkg_install() {
     fi
 }
 
-case "$TARGET" in
-    "centos")
-        AMI=ami-ae7bfdb8
-        REGION=us-east-1
-        SSH_USERNAME=centos
-        ;;
-    "trusty")
-        AMI=ami-ff427095
-        REGION=us-east-1
-        SSH_USERNAME=ubuntu
-        ;;
-    "xenial")
-        AMI=ami-da05a4a0
-        REGION=us-east-1
-        SSH_USERNAME=ubuntu
-        ;;
-    *)
-        echo "build_ami.sh does not supported this distribution."
-        exit 1
-        ;;
-esac
+AMI=ami-ae7bfdb8
+REGION=us-east-1
+SSH_USERNAME=centos
 
 if [ $LOCALRPM -eq 1 ]; then
     sudo rm -rf build/*
-    REPO=`./scripts/scylla_current_repo --target $TARGET`
+    REPO=`./scripts/scylla_current_repo --target centos`
     INSTALL_ARGS="$INSTALL_ARGS --localrpm --repo $REPO"
     if [ ! -f /usr/bin/git ]; then
         pkg_install git
     fi
 
-    if [ "$TARGET" = "centos" ]; then
-        if [ ! -f dist/ami/files/scylla.x86_64.rpm ] || [ ! -f dist/ami/files/scylla-kernel-conf.x86_64.rpm ] || [ ! -f dist/ami/files/scylla-conf.x86_64.rpm ] || [ ! -f dist/ami/files/scylla-server.x86_64.rpm ] || [ ! -f dist/ami/files/scylla-debuginfo.x86_64.rpm ]; then
-            dist/redhat/build_rpm.sh --dist --target epel-7-x86_64
-            cp build/rpms/scylla-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla.x86_64.rpm
-            cp build/rpms/scylla-kernel-conf-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla-kernel-conf.x86_64.rpm
-            cp build/rpms/scylla-conf-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla-conf.x86_64.rpm
-            cp build/rpms/scylla-server-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla-server.x86_64.rpm
-            cp build/rpms/scylla-debuginfo-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla-debuginfo.x86_64.rpm
-        fi
-        if [ ! -f dist/ami/files/scylla-jmx.noarch.rpm ]; then
-            cd build
-            git clone --depth 1 https://github.com/scylladb/scylla-jmx.git
-            cd scylla-jmx
-            dist/redhat/build_rpm.sh --target epel-7-x86_64
-            cd ../..
-            cp build/scylla-jmx/build/rpms/scylla-jmx-`cat build/scylla-jmx/build/SCYLLA-VERSION-FILE`-`cat build/scylla-jmx/build/SCYLLA-RELEASE-FILE`.*.noarch.rpm dist/ami/files/scylla-jmx.noarch.rpm
-        fi
-        if [ ! -f dist/ami/files/scylla-tools.noarch.rpm ] || [ ! -f dist/ami/files/scylla-tools-core.noarch.rpm ]; then
-            cd build
-            git clone --depth 1 https://github.com/scylladb/scylla-tools-java.git
-            cd scylla-tools-java
-            dist/redhat/build_rpm.sh --target epel-7-x86_64
-            cd ../..
-            cp build/scylla-tools-java/build/rpms/scylla-tools-`cat build/scylla-tools-java/build/SCYLLA-VERSION-FILE`-`cat build/scylla-tools-java/build/SCYLLA-RELEASE-FILE`.*.noarch.rpm dist/ami/files/scylla-tools.noarch.rpm
-            cp build/scylla-tools-java/build/rpms/scylla-tools-core-`cat build/scylla-tools-java/build/SCYLLA-VERSION-FILE`-`cat build/scylla-tools-java/build/SCYLLA-RELEASE-FILE`.*.noarch.rpm dist/ami/files/scylla-tools-core.noarch.rpm
-        fi
-    else
-        if [ ! -f dist/ami/files/scylla-server_amd64.deb ]; then
-            ./scripts/git-archive-all --force-submodules --prefix scylla build/scylla.tar
-            tar -C build/ -xvpf build/scylla.tar
-            cd build/scylla
-            dist/debian/build_deb.sh --dist --target $TARGET
-            cd ../..
-            cp build/scylla/build/debs/scylla_`cat build/SCYLLA-VERSION-FILE | sed 's/\.rc/~rc/'`-`cat build/SCYLLA-RELEASE-FILE`-0ubuntu1~${TARGET}_amd64.deb dist/ami/files/scylla_amd64.deb
-            cp build/scylla/build/debs/scylla-kernel-conf_`cat build/SCYLLA-VERSION-FILE | sed 's/\.rc/~rc/'`-`cat build/SCYLLA-RELEASE-FILE`-0ubuntu1~${TARGET}_amd64.deb dist/ami/files/scylla-kernel-conf_amd64.deb
-            cp build/scylla/build/debs/scylla-conf_`cat build/SCYLLA-VERSION-FILE | sed 's/\.rc/~rc/'`-`cat build/SCYLLA-RELEASE-FILE`-0ubuntu1~${TARGET}_amd64.deb dist/ami/files/scylla-conf_amd64.deb
-            cp build/scylla/build/debs/scylla-server_`cat build/SCYLLA-VERSION-FILE | sed 's/\.rc/~rc/'`-`cat build/SCYLLA-RELEASE-FILE`-0ubuntu1~${TARGET}_amd64.deb dist/ami/files/scylla-server_amd64.deb
-            cp build/scylla/build/debs/scylla-server-dbg_`cat build/SCYLLA-VERSION-FILE | sed 's/\.rc/~rc/'`-`cat build/SCYLLA-RELEASE-FILE`-0ubuntu1~${TARGET}_amd64.deb dist/ami/files/scylla-server-dbg_amd64.deb
-        fi
-        if [ ! -f dist/ami/files/scylla-jmx_all.deb ]; then
-            cd build
-            git clone --depth 1 https://github.com/scylladb/scylla-jmx.git
-            cd scylla-jmx
-            dist/debian/build_deb.sh --target $TARGET
-            cd ../..
-            cp build/scylla-jmx/build/debs/scylla-jmx_`cat build/scylla-jmx/build/SCYLLA-VERSION-FILE | sed 's/\.rc/~rc/'`-`cat build/scylla-jmx/build/SCYLLA-RELEASE-FILE`-0ubuntu1~${TARGET}_all.deb dist/ami/files/scylla-jmx_all.deb
-        fi
-        if [ ! -f dist/ami/files/scylla-tools_all.deb ]; then
-            cd build
-            git clone --depth 1 https://github.com/scylladb/scylla-tools-java.git
-            cd scylla-tools-java
-            dist/debian/build_deb.sh --target $TARGET
-            cd ../..
-            cp build/scylla-tools-java/build/debs/scylla-tools_`cat build/scylla-tools-java/build/SCYLLA-VERSION-FILE | sed 's/\.rc/~rc/'`-`cat build/scylla-tools-java/build/SCYLLA-RELEASE-FILE`-0ubuntu1~${TARGET}_all.deb dist/ami/files/scylla-tools_all.deb
-        fi
+    if [ ! -f dist/ami/files/scylla.x86_64.rpm ] || [ ! -f dist/ami/files/scylla-kernel-conf.x86_64.rpm ] || [ ! -f dist/ami/files/scylla-conf.x86_64.rpm ] || [ ! -f dist/ami/files/scylla-server.x86_64.rpm ] || [ ! -f dist/ami/files/scylla-debuginfo.x86_64.rpm ]; then
+        dist/redhat/build_rpm.sh --dist --target epel-7-x86_64
+        cp build/rpms/scylla-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla.x86_64.rpm
+        cp build/rpms/scylla-kernel-conf-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla-kernel-conf.x86_64.rpm
+        cp build/rpms/scylla-conf-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla-conf.x86_64.rpm
+        cp build/rpms/scylla-server-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla-server.x86_64.rpm
+        cp build/rpms/scylla-debuginfo-`cat build/SCYLLA-VERSION-FILE`-`cat build/SCYLLA-RELEASE-FILE`.*.x86_64.rpm dist/ami/files/scylla-debuginfo.x86_64.rpm
+    fi
+    if [ ! -f dist/ami/files/scylla-jmx.noarch.rpm ]; then
+        cd build
+        git clone --depth 1 https://github.com/scylladb/scylla-jmx.git
+        cd scylla-jmx
+        dist/redhat/build_rpm.sh --target epel-7-x86_64
+        cd ../..
+        cp build/scylla-jmx/build/rpms/scylla-jmx-`cat build/scylla-jmx/build/SCYLLA-VERSION-FILE`-`cat build/scylla-jmx/build/SCYLLA-RELEASE-FILE`.*.noarch.rpm dist/ami/files/scylla-jmx.noarch.rpm
+    fi
+    if [ ! -f dist/ami/files/scylla-tools.noarch.rpm ] || [ ! -f dist/ami/files/scylla-tools-core.noarch.rpm ]; then
+        cd build
+        git clone --depth 1 https://github.com/scylladb/scylla-tools-java.git
+        cd scylla-tools-java
+        dist/redhat/build_rpm.sh --target epel-7-x86_64
+        cd ../..
+        cp build/scylla-tools-java/build/rpms/scylla-tools-`cat build/scylla-tools-java/build/SCYLLA-VERSION-FILE`-`cat build/scylla-tools-java/build/SCYLLA-RELEASE-FILE`.*.noarch.rpm dist/ami/files/scylla-tools.noarch.rpm
+        cp build/scylla-tools-java/build/rpms/scylla-tools-core-`cat build/scylla-tools-java/build/SCYLLA-VERSION-FILE`-`cat build/scylla-tools-java/build/SCYLLA-RELEASE-FILE`.*.noarch.rpm dist/ami/files/scylla-tools-core.noarch.rpm
     fi
 fi
 

dist/ami/files/.bash_profile

@@ -7,121 +7,8 @@ fi
 
 # User specific environment and startup programs
 
-. /usr/lib/scylla/scylla_lib.sh
-
 PATH=$PATH:$HOME/.local/bin:$HOME/bin
 
 export PATH
 
-echo
-echo '   _____            _ _       _____  ____  '
-echo '  / ____|          | | |     |  __ \|  _ \ '
-echo ' | (___   ___ _   _| | | __ _| |  | | |_) |'
-echo '  \___ \ / __| | | | | |/ _` | |  | |  _ < '
-echo '  ____) | (__| |_| | | | (_| | |__| | |_) |'
-echo ' |_____/ \___|\__, |_|_|\__,_|_____/|____/ '
-echo '               __/ |                       '
-echo '              |___/                        '
-echo ''
-echo ''
-echo 'Nodetool:'
-echo '	nodetool help'
-echo 'CQL Shell:'
-echo '	cqlsh'
-echo 'More documentation available at: '
-echo '	http://www.scylladb.com/doc/'
-echo 'By default, Scylla sends certain information about this node to a data collection server. For information, see http://www.scylladb.com/privacy/'
-echo
-
-if [ `ec2_is_supported_instance_type` -eq 0 ]; then
-	TYPE=`curl -s http://169.254.169.254/latest/meta-data/instance-type`
-	tput setaf 1
-	tput bold
-	echo "    $TYPE is not supported instance type!"
-	tput sgr0
-	echo -n "To continue startup ScyllaDB on this instance, run 'sudo scylla_io_setup' "
-	if ! is_systemd; then
-		echo "then 'initctl start scylla-server'."
-	else
-		echo "then 'systemctl start scylla-server'."
-	fi
-	echo "For a list of optimized instance types and more EC2 instructions see http://www.scylladb.com/doc/getting-started-amazon/"
-	echo
-else
-	SETUP=
-	if is_systemd; then
-		SETUP=`systemctl is-active scylla-ami-setup`
-	fi
-	if [ "$SETUP" == "activating" ]; then
-		tput setaf 4
-		tput bold
-		echo "    Constructing RAID volume..."
-		tput sgr0
-		echo
-		echo "Please wait for setup. To see status, run "
-		echo " 'systemctl status scylla-ami-setup'"
-		echo
-		echo "After setup finished, scylla-server service will launch."
-		echo "To see status of scylla-server, run "
-		echo " 'systemctl status scylla-server'"
-		echo
-	elif [ "$SETUP" == "failed" ]; then
-		tput setaf 1
-		tput bold
-		echo "    AMI initial configuration failed!"
-		tput sgr0
-		echo
-		echo "To see status, run "
-		echo " 'systemctl status scylla-ami-setup'"
-		echo
-	else
-		if is_systemd; then
-			SCYLLA=`systemctl is-active scylla-server`
-		else
-			if [ "`initctl status scylla-server|grep "running, process"`" != "" ]; then
-				SCYLLA="active"
-			else
-				SCYLLA="failed"
-			fi
-		fi
-		if [ "$SCYLLA" == "activating" ]; then
-			tput setaf 4
-			tput bold
-			echo "    ScyllaDB is starting..."
-			tput sgr0
-			echo
-			echo "Please wait for start. To see status, run "
-			echo " 'systemctl status scylla-server'"
-			echo
-		elif [ "$SCYLLA" == "active" ]; then
-			tput setaf 4
-			tput bold
-			echo "    ScyllaDB is active."
-			tput sgr0
-			echo
-			echo "$ nodetool status"
-			echo
-			nodetool status
-		else
-			tput setaf 1
-			tput bold
-			echo "    ScyllaDB is not started!"
-			tput sgr0
-			echo "Please wait for startup. To see status of ScyllaDB, run "
-			if ! is_systemd; then
-				echo " 'initctl status scylla-server'"
-				echo "and"
-				echo " 'sudo cat /var/log/upstart/scylla-server.log'"
-				echo
-			else
-				echo " 'systemctl status scylla-server'"
-				echo
-			fi
-		fi
-	fi
-	echo -n "    "
-	/usr/lib/scylla/scylla_ec2_check
-	if [ $? -eq 0 ]; then
-	    echo
-	fi
-fi
+~/.scylla_ami_login

dist/ami/files/.scylla_ami_login

@@ -0,0 +1,118 @@
+#!/usr/bin/python3
+#
+# Copyright 2018 ScyllaDB
+#
+
+#
+# This file is part of Scylla.
+#
+# Scylla is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# Scylla is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
+
+import os
+import sys
+import argparse
+sys.path.append('/usr/lib/scylla')
+from scylla_util import *
+
+MSG_HEADER = '''
+
+   _____            _ _       _____  ____  
+  / ____|          | | |     |  __ \|  _ \ 
+ | (___   ___ _   _| | | __ _| |  | | |_) |
+  \___ \ / __| | | | | |/ _` | |  | |  _ < 
+  ____) | (__| |_| | | | (_| | |__| | |_) |
+ |_____/ \___|\__, |_|_|\__,_|_____/|____/ 
+               __/ |                       
+              |___/                        
+
+
+Nodetool:
+	nodetool help
+CQL Shell:
+	cqlsh
+More documentation available at: 
+	http://www.scylladb.com/doc/
+By default, Scylla sends certain information about this node to a data collection server. For information, see http://www.scylladb.com/privacy/
+
+'''[1:-1]
+MSG_UNSUPPORTED_INSTANCE_TYPE = '''
+    {red}{type} is not supported instance type!{nocolor}
+To continue startup ScyllaDB on this instance, run 'sudo scylla_io_setup' then 'systemctl start scylla-server'.
+For a list of optimized instance types and more EC2 instructions see http://www.scylladb.com/doc/getting-started-amazon/"
+
+'''[1:-1]
+MSG_SETUP_ACTIVATING = '''
+    {green}Constructing RAID volume...{nocolor}
+
+Please wait for setup. To see status, run 
+ 'systemctl status scylla-ami-setup'
+
+After setup finished, scylla-server service will launch.
+To see status of scylla-server, run 
+ 'systemctl status scylla-server'
+
+'''[1:-1]
+MSG_SETUP_FAILED = '''
+    {red}AMI initial configuration failed!{nocolor}
+
+To see status, run 
+ 'systemctl status scylla-ami-setup'
+
+'''[1:-1]
+MSG_SCYLLA_ACTIVATING = '''
+    {green}ScyllaDB is starting...{nocolor}
+
+Please wait for start. To see status, run 
+ 'systemctl status scylla-server'
+
+'''[1:-1]
+MSG_SCYLLA_FAILED = '''
+    {red}ScyllaDB is not started!{nocolor}
+Please wait for startup. To see status of ScyllaDB, run 
+ 'systemctl status scylla-server'
+
+'''[1:-1]
+MSG_SCYLLA_ACTIVE = '''
+    {green}ScyllaDB is active.{nocolor}
+
+$ nodetool status
+
+'''[1:-1]
+
+if __name__ == '__main__':
+    colorprint(MSG_HEADER)
+    aws = aws_instance()
+    if not aws.is_supported_instance_class():
+        colorprint(MSG_UNSUPPORTED_INSTANCE_TYPE, type=aws.instance_class())
+    else:
+        setup = systemd_unit('scylla-ami-setup.service')
+        res = setup.is_active()
+        if res == 'activating':
+            colorprint(MSG_SETUP_ACTIVATING)
+        elif res == 'failed':
+            colorprint(MSG_SETUP_FAILED)
+        else:
+            server = systemd_unit('scylla-server.service')
+            res = server.is_active()
+            if res == 'activating':
+                colorprint(MSG_SCYLLA_ACTIVATING)
+            elif res == 'failed':
+                colorprint(MSG_SCYLLA_FAILED)
+            else:
+                colorprint(MSG_SCYLLA_ACTIVE)
+                run('nodetool status', exception=False)
+        print('    ', end='')
+        res = run('/usr/lib/scylla/scylla_ec2_check --nic eth0', exception=False)
+        if res == 0:
+            print('')

dist/ami/scylla.json

@@ -64,14 +64,11 @@
       "source": "files/",
       "destination": "/home/{{user `ssh_username`}}/"
     },
-    {
-      "type": "file",
-      "source": "../../scripts/scylla_install_pkg",
-      "destination": "/home/{{user `ssh_username`}}/scylla_install_pkg"
-    },
     {
       "type": "shell",
       "inline": [
+         "sudo yum install -y epel-release",
+         "sudo yum install -y python36",
          "sudo /home/{{user `ssh_username`}}/scylla-ami/scylla_install_ami {{ user `install_args` }}"
        ]
     }

dist/common/scripts/node_exporter_install

@@ -1,6 +1,8 @@
-#!/bin/sh
+#!/usr/bin/python3
 #
-# Copyright 2016 ScyllaDB
+# Copyright 2018 ScyllaDB
+#
+
 #
 # This file is part of Scylla.
 #
@@ -17,42 +19,46 @@
 # You should have received a copy of the GNU General Public License
 # along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
 
-if [ "`id -u`" -ne 0 ]; then
-    echo "Requires root permission."
-    exit 1
-fi
+import os
+import sys
+import tempfile
+import tarfile
+from scylla_util import *
 
-if [ -f /usr/bin/node_exporter ] || [ -f /usr/bin/prometheus-node_exporter ]; then
-    echo "node_exporter already installed"
-    exit 1
-fi
+VERSION='0.14.0'
+INSTALL_DIR='/usr/lib/scylla/Prometheus/node_exporter'
 
-. /usr/lib/scylla/scylla_lib.sh
+if __name__ == '__main__':
+    if os.getuid() > 0:
+        print('Requires root permission.')
+        sys.exit(1)
 
-if is_gentoo_variant; then
-    emerge -uq app-metrics/node_exporter
-    if is_systemd; then
-       echo "app-metrics/node_exporter does not install systemd service files, please fill a bug if you need them."
-    else
-        rc-update add node_exporter default
-        rc-service node_exporter start
-    fi
-else
-    version=0.14.0
-    dir=/usr/lib/scylla/Prometheus/node_exporter
-    mkdir -p $dir
-    cd $dir
-    curl -L https://github.com/prometheus/node_exporter/releases/download/v$version/node_exporter-$version.linux-amd64.tar.gz -o $dir/node_exporter-$version.linux-amd64.tar.gz
-    tar -xvzf $dir/node_exporter-$version.linux-amd64.tar.gz
-    rm $dir/node_exporter-$version.linux-amd64.tar.gz
-    ln -s $dir/node_exporter-$version.linux-amd64/node_exporter /usr/bin
-    . /etc/os-release
+    if os.path.exists('/usr/bin/node_exporter') or os.path.exists('/usr/bin/prometheus-node_exporter'):
+        print('node_exporter already installed')
+        sys.exit(1)
 
-     if is_systemd; then
-        systemctl enable node-exporter
-        systemctl start node-exporter
-    else
-        cat <<EOT >> /etc/init/node_exporter.conf
+    if is_gentoo_variant():
+        run('emerge -uq app-metrics/node_exporter')
+        if is_systemd():
+            print('app-metrics/node_exporter does not install systemd service files, please fill a bug if you need them.')
+            sys.exit(1)
+        else:
+            run('rc-update add node_exporter default')
+            run('rc-service node_exporter start')
+    else:
+        data = curl('https://github.com/prometheus/node_exporter/releases/download/v{version}/node_exporter-{version}.linux-amd64.tar.gz'.format(version=VERSION), byte=True)
+        with open('/var/tmp/node_exporter-{version}.linux-amd64.tar.gz'.format(version=VERSION), 'wb') as f:
+            f.write(data)
+        with tarfile.open('/var/tmp/node_exporter-{version}.linux-amd64.tar.gz'.format(version=VERSION)) as tf:
+            tf.extractall(INSTALL_DIR)
+        os.remove('/var/tmp/node_exporter-{version}.linux-amd64.tar.gz'.format(version=VERSION))
+        os.symlink('{install_dir}/node_exporter-{version}.linux-amd64/node_exporter'.format(install_dir=INSTALL_DIR, version=VERSION), '/usr/bin/node_exporter')
+        if is_systemd():
+            node_exporter = systemd_unit('node-exporter.service')
+            node_exporter.enable()
+            node_exporter.start()
+        else:
+            conf = '''
 # Run node_exporter
 
 start on startup
@@ -60,9 +66,9 @@ start on startup
 script
    /usr/bin/node_exporter
 end script
-EOT
-        service node_exporter start
-    fi
-fi
+'''[1:-1]
+            with open('/etc/init/node_exporter.conf', 'w') as f:
+                f.write(conf)
+            run('service node_exporter start')
 
-printf "node_exporter successfully installed\n"
+    print('node_exporter successfully installed')

dist/common/scripts/node_health_check

@@ -28,6 +28,7 @@ OUTPUT_PATH4="$OUTPUT_PATH/data_model"
 OUTPUT_PATH5="$OUTPUT_PATH/network_checks"
 IS_FEDORA="0"
 IS_DEBIAN="0"
+IS_GENTOO="0"
 JMX_PORT="7199"
 CQL_PORT="9042"
 PRINT_DM=NO
@@ -75,7 +76,7 @@ while getopts ":hdncap:q:" opt; do
 done
 
 
-##Check server release (Fedora/Oracle/Debian)##
+##Check server release (Fedora/Oracle/Debian/Gentoo)##
 cat /etc/os-release | grep -i fedora &> /dev/null
 if [ $? -ne 0 ]; then
     cat /etc/os-release | grep -i oracle &> /dev/null
@@ -89,7 +90,12 @@ if [ $? -ne 0 ]; then
     IS_DEBIAN="1"
 fi
 
-if [ "$IS_FEDORA" == "1" ] && [ "$IS_DEBIAN" == "1" ]; then
+cat /etc/os-release | grep -i gentoo &> /dev/null
+if [ $? -ne 0 ]; then
+    IS_GENTOO="1"
+fi
+
+if [ "$IS_FEDORA" == "1" ] && [ "$IS_DEBIAN" == "1" ] && [ "$IS_GENTOO" == "1" ]; then
     echo "This s a Non-Supported OS, Please Review the Support Matrix"
     exit 222
 fi
@@ -108,7 +114,7 @@ if [ $? -ne 0 ]; then
 else
     echo "Scylla-server Service: OK"
     echo "--------------------------------------------------"
-fi 
+fi
 
 
 ##Scylla-JMX service status##
@@ -125,7 +131,7 @@ if [ $? -ne 0 ]; then
 else
     echo "Scylla-JMX Service (nodetool): OK"
     echo "--------------------------------------------------"
-fi 
+fi
 
 
 #Install 'net-tools' pkg, to be used for netstat command#
@@ -141,6 +147,9 @@ if [ "$IS_DEBIAN" == "0" ]; then
     sudo apt-get install net-tools -y | grep already
 fi
 
+if [ "$IS_GENTOO" == "0" ]; then
+    sudo emerge -1uq sys-apps/ethtool sys-apps/net-tools
+fi
 
 #Create dir structure to save output_files#
 echo "--------------------------------------------------"
@@ -182,6 +191,12 @@ if [ "$IS_DEBIAN" == "0" ]; then
     cp -p /etc/default/scylla-server $OUTPUT_PATH2
 fi
 
+if [ "$IS_GENTOO" == "0" ]; then
+    sudo emerge -1uq app-portage/portage-utils
+    sudo qlist -ICv scylla > $OUTPUT_PATH2/scylla-pkgs.txt
+    cp -p /etc/default/scylla-server $OUTPUT_PATH2
+fi
+
 
 #Scylla Logs#
 echo "--------------------------------------------------"
@@ -192,7 +207,11 @@ journalctl --help &> /dev/null
 if [ $? -eq 0 ]; then
     journalctl -t scylla > $OUTPUT_PATH/scylla-logs.txt
 else
-    cat /var/log/syslog | grep -i scylla > $OUTPUT_PATH/scylla-logs.txt
+    if [ "$IS_GENTOO" == "0" ]; then
+        cat /var/log/scylla/scylla.log > $OUTPUT_PATH/scylla-logs.txt
+    else
+        cat /var/log/syslog | grep -i scylla > $OUTPUT_PATH/scylla-logs.txt
+    fi
 fi
 
 gzip -f $OUTPUT_PATH/scylla-logs.txt
@@ -224,6 +243,7 @@ if [ "$SCYLLA_SERVICE" == "1" ]; then
     echo "Skipping Data Model Info Collection"
     echo "--------------------------------------------------"
 else
+    # TODO: handle connecting with authentication
     cqlsh `hostname -i` $CQL_PORT -e "HELP" &> /dev/null
     if [ $? -eq 0 ]; then
         echo "Collecting Data Model Info (using port $CQL_PORT)"
@@ -357,7 +377,7 @@ if [ "$IS_FEDORA" == "0" ]; then
     echo "## /etc/sysconfig/scylla-server ##" >> $REPORT
 fi
 
-if [ "$IS_DEBIAN" == "0" ]; then
+if [ "$IS_DEBIAN" == "0" ] || [ "$IS_GENTOO" == "0" ]; then
     echo "## /etc/default/scylla-server ##" >> $REPORT
 fi
 

dist/common/scripts/scylla_coredump_setup

@@ -23,7 +23,6 @@ import os
 import sys
 import argparse
 import subprocess
-import shutil
 from scylla_util import *
 
 if __name__ == '__main__':
@@ -62,7 +61,7 @@ ExternalSizeMax=1024G
         with open('/etc/systemd/coredump.conf', 'w') as f:
             conf = f.write(conf_data)
         if args.dump_to_raiddir:
-            shutil.rmtree('/var/lib/systemd/coredump')
+            rmtree('/var/lib/systemd/coredump')
             makedirs('/var/lib/scylla/coredump')
             os.symlink('/var/lib/scylla/coredump', '/var/lib/systemd/coredump')
         run('systemctl daemon-reload')

dist/common/scripts/scylla_ec2_check

@@ -24,46 +24,38 @@ import sys
 import argparse
 from scylla_util import *
 
-def get_en_interface_type():
-    type, subtype = curl('http://169.254.169.254/latest/meta-data/instance-type').split('.')
-    if type in ['c3', 'c4', 'd4', 'd2', 'i2', 'r3']:
-        return 'ixgbevf'
-    if type in ['i3', 'p2', 'r4', 'x1']:
-        return 'ena'
-    if type == 'm4':
-        if subtype == '16xlarge':
-            return 'ena'
-        else:
-            return 'ixgbevf'
-
-def is_vpc_enabled():
-    with open('/sys/class/net/eth0/address') as f:
-        mac = f.read().strip()
-    mac_stat = curl('http://169.254.169.254/latest/meta-data/network/interfaces/macs/{}/'.format(mac))
-    return True if re.search(r'^vpc-id$', mac_stat, flags=re.MULTILINE) else False
-
-
 if __name__ == '__main__':
     if not is_ec2():
         sys.exit(0)
+    parser = argparse.ArgumentParser(description='Verify EC2 configuration is optimized.')
+    parser.add_argument('--nic', default='eth0',
+                        help='specify NIC')
+    args = parser.parse_args()
 
-    type = curl('http://169.254.169.254/latest/meta-data/instance-type')
-    en = get_en_interface_type()
-    match = re.search(r'^driver: (\S+)$', out('ethtool -i eth0'), flags=re.MULTILINE)
+    if not is_valid_nic(args.nic):
+        print('NIC {} doesn\'t exist.'.format(args.nic))
+        sys.exit(1)
+
+    aws = aws_instance()
+    instance_class = aws.instance_class()
+    en = aws.get_en_interface_type()
+    match = re.search(r'^driver: (\S+)$', out('ethtool -i {}'.format(args.nic)), flags=re.MULTILINE)
     driver = match.group(1)
 
     if not en:
-        print('{bold_red}{type} doesn\'t support enahanced networking!{no_color}'.format(bold_red=concolor.BOLD_RED, type=type, no_color=concolor.NO_COLOR))
+        colorprint('{red}{instance_class} doesn\'t support enahanced networking!{nocolor}', instance_class=instance_class)
         print('''To enable enhanced networking, please use the instance type which supports it.
 More documentation available at:
 http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/enhanced-networking.html#enabling_enhanced_networking''')
         sys.exit(1)
-    elif not is_vpc_enabled():
-        print('{bold_red}VPC is not enabled!{no_color}'.format(bold_red=concolor.BOLD_RED, no_color=concolor.NO_COLOR))
+    elif not aws.is_vpc_enabled(args.nic):
+        colorprint('{red}VPC is not enabled!{nocolor}')
         print('To enable enhanced networking, please enable VPC.')
         sys.exit(1)
     elif driver != en:
-        print('{bold_red}Enhanced networking is disabled!{no_color}'.format(bold_red=concolor.BOLD_RED, no_color=concolor.NO_COLOR))
+        colorprint('{red}Enhanced networking is disabled!{nocolor}')
         print('''More documentation available at:
 http://docs.aws.amazon.com/AWSEC2/latest/UserGuide/enhanced-networking.html''')
         sys.exit(1)
+
+    colorprint('{green}This EC2 instance is optimized for Scylla.{nocolor}')

dist/common/scripts/scylla_fstrim_setup

@@ -28,6 +28,8 @@ if __name__ == '__main__':
     if os.getuid() > 0:
         print('Requires root permission.')
         sys.exit(1)
+    if is_systemd():
+        systemd_unit('scylla-fstrim.timer').unmask()
     if is_redhat_variant():
         systemd_unit('fstrim.timer').disable()
     if dist_name() == 'Ubuntu' and os.path.exists('/etc/cron.weekly/fstrim'):

dist/common/scripts/scylla_lib.sh

@@ -1,122 +0,0 @@
-#
-#  Copyright (C) 2016 ScyllaDB
-
-is_debian_variant() {
-    [ -f /etc/debian_version ]
-}
-
-is_redhat_variant() {
-    [ -f /etc/redhat-release ]
-}
-
-is_gentoo_variant() {
-    [ -f /etc/gentoo-release ]
-}
-
-is_systemd() {
-    grep -q '^systemd$' /proc/1/comm
-}
-
-is_ec2() {
-    [ -f /sys/hypervisor/uuid ] && [ "$(head -c 3 /sys/hypervisor/uuid)" = "ec2" ]
-}
-
-is_selinux_enabled() {
-    STATUS=`getenforce`
-    if [ "$STATUS" = "Disabled" ]; then
-        return 0
-    else
-        return 1
-    fi
-}
-
-ec2_is_supported_instance_type() {
-    TYPE=`curl -s http://169.254.169.254/latest/meta-data/instance-type|cut -d . -f 1`
-    case $TYPE in
-           "i2"|"i3") echo 1;;
-            *) echo 0;;
-    esac
-}
-
-verify_args() {
-     if [ -z "$2" ] || [[ "$2" =~ ^--+ ]]; then
-        echo "Requires more parameter for $1."
-        print_usage
-        exit 1
-    fi
-}
-
-#
-#  get_mode_cpu_set <mode name, e.g. 'mq', 'sq', 'sq_split'>
-#
-get_mode_cpu_set() {
-    local mode=$1
-    local mode_cpu_mask=`/usr/lib/scylla/perftune.py --tune net --nic "$nic" --mode "$mode" --get-cpu-mask` 2>&-
-
-    # If the given mode is not supported - return invalid CPU set
-    if [[ "$?" -ne "0" ]]; then
-        echo "-1"
-    else
-        echo "$mode_cpu_mask" | /usr/lib/scylla/hex2list.py
-    fi
-}
-
-#
-# check_cpuset_conf <NIC name>
-#
-get_tune_mode() {
-    local nic=$1
-
-    # if cpuset.conf doesn't exist use the default mode
-    [[ ! -e '/etc/scylla.d/cpuset.conf' ]] && return
-
-    local cur_cpuset=`cat /etc/scylla.d/cpuset.conf | cut -d "\"" -f2- | cut -d" " -f2`
-    local mq_cpuset=`get_mode_cpu_set 'mq'`
-    local sq_cpuset=`get_mode_cpu_set 'sq'`
-    local sq_split_cpuset=`get_mode_cpu_set 'sq_split'`
-    local tune_mode=""
-
-    case "$cur_cpuset" in
-        "$mq_cpuset")
-            tune_mode="--mode mq"
-            ;;
-        "$sq_cpuset")
-            tune_mode="--mode sq"
-            ;;
-        "$sq_split_cpuset")
-            tune_mode="--mode sq_split"
-            ;;
-    esac
-
-    # if cpuset is something different from what we expect - use the default mode
-    echo "$tune_mode"
-}
-
-#
-# create_perftune_conf [<NIC name>]
-#
-create_perftune_conf() {
-    local nic=$1
-    [[ -z "$nic" ]] && nic='eth0'
-
-    # if exists - do nothing
-    [[ -e '/etc/scylla.d/perftune.yaml' ]] && return
-
-    local mode=`get_tune_mode "$nic"`
-    /usr/lib/scylla/perftune.py --tune net --nic "$nic" $mode --dump-options-file > /etc/scylla.d/perftune.yaml
-}
-
-. /etc/os-release
-if is_debian_variant || is_gentoo_variant; then
-    SYSCONFIG=/etc/default
-else
-    SYSCONFIG=/etc/sysconfig
-fi
-. $SYSCONFIG/scylla-server
-
-for i in /etc/scylla.d/*.conf; do
-    if [ "$i" = "/etc/scylla.d/*.conf" ]; then
-        break
-    fi
-    . "$i"
-done

dist/common/scripts/scylla_ntp_setup

@@ -49,7 +49,8 @@ if __name__ == '__main__':
         if is_systemd():
             ntp = systemd_unit('ntp.service')
             ntp.stop()
-            run('ntpdate ntp.ubuntu.com')
+            # ignore error, ntpd may able to adjust clock later
+            run('ntpdate ntp.ubuntu.com', exception=False)
             ntp.start()
         else:
             run('service ntp stop')
@@ -70,7 +71,8 @@ if __name__ == '__main__':
             sntpd.start()
         else:
             run('rc-service ntpd stop', exception=False)
-            run('ntpdate {}'.format(server))
+            # ignore error, ntpd may able to adjust clock later
+            run('ntpdate {}'.format(server), exception=False)
             run('rc-update add ntpd default')
             run('rc-service ntpd start')
 
@@ -87,6 +89,7 @@ if __name__ == '__main__':
         server = match.group(1)
         ntpd = systemd_unit('ntpd.service')
         ntpd.stop()
-        run('ntpdate {}'.format(server))
+        # ignore error, ntpd may able to adjust clock later
+        run('ntpdate {}'.format(server), exception=False)
         ntpd.enable()
         ntpd.start()

dist/common/scripts/scylla_prepare

@@ -1,33 +1,71 @@
-#!/bin/bash -e
+#!/usr/bin/python3
+#
+# Copyright 2018 ScyllaDB
+#
 
-. /usr/lib/scylla/scylla_lib.sh
+#
+# This file is part of Scylla.
+#
+# Scylla is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# Scylla is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
 
-if [ "$AMI" = "yes" ] && [ -f /etc/scylla/ami_disabled ]; then
-    rm /etc/scylla/ami_disabled
-    exit 1
-fi
+import os
+import sys
+import glob
+from scylla_util import *
 
-if [ "$NETWORK_MODE" = "virtio" ]; then
-    ip tuntap del mode tap dev $TAP
-    ip tuntap add mode tap dev $TAP user $USER one_queue vnet_hdr
-    ip link set dev $TAP up
-    ip link set dev $TAP master $BRIDGE
-    chown $USER.$GROUP /dev/vhost-net
-elif [ "$NETWORK_MODE" = "dpdk" ]; then
-    modprobe uio
-    modprobe uio_pci_generic
-    /usr/lib/scylla/dpdk-devbind.py --force --bind=uio_pci_generic $ETHPCIID
-    for n in /sys/devices/system/node/node?; do
-        echo $NR_HUGEPAGES > $n/hugepages/hugepages-2048kB/nr_hugepages
-    done
-    if [ "$ID" = "ubuntu" ]; then
-        hugeadm --create-mounts
-    fi
-else # NETWORK_MODE = posix
-    if [ "$SET_NIC" = "yes" ]; then
-        create_perftune_conf "$IFNAME"
-        /usr/lib/scylla/posix_net_conf.sh $IFNAME --options-file /etc/scylla.d/perftune.yaml
-    fi
-fi
+if __name__ == '__main__':
+    if os.getuid() > 0:
+        print('Requires root permission.')
+        sys.exit(1)
+    if is_redhat_variant():
+        cfg = sysconfig_parser('/etc/sysconfig/scylla-server')
+    else:
+        cfg = sysconfig_parser('/etc/default/scylla-server')
+    ami = cfg.get('AMI')
+    mode = cfg.get('NETWORK_MODE')
 
-/usr/lib/scylla/scylla-blocktune
+    if ami == 'yes' and os.path.exists('/etc/scylla/ami_disabled'):
+        os.remove('/etc/scylla/ami_disabled')
+        sys.exit(1)
+
+    if mode == 'virtio':
+        tap = cfg.get('TAP')
+        user = cfg.get('USER')
+        group = cfg.get('GROUP')
+        bridge = cfg.get('BRIDGE')
+        run('ip tuntap del mode tap dev {TAP}'.format(TAP=tap))
+        run('ip tuntap add mode tap dev {TAP} user {USER} one_queue vnet_hdr'.format(TAP=tap, USER=user))
+        run('ip link set dev {TAP} up'.format(TAP=tap))
+        run('ip link set dev {TAP} master {BRIDGE}'.format(TAP=tap, BRIDGE=bridge))
+        run('chown {USER}.{GROUP} /dev/vhost-net'.format(USER=user, GROUP=group))
+    elif mode == 'dpdk':
+        ethpcciid = cfg.get('ETHPCIID')
+        nr_hugepages = cfg.get('NR_HUGEPAGES')
+        run('modprobe uio')
+        run('modprobe uio_pci_generic')
+        run('/usr/lib/scylla/dpdk-devbind.py --force --bind=uio_pci_generic {ETHPCIID}'.format(ETHPCIID=ethpciid))
+        for n in glob.glob('/sys/devices/system/node/node?'):
+            with open('{n}/hugepages/hugepages-2048kB/nr_hugepages'.format(n=n), 'w') as f:
+                f.write(nr_hugepages)
+        if dist_name() == 'Ubuntu': 
+            run('hugeadm --create-mounts')
+        fi
+    else:
+        set_nic = cfg.get('SET_NIC')
+        ifname = cfg.get('IFNAME')
+        if set_nic  == 'yes':
+            create_perftune_conf(ifname)
+            run('/usr/lib/scylla/posix_net_conf.sh {IFNAME} --options-file /etc/scylla.d/perftune.yaml'.format(IFNAME=ifname))
+
+    run('/usr/lib/scylla/scylla-blocktune')

dist/common/scripts/scylla_raid_setup

@@ -23,6 +23,8 @@ import os
 import argparse
 import pwd
 import grp
+import sys
+import stat
 from scylla_util import *
 
 if __name__ == '__main__':
@@ -40,6 +42,8 @@ if __name__ == '__main__':
                         help='specify the root of the tree')
     parser.add_argument('--volume-role', default='all',
                         help='specify how will this device be used (data, commitlog, or all)')
+    parser.add_argument('--force-raid', action='store_true', default=False,
+                        help='force constructing RAID when only one disk is specified')
 
     args = parser.parse_args()
 
@@ -60,6 +64,12 @@ if __name__ == '__main__':
         if not os.path.exists(disk):
             print('{} is not found'.format(disk))
             sys.exit(1)
+        if not stat.S_ISBLK(os.stat(disk).st_mode):
+            print('{} is not block device'.format(disk))
+            sys.exit(1)
+        if not is_unused_disk(disk):
+            print('{} is busy'.format(disk))
+            sys.exit(1)
 
     if os.path.exists(args.raiddev):
         print('{} is already using'.format(args.raiddev))
@@ -74,12 +84,20 @@ if __name__ == '__main__':
     elif is_gentoo_variant():
         run('emerge -uq sys-fs/mdadm sys-fs/xfsprogs')
 
-    print('Creating RAID0 for scylla using {nr_disk} disk(s): {disks}'.format(nr_disk=len(disks), disks=args.disks))
+    if len(disks) == 1 and not args.force_raid:
+        raid = False
+        fsdev = disks[0]
+    else:
+        raid = True
+        fsdev = args.raiddev
+
+    print('Creating {type} for scylla using {nr_disk} disk(s): {disks}'.format(type='RAID0' if raid else 'XFS volume', nr_disk=len(disks), disks=args.disks))
     if dist_name() == 'Ubuntu' and dist_ver() == '14.04':
-        run('udevadm settle')
-        run('mdadm --create --verbose --force --run {raid} --level=0 -c1024 --raid-devices={nr_disk} {disks}'.format(raid=args.raiddev, nr_disk=len(disks), disks=args.disks.replace(',', ' ')))
-        run('udevadm settle')
-        run('mkfs.xfs {} -f'.format(args.raiddev))
+        if raid:
+            run('udevadm settle')
+            run('mdadm --create --verbose --force --run {raid} --level=0 -c1024 --raid-devices={nr_disk} {disks}'.format(raid=fsdev, nr_disk=len(disks), disks=args.disks.replace(',', ' ')))
+            run('udevadm settle')
+        run('mkfs.xfs {} -f'.format(fsdev))
     else:
         procs=[]
         for disk in disks:
@@ -93,22 +111,24 @@ if __name__ == '__main__':
                     procs.append(proc)
         for proc in procs:
             proc.wait()
-        run('udevadm settle')
-        run('mdadm --create --verbose --force --run {raid} --level=0 -c1024 --raid-devices={nr_disk} {disks}'.format(raid=args.raiddev, nr_disk=len(disks), disks=args.disks.replace(',', ' ')))
-        run('udevadm settle')
-        run('mkfs.xfs {} -f -K'.format(args.raiddev))
+        if raid:
+            run('udevadm settle')
+            run('mdadm --create --verbose --force --run {raid} --level=0 -c1024 --raid-devices={nr_disk} {disks}'.format(raid=fsdev, nr_disk=len(disks), disks=args.disks.replace(',', ' ')))
+            run('udevadm settle')
+        run('mkfs.xfs {} -f -K'.format(fsdev))
 
     if is_debian_variant():
         confpath = '/etc/mdadm/mdadm.conf'
     else:
         confpath = '/etc/mdadm.conf'
 
-    res = out('mdadm --detail --scan')
-    with open(confpath, 'w') as f:
-        f.write(res)
+    if raid:
+        res = out('mdadm --detail --scan')
+        with open(confpath, 'w') as f:
+            f.write(res)
 
     makedirs(mount_at)
-    run('mount -t xfs -o noatime {raid} "{mount_at}"'.format(raid=args.raiddev, mount_at=mount_at))
+    run('mount -t xfs -o noatime {raid} "{mount_at}"'.format(raid=fsdev, mount_at=mount_at))
 
     makedirs('{}/data'.format(root))
     makedirs('{}/commitlog'.format(root))
@@ -122,11 +142,19 @@ if __name__ == '__main__':
     os.chown('{}/coredump'.format(root), uid, gid)
 
     if args.update_fstab:
-        res = out('blkid {}'.format(args.raiddev))
+        res = out('blkid {}'.format(fsdev))
         match = re.search(r'^/dev/\S+: (UUID="\S+")', res.strip())
         uuid = match.group(1)
         with open('/etc/fstab', 'a') as f:
-            f.write('{uuid} {mount_at} xfs noatime 0 0\n'.format(uuid=uuid, mount_at=mount_at))
+            f.write('{uuid} {mount_at} xfs noatime,nofail 0 0\n'.format(uuid=uuid, mount_at=mount_at))
+        mounts_conf = '/etc/systemd/system/scylla-server.service.d/mounts.conf'
+        if not os.path.exists(mounts_conf):
+            makedirs('/etc/systemd/system/scylla-server.service.d/')
+            with open(mounts_conf, 'w') as f:
+                f.write('[Unit]\nRequiresMountsFor={mount_at}\n'.format(mount_at=mount_at))
+        else:
+            with open(mounts_conf, 'a') as f:
+                f.write('RequiresMountsFor={mount_at}\n'.format(mount_at=mount_at))
 
     if is_debian_variant():
         run('update-initramfs -u')

dist/common/scripts/scylla_setup

@@ -22,7 +22,6 @@
 import os
 import sys
 import argparse
-import logging
 import glob
 import shutil
 import io
@@ -49,11 +48,28 @@ def interactive_ask_service(msg1, msg2, default = None):
         elif ans == 'no' or ans =='n':
             return False
 
+def interactive_choose_nic():
+    nics = [os.path.basename(n) for n in glob.glob('/sys/class/net/*') if n != '/sys/class/net/lo']
+    if len(nics) == 0:
+        print('A NIC was not found.')
+        sys.exit(1)
+    elif len(nics) == 1:
+        return nics[0]
+    else:
+        print('Please select a NIC from the following list:')
+        while True:
+            print(nics)
+            n = input('> ')
+            if is_valid_nic(n):
+                return n
+
 def do_verify_package(pkg):
     if is_debian_variant():
         res = run('dpkg -s {}'.format(pkg), silent=True, exception=False)
     elif is_redhat_variant():
         res = run('rpm -q {}'.format(pkg), silent=True, exception=False)
+    elif is_gentoo_variant():
+        res = 1 if len(glob.glob('/var/db/pkg/*/{}-*'.format(pkg))) else 0
     if res != 0:
         print('{} package is not installed.'.format(pkg))
         sys.exit(1)
@@ -67,22 +83,18 @@ def list_block_devices():
         devices = []
         for p in ['/dev/sd*', '/dev/hd*', '/dev/xvd*', '/dev/nvme*', '/dev/mapper/*']:
             devices.extend([d for d in glob.glob(p) if d != '/dev/mapper/control'])
-        return devices
+    return devices
 
 def get_unused_disks():
     unused = []
     for dev in list_block_devices():
-        with open('/proc/mounts') as f:
-            s = f.read().strip()
-        count_raw = len(re.findall('^{} '.format(dev), s, flags=re.MULTILINE))
-        count_pvs = 0
-        if shutil.which('pvs'):
-            s = out('pvs -o pv_name --nohead')
-            count_pvs = len(re.findall(dev, s, flags=re.MULTILINE))
-        s = out('swapon --show=NAME --noheadings')
-        count_swap = len(re.findall(dev, s, flags=re.MULTILINE))
-        if count_raw + count_pvs + count_swap == 0:
-            unused.append(dev)
+        # dev contains partitions
+        if len(glob.glob('/sys/class/block/{dev}/{dev}*'.format(dev=dev.replace('/dev/','')))) > 0:
+            continue
+        # dev is used
+        if not is_unused_disk(dev):
+            continue
+        unused.append(dev)
     return unused
 
 def run_setup_script(name, script):
@@ -90,7 +102,7 @@ def run_setup_script(name, script):
     res = run(script, exception=False)
     if res != 0:
         if interactive:
-            print('{red}{name} setup failed. Press any key to continue...{no_color}'.format(red=concolor.BOLD_RED, name=name, no_color=concolor.NO_COLOR))
+            colorprint('{red}{name} setup failed. Press any key to continue...{nocolor}', name=name)
             input()
         else:
             print('{} setup failed.'.format(name))
@@ -99,12 +111,12 @@ def run_setup_script(name, script):
 
 if __name__ == '__main__':
     if os.getuid() > 0:
-        logging.error('Requires root permission.')
+        print('Requires root permission.')
         sys.exit(1)
     parser = argparse.ArgumentParser(description='Configure environment for Scylla.')
     parser.add_argument('--disks',
                         help='specify disks for RAID')
-    parser.add_argument('--nic',
+    parser.add_argument('--nic', default='eth0',
                         help='specify NIC')
     parser.add_argument('--ntp-domain',
                         help='specify NTP domain')
@@ -115,7 +127,7 @@ if __name__ == '__main__':
     parser.add_argument('--developer-mode', action='store_true', default=False,
                         help='enable developer mode')
     parser.add_argument('--no-ec2-check', action='store_true', default=False,
-                        help='skip EC2 configuration check(only on EC2)')
+                        help='skip EC2 configuration check')
     parser.add_argument('--no-kernel-check', action='store_true', default=False,
                         help='skip kernel version check')
     parser.add_argument('--no-verify-package', action='store_true', default=False,
@@ -150,12 +162,14 @@ if __name__ == '__main__':
     if len(sys.argv) == 1:
         interactive = True
 
-    if not interactive and not args.no_raid_setup and not args.disks:
-        parser.print_help()
-        sys.exit(1)
-    if not interactive and not args.no_sysconfig_setup and not args.nic:
-        parser.print_help()
-        sys.exit(1)
+    if not interactive:
+        if not args.no_raid_setup and not args.disks:
+            parser.print_help()
+            sys.exit(1)
+        if not args.no_sysconfig_setup or (is_ec2() and not args.no_ec2_check):
+            if not is_valid_nic(args.nic):
+                print('NIC {} doesn\'t exist.'.format(args.nic))
+                sys.exit(1)
 
     disks = args.disks
     nic = args.nic
@@ -178,13 +192,16 @@ if __name__ == '__main__':
     fstrim_setup = not args.no_fstrim_setup
     selinux_reboot_required = False
 
-    print('{green}Skip any of the following steps by answering \'no\'{no_color}'.format(green=concolor.GREEN, no_color=concolor.NO_COLOR))
+    if interactive:
+        colorprint('{green}Skip any of the following steps by answering \'no\'{nocolor}')
 
     if is_ec2():
         if interactive:
-            ec2_check = interactive_ask_service('Do you want to run Amazon EC2 configuration check?', 'Yes - runs a script to verify that this instance is optimized for running Scylls. No - skips the configuration check.', 'yes')
+            ec2_check = interactive_ask_service('Do you want to run Amazon EC2 configuration check?', 'Yes - runs a script to verify that this instance is optimized for running Scylla. No - skips the configuration check.', 'yes')
+            if ec2_check:
+                nic = interactive_choose_nic()
         if ec2_check:
-            run('/usr/lib/scylla/scylla_ec2_check')
+            run('/usr/lib/scylla/scylla_ec2_check --nic {}'.format(nic))
 
     if interactive:
         kernel_check = interactive_ask_service('Do you want to run check your kernel version?', 'Yes - runs a  script to verify that the kernel for this instance qualifies to run Scylla. No - skips the kernel check.', 'yes')
@@ -202,11 +219,10 @@ if __name__ == '__main__':
     if enable_service:
         if is_systemd():
             systemd_unit('scylla-server.service').enable()
-            systemd_unit('scylla-fstrim.timer').unmask()
         elif is_gentoo_variant():
             run('rc-update add scylla-server default')
 
-    if interactive:
+    if interactive and not os.path.exists('/etc/scylla.d/housekeeping.cfg'):
         version_check = interactive_ask_service('Do you want to enable Scylla to check if there is a newer version of Scylla available?', 'Yes - start the Scylla-housekeeping service to check for a newer version. This check runs periodically. No - skips this step.', 'yes')
     if version_check:
         with open('/etc/scylla.d/housekeeping.cfg', 'w') as f:
@@ -277,10 +293,14 @@ if __name__ == '__main__':
             else:
                 print('Please select unmounted disks from the following list: {}'.format(devices))
             selected = []
-            while len(devices):
+            dsklist = []
+            while True:
                 print('type \'cancel\' to cancel RAID/XFS setup.')
                 print('type \'done\' to finish selection. Selected: {}'.format(selected))
-                dsk = input('> ')
+                if len(dsklist) > 0:
+                    dsk = dsklist.pop(0)
+                else:
+                    dsk = input('> ')
                 if dsk == 'cancel':
                     raid_setup = 0
                     break
@@ -290,12 +310,16 @@ if __name__ == '__main__':
                     break
                 if dsk == '':
                     continue
+                if dsk.find(',') > 0:
+                    dsklist = dsk.split(',')
+                    continue
                 if not os.path.exists(dsk):
                     print('{} not found'.format(dsk))
                     continue
                 if not stat.S_ISBLK(os.stat(dsk).st_mode):
                     print('{} is not block device'.format(dsk))
-                selected += dsk
+                    continue
+                selected.append(dsk)
                 devices.remove(dsk)
             disks = ','.join(selected)
     if raid_setup:
@@ -312,21 +336,9 @@ if __name__ == '__main__':
     if interactive:
         sysconfig_setup = interactive_ask_service('Do you want to setup a system-wide customized configuration for Scylla?', 'Yes - setup the sysconfig file. No - skips this step.', 'yes')
     if sysconfig_setup:
-        nics = [os.path.basename(n) for n in glob.glob('/sys/class/net/*') if n != '/sys/class/net/lo']
-        if len(nics) == 0:
-            print('A NIC was not found.')
-            sys.exit(1)
-        elif len(nics) == 1:
-            nic=nics[0]
-        else:
-            print('Please select a NIC from the following list:')
-            while True:
-                print(nics)
-                n = input('> ')
-                if os.path.exists('/sys/class/net/{}'.format(n)):
-                    nic = n
-                    break
-        set_nic = interactive_ask_service('Do you want to enable Network Interface Card (NIC) optimization?', 'Yes - optimize the NIC queue settings. Selecting Yes greatly improves performance. No - skip this step.', 'yes')
+        nic = interactive_choose_nic()
+        if interactive:
+            set_nic = interactive_ask_service('Do you want to enable Network Interface Card (NIC) optimization?', 'Yes - optimize the NIC queue settings. Selecting Yes greatly improves performance. No - skip this step.', 'yes')
     if sysconfig_setup:
         setup_args = '--setup-nic' if set_nic else ''
         run_setup_script('NIC queue', '/usr/lib/scylla/scylla_sysconfig_setup --nic {nic} {setup_args}'.format(nic=nic, setup_args=setup_args))

dist/common/scripts/scylla_stop

@@ -1,10 +1,40 @@
-#!/bin/bash -e
+#!/usr/bin/python3
+#
+# Copyright 2018 ScyllaDB
+#
 
-. /usr/lib/scylla/scylla_lib.sh
+#
+# This file is part of Scylla.
+#
+# Scylla is free software: you can redistribute it and/or modify
+# it under the terms of the GNU Affero General Public License as published by
+# the Free Software Foundation, either version 3 of the License, or
+# (at your option) any later version.
+#
+# Scylla is distributed in the hope that it will be useful,
+# but WITHOUT ANY WARRANTY; without even the implied warranty of
+# MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+# GNU General Public License for more details.
+#
+# You should have received a copy of the GNU General Public License
+# along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
 
-if [ "$NETWORK_MODE" = "virtio" ]; then
-    ip tuntap del mode tap dev $TAP
-elif [ "$NETWORK_MODE" = "dpdk" ]; then
-    /usr/lib/scylla/dpdk-devbind.py -u $ETHPCIID
-    /usr/lib/scylla/dpdk-devbind.py -b $ETHDRV $ETHPCIID
-fi
+import os
+import sys
+from scylla_util import *
+
+if __name__ == '__main__':
+    if os.getuid() > 0:
+        print('Requires root permission.')
+        sys.exit(1)
+    if is_redhat_variant():
+        cfg = sysconfig_parser('/etc/sysconfig/scylla-server')
+    else:
+        cfg = sysconfig_parser('/etc/default/scylla-server')
+
+    
+    if cfg.get('NETWORK_MODE') == 'virtio':
+        run('ip tuntap del mode tap dev {TAP}'.format(TAP=cfg.get('TAP')))
+    elif cfg.get('NETWORK_MODE') == 'dpdk':
+        run('/usr/lib/scylla/dpdk-devbind.py -u {ETHPCIID}'.format(ETHPCIID=cfg.get('ETHPCIID')))
+        run('/usr/lib/scylla/dpdk-devbind.py -b {ETHDRV} {ETHPCIID}'.format(ETHDRV=cfg.get('ETHDRV'), ETHPCIID=cfg.get('ETHPCIID')))

dist/common/scripts/scylla_sysconfig_setup

@@ -64,6 +64,10 @@ if __name__ == '__main__':
                         help='AMI instance mode')
     args = parser.parse_args()
 
+    if args.nic and not is_valid_nic(args.nic):
+            print('NIC {} not found.'.format(args.nic))
+            sys.exit(1)
+
     ifname = args.nic if args.nic else cfg.get('IFNAME')
     network_mode = args.mode if args.mode else cfg.get('NETWORK_MODE')
 

dist/common/scripts/scylla_util.py

@@ -27,14 +27,19 @@ import platform
 import configparser
 import io
 import shlex
+import shutil
 
-def curl(url):
+def curl(url, byte=False):
     max_retries = 5
     retries = 0
     while True:
         try:
             req = urllib.request.Request(url)
-            return urllib.request.urlopen(req).read().decode('utf-8')
+            with urllib.request.urlopen(req) as res:
+                if byte:
+                    return res.read()
+                else:
+                    return res.read().decode('utf-8')
         except urllib.error.HTTPError:
             logging.warn("Failed to grab %s..." % url)
             time.sleep(5)
@@ -79,6 +84,10 @@ class aws_instance:
                 continue
             self._disks[t] += [ self.__xenify(dev) ]
 
+    def __mac_address(self, nic='eth0'):
+        with open('/sys/class/net/{}/address'.format(nic)) as f:
+            return f.read().strip()
+
     def __init__(self):
         self._type = self.__instance_metadata("instance-type")
         self.__populate_disks()
@@ -95,6 +104,25 @@ class aws_instance:
         """Returns the class of the instance we are running in. i.e.: i3"""
         return self._type.split(".")[0]
 
+    def is_supported_instance_class(self):
+        if self.instance_class() in ['i2', 'i3']:
+            return True
+        return False
+
+    def get_en_interface_type(self):
+        instance_class = self.instance_class()
+        instance_size = self.instance_size()
+        if instance_class in ['c3', 'c4', 'd2', 'i2', 'r3']:
+            return 'ixgbevf'
+        if instance_class in ['c5', 'c5d', 'f1', 'g3', 'h1', 'i3', 'm5', 'm5d', 'p2', 'p3', 'r4', 'x1']:
+            return 'ena'
+        if instance_class == 'm4':
+            if instance_size == '16xlarge':
+                return 'ena'
+            else:
+                return 'ixgbevf'
+        return None
+
     def disks(self):
         """Returns all disks in the system, as visible from the AWS registry"""
         disks = set()
@@ -133,6 +161,11 @@ class aws_instance:
         """Returns the private IPv4 address of this instance"""
         return self.__instance_metadata("local-ipv4")
 
+    def is_vpc_enabled(self, nic='eth0'):
+        mac = self.__mac_address(nic)
+        mac_stat = self.__instance_metadata('network/interfaces/macs/{}'.format(mac))
+        return True if re.search(r'^vpc-id$', mac_stat, flags=re.MULTILINE) else False
+
 
 ## Regular expression helpers
 # non-advancing comment matcher
@@ -222,37 +255,24 @@ class scylla_cpuinfo:
             return len(self._cpu_data["system"])
 
 def run(cmd, shell=False, silent=False, exception=True):
-    stdout=None
-    stderr=None
-    if silent:
-        stdout=subprocess.DEVNULL
-        stderr=subprocess.DEVNULL
-    if shell:
-        if exception:
-            return subprocess.check_call(cmd, shell=True, stdout=stdout, stderr=stderr)
-        else:
-            p = subprocess.Popen(cmd, shell=True, stdout=stdout, stderr=stderr)
-            return p.wait()
+    stdout=subprocess.DEVNULL if silent else None
+    stderr=subprocess.DEVNULL if silent else None
+    if not shell:
+        cmd = shlex.split(cmd)
+    if exception:
+        return subprocess.check_call(cmd, shell=shell, stdout=stdout, stderr=stderr)
     else:
-        if exception:
-            return subprocess.check_call(shlex.split(cmd), stdout=stdout, stderr=stderr)
-        else:
-            p = subprocess.Popen(shlex.split(cmd), stdout=stdout, stderr=stderr)
-            return p.wait()
+        p = subprocess.Popen(cmd, shell=shell, stdout=stdout, stderr=stderr)
+        return p.wait()
 
 def out(cmd, shell=False, exception=True):
-    if shell:
-        if exception:
-            return subprocess.check_output(cmd, shell=True).strip().decode('utf-8')
-        else:
-            p = subprocess.Popen(cmd, shell=True, stdout=subprocess.PIPE)
-            return p.communicate()[0].strip().decode('utf-8')
+    if not shell:
+        cmd = shlex.split(cmd)
+    if exception:
+        return subprocess.check_output(cmd, shell=shell).strip().decode('utf-8')
     else:
-        if exception:
-            return subprocess.check_output(shlex.split(cmd)).strip().decode('utf-8')
-        else:
-            p = subprocess.Popen(shlex.split(cmd), stdout=subprocess.PIPE)
-            return p.communicate()[0].strip().decode('utf-8')
+        p = subprocess.Popen(cmd, shell=shell, stdout=subprocess.PIPE)
+        return p.communicate()[0].strip().decode('utf-8')
 
 def is_debian_variant():
     return os.path.exists('/etc/debian_version')
@@ -306,17 +326,82 @@ def makedirs(name):
     if not os.path.isdir(name):
         os.makedirs(name)
 
+def rmtree(path):
+    if not os.path.islink(path):
+        shutil.rmtree(path)
+    else:
+        os.remove(path)
+
 def dist_name():
     return platform.dist()[0]
 
 def dist_ver():
     return platform.dist()[1]
 
+def is_unused_disk(dev):
+    # dev is not in /sys/class/block/, like /dev/nvme[0-9]+
+    if not os.path.isdir('/sys/class/block/{dev}'.format(dev=dev.replace('/dev/',''))):
+        return False
+    try:
+        fd = os.open(dev, os.O_EXCL)
+        os.close(fd)
+        return True
+    except OSError:
+        return False
+
+CONCOLORS = {'green':'\033[1;32m', 'red':'\033[1;31m', 'nocolor':'\033[0m'}
+def colorprint(msg, **kwargs):
+    fmt = dict(CONCOLORS)
+    fmt.update(kwargs)
+    print(msg.format(**fmt))
+
+def get_mode_cpuset(nic, mode):
+    try:
+        mode_cpu_mask=out('/usr/lib/scylla/perftune.py --tune net --nic "{nic}" --mode "{mode}" --get-cpu-mask'.format(nic=nic, mode=mode))
+        return hex2list(mode_cpu_mask)
+    except subprocess.CalledProcessError:
+        return '-1'
+
+def get_cur_cpuset():
+    cfg = sysconfig_parser('/etc/scylla.d/cpuset.conf')
+    cpuset=cfg.get('CPUSET')
+    return re.sub(r'^--cpuset (.+)$', r'\1', cpuset).strip()
+
+def get_tune_mode(nic):
+    if not os.path.exists('/etc/scylla.d/cpuset.conf'):
+        return
+    cur_cpuset=get_cur_cpuset()
+    mq_cpuset=get_mode_cpuset(nic, 'mq')
+    sq_cpuset=get_mode_cpuset(nic, 'sq')
+    sq_split_cpuset=get_mode_cpuset(nic, 'sq_split')
+
+    if cur_cpuset == mq_cpuset:
+        return 'mq'
+    elif cur_cpuset == sq_cpuset:
+        return 'sq'
+    elif cur_cpuset == sq_split_cpuset:
+        return 'sq_split'
+
+def create_perftune_conf(nic='eth0'):
+    if os.path.exists('/etc/scylla.d/perftune.yaml'):
+        return
+    mode=get_tune_mode(nic)
+    yaml=out('/usr/lib/scylla/perftune.py --tune net --nic "{nic}" --mode {mode} --dump-options-file'.format(nic=nic, mode=mode))
+    with open('/etc/scylla.d/perftune.yaml', 'w') as f:
+        f.write(yaml)
+
+def is_valid_nic(nic):
+    return os.path.exists('/sys/class/net/{}'.format(nic))
+
 class SystemdException(Exception):
     pass
 
 class systemd_unit:
     def __init__(self, unit):
+        try:
+            run('systemctl cat {}'.format(unit), silent=True)
+        except subprocess.CalledProcessError:
+            raise SystemdException('unit {} not found'.format(unit))
         self._unit = unit
 
     def start(self):
@@ -336,8 +421,7 @@ class systemd_unit:
         return run('systemctl disable {}'.format(self._unit))
 
     def is_active(self):
-        res = out('systemctl is-active {}'.format(self._unit), exception=False)
-        return True if re.match(r'^active', res, flags=re.MULTILINE) else False
+        return out('systemctl is-active {}'.format(self._unit), exception=False)
 
     def mask(self):
         return run('systemctl mask {}'.format(self._unit))
@@ -368,7 +452,7 @@ class sysconfig_parser:
         self.__load()
 
     def get(self, key):
-        return self._cfg.get('global', key)
+        return self._cfg.get('global', key).strip('"')
 
     def set(self, key, val):
         if not self._cfg.has_option('global', key):
@@ -379,9 +463,3 @@ class sysconfig_parser:
     def commit(self):
         with open(self._filename, 'w') as f:
             f.write(self._data)
-
-class concolor:
-    GREEN = '\033[0;32m'
-    RED = '\033[0;31m'
-    BOLD_RED = '\033[1;31m'
-    NO_COLOR = '\033[0m'

dist/common/systemd/scylla-housekeeping-daily.service.mustache

@@ -10,7 +10,7 @@ Group=scylla
 ExecStart=/usr/lib/scylla/scylla-housekeeping --uuid-file /var/lib/scylla-housekeeping/housekeeping.uuid -q -c /etc/scylla.d/housekeeping.cfg --repo-files '/etc/apt/sources.list.d/scylla*.list' version --mode d
 {{/debian}}
 {{#redhat}}
-ExecStart=/usr/lib/scylla/scylla-housekeeping --uuid-file /var/lib/scylla-housekeeping/housekeeping.uuid -q -c /etc/scylla.d/housekeeping.cfg --repo-files @@REPOFILES@@ version --mode d
+ExecStart=/usr/lib/scylla/scylla-housekeeping --uuid-file /var/lib/scylla-housekeeping/housekeeping.uuid -q -c /etc/scylla.d/housekeeping.cfg --repo-files '/etc/yum.repos.d/scylla*.repo' version --mode d
 {{/redhat}}
 
 [Install]

dist/common/systemd/scylla-housekeeping-restart.service.mustache

@@ -6,7 +6,7 @@ After=network.target
 Type=simple
 User=scylla
 Group=scylla
-ExecStart=/usr/lib/scylla/scylla-housekeeping --uuid-file /var/lib/scylla-housekeeping/housekeeping.uuid -q -c /etc/scylla.d/housekeeping.cfg --repo-files @@REPOFILES@@ version --mode r
+ExecStart=/usr/lib/scylla/scylla-housekeeping --uuid-file /var/lib/scylla-housekeeping/housekeeping.uuid -q -c /etc/scylla.d/housekeeping.cfg --repo-files '/etc/yum.repos.d/scylla*.repo' version --mode r
 
 [Install]
 WantedBy=multi-user.target

dist/debian/build_deb.sh

@@ -51,6 +51,18 @@ is_redhat_variant() {
 is_debian_variant() {
     [ -f /etc/debian_version ]
 }
+is_debian() {
+    case "$1" in
+        jessie|stretch) return 0;;
+        *) return 1;;
+    esac
+}
+is_ubuntu() {
+    case "$1" in
+        trusty|xenial|bionic) return 0;;
+        *) return 1;;
+    esac
+}
 
 
 pkg_install() {
@@ -99,11 +111,14 @@ if [ ! -f /usr/bin/dh_testdir ]; then
 fi
 if [ ! -f /usr/bin/pystache ]; then
     if is_redhat_variant; then
-        sudo yum install -y python2-pystache || sudo yum install -y pystache
+        sudo yum install -y /usr/bin/pystache
     elif is_debian_variant; then
         sudo apt-get install -y python-pystache
     fi
 fi
+if is_debian_variant && [ ! -f /usr/share/doc/python-pkg-resources/copyright ]; then
+    sudo apt-get install -y python-pkg-resources
+fi
 
 if [ -z "$TARGET" ]; then
     if is_debian_variant; then
@@ -125,12 +140,12 @@ echo $VERSION > version
 
 cp -a dist/debian/debian debian
 cp dist/common/sysconfig/scylla-server debian/scylla-server.default
-if [ "$TARGET" = "jessie" ] || [ "$TARGET" = "stretch" ]; then
-    REVISION="1~$TARGET"
-elif [ "$TARGET" = "trusty" ]; then
+if [ "$TARGET" = "trusty" ]; then
     cp dist/debian/scylla-server.cron.d debian/
-    REVISION="0ubuntu1~$TARGET"
-elif [ "$TARGET" = "xenial" ] || [ "$TARGET" = "bionic" ]; then
+fi
+if is_debian $TARGET; then
+    REVISION="1~$TARGET"
+elif is_ubuntu $TARGET; then
     REVISION="0ubuntu1~$TARGET"
 else
    echo "Unknown distribution: $TARGET"
@@ -145,8 +160,8 @@ chmod a+rx debian/rules
 
 if [ "$TARGET" != "trusty" ]; then
     pystache dist/common/systemd/scylla-server.service.mustache "{ $MUSTACHE_DIST }" > debian/scylla-server.service
-    pystache dist/common/systemd/scylla-housekeeping-daily.service.mustache "{ $MUSTACHE_DIST }" > debian/scylla-housekeeping-daily.service
-    pystache dist/common/systemd/scylla-housekeeping-restart.service.mustache "{ $MUSTACHE_DIST }" > debian/scylla-housekeeping-restart.service
+    pystache dist/common/systemd/scylla-housekeeping-daily.service.mustache "{ $MUSTACHE_DIST }" > debian/scylla-server.scylla-housekeeping-daily.service
+    pystache dist/common/systemd/scylla-housekeeping-restart.service.mustache "{ $MUSTACHE_DIST }" > debian/scylla-server.scylla-housekeeping-restart.service
     cp dist/common/systemd/scylla-fstrim.service debian/scylla-server.scylla-fstrim.service
     cp dist/common/systemd/node-exporter.service debian/scylla-server.node-exporter.service
 fi

dist/docker/redhat/Dockerfile

@@ -26,14 +26,14 @@ ADD commandlineparser.py /commandlineparser.py
 ADD docker-entrypoint.py /docker-entrypoint.py
 
 # Install Scylla:
-RUN curl http://downloads.scylladb.com/rpm/unstable/centos/master/latest/scylla.repo -o /etc/yum.repos.d/scylla.repo && \
+RUN curl http://downloads.scylladb.com/rpm/centos/scylla-2.3.repo -o /etc/yum.repos.d/scylla.repo && \
     yum -y install epel-release && \
     yum -y clean expire-cache && \
     yum -y update && \
     yum -y remove boost-thread boost-system && \
     yum -y install scylla hostname supervisor && \
     yum clean all && \
-    yum -y install python34 python34-PyYAML && \
+    yum -y install python36 python36-PyYAML && \
     cat /scylla_bashrc >> /etc/bashrc && \
     mkdir -p /etc/supervisor.conf.d && \
     mkdir -p /var/log/scylla && \

dist/docker/redhat/scylla-service.sh

@@ -1,7 +1,13 @@
 #!/bin/bash
 
-. /usr/lib/scylla/scylla_prepare
+/usr/lib/scylla/scylla_prepare
+
+. /etc/sysconfig/scylla-server
 
 export SCYLLA_HOME SCYLLA_CONF
 
+for f in /etc/scylla.d/*.conf; do
+    . "$f"
+done
+
 exec /usr/bin/scylla $SCYLLA_ARGS $SEASTAR_IO $DEV_MODE $CPUSET $SCYLLA_DOCKER_ARGS

dist/redhat/build_rpm.sh

@@ -98,12 +98,19 @@ rm -f version
 
 pystache dist/redhat/scylla.spec.mustache "{ \"version\": \"$SCYLLA_VERSION\", \"release\": \"$SCYLLA_RELEASE\", \"housekeeping\": $DIST }" > build/scylla.spec
 
+# mock generates files owned by root, fix this up
+fix_ownership() {
+    sudo chown "$(id -u):$(id -g)" -R "$@"
+}
+
 if [ $JOBS -gt 0 ]; then
     RPM_JOBS_OPTS=(--define="_smp_mflags -j$JOBS")
 fi
 sudo mock --buildsrpm --root=$TARGET --resultdir=`pwd`/build/srpms --spec=build/scylla.spec --sources=build/scylla-$VERSION.tar $SRPM_OPTS "${RPM_JOBS_OPTS[@]}"
+fix_ownership build/srpms
 if [[ "$TARGET" =~ ^epel-7- ]]; then
     TARGET=scylla-$TARGET
     RPM_OPTS="$RPM_OPTS --configdir=dist/redhat/mock"
 fi
 sudo mock --rebuild --root=$TARGET --resultdir=`pwd`/build/rpms $RPM_OPTS "${RPM_JOBS_OPTS[@]}" build/srpms/scylla-$VERSION*.src.rpm
+fix_ownership build/rpms

dist/redhat/scylla.spec.mustache

@@ -56,9 +56,9 @@ License:        AGPLv3
 URL:            http://www.scylladb.com/
 BuildRequires:  libaio-devel libstdc++-devel cryptopp-devel hwloc-devel numactl-devel libpciaccess-devel libxml2-devel zlib-devel thrift-devel yaml-cpp-devel lz4-devel snappy-devel jsoncpp-devel systemd-devel xz-devel pcre-devel elfutils-libelf-devel bzip2-devel keyutils-libs-devel xfsprogs-devel make gnutls-devel systemd-devel lksctp-tools-devel protobuf-devel protobuf-compiler systemtap-sdt-devel ninja-build cmake python ragel grep kernel-headers
 %{?fedora:BuildRequires: boost-devel antlr3-tool antlr3-C++-devel python3 gcc-c++ libasan libubsan python3-pyparsing dnf-yum python2-pystache}
-%{?rhel:BuildRequires: scylla-libstdc++73-static scylla-boost163-devel scylla-boost163-static scylla-antlr35-tool scylla-antlr35-C++-devel python34 scylla-gcc73-c++, scylla-python34-pyparsing20 yaml-cpp-static pystache python-setuptools}
+%{?rhel:BuildRequires: scylla-libstdc++73-static scylla-boost163-devel scylla-boost163-static scylla-antlr35-tool scylla-antlr35-C++-devel python36 scylla-gcc73-c++, scylla-python36-pyparsing20 yaml-cpp-static pystache python-setuptools}
 Requires:       scylla-conf systemd-libs hwloc PyYAML python-urwid pciutils pyparsing python-requests curl util-linux python-setuptools pciutils python3-pyudev mdadm xfsprogs
-%{?rhel:Requires: python34 python34-PyYAML kernel >= 3.10.0-514}
+%{?rhel:Requires: python36 python36-PyYAML kernel >= 3.10.0-514}
 %{?fedora:Requires: python3 python3-PyYAML}
 Conflicts:      abrt
 %ifarch x86_64
@@ -97,7 +97,7 @@ cflags="--cflags=${defines[*]}"
 %endif
 %if 0%{?rhel}
 . /etc/profile.d/scylla.sh
-python3.4 ./configure.py %{?configure_opt} --with=scylla --with=iotune --mode=release "$cflags" --static-boost --static-yaml-cpp --compiler=/opt/scylladb/bin/g++-7.3 --python python3.4 --ldflag=-Wl,-rpath=/opt/scylladb/lib64
+python3.6 ./configure.py %{?configure_opt} --with=scylla --with=iotune --mode=release "$cflags" --static-boost --static-yaml-cpp --compiler=/opt/scylladb/bin/g++-7.3 --python python3.6 --ldflag=-Wl,-rpath=/opt/scylladb/lib64
 %endif
 ninja-build %{?_smp_mflags} build/release/scylla build/release/iotune
 
@@ -201,7 +201,6 @@ rm -rf $RPM_BUILD_ROOT
 %{_prefix}/lib/scylla/api/api-doc/*
 %{_prefix}/lib/scylla/scyllatop/*
 %{_prefix}/lib/scylla/scylla_config_get.py
-%{_prefix}/lib/scylla/scylla_lib.sh
 %{_prefix}/lib/scylla/scylla_util.py
 %if 0%{?fedora} >= 27
 %{_prefix}/lib/scylla/scylla-gdb.py

docs/protocol-extensions.md

@@ -0,0 +1,82 @@
+Protocol extensions to the Cassandra Native Protocol
+====================================================
+
+This document specifies extensions to the protocol defined
+by Cassandra's native_protocol_v4.spec and native_protocol_v5.spec.
+The extensions are designed so that a driver supporting them can
+continue to interoperate with Cassandra and other compatible servers
+with no configuration needed; the driver can discover the extensions
+and enable them conditionally.
+
+An extension can be discovered by using the OPTIONS request; the
+returned SUPPORTED response will have zero or more options beginning
+with SCYLLA indicating extensions defined in this documented, in
+addition to options documented by Cassandra. How to use the extension
+is further explained in this document.
+
+# Intranode sharding
+
+This extension allows the driver to discover how Scylla internally
+partitions data among logical cores. It can then create at least
+one connection per logical core, and send queries directly to the
+logical core that will serve them, greatly improving load balancing
+and efficiency.
+
+To use the extension, send the OPTIONS message. The data is returned
+in the SUPPORTED message, as a set of key/value options. Numeric values
+are returned as their base-10 ASCII representation.
+
+The keys and values are:
+  - `SCYLLA_SHARD` is an integer, the zero-based shard number this connection
+    is connected to (for example, `3`).
+  - `SCYLLA_NR_SHARDS` is an integer containing the number of shards on this
+    node (for example, `12`). All shard numbers are smaller than this number.
+  - `SCYLLA_PARTITIONER` is a the fully-qualified name of the partitioner in use (i.e.
+    `org.apache.cassandra.partitioners.Murmur3Partitioner`).
+  - `SCYLLA_SHARDING_ALGORITHM` is the name of an algorithm used to select how
+    partitions are mapped into shards (described below)
+  - `SCYLLA_SHARDING_IGNORE_MSB` is an integer parameter to the algorithm (also
+    described below)
+
+Currently, one `SCYLLA_SHARDING_ALGORITHM` is defined,
+`biased-token-round-robin`. To apply the algorithm,
+perform the following steps (assuming infinite-precision arithmetic):
+
+  - subtract the minimum token value from the partition's token
+    in order to bias it: `biased_token = token - (-2**63)`
+  - shift `biased_token` left by `ignore_msb` bits, discarding any
+    bits beyond the 63rd:
+      `biased_token = (biased_token << SCYLLA_SHARDING_IGNORE_MSB) % (2**64)`
+  - multiply by `SCYLLA_NR_SHARDS` and perform a truncating division by 2**64:
+    `shard = (biased_token * SCYLLA_NR_SHARDS) / 2**64`
+
+(this apparently convoluted algorithm replaces a slow division instruction with
+a fast multiply instruction).
+
+in C with 128-bit arithmetic support, these operations can be efficiently
+performed in three steps:
+
+```c++
+    uint64_t biased_token = token + ((uint64_t)1 << 63);
+    biased_token <<= ignore_msb;
+    int shard = ((unsigned __int128)biased_token * nr_shards) >> 64;
+```
+
+In languages without 128-bit arithmetic support, use the following (this example
+is for Java):
+
+```Java
+    private int scyllaShardOf(long token) {
+        token += Long.MIN_VALUE;
+        token <<= ignoreMsb;
+        long tokLo = token & 0xffffffffL;
+        long tokHi = (token >>> 32) & 0xffffffffL;
+        long mul1 = tokLo * nrShards;
+        long mul2 = tokHi * nrShards;
+        long sum = (mul1 >>> 32) + mul2;
+        return (int)(sum >>> 32);
+    }
+```
+
+It is recommended that drivers open connections until they have at
+least one connection per shard, then close excess connections.

flat_mutation_reader.hh

@@ -449,9 +449,13 @@ GCC6_CONCEPT(requires requires(StopCondition stop, ConsumeMutationFragment consu
     { consume_mf(std::move(mf)) } -> void;
     { consume_eos() } -> future<>;
 })
-future<> consume_mutation_fragments_until(flat_mutation_reader& r, StopCondition&& stop,
-                                          ConsumeMutationFragment&& consume_mf, ConsumeEndOfStream&& consume_eos) {
-    return do_until([stop] { return stop(); }, [&r, stop, consume_mf, consume_eos] {
+future<> consume_mutation_fragments_until(
+        flat_mutation_reader& r,
+        StopCondition&& stop,
+        ConsumeMutationFragment&& consume_mf,
+        ConsumeEndOfStream&& consume_eos,
+        db::timeout_clock::time_point timeout) {
+    return do_until([stop] { return stop(); }, [&r, stop, consume_mf, consume_eos, timeout] {
         while (!r.is_buffer_empty()) {
             consume_mf(r.pop_mutation_fragment());
             if (stop()) {
@@ -461,7 +465,7 @@ future<> consume_mutation_fragments_until(flat_mutation_reader& r, StopCondition
         if (r.is_end_of_stream()) {
             return consume_eos();
         }
-        return r.fill_buffer();
+        return r.fill_buffer(timeout);
     });
 }
 

gms/endpoint_state.hh

@@ -129,26 +129,8 @@ public:
         update_is_normal();
     }
 
-    void apply_application_state(application_state key, versioned_value&& value) {
-        auto&& e = _application_state[key];
-        if (e.version < value.version) {
-            e = std::move(value);
-        }
-        update_is_normal();
-    }
-
-    void apply_application_state(application_state key, const versioned_value& value) {
-        auto&& e = _application_state[key];
-        if (e.version < value.version) {
-            e = value;
-        }
-        update_is_normal();
-    }
-
-    void apply_application_state(const endpoint_state& es) {
-        for (auto&& e : es._application_state) {
-            apply_application_state(e.first, e.second);
-        }
+    void add_application_state(const endpoint_state& es) {
+        _application_state = es._application_state;
         update_is_normal();
     }
 

gms/gossiper.cc

@@ -923,7 +923,7 @@ void gossiper::make_random_gossip_digest(std::vector<gossip_digest>& g_digests)
 future<> gossiper::replicate(inet_address ep, const endpoint_state& es) {
     return container().invoke_on_all([ep, es, orig = engine().cpu_id(), self = shared_from_this()] (gossiper& g) {
         if (engine().cpu_id() != orig) {
-            g.endpoint_state_map[ep].apply_application_state(es);
+            g.endpoint_state_map[ep].add_application_state(es);
         }
     });
 }
@@ -932,7 +932,7 @@ future<> gossiper::replicate(inet_address ep, const std::map<application_state,
     return container().invoke_on_all([ep, &src, &changed, orig = engine().cpu_id(), self = shared_from_this()] (gossiper& g) {
         if (engine().cpu_id() != orig) {
             for (auto&& key : changed) {
-                g.endpoint_state_map[ep].apply_application_state(key, src.at(key));
+                g.endpoint_state_map[ep].add_application_state(key, src.at(key));
             }
         }
     });
@@ -941,7 +941,7 @@ future<> gossiper::replicate(inet_address ep, const std::map<application_state,
 future<> gossiper::replicate(inet_address ep, application_state key, const versioned_value& value) {
     return container().invoke_on_all([ep, key, &value, orig = engine().cpu_id(), self = shared_from_this()] (gossiper& g) {
         if (engine().cpu_id() != orig) {
-            g.endpoint_state_map[ep].apply_application_state(key, value);
+            g.endpoint_state_map[ep].add_application_state(key, value);
         }
     });
 }
@@ -1168,11 +1168,13 @@ stdx::optional<endpoint_state> gossiper::get_endpoint_state_for_endpoint(inet_ad
     }
 }
 
-void gossiper::reset_endpoint_state_map() {
-    endpoint_state_map.clear();
+future<> gossiper::reset_endpoint_state_map() {
     _unreachable_endpoints.clear();
     _live_endpoints.clear();
     _live_endpoints_just_added.clear();
+    return container().invoke_on_all([] (gossiper& g) {
+        g.endpoint_state_map.clear();
+    });
 }
 
 std::unordered_map<inet_address, endpoint_state>& gms::gossiper::get_endpoint_states() {
@@ -1655,6 +1657,7 @@ void gossiper::maybe_initialize_local_state(int generation_nbr) {
     }
 }
 
+// Runs inside seastar::async context
 void gossiper::add_saved_endpoint(inet_address ep) {
     if (ep == get_broadcast_address()) {
         logger.debug("Attempt to add self as saved endpoint");
@@ -1680,6 +1683,7 @@ void gossiper::add_saved_endpoint(inet_address ep) {
     }
     ep_state.mark_dead();
     endpoint_state_map[ep] = ep_state;
+    replicate(ep, ep_state).get();
     _unreachable_endpoints[ep] = now();
     logger.trace("Adding saved endpoint {} {}", ep, ep_state.get_heart_beat_state().get_generation());
 }
@@ -1915,6 +1919,7 @@ void gossiper::mark_as_shutdown(const inet_address& endpoint) {
         auto& ep_state = *es;
         ep_state.add_application_state(application_state::STATUS, storage_service_value_factory().shutdown(true));
         ep_state.get_heart_beat_state().force_highest_possible_version_unsafe();
+        replicate(endpoint, ep_state).get();
         mark_dead(endpoint, ep_state);
         get_local_failure_detector().force_conviction(endpoint);
     }

gms/gossiper.hh

@@ -418,7 +418,7 @@ public:
     stdx::optional<endpoint_state> get_endpoint_state_for_endpoint(inet_address ep) const;
 
     // removes ALL endpoint states; should only be called after shadow gossip
-    void reset_endpoint_state_map();
+    future<> reset_endpoint_state_map();
 
     std::unordered_map<inet_address, endpoint_state>& get_endpoint_states();
 

imr/alloc.hh

@@ -84,6 +84,8 @@ template<typename Structure, typename CtxFactory>
 GCC6_CONCEPT(requires ContextFactory<CtxFactory>)
 class lsa_migrate_fn final : public migrate_fn_type, CtxFactory {
 public:
+    using structure = Structure;
+
     explicit lsa_migrate_fn(CtxFactory context_factory)
         : migrate_fn_type(1)
         , CtxFactory(std::move(context_factory))
@@ -201,8 +203,21 @@ public:
         /// arguments are passed to `T::size_when_serialized`.
         ///
         /// \return null pointer of type `uint8_t*`.
+        template<typename T, typename MigrateFn, typename... Args>
+        uint8_t* allocate(MigrateFn* migrate_fn, Args&&... args) noexcept {
+            static_assert(std::is_same_v<typename MigrateFn::structure, T>);
+            return do_allocate<T>(migrate_fn, std::forward<Args>(args)...);
+        }
+
+        template<typename T, typename MigrateFn, typename... Args>
+        auto allocate_nested(MigrateFn* migrate_fn, Args&&... args) noexcept {
+            static_assert(std::is_same_v<typename MigrateFn::structure, T>);
+            return do_allocate_nested<T>(migrate_fn, std::forward<Args>(args)...);
+        }
+
+    private:
         template<typename T, typename... Args>
-        uint8_t* allocate(migrate_fn_type* migrate_fn, Args&& ... args) noexcept {
+        uint8_t* do_allocate(migrate_fn_type* migrate_fn, Args&&... args) noexcept {
             auto size = T::size_when_serialized(std::forward<Args>(args)...);
             _parent.request(size, migrate_fn);
 
@@ -216,7 +231,7 @@ public:
         }
 
         template<typename T, typename... Args>
-        auto allocate_nested(migrate_fn_type* migrate_fn, Args&& ... args) noexcept {
+        auto do_allocate_nested(migrate_fn_type* migrate_fn, Args&& ... args) noexcept {
             auto n = _parent.request(0, migrate_fn);
             return T::get_sizer(continuation(_parent, n),
                                 std::forward<Args>(args)...);
@@ -244,15 +259,28 @@ public:
         /// to the buffer requested in the sizing phase. Arguments are passed
         /// to `T::serialize`.
         /// \return pointer to the IMR object
+        template<typename T, typename MigrateFn, typename... Args>
+        uint8_t* allocate(MigrateFn* migrate_fn, Args&&... args) noexcept {
+            static_assert(std::is_same_v<typename MigrateFn::structure, T>);
+            return do_allocate<T>(migrate_fn, std::forward<Args>(args)...);
+        }
+
+        template<typename T, typename MigrateFn, typename... Args>
+        auto allocate_nested(MigrateFn* migrate_fn, Args&&... args) noexcept {
+            static_assert(std::is_same_v<typename MigrateFn::structure, T>);
+            return do_allocate_nested<T>(migrate_fn, std::forward<Args>(args)...);
+        }
+
+    private:
         template<typename T, typename... Args>
-        uint8_t* allocate(migrate_fn_type* migrate_fn, Args&& ... args) noexcept {
+        uint8_t* do_allocate(migrate_fn_type* migrate_fn, Args&&... args) noexcept {
             auto ptr = _parent.next_object();
             T::serialize(ptr, std::forward<Args>(args)...);
             return ptr;
         }
 
         template<typename T, typename... Args>
-        auto allocate_nested(migrate_fn_type*, Args&& ... args) noexcept {
+        auto do_allocate_nested(migrate_fn_type*, Args&& ... args) noexcept {
             auto ptr = _parent.next_object();
             return T::get_serializer(ptr,
                                      continuation(ptr),

imr/utils.hh

@@ -61,8 +61,12 @@ private:
 public:
     object_context(const uint8_t*, State... state) : _state { state... } { }
     template<typename Tag, typename... Args>
-    Context context_for(const uint8_t* ptr, Args&&... args) const noexcept {
-        return create(ptr, std::index_sequence_for<State...>());
+    auto context_for(const uint8_t* ptr, Args&&... args) const noexcept {
+        if constexpr (std::is_same_v<Tag, basic_object::tags::back_pointer>) {
+            return no_context_t();
+        } else {
+            return create(ptr, std::index_sequence_for<State...>());
+        }
     }
 };
 
@@ -92,7 +96,7 @@ public:
                         imr::member<tags::back_pointer, imr::tagged_type<tags::back_pointer, imr::pod<basic_object*>>>,
                         imr::member<tags::object, Structure>
                       >;
-
+    static constexpr size_t size_overhead = sizeof(basic_object*);
 private:
     explicit object(uint8_t* ptr) noexcept
         : basic_object(ptr)
@@ -158,13 +162,22 @@ public:
     }
 
     /// Create an IMR objects
-    template<typename Writer>
+    template<typename Writer, typename MigrateFn>
     GCC6_CONCEPT(requires WriterAllocator<Writer, Structure>)
     static object make(Writer&& object_writer,
-                       allocation_strategy::migrate_fn migrate = &imr::alloc::default_lsa_migrate_fn<structure>::migrate_fn) {
+                       MigrateFn* migrate = &imr::alloc::default_lsa_migrate_fn<structure>::migrate_fn) {
+        static_assert(std::is_same_v<typename MigrateFn::structure, structure>);
+        return do_make(std::forward<Writer>(object_writer), migrate);
+    }
+private:
+    template<typename Writer>
+    GCC6_CONCEPT(requires WriterAllocator<Writer, Structure>)
+    static object do_make(Writer&& object_writer, allocation_strategy::migrate_fn migrate) {
         struct alloc_deleter {
+            size_t _size;
+
             void operator()(uint8_t* ptr) {
-                current_allocator().free(ptr);
+                current_allocator().free(ptr, _size);
             }
         };
         using alloc_unique_ptr = std::unique_ptr<uint8_t[], alloc_deleter>;
@@ -176,7 +189,7 @@ public:
         auto& alloc = current_allocator();
         alloc::object_allocator allocator(alloc);
         auto obj_size = structure::size_when_serialized(writer, allocator.get_sizer());
-        auto ptr = alloc_unique_ptr(static_cast<uint8_t*>(alloc.alloc(migrate, obj_size, 1)));
+        auto ptr = alloc_unique_ptr(static_cast<uint8_t*>(alloc.alloc(migrate, obj_size, 1)), alloc_deleter { obj_size });
         allocator.allocate_all();
         structure::serialize(ptr.get(), writer, allocator.get_serializer());
         return object(ptr.release());

install-dependencies.sh

@@ -42,5 +42,5 @@ elif [ "$ID" = "fedora" ]; then
     yum install -y yaml-cpp-devel thrift-devel antlr3-tool antlr3-C++-devel jsoncpp-devel snappy-devel
 elif [ "$ID" = "centos" ]; then
     yum install -y yaml-cpp-devel thrift-devel scylla-antlr35-tool scylla-antlr35-C++-devel jsoncpp-devel snappy-devel scylla-boost163-static scylla-python34-pyparsing20 systemd-devel
-    echo -e "Configure example:\n\tpython3.4 ./configure.py --enable-dpdk --mode=release --static-boost --compiler=/opt/scylladb/bin/g++-7.3 --python python3.4 --ldflag=-Wl,-rpath=/opt/scylladb/lib64 --cflags=-I/opt/scylladb/include --with-antlr3=/opt/scylladb/bin/antlr3"
+    echo -e "Configure example:\n\tpython3.6 ./configure.py --enable-dpdk --mode=release --static-boost --compiler=/opt/scylladb/bin/g++-7.3 --python python3.6 --ldflag=-Wl,-rpath=/opt/scylladb/lib64 --cflags=-I/opt/scylladb/include --with-antlr3=/opt/scylladb/bin/antlr3"
 fi

json.cc

@@ -0,0 +1,80 @@
+/*
+ * Copyright (C) 2018 ScyllaDB
+ */
+
+/*
+ * This file is part of Scylla.
+ *
+ * Scylla is free software: you can redistribute it and/or modify
+ * it under the terms of the GNU Affero General Public License as published by
+ * the Free Software Foundation, either version 3 of the License, or
+ * (at your option) any later version.
+ *
+ * Scylla is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with Scylla.  If not, see <http://www.gnu.org/licenses/>.
+ */
+
+#include "json.hh"
+
+namespace seastar {
+namespace json {
+
+static inline bool is_control_char(char c) {
+    return c >= 0 && c <= 0x1F;
+}
+
+static inline bool needs_escaping(const sstring& s) {
+    return std::any_of(s.begin(), s.end(), [](char c) {return is_control_char(c) || c == '"' || c == '\\';});
+}
+
+sstring value_to_quoted_string(const sstring& value) {
+    if (!needs_escaping(value)) {
+        return sprint("\"%s\"", value);
+    }
+    std::ostringstream oss;
+    oss << std::hex << std::uppercase << std::setfill('0');
+    oss.put('"');
+    for (char c : value) {
+        switch (c) {
+        case '"':
+            oss.put('\\').put('"');
+            break;
+        case '\\':
+            oss.put('\\').put('\\');
+            break;
+        case '\b':
+            oss.put('\\').put('b');
+            break;
+        case '\f':
+            oss.put('\\').put('f');
+            break;
+        case '\n':
+            oss.put('\\').put('n');
+            break;
+        case '\r':
+            oss.put('\\').put('r');
+            break;
+        case '\t':
+            oss.put('\\').put('t');
+            break;
+        default:
+            if (is_control_char(c)) {
+                oss.put('\\').put('u') << std::setw(4) << static_cast<int>(c);
+            } else {
+                oss.put(c);
+            }
+            break;
+        }
+    }
+    oss.put('"');
+    return oss.str();
+}
+
+}
+
+}

json.hh

@@ -95,6 +95,8 @@ inline std::map<sstring, sstring> to_map(const sstring& raw) {
     return to_map(raw, std::map<sstring, sstring>());
 }
 
+sstring value_to_quoted_string(const sstring& value);
+
 }
 
 }

keys.hh

@@ -748,6 +748,10 @@ public:
     static const compound& get_compound_type(const schema& s) {
         return s.clustering_key_prefix_type();
     }
+
+    static clustering_key_prefix_view make_empty() {
+        return { bytes_view() };
+    }
 };
 
 class clustering_key_prefix : public prefix_compound_wrapper<clustering_key_prefix, clustering_key_prefix_view, clustering_key> {

locator/abstract_replication_strategy.cc

@@ -119,9 +119,17 @@ insert_token_range_to_sorted_container_while_unwrapping(
         const dht::token& tok,
         dht::token_range_vector& ret) {
     if (prev_tok < tok) {
-        ret.emplace_back(
-                dht::token_range::bound(prev_tok, false),
-                dht::token_range::bound(tok, true));
+        auto pos = ret.end();
+        if (!ret.empty() && !std::prev(pos)->end()) {
+            // We inserted a wrapped range (a, b] previously as
+            // (-inf, b], (a, +inf). So now we insert in the next-to-last
+            // position to keep the last range (a, +inf) at the end.
+            pos = std::prev(pos);
+        }
+        ret.insert(pos,
+                dht::token_range{
+                        dht::token_range::bound(prev_tok, false),
+                        dht::token_range::bound(tok, true)});
     } else {
         ret.emplace_back(
                 dht::token_range::bound(prev_tok, false),
@@ -164,6 +172,30 @@ abstract_replication_strategy::get_primary_ranges(inet_address ep) {
     return ret;
 }
 
+dht::token_range_vector
+abstract_replication_strategy::get_primary_ranges_within_dc(inet_address ep) {
+    dht::token_range_vector ret;
+    sstring local_dc = _snitch->get_datacenter(ep);
+    std::unordered_set<inet_address> local_dc_nodes = _token_metadata.get_topology().get_datacenter_endpoints().at(local_dc);
+    auto prev_tok = _token_metadata.sorted_tokens().back();
+    for (auto tok : _token_metadata.sorted_tokens()) {
+        auto&& eps = calculate_natural_endpoints(tok, _token_metadata);
+        // Unlike get_primary_ranges() which checks if ep is the first
+        // owner of this range, here we check if ep is the first just
+        // among nodes which belong to the local dc of ep.
+        for (auto& e : eps) {
+            if (local_dc_nodes.count(e)) {
+                if (e == ep) {
+                    insert_token_range_to_sorted_container_while_unwrapping(prev_tok, tok, ret);
+                }
+                break;
+            }
+        }
+        prev_tok = tok;
+    }
+    return ret;
+}
+
 std::unordered_multimap<inet_address, dht::token_range>
 abstract_replication_strategy::get_address_ranges(token_metadata& tm) const {
     std::unordered_multimap<inet_address, dht::token_range> ret;

locator/abstract_replication_strategy.hh

@@ -113,6 +113,10 @@ public:
     // This function is the analogue of Origin's
     // StorageService.getPrimaryRangesForEndpoint().
     dht::token_range_vector get_primary_ranges(inet_address ep);
+    // get_primary_ranges_within_dc() is similar to get_primary_ranges()
+    // except it assigns a primary node for each range within each dc,
+    // instead of one node globally.
+    dht::token_range_vector get_primary_ranges_within_dc(inet_address ep);
 
     std::unordered_multimap<inet_address, dht::token_range> get_address_ranges(token_metadata& tm) const;
 

main.cc

@@ -763,8 +763,11 @@ int main(int ac, char** av) {
                 return service::get_local_storage_service().drain_on_shutdown();
             });
 
-            engine().at_exit([] {
-                return view_builder.stop();
+            engine().at_exit([cfg] {
+                if (cfg->view_building()) {
+                    return view_builder.stop();
+                }
+                return make_ready_future<>();
             });
 
             engine().at_exit([&db] {

message/messaging_service.cc

@@ -262,11 +262,12 @@ void messaging_service::start_listen() {
     // FIXME: we don't set so.tcp_nodelay, because we can't tell at this point whether the connection will come from a
     //        local or remote datacenter, and whether or not the connection will be used for gossip. We can fix
     //        the first by wrapping its server_socket, but not the second.
+    auto limits = rpc_resource_limits(_mcfg.rpc_memory_limit);
     if (!_server[0]) {
         auto listen = [&] (const gms::inet_address& a) {
             auto addr = ipv4_addr{a.raw_addr(), _port};
             return std::unique_ptr<rpc_protocol_server_wrapper>(new rpc_protocol_server_wrapper(*_rpc,
-                    so, addr, rpc_resource_limits(_mcfg.rpc_memory_limit)));
+                    so, addr, limits));
         };
         _server[0] = listen(_listen_address);
         if (listen_to_bc) {
@@ -277,7 +278,7 @@ void messaging_service::start_listen() {
     if (!_server_tls[0]) {
         auto listen = [&] (const gms::inet_address& a) {
             return std::unique_ptr<rpc_protocol_server_wrapper>(
-                    [this, &so, &a] () -> std::unique_ptr<rpc_protocol_server_wrapper>{
+                    [this, &so, &a, limits] () -> std::unique_ptr<rpc_protocol_server_wrapper>{
                 if (_encrypt_what == encrypt_what::none) {
                     return nullptr;
                 }
@@ -285,7 +286,7 @@ void messaging_service::start_listen() {
                 lo.reuse_address = true;
                 auto addr = make_ipv4_address(ipv4_addr{a.raw_addr(), _ssl_port});
                 return std::make_unique<rpc_protocol_server_wrapper>(*_rpc,
-                        so, seastar::tls::listen(_credentials, addr, lo));
+                        so, seastar::tls::listen(_credentials, addr, lo), limits);
             }());
         };
         _server_tls[0] = listen(_listen_address);

mutation_fragment.hh

@@ -125,7 +125,7 @@ public:
         return _ck.equal(s, other._ck)
                && _t == other._t
                && _marker == other._marker
-               && _cells.equal(column_kind::static_column, s, other._cells, s);
+               && _cells.equal(column_kind::regular_column, s, other._cells, s);
     }
 
     friend std::ostream& operator<<(std::ostream& os, const clustering_row& row);

mutation_partition.cc

@@ -144,7 +144,14 @@ mutation_partition::mutation_partition(const schema& s, const mutation_partition
         , _static_row(s, column_kind::static_column, x._static_row)
         , _static_row_continuous(x._static_row_continuous)
         , _rows()
-        , _row_tombstones(x._row_tombstones) {
+        , _row_tombstones(x._row_tombstones)
+#ifdef SEASTAR_DEBUG
+        , _schema_version(s.version())
+#endif
+{
+#ifdef SEASTAR_DEBUG
+    assert(x._schema_version == _schema_version);
+#endif
     auto cloner = [&s] (const auto& x) {
         return current_allocator().construct<rows_entry>(s, x);
     };
@@ -157,7 +164,14 @@ mutation_partition::mutation_partition(const mutation_partition& x, const schema
         , _static_row(schema, column_kind::static_column, x._static_row)
         , _static_row_continuous(x._static_row_continuous)
         , _rows()
-        , _row_tombstones(x._row_tombstones, range_tombstone_list::copy_comparator_only()) {
+        , _row_tombstones(x._row_tombstones, range_tombstone_list::copy_comparator_only())
+#ifdef SEASTAR_DEBUG
+        , _schema_version(schema.version())
+#endif
+{
+#ifdef SEASTAR_DEBUG
+    assert(x._schema_version == _schema_version);
+#endif
     try {
         for(auto&& r : ck_ranges) {
             for (const rows_entry& e : x.range(schema, r)) {
@@ -180,7 +194,13 @@ mutation_partition::mutation_partition(mutation_partition&& x, const schema& sch
     , _static_row_continuous(x._static_row_continuous)
     , _rows(std::move(x._rows))
     , _row_tombstones(std::move(x._row_tombstones))
+#ifdef SEASTAR_DEBUG
+    , _schema_version(schema.version())
+#endif
 {
+#ifdef SEASTAR_DEBUG
+    assert(x._schema_version == _schema_version);
+#endif
     {
         auto deleter = current_deleter<rows_entry>();
         auto it = _rows.begin();
@@ -220,6 +240,7 @@ mutation_partition::operator=(mutation_partition&& x) noexcept {
 }
 
 void mutation_partition::ensure_last_dummy(const schema& s) {
+    check_schema(s);
     if (_rows.empty() || !_rows.rbegin()->is_last_dummy()) {
         _rows.insert_before(_rows.end(),
             *current_allocator().construct<rows_entry>(s, rows_entry::last_dummy_tag(), is_continuous::yes));
@@ -276,11 +297,16 @@ void deletable_row::apply(const schema& s, clustering_row cr) {
 
 void
 mutation_partition::apply(const schema& s, const mutation_fragment& mf) {
+    check_schema(s);
     mutation_fragment_applier applier{s, *this};
     mf.visit(applier);
 }
 
 void mutation_partition::apply_monotonically(const schema& s, mutation_partition&& p, cache_tracker* tracker) {
+#ifdef SEASTAR_DEBUG
+    assert(s.version() == _schema_version);
+    assert(p._schema_version == _schema_version);
+#endif
     _tombstone.apply(p._tombstone);
     _row_tombstones.apply_monotonically(s, std::move(p._row_tombstones));
     _static_row.apply_monotonically(s, column_kind::static_column, std::move(p._static_row));
@@ -356,6 +382,7 @@ void mutation_partition::apply_weak(const schema& s, mutation_partition&& p) {
 
 tombstone
 mutation_partition::range_tombstone_for_row(const schema& schema, const clustering_key& key) const {
+    check_schema(schema);
     tombstone t = _tombstone;
     if (!_row_tombstones.empty()) {
         auto found = _row_tombstones.search_tombstone_covering(schema, key);
@@ -366,6 +393,7 @@ mutation_partition::range_tombstone_for_row(const schema& schema, const clusteri
 
 row_tombstone
 mutation_partition::tombstone_for_row(const schema& schema, const clustering_key& key) const {
+    check_schema(schema);
     row_tombstone t = row_tombstone(range_tombstone_for_row(schema, key));
 
     auto j = _rows.find(key, rows_entry::compare(schema));
@@ -378,6 +406,7 @@ mutation_partition::tombstone_for_row(const schema& schema, const clustering_key
 
 row_tombstone
 mutation_partition::tombstone_for_row(const schema& schema, const rows_entry& e) const {
+    check_schema(schema);
     row_tombstone t = e.row().deleted_at();
     t.apply(range_tombstone_for_row(schema, e.key()));
     return t;
@@ -385,6 +414,7 @@ mutation_partition::tombstone_for_row(const schema& schema, const rows_entry& e)
 
 void
 mutation_partition::apply_row_tombstone(const schema& schema, clustering_key_prefix prefix, tombstone t) {
+    check_schema(schema);
     assert(!prefix.is_full(schema));
     auto start = prefix;
     _row_tombstones.apply(schema, {std::move(start), std::move(prefix), std::move(t)});
@@ -392,11 +422,13 @@ mutation_partition::apply_row_tombstone(const schema& schema, clustering_key_pre
 
 void
 mutation_partition::apply_row_tombstone(const schema& schema, range_tombstone rt) {
+    check_schema(schema);
     _row_tombstones.apply(schema, std::move(rt));
 }
 
 void
 mutation_partition::apply_delete(const schema& schema, const clustering_key_prefix& prefix, tombstone t) {
+    check_schema(schema);
     if (prefix.is_empty(schema)) {
         apply(t);
     } else if (prefix.is_full(schema)) {
@@ -408,6 +440,7 @@ mutation_partition::apply_delete(const schema& schema, const clustering_key_pref
 
 void
 mutation_partition::apply_delete(const schema& schema, range_tombstone rt) {
+    check_schema(schema);
     if (range_tombstone::is_single_clustering_row_tombstone(schema, rt.start, rt.start_kind, rt.end, rt.end_kind)) {
         apply_delete(schema, std::move(rt.start), std::move(rt.tomb));
         return;
@@ -417,6 +450,7 @@ mutation_partition::apply_delete(const schema& schema, range_tombstone rt) {
 
 void
 mutation_partition::apply_delete(const schema& schema, clustering_key&& prefix, tombstone t) {
+    check_schema(schema);
     if (prefix.is_empty(schema)) {
         apply(t);
     } else if (prefix.is_full(schema)) {
@@ -428,6 +462,7 @@ mutation_partition::apply_delete(const schema& schema, clustering_key&& prefix,
 
 void
 mutation_partition::apply_delete(const schema& schema, clustering_key_prefix_view prefix, tombstone t) {
+    check_schema(schema);
     if (prefix.is_empty(schema)) {
         apply(t);
     } else if (prefix.is_full(schema)) {
@@ -451,12 +486,14 @@ void mutation_partition::insert_row(const schema& s, const clustering_key& key,
 }
 
 void mutation_partition::insert_row(const schema& s, const clustering_key& key, const deletable_row& row) {
+    check_schema(s);
     auto e = current_allocator().construct<rows_entry>(s, key, row);
     _rows.insert(_rows.end(), *e, rows_entry::compare(s));
 }
 
 const row*
 mutation_partition::find_row(const schema& s, const clustering_key& key) const {
+    check_schema(s);
     auto i = _rows.find(key, rows_entry::compare(s));
     if (i == _rows.end()) {
         return nullptr;
@@ -466,6 +503,7 @@ mutation_partition::find_row(const schema& s, const clustering_key& key) const {
 
 deletable_row&
 mutation_partition::clustered_row(const schema& s, clustering_key&& key) {
+    check_schema(s);
     auto i = _rows.find(key, rows_entry::compare(s));
     if (i == _rows.end()) {
         auto e = current_allocator().construct<rows_entry>(std::move(key));
@@ -477,6 +515,7 @@ mutation_partition::clustered_row(const schema& s, clustering_key&& key) {
 
 deletable_row&
 mutation_partition::clustered_row(const schema& s, const clustering_key& key) {
+    check_schema(s);
     auto i = _rows.find(key, rows_entry::compare(s));
     if (i == _rows.end()) {
         auto e = current_allocator().construct<rows_entry>(key);
@@ -488,6 +527,7 @@ mutation_partition::clustered_row(const schema& s, const clustering_key& key) {
 
 deletable_row&
 mutation_partition::clustered_row(const schema& s, clustering_key_view key) {
+    check_schema(s);
     auto i = _rows.find(key, rows_entry::compare(s));
     if (i == _rows.end()) {
         auto e = current_allocator().construct<rows_entry>(key);
@@ -499,6 +539,7 @@ mutation_partition::clustered_row(const schema& s, clustering_key_view key) {
 
 deletable_row&
 mutation_partition::clustered_row(const schema& s, position_in_partition_view pos, is_dummy dummy, is_continuous continuous) {
+    check_schema(s);
     auto i = _rows.find(pos, rows_entry::compare(s));
     if (i == _rows.end()) {
         auto e = current_allocator().construct<rows_entry>(s, pos, dummy, continuous);
@@ -510,6 +551,7 @@ mutation_partition::clustered_row(const schema& s, position_in_partition_view po
 
 mutation_partition::rows_type::const_iterator
 mutation_partition::lower_bound(const schema& schema, const query::clustering_range& r) const {
+    check_schema(schema);
     if (!r.start()) {
         return std::cbegin(_rows);
     }
@@ -518,6 +560,7 @@ mutation_partition::lower_bound(const schema& schema, const query::clustering_ra
 
 mutation_partition::rows_type::const_iterator
 mutation_partition::upper_bound(const schema& schema, const query::clustering_range& r) const {
+    check_schema(schema);
     if (!r.end()) {
         return std::cend(_rows);
     }
@@ -526,6 +569,7 @@ mutation_partition::upper_bound(const schema& schema, const query::clustering_ra
 
 boost::iterator_range<mutation_partition::rows_type::const_iterator>
 mutation_partition::range(const schema& schema, const query::clustering_range& r) const {
+    check_schema(schema);
     return boost::make_iterator_range(lower_bound(schema, r), upper_bound(schema, r));
 }
 
@@ -562,6 +606,7 @@ mutation_partition::upper_bound(const schema& schema, const query::clustering_ra
 template<typename Func>
 void mutation_partition::for_each_row(const schema& schema, const query::clustering_range& row_range, bool reversed, Func&& func) const
 {
+    check_schema(schema);
     auto r = range(schema, row_range);
     if (!reversed) {
         for (const auto& e : r) {
@@ -778,6 +823,7 @@ bool has_any_live_data(const schema& s, column_kind kind, const row& cells, tomb
 
 void
 mutation_partition::query_compacted(query::result::partition_writer& pw, const schema& s, uint32_t limit) const {
+    check_schema(s);
     const query::partition_slice& slice = pw.slice();
     max_timestamp max_ts{pw.last_modified()};
 
@@ -996,6 +1042,10 @@ bool mutation_partition::equal(const schema& s, const mutation_partition& p) con
 }
 
 bool mutation_partition::equal(const schema& this_schema, const mutation_partition& p, const schema& p_schema) const {
+#ifdef SEASTAR_DEBUG
+    assert(_schema_version == this_schema.version());
+    assert(p._schema_version == p_schema.version());
+#endif
     if (_tombstone != p._tombstone) {
         return false;
     }
@@ -1095,7 +1145,7 @@ row::apply_monotonically(const column_definition& column, atomic_cell_or_collect
     if (_type == storage_type::vector && id < max_vector_size) {
         if (id >= _storage.vector.v.size()) {
             _storage.vector.v.resize(id);
-            _storage.vector.v.emplace_back(cell_and_hash{std::move(value), std::move(hash)});
+            _storage.vector.v.emplace_back(std::move(value), std::move(hash));
             _storage.vector.present.set(id);
             _size++;
         } else if (auto& cell_and_hash = _storage.vector.v[id]; !bool(cell_and_hash.cell)) {
@@ -1124,6 +1174,7 @@ row::apply_monotonically(const column_definition& column, atomic_cell_or_collect
 void
 row::append_cell(column_id id, atomic_cell_or_collection value) {
     if (_type == storage_type::vector && id < max_vector_size) {
+        assert(_storage.vector.v.size() <= id);
         _storage.vector.v.resize(id);
         _storage.vector.v.emplace_back(cell_and_hash{std::move(value), cell_hash_opt()});
         _storage.vector.present.set(id);
@@ -1162,7 +1213,7 @@ row::find_cell(column_id id) const {
 size_t row::external_memory_usage(const schema& s, column_kind kind) const {
     size_t mem = 0;
     if (_type == storage_type::vector) {
-        mem += _storage.vector.v.external_memory_usage();
+        mem += _storage.vector.v.used_space_external_memory_usage();
         column_id id = 0;
         for (auto&& c_a_h : _storage.vector.v) {
             auto& cdef = s.column_at(kind, id++);
@@ -1188,6 +1239,7 @@ size_t rows_entry::memory_usage(const schema& s) const {
 }
 
 size_t mutation_partition::external_memory_usage(const schema& s) const {
+    check_schema(s);
     size_t sum = 0;
     sum += static_row().external_memory_usage(s, column_kind::static_column);
     for (auto& clr : clustered_rows()) {
@@ -1206,6 +1258,7 @@ void mutation_partition::trim_rows(const schema& s,
     const std::vector<query::clustering_range>& row_ranges,
     Func&& func)
 {
+    check_schema(s);
     static_assert(std::is_same<stop_iteration, std::result_of_t<Func(rows_entry&)>>::value, "Bad func signature");
 
     stop_iteration stop = stop_iteration::no;
@@ -1250,6 +1303,7 @@ uint32_t mutation_partition::do_compact(const schema& s,
     uint32_t row_limit,
     can_gc_fn& can_gc)
 {
+    check_schema(s);
     assert(row_limit > 0);
 
     auto gc_before = saturating_subtract(query_time, s.gc_grace_seconds());
@@ -1315,12 +1369,14 @@ mutation_partition::compact_for_query(
     bool reverse,
     uint32_t row_limit)
 {
+    check_schema(s);
     return do_compact(s, query_time, row_ranges, reverse, row_limit, always_gc);
 }
 
 void mutation_partition::compact_for_compaction(const schema& s,
     can_gc_fn& can_gc, gc_clock::time_point compaction_time)
 {
+    check_schema(s);
     static const std::vector<query::clustering_range> all_rows = {
         query::clustering_range::make_open_ended_both_sides()
     };
@@ -1354,11 +1410,13 @@ row::is_live(const schema& s, column_kind kind, tombstone base_tombstone, gc_clo
 
 bool
 mutation_partition::is_static_row_live(const schema& s, gc_clock::time_point query_time) const {
+    check_schema(s);
     return has_any_live_data(s, column_kind::static_column, static_row(), _tombstone, query_time);
 }
 
 size_t
 mutation_partition::live_row_count(const schema& s, gc_clock::time_point query_time) const {
+    check_schema(s);
     size_t count = 0;
 
     for (const rows_entry& e : non_dummy_rows()) {
@@ -1396,12 +1454,17 @@ row::row(const schema& s, column_kind kind, const row& o)
     if (_type == storage_type::vector) {
         auto& other_vec = o._storage.vector;
         auto& vec = *new (&_storage.vector) vector_storage;
-        vec.present = other_vec.present;
-        vec.v.reserve(other_vec.v.size());
-        column_id id = 0;
-        for (auto& cell : other_vec.v) {
-            auto& cdef = s.column_at(kind, id++);
-            vec.v.emplace_back(cell_and_hash { cell.cell.copy(*cdef.type), cell.hash });
+        try {
+            vec.present = other_vec.present;
+            vec.v.reserve(other_vec.v.size());
+            column_id id = 0;
+            for (auto& cell : other_vec.v) {
+                auto& cdef = s.column_at(kind, id++);
+                vec.v.emplace_back(cell_and_hash{cell.cell.copy(*cdef.type), cell.hash});
+            }
+        } catch (...) {
+            _storage.vector.~vector_storage();
+            throw;
         }
     } else {
         auto cloner = [&] (const auto& x) {
@@ -1699,6 +1762,7 @@ row row::difference(const schema& s, column_kind kind, const row& other) const
 
 mutation_partition mutation_partition::difference(schema_ptr s, const mutation_partition& other) const
 {
+    check_schema(*s);
     mutation_partition mp(s);
     if (_tombstone > other._tombstone) {
         mp.apply(_tombstone);
@@ -1729,6 +1793,7 @@ mutation_partition mutation_partition::difference(schema_ptr s, const mutation_p
 }
 
 void mutation_partition::accept(const schema& s, mutation_partition_visitor& v) const {
+    check_schema(s);
     v.accept_partition_tombstone(_tombstone);
     _static_row.for_each_cell([&] (column_id id, const atomic_cell_or_collection& cell) {
         const column_definition& def = s.static_column_at(id);
@@ -1812,9 +1877,10 @@ void mutation_querier::query_static_row(const row& r, tombstone current_tombston
         } else if (_short_reads_allowed) {
             seastar::measuring_output_stream stream;
             ser::qr_partition__static_row__cells<seastar::measuring_output_stream> out(stream, { });
+            auto start = stream.size();
             get_compacted_row_slice(_schema, slice, column_kind::static_column,
-                                    r, slice.static_columns, _static_cells_wr);
-            _memory_accounter.update(stream.size());
+                                    r, slice.static_columns, out);
+            _memory_accounter.update(stream.size() - start);
         }
         if (_pw.requested_digest()) {
             max_timestamp max_ts{_pw.last_modified()};
@@ -1875,8 +1941,9 @@ stop_iteration mutation_querier::consume(clustering_row&& cr, row_tombstone curr
     } else if (_short_reads_allowed) {
         seastar::measuring_output_stream stream;
         ser::qr_partition__rows<seastar::measuring_output_stream> out(stream, { });
+        auto start = stream.size();
         write_row(out);
-        stop = _memory_accounter.update_and_check(stream.size());
+        stop = _memory_accounter.update_and_check(stream.size() - start);
     }
 
     _live_clustering_rows++;
@@ -2160,6 +2227,9 @@ mutation_partition::mutation_partition(mutation_partition::incomplete_tag, const
     , _static_row_continuous(!s.has_static_columns())
     , _rows()
     , _row_tombstones(s)
+#ifdef SEASTAR_DEBUG
+    , _schema_version(s.version())
+#endif
 {
     _rows.insert_before(_rows.end(),
         *current_allocator().construct<rows_entry>(s, rows_entry::last_dummy_tag(), is_continuous::no));
@@ -2191,6 +2261,7 @@ void mutation_partition::make_fully_continuous() {
 }
 
 clustering_interval_set mutation_partition::get_continuity(const schema& s, is_continuous cont) const {
+    check_schema(s);
     clustering_interval_set result;
     auto i = _rows.begin();
     auto prev_pos = position_in_partition::before_all_clustered_rows();
@@ -2240,6 +2311,7 @@ stop_iteration mutation_partition::clear_gently(cache_tracker* tracker) noexcept
 
 bool
 mutation_partition::check_continuity(const schema& s, const position_range& r, is_continuous cont) const {
+    check_schema(s);
     auto less = rows_entry::compare(s);
     auto i = _rows.lower_bound(r.start(), less);
     auto end = _rows.lower_bound(r.end(), less);

mutation_partition.hh

@@ -74,6 +74,15 @@ using cell_hash_opt = seastar::optimized_optional<cell_hash>;
 struct cell_and_hash {
     atomic_cell_or_collection cell;
     mutable cell_hash_opt hash;
+
+    cell_and_hash() = default;
+    cell_and_hash(cell_and_hash&&) noexcept = default;
+    cell_and_hash& operator=(cell_and_hash&&) noexcept = default;
+
+    cell_and_hash(atomic_cell_or_collection&& cell, cell_hash_opt hash)
+        : cell(std::move(cell))
+        , hash(hash)
+    { }
 };
 
 //
@@ -896,6 +905,9 @@ private:
     // Contains only strict prefixes so that we don't have to lookup full keys
     // in both _row_tombstones and _rows.
     range_tombstone_list _row_tombstones;
+#ifdef SEASTAR_DEBUG
+    table_schema_version _schema_version;
+#endif
 
     friend class mutation_partition_applier;
     friend class converting_mutation_partition_applier;
@@ -910,10 +922,16 @@ public:
     mutation_partition(schema_ptr s)
         : _rows()
         , _row_tombstones(*s)
+#ifdef SEASTAR_DEBUG
+        , _schema_version(s->version())
+#endif
     { }
     mutation_partition(mutation_partition& other, copy_comparators_only)
         : _rows()
         , _row_tombstones(other._row_tombstones, range_tombstone_list::copy_comparator_only())
+#ifdef SEASTAR_DEBUG
+        , _schema_version(other._schema_version)
+#endif
     { }
     mutation_partition(mutation_partition&&) = default;
     mutation_partition(const schema& s, const mutation_partition&);
@@ -1113,6 +1131,12 @@ private:
     template<typename Func>
     void for_each_row(const schema& schema, const query::clustering_range& row_range, bool reversed, Func&& func) const;
     friend class counter_write_query_result_builder;
+
+    void check_schema(const schema& s) const {
+#ifdef SEASTAR_DEBUG
+        assert(s.version() == _schema_version);
+#endif
+    }
 };
 
 inline

partition_version.cc

@@ -312,7 +312,7 @@ partition_version& partition_entry::add_version(const schema& s, cache_tracker*
 
 void partition_entry::apply(const schema& s, const mutation_partition& mp, const schema& mp_schema)
 {
-    apply(s, mutation_partition(s, mp), mp_schema);
+    apply(s, mutation_partition(mp_schema, mp), mp_schema);
 }
 
 void partition_entry::apply(const schema& s, mutation_partition&& mp, const schema& mp_schema)
@@ -457,7 +457,10 @@ coroutine partition_entry::apply_to_incomplete(const schema& s,
         pe.upgrade(pe_schema.shared_from_this(), s.shared_from_this(), pe_cleaner, no_cache_tracker);
     }
 
-    bool can_move = !pe._snapshot;
+    // When preemptible, later memtable reads could start using the snapshot before
+    // snapshot's writes are made visible in cache, which would cause them to miss those writes.
+    // So we cannot allow erasing when preemptible.
+    bool can_move = !preemptible && !pe._snapshot;
 
     auto src_snp = pe.read(reg, pe_cleaner, s.shared_from_this(), no_cache_tracker);
     lw_shared_ptr<partition_snapshot> prev_snp;

position_in_partition.hh

@@ -273,6 +273,11 @@ public:
         return is_partition_end() || (_ck && _ck->is_empty(s) && _bound_weight > 0);
     }
 
+    bool is_before_all_clustered_rows(const schema& s) const {
+        return _type < partition_region::clustered
+               || (_type == partition_region::clustered && _ck->is_empty(s) && _bound_weight < 0);
+    }
+
     template<typename Hasher>
     void feed_hash(Hasher& hasher, const schema& s) const {
         ::feed_hash(hasher, _bound_weight);

read_context.hh

@@ -38,7 +38,7 @@ class autoupdating_underlying_reader final {
     row_cache& _cache;
     read_context& _read_context;
     stdx::optional<flat_mutation_reader> _reader;
-    utils::phased_barrier::phase_type _reader_creation_phase;
+    utils::phased_barrier::phase_type _reader_creation_phase = 0;
     dht::partition_range _range = { };
     stdx::optional<dht::decorated_key> _last_key;
     stdx::optional<dht::decorated_key> _new_last_key;
@@ -105,7 +105,6 @@ public:
         return make_ready_future<>();
     }
     utils::phased_barrier::phase_type creation_phase() const {
-        assert(_reader);
         return _reader_creation_phase;
     }
     const dht::partition_range& range() const {

repair/repair.cc

@@ -1004,6 +1004,22 @@ static dht::token_range_vector get_primary_ranges(
             utils::fb_utilities::get_broadcast_address());
 }
 
+// get_primary_ranges_within_dc() is similar to get_primary_ranges(),
+// but instead of each range being assigned just one primary owner
+// across the entire cluster, here each range is assigned a primary
+// owner in each of the clusters.
+static dht::token_range_vector get_primary_ranges_within_dc(
+        database& db, sstring keyspace) {
+    auto& rs = db.find_keyspace(keyspace).get_replication_strategy();
+    return rs.get_primary_ranges_within_dc(
+            utils::fb_utilities::get_broadcast_address());
+}
+
+static sstring get_local_dc() {
+    return locator::i_endpoint_snitch::get_local_snitch_ptr()->get_datacenter(
+            utils::fb_utilities::get_broadcast_address());
+}
+
 
 struct repair_options {
     // If primary_range is true, we should perform repair only on this node's
@@ -1256,21 +1272,14 @@ static int do_repair_start(seastar::sharded<database>& db, sstring keyspace,
         rlogger.info("primary-range repair");
         // when "primary_range" option is on, neither data_centers nor hosts
         // may be set, except data_centers may contain only local DC (-local)
-#if 0
         if (options.data_centers.size() == 1 &&
-                options.data_centers[0] == DatabaseDescriptor.getLocalDataCenter()) {
+            options.data_centers[0] == get_local_dc()) {
             ranges = get_primary_ranges_within_dc(db.local(), keyspace);
-        } else
-#endif
-#if 0
-        if (options.data_centers.size() > 0 || options.hosts.size() > 0) {
+        } else if (options.data_centers.size() > 0 || options.hosts.size() > 0) {
             throw std::runtime_error("You need to run primary range repair on all nodes in the cluster.");
         } else {
-#endif
             ranges = get_primary_ranges(db.local(), keyspace);
-#if 0
         }
-#endif
     } else {
         ranges = get_local_ranges(db.local(), keyspace);
     }

scripts/scylla_install_pkg

@@ -1,109 +0,0 @@
-#!/bin/bash -e
-#
-#  Copyright (C) 2015 ScyllaDB
-
-if [ "`id -u`" -ne 0 ]; then
-    echo "Requires root permission."
-    exit 1
-fi
-
-print_usage() {
-    echo "scylla_install_pkg --local-pkg /home/scylla/rpms --repo [URL]"
-    echo "  --local-pkg	install locally built .rpm/.deb on specified directory"
-    echo "  --repo  repository for both install and update, specify .repo/.list file URL"
-    echo "  --repo-for-install  repository for install, specify .repo/.list file URL"
-    echo "  --repo-for-update  repository for update, specify .repo/.list file URL"
-    exit 1
-}
-
-LOCAL_PKG=
-UNSTABLE=0
-REPO_FOR_INSTALL=
-REPO_FOR_UPDATE=
-while [ $# -gt 0 ]; do
-    case "$1" in
-        "--local-pkg")
-            LOCAL_PKG=$2
-            shift 2
-            ;;
-        "--repo")
-            REPO_FOR_INSTALL=$2
-            REPO_FOR_UPDATE=$2
-            shift 2
-            ;;
-        "--repo-for-install")
-            REPO_FOR_INSTALL=$2
-            shift 2
-            ;;
-        "--repo-for-update")
-            REPO_FOR_UPDATE=$2
-            shift 2
-            ;;
-        *)
-            print_usage
-            shift 1
-            ;;
-    esac
-done
-
-. /etc/os-release
-
-if [ -f /etc/debian_version ]; then
-    echo "#!/bin/sh" >> /usr/sbin/policy-rc.d
-    echo "exit 101" >> /usr/sbin/policy-rc.d
-    chmod +x /usr/sbin/policy-rc.d
-    cp /etc/hosts /etc/hosts.orig
-    echo 127.0.0.1 `hostname` >> /etc/hosts
-    if [ "$REPO_FOR_INSTALL" != "" ]; then
-        curl -L -o /etc/apt/sources.list.d/scylla_install.list $REPO_FOR_INSTALL
-    fi
-    apt-get -o Acquire::AllowInsecureRepositories=true \
-            -o Acquire::AllowDowngradeToInsecureRepositories=true update
-    if [ "$LOCAL_PKG" = "" ]; then
-        apt-get install -o APT::Get::AllowUnauthenticated=true \
-                        -y --force-yes scylla
-    else
-        if [ ! -f /usr/bin/gdebi ]; then
-            apt-get install -y --force-yes gdebi-core
-        fi
-        echo Y | gdebi $LOCAL_PKG/scylla-kernel-conf*.deb
-        echo Y | gdebi $LOCAL_PKG/scylla-conf*.deb
-        echo Y | gdebi $LOCAL_PKG/scylla-server_*.deb
-        echo Y | gdebi $LOCAL_PKG/scylla-server-dbg*.deb
-        echo Y | gdebi $LOCAL_PKG/scylla-jmx*.deb
-        echo Y | gdebi $LOCAL_PKG/scylla-tools*.deb
-        echo Y | gdebi $LOCAL_PKG/scylla_*.deb
-    fi
-    mv /etc/hosts.orig /etc/hosts
-    rm /usr/sbin/policy-rc.d
-    rm /etc/apt/sources.list.d/scylla_install.list
-    if [ "$REPO_FOR_UPDATE" != "" ]; then
-        curl -L -o /etc/apt/sources.list.d/scylla.list $REPO_FOR_UPDATE
-    fi
-    apt-get -o Acquire::AllowInsecureRepositories=true \
-            -o Acquire::AllowDowngradeToInsecureRepositories=true update
-else
-    if [ "$REPO_FOR_INSTALL" != "" ]; then
-        curl -L -o /etc/yum.repos.d/scylla_install.repo $REPO_FOR_INSTALL
-    fi
-
-    if [ "$ID" = "centos" ]; then
-            yum install -y epel-release
-    elif [ "$ID" = "rhel" ]; then
-        rpm -ivh http://download.fedoraproject.org/pub/epel/7/x86_64/e/epel-release-7-7.noarch.rpm
-    else
-        echo "Unsupported distribution"
-        exit 1
-    fi
-
-    if [ "$LOCAL_PKG" = "" ]; then
-        yum install -y scylla
-    else
-        yum install -y $LOCAL_PKG/scylla*.*.rpm
-    fi
-
-    rm /etc/yum.repos.d/scylla_install.repo
-    if [ "$REPO_FOR_UPDATE" != "" ]; then
-        curl -L -o /etc/yum.repos.d/scylla.repo $REPO_FOR_UPDATE
-    fi
-fi

service/migration_task.cc

@@ -54,7 +54,7 @@ static logging::logger mlogger("migration_task");
 future<> migration_task::run_may_throw(distributed<service::storage_proxy>& proxy, const gms::inet_address& endpoint)
 {
     if (!gms::get_failure_detector().local().is_alive(endpoint)) {
-        mlogger.error("Can't send migration request: node {} is down.", endpoint);
+        mlogger.warn("Can't send migration request: node {} is down.", endpoint);
         return make_ready_future<>();
     }
     netw::messaging_service::msg_addr id{endpoint, 0};

service/misc_services.cc

@@ -144,7 +144,11 @@ future<lowres_clock::duration> cache_hitrate_calculator::recalculate_hitrates()
         return _db.invoke_on_all([this, rates = std::move(rates), cpuid = engine().cpu_id()] (database& db) {
             sstring gstate;
             for (auto& cf : db.get_column_families() | boost::adaptors::filtered(non_system_filter)) {
-                stat s = rates.at(cf.first);
+                auto it = rates.find(cf.first);
+                if (it == rates.end()) { // a table may be added before map/reduce compltes and this code runs
+                    continue;
+                }
+                stat s = it->second;
                 float rate = 0;
                 if (s.h) {
                     rate = s.h / (s.h + s.m);

service/pager/query_pagers.cc

@@ -85,7 +85,7 @@ static bool has_clustering_keys(const schema& s, const query::read_command& cmd)
             _query_read_repair_decision = state->get_query_read_repair_decision();
         } else {
             // Reusing readers is currently only supported for singular queries.
-            if (_ranges.front().is_singular()) {
+            if (!_ranges.empty() && query::is_single_partition(_ranges.front())) {
                 _cmd->query_uuid = utils::make_random_uuid();
             }
             _cmd->is_first_page = true;

service/storage_proxy.cc

@@ -211,7 +211,7 @@ protected:
 
 protected:
     virtual bool waited_for(gms::inet_address from) = 0;
-    virtual void signal(gms::inet_address from) {
+    void signal(gms::inet_address from) {
         if (waited_for(from)) {
             signal();
         }
@@ -221,7 +221,7 @@ public:
     abstract_write_response_handler(shared_ptr<storage_proxy> p, keyspace& ks, db::consistency_level cl, db::write_type type,
             std::unique_ptr<mutation_holder> mh, std::unordered_set<gms::inet_address> targets, tracing::trace_state_ptr trace_state,
             storage_proxy::write_stats& stats, size_t pending_endpoints = 0, std::vector<gms::inet_address> dead_endpoints = {})
-            : _id(p->_next_response_id++), _proxy(std::move(p)), _trace_state(trace_state), _cl(cl), _type(type), _mutation_holder(std::move(mh)), _targets(std::move(targets)),
+            : _id(p->get_next_response_id()), _proxy(std::move(p)), _trace_state(trace_state), _cl(cl), _type(type), _mutation_holder(std::move(mh)), _targets(std::move(targets)),
               _dead_endpoints(std::move(dead_endpoints)), _stats(stats) {
         // original comment from cassandra:
         // during bootstrap, include pending endpoints in the count
@@ -285,10 +285,13 @@ public:
     }
     // return true on last ack
     bool response(gms::inet_address from) {
-        signal(from);
         auto it = _targets.find(from);
-        assert(it != _targets.end());
-        _targets.erase(it);
+        if (it != _targets.end()) {
+            signal(from);
+            _targets.erase(it);
+        } else {
+            slogger.warn("Receive outdated write ack from {}", from);
+        }
         return _targets.size() == 0;
     }
     future<> wait() {
@@ -632,9 +635,12 @@ void storage_proxy_stats::split_stats::register_metrics_for(gms::inet_address ep
     }
 }
 
+using namespace std::literals::chrono_literals;
+
 storage_proxy::~storage_proxy() {}
 storage_proxy::storage_proxy(distributed<database>& db, storage_proxy::config cfg)
     : _db(db)
+    , _next_response_id(std::chrono::system_clock::now().time_since_epoch()/1ms)
     , _hints_resource_manager(cfg.available_memory / 10)
     , _hints_for_views_manager(_db.local().get_config().data_file_directories()[0] + "/view_pending_updates", {}, _db.local().get_config().max_hint_window_in_ms(), _hints_resource_manager, _db)
     , _background_write_throttle_threahsold(cfg.available_memory / 10) {
@@ -1974,18 +1980,21 @@ public:
         _timeout.arm(timeout);
     }
     virtual ~abstract_read_resolver() {};
-    virtual void on_error(gms::inet_address ep) = 0;
+    virtual void on_error(gms::inet_address ep, bool disconnect) = 0;
     future<> done() {
         return _done_promise.get_future();
     }
     void error(gms::inet_address ep, std::exception_ptr eptr) {
         sstring why;
+        bool disconnect = false;
         try {
             std::rethrow_exception(eptr);
         } catch (rpc::closed_error&) {
-            return; // do not report connection closed exception, gossiper does that
+            // do not report connection closed exception, gossiper does that
+            disconnect = true;
         } catch (rpc::timeout_error&) {
-            return; // do not report timeouts, the whole operation will timeout and be reported
+            // do not report timeouts, the whole operation will timeout and be reported
+            return; // also do not report timeout as replica failure for the same reason
         } catch(std::exception& e) {
             why = e.what();
         } catch(...) {
@@ -1993,10 +2002,12 @@ public:
         }
 
         if (!_request_failed) { // request may fail only once.
-            on_error(ep);
+            on_error(ep, disconnect);
         }
 
-        slogger.error("Exception when communicating with {}: {}", ep, why);
+        if (why.length()) {
+            slogger.error("Exception when communicating with {}: {}", ep, why);
+        }
     }
 };
 
@@ -2071,10 +2082,16 @@ public:
             _done_promise.set_value();
         }
     }
-    void on_error(gms::inet_address ep) override {
+    void on_error(gms::inet_address ep, bool disconnect) override {
         if (waiting_for(ep)) {
             _failed++;
         }
+        if (disconnect && _block_for == _target_count_for_cl) {
+            // if the error is because of a connection disconnect and there is no targets to speculate
+            // wait for timeout in hope that the client will issue speculative read
+            // FIXME: resolver should have access to all replicas and try another one in this case
+            return;
+        }
         if (_block_for + _failed > _target_count_for_cl) {
             fail_request(std::make_exception_ptr(read_failure_exception(_schema->ks_name(), _schema->cf_name(), _cl, _cl_responses, _failed, _block_for, _data_result)));
         }
@@ -2400,7 +2417,7 @@ public:
             }
         }
     }
-    void on_error(gms::inet_address ep) override {
+    void on_error(gms::inet_address ep, bool disconnect) override {
         fail_request(std::make_exception_ptr(read_failure_exception(_schema->ks_name(), _schema->cf_name(), _cl, response_count(), 1, _targets_count, response_count() != 0)));
     }
     uint32_t max_live_count() const {
@@ -3323,9 +3340,22 @@ storage_proxy::query_partition_key_range(lw_shared_ptr<query::read_command> cmd,
     slogger.debug("Estimated result rows per range: {}; requested rows: {}, ranges.size(): {}; concurrent range requests: {}",
             result_rows_per_range, cmd->row_limit, ranges.size(), concurrency_factor);
 
+    // The call to `query_partition_key_range_concurrent()` below
+    // updates `cmd` directly when processing the results. Under
+    // some circumstances, when the query executes without deferring,
+    // this updating will happen before the lambda object is constructed
+    // and hence the updates will be visible to the lambda. This will
+    // result in the merger below trimming the results according to the
+    // updated (decremented) limits and causing the paging logic to
+    // declare the query exhausted due to the non-full page. To avoid
+    // this save the original values of the limits here and pass these
+    // to the lambda below.
+    const auto row_limit = cmd->row_limit;
+    const auto partition_limit = cmd->partition_limit;
+
     return query_partition_key_range_concurrent(query_options.timeout(*this), std::move(results), cmd, cl, ranges.begin(), std::move(ranges),
             concurrency_factor, std::move(query_options.trace_state), cmd->row_limit, cmd->partition_limit)
-            .then([row_limit = cmd->row_limit, partition_limit = cmd->partition_limit](std::vector<foreign_ptr<lw_shared_ptr<query::result>>> results) {
+            .then([row_limit, partition_limit](std::vector<foreign_ptr<lw_shared_ptr<query::result>>> results) {
         query::result_merger merger(row_limit, partition_limit);
         merger.reserve(results.size());
 
@@ -3682,6 +3712,7 @@ future<> storage_proxy::truncate_blocking(sstring keyspace, sstring cfname) {
            std::rethrow_exception(ep);
        } catch (rpc::timeout_error& e) {
            slogger.trace("Truncation of {} timed out: {}", cfname, e.what());
+           throw;
        } catch (...) {
            throw;
        }

service/storage_proxy.hh

@@ -143,7 +143,7 @@ public:
     };
 private:
     distributed<database>& _db;
-    response_id_type _next_response_id = 1; // 0 is reserved for unique_response_handler
+    response_id_type _next_response_id;
     std::unordered_map<response_id_type, rh_entry> _response_handlers;
     // This buffer hold ids of throttled writes in case resource consumption goes
     // below the threshold and we want to unthrottle some of them. Without this throttled
@@ -263,6 +263,13 @@ public:
         return _db;
     }
 
+    response_id_type get_next_response_id() {
+        auto next = _next_response_id++;
+        if (next == 0) { // 0 is reserved for unique_response_handler
+            next = _next_response_id++;
+        }
+        return next;
+    }
     void init_messaging_service();
 
     // Applies mutation on this node.

service/storage_service.cc

@@ -303,7 +303,7 @@ void storage_service::prepare_to_join(std::vector<inet_address> loaded_endpoints
                         gossiper.check_knows_remote_features(local_features, peer_features);
                     }
 
-                    gossiper.reset_endpoint_state_map();
+                    gossiper.reset_endpoint_state_map().get();
                     for (auto ep : loaded_endpoints) {
                         gossiper.add_saved_endpoint(ep);
                     }
@@ -317,7 +317,7 @@ void storage_service::prepare_to_join(std::vector<inet_address> loaded_endpoints
             slogger.info("Checking remote features with gossip");
             gossiper.do_shadow_round().get();
             gossiper.check_knows_remote_features(local_features);
-            gossiper.reset_endpoint_state_map();
+            gossiper.reset_endpoint_state_map().get();
             for (auto ep : loaded_endpoints) {
                 gossiper.add_saved_endpoint(ep);
             }
@@ -419,13 +419,9 @@ void storage_service::join_token_ring(int delay) {
             db::system_keyspace::set_bootstrap_state(db::system_keyspace::bootstrap_state::IN_PROGRESS).get();
         }
         set_mode(mode::JOINING, "waiting for ring information", true);
-        // first sleep the delay to make sure we see all our peers
-        for (int i = 0; i < delay; i += 1000) {
-            // if we see schema, we can proceed to the next check directly
-            if (_db.local().get_version() != database::empty_version) {
-                slogger.debug("got schema: {}", _db.local().get_version());
-                break;
-            }
+        auto& gossiper = gms::get_gossiper().local();
+        // first sleep the delay to make sure we see *at least* one other node
+        for (int i = 0; i < delay && gossiper.get_live_members().size() < 2; i += 1000) {
             sleep(std::chrono::seconds(1)).get();
         }
         // if our schema hasn't matched yet, keep sleeping until it does
@@ -484,7 +480,6 @@ void storage_service::join_token_ring(int delay) {
                 for (auto token : _bootstrap_tokens) {
                     auto existing = _token_metadata.get_endpoint(token);
                     if (existing) {
-                        auto& gossiper = gms::get_local_gossiper();
                         auto* eps = gossiper.get_endpoint_state_for_endpoint_ptr(*existing);
                         if (eps && eps->get_update_timestamp() > gms::gossiper::clk::now() - std::chrono::milliseconds(delay)) {
                             throw std::runtime_error("Cannot replace a live node...");
@@ -622,6 +617,8 @@ void storage_service::bootstrap(std::unordered_set<token> tokens) {
     db::system_keyspace::update_tokens(tokens).get();
     auto& gossiper = gms::get_local_gossiper();
     if (!db().local().is_replacing()) {
+        // Wait until we know tokens of existing node before announcing join status.
+        gossiper.wait_for_range_setup().get();
         // if not an existing token then bootstrap
         gossiper.add_local_application_state({
             { gms::application_state::TOKENS, value_factory.tokens(tokens) },
@@ -1541,7 +1538,7 @@ future<> storage_service::check_for_endpoint_collision() {
                             throw std::runtime_error("Other bootstrapping/leaving/moving nodes detected, cannot bootstrap while consistent_rangemovement is true (check_for_endpoint_collision)");
                         } else {
                             gossiper.goto_shadow_round();
-                            gossiper.reset_endpoint_state_map();
+                            gossiper.reset_endpoint_state_map().get();
                             found_bootstrapping_node = true;
                             auto elapsed = std::chrono::duration_cast<std::chrono::seconds>(gms::gossiper::clk::now() - t).count();
                             slogger.info("Checking bootstrapping/leaving/moving nodes: node={}, status={}, sleep 1 second and check again ({} seconds elapsed) (check_for_endpoint_collision)", addr, state, elapsed);
@@ -1553,7 +1550,7 @@ future<> storage_service::check_for_endpoint_collision() {
             }
         } while (found_bootstrapping_node);
         slogger.info("Checking bootstrapping/leaving/moving nodes: ok (check_for_endpoint_collision)");
-        gossiper.reset_endpoint_state_map();
+        gossiper.reset_endpoint_state_map().get();
     });
 }
 
@@ -1603,8 +1600,9 @@ future<std::unordered_set<token>> storage_service::prepare_replacement_info() {
         auto tokens = get_tokens_for(replace_address);
         // use the replacee's host Id as our own so we receive hints, etc
         return db::system_keyspace::set_local_host_id(host_id).discard_result().then([replace_address, tokens = std::move(tokens)] {
-            gms::get_local_gossiper().reset_endpoint_state_map(); // clean up since we have what we need
-            return make_ready_future<std::unordered_set<token>>(std::move(tokens));
+            return gms::get_local_gossiper().reset_endpoint_state_map().then([tokens = std::move(tokens)] { // clean up since we have what we need
+                return make_ready_future<std::unordered_set<token>>(std::move(tokens));
+            });
         });
     });
 }
@@ -2643,14 +2641,20 @@ future<> storage_service::send_replication_notification(inet_address remote) {
     // notify the remote token
     auto done = make_shared<bool>(false);
     auto local = get_broadcast_address();
+    auto sent = make_lw_shared<int>(0);
     slogger.debug("Notifying {} of replication completion", remote);
     return do_until(
-        [done, remote] {
-            return *done || !gms::get_local_failure_detector().is_alive(remote);
+        [done, sent, remote] {
+            // The node can send REPLICATION_FINISHED to itself, in which case
+            // is_alive will be true. If the messaging_service is stopped,
+            // REPLICATION_FINISHED can be sent infinitely here. To fix, limit
+            // the number of retries.
+            return *done || !gms::get_local_failure_detector().is_alive(remote) || *sent >= 3;
         },
-        [done, remote, local] {
+        [done, sent, remote, local] {
             auto& ms = netw::get_local_messaging_service();
             netw::msg_addr id{remote, 0};
+            (*sent)++;
             return ms.send_replication_finished(id, local).then_wrapped([id, done] (auto&& f) {
                 try {
                     f.get();

sstables/compaction.cc

@@ -179,6 +179,8 @@ struct compaction_read_monitor_generator final : public read_monitor_generator {
         void remove_sstable(bool is_tracking) {
             if (is_tracking) {
                 _cf.get_compaction_strategy().get_backlog_tracker().remove_sstable(_sst);
+            } else if (_sst) {
+                _cf.get_compaction_strategy().get_backlog_tracker().revert_charges(_sst);
             }
             _sst = {};
         }
@@ -303,6 +305,7 @@ public:
 class compaction {
 protected:
     column_family& _cf;
+    schema_ptr _schema;
     std::vector<shared_sstable> _sstables;
     uint64_t _max_sstable_size;
     uint32_t _sstable_level;
@@ -313,6 +316,7 @@ protected:
 protected:
     compaction(column_family& cf, std::vector<shared_sstable> sstables, uint64_t max_sstable_size, uint32_t sstable_level)
         : _cf(cf)
+        , _schema(cf.schema())
         , _sstables(std::move(sstables))
         , _max_sstable_size(max_sstable_size)
         , _sstable_level(sstable_level)
@@ -361,10 +365,9 @@ private:
     virtual flat_mutation_reader make_sstable_reader(lw_shared_ptr<sstables::sstable_set> ssts) const = 0;
 
     flat_mutation_reader setup() {
-        auto ssts = make_lw_shared<sstables::sstable_set>(_cf.get_compaction_strategy().make_sstable_set(_cf.schema()));
-        auto schema = _cf.schema();
+        auto ssts = make_lw_shared<sstables::sstable_set>(_cf.get_compaction_strategy().make_sstable_set(_schema));
         sstring formatted_msg = "[";
-        auto fully_expired = get_fully_expired_sstables(_cf, _sstables, gc_clock::now() - schema->gc_grace_seconds());
+        auto fully_expired = get_fully_expired_sstables(_cf, _sstables, gc_clock::now() - _schema->gc_grace_seconds());
 
         for (auto& sst : _sstables) {
             // Compacted sstable keeps track of its ancestors.
@@ -396,8 +399,8 @@ private:
         }
         formatted_msg += "]";
         _info->sstables = _sstables.size();
-        _info->ks = schema->ks_name();
-        _info->cf = schema->cf_name();
+        _info->ks = _schema->ks_name();
+        _info->cf = _schema->cf_name();
         report_start(formatted_msg);
 
         return make_sstable_reader(std::move(ssts));
@@ -462,7 +465,7 @@ private:
     }
 
     const schema_ptr& schema() const {
-        return _cf.schema();
+        return _schema;
     }
 public:
     static future<compaction_info> run(std::unique_ptr<compaction> c);
@@ -518,10 +521,10 @@ public:
     }
 
     flat_mutation_reader make_sstable_reader(lw_shared_ptr<sstables::sstable_set> ssts) const override {
-        return ::make_local_shard_sstable_reader(_cf.schema(),
+        return ::make_local_shard_sstable_reader(_schema,
                 std::move(ssts),
                 query::full_partition_range,
-                _cf.schema()->full_slice(),
+                _schema->full_slice(),
                 service::get_local_compaction_priority(),
                 no_resource_tracking(),
                 nullptr,
@@ -570,7 +573,7 @@ public:
             cfg.monitor = &_active_write_monitors.back();
             cfg.large_partition_handler = _cf.get_large_partition_handler();
             // TODO: calculate encoding_stats based on statistics of compacted sstables
-            _writer.emplace(_sst->get_writer(*_cf.schema(), partitions_per_sstable(), cfg, encoding_stats{}, priority));
+            _writer.emplace(_sst->get_writer(*_schema, partitions_per_sstable(), cfg, encoding_stats{}, priority));
         }
         return &*_writer;
     }
@@ -610,7 +613,7 @@ public:
     }
 
     std::function<bool(const dht::decorated_key&)> filter_func() const override {
-        dht::token_range_vector owned_ranges = service::get_local_storage_service().get_local_ranges(_cf.schema()->ks_name());
+        dht::token_range_vector owned_ranges = service::get_local_storage_service().get_local_ranges(_schema->ks_name());
 
         return [this, owned_ranges = std::move(owned_ranges)] (const dht::decorated_key& dk) {
             if (dht::shard_of(dk.token()) != engine().cpu_id()) {
@@ -684,10 +687,10 @@ public:
 
     // Use reader that makes sure no non-local mutation will not be filtered out.
     flat_mutation_reader make_sstable_reader(lw_shared_ptr<sstables::sstable_set> ssts) const override {
-        return ::make_range_sstable_reader(_cf.schema(),
+        return ::make_range_sstable_reader(_schema,
                 std::move(ssts),
                 query::full_partition_range,
-                _cf.schema()->full_slice(),
+                _schema->full_slice(),
                 service::get_local_compaction_priority(),
                 no_resource_tracking(),
                 nullptr,
@@ -719,7 +722,7 @@ public:
             cfg.large_partition_handler = _cf.get_large_partition_handler();
             auto&& priority = service::get_local_compaction_priority();
             // TODO: calculate encoding_stats based on statistics of compacted sstables
-            writer.emplace(sst->get_writer(*_cf.schema(), partitions_per_sstable(_shard), cfg, encoding_stats{}, priority, _shard));
+            writer.emplace(sst->get_writer(*_schema, partitions_per_sstable(_shard), cfg, encoding_stats{}, priority, _shard));
         }
         return &*writer;
     }

sstables/compaction_manager.cc

@@ -66,6 +66,14 @@ public:
             _cm->deregister_compacting_sstables(_compacting);
         }
     }
+
+    // Explicitly release compacting sstables
+    void release_compacting(const std::vector<sstables::shared_sstable>& sstables) {
+        _cm->deregister_compacting_sstables(sstables);
+        for (auto& sst : sstables) {
+            _compacting.erase(boost::remove(_compacting, sst), _compacting.end());
+        }
+    }
 };
 
 compaction_weight_registration::compaction_weight_registration(compaction_manager* cm, int weight)
@@ -564,17 +572,23 @@ future<> compaction_manager::perform_cleanup(column_family* cf) {
             return make_ready_future<stop_iteration>(stop_iteration::yes);
         }
         column_family& cf = *task->compacting_cf;
-        sstables::compaction_descriptor descriptor = sstables::compaction_descriptor(get_candidates(cf));
-        auto compacting = compacting_sstable_registration(this, descriptor.sstables);
+        auto sstables = get_candidates(cf);
+        auto compacting = make_lw_shared<compacting_sstable_registration>(this, sstables);
 
         _stats.pending_tasks--;
         _stats.active_tasks++;
         compaction_backlog_tracker user_initiated(std::make_unique<user_initiated_backlog_tracker>(_compaction_controller.backlog_of_shares(200), _available_memory));
-        return do_with(std::move(user_initiated), [this, &cf, descriptor = std::move(descriptor)] (compaction_backlog_tracker& bt) mutable {
-            return with_scheduling_group(_scheduling_group, [this, &cf, descriptor = std::move(descriptor)] () mutable {
-                return cf.cleanup_sstables(std::move(descriptor));
+        return do_with(std::move(user_initiated), std::move(sstables), [this, &cf, compacting] (compaction_backlog_tracker& bt,
+                std::vector<sstables::shared_sstable>& sstables) mutable {
+            return with_scheduling_group(_scheduling_group, [this, &cf, &sstables, compacting] () mutable {
+                return do_for_each(sstables, [this, &cf, compacting] (auto& sst) {
+                    return cf.cleanup_sstables(sstables::compaction_descriptor({sst})).then([&sst, compacting] {
+                        // Releases reference to cleaned sstable such that respective used disk space can be freed.
+                        compacting->release_compacting({std::move(sst)});
+                    });
+                });
             });
-        }).then_wrapped([this, task, compacting = std::move(compacting)] (future<> f) mutable {
+        }).then_wrapped([this, task, compacting] (future<> f) mutable {
             _stats.active_tasks--;
             if (!can_proceed(task)) {
                 maybe_stop_on_error(std::move(f));

sstables/compaction_strategy.cc

@@ -404,11 +404,6 @@ public:
             auto itw = writes_per_window.find(bound);
             if (itw != writes_per_window.end()) {
                 ow_this_window = &itw->second;
-                // We will erase here so we can keep track of which
-                // writes belong to existing windows. Writes that don't belong to any window
-                // are writes in progress to new windows and will be accounted in the final
-                // loop before we return
-                writes_per_window.erase(itw);
             }
             auto* oc_this_window = &no_oc;
             auto itc = compactions_per_window.find(bound);
@@ -416,6 +411,13 @@ public:
                 oc_this_window = &itc->second;
             }
             b += windows.second.backlog(*ow_this_window, *oc_this_window);
+            if (itw != writes_per_window.end()) {
+                // We will erase here so we can keep track of which
+                // writes belong to existing windows. Writes that don't belong to any window
+                // are writes in progress to new windows and will be accounted in the final
+                // loop before we return
+                writes_per_window.erase(itw);
+            }
         }
 
         // Partial writes that don't belong to any window are accounted here.

sstables/compress.cc

@@ -33,6 +33,7 @@
 #include "unimplemented.hh"
 #include "stdx.hh"
 #include "segmented_compress_params.hh"
+#include "utils/class_registrator.hh"
 
 namespace sstables {
 
@@ -299,7 +300,8 @@ size_t local_compression::compress_max_size(size_t input_len) const {
 
 void compression::set_compressor(compressor_ptr c) {
     if (c) {
-        auto& cn = c->name();
+        unqualified_name uqn(compressor::namespace_prefix, c->name());
+        const sstring& cn = uqn;
         name.value = bytes(cn.begin(), cn.end());
         for (auto& p : c->options()) {
             if (p.first != compression_parameters::SSTABLE_COMPRESSION) {

sstables/mp_row_consumer.cc

@@ -53,7 +53,8 @@ atomic_cell make_counter_cell(api::timestamp_type timestamp, bytes_view value) {
             throw marshal_exception("encountered a local shard in a counter cell");
         }
     }
-    auto shard_count = value.size() / shard_size;
+    auto header_length = (size_t(header_size) + 1) * sizeof(int16_t);
+    auto shard_count = (value.size() - header_length) / shard_size;
     if (shard_count != size_t(header_size)) {
         throw marshal_exception("encountered remote shards in a counter cell");
     }

sstables/mp_row_consumer.hh

@@ -695,9 +695,12 @@ public:
     // Sets streamed_mutation::_end_of_range when there are no more fragments for the query range.
     // Returns information whether the parser should continue to parse more
     // input and produce more fragments or we have collected enough and should yield.
+    // Returns proceed:yes only when all pending fragments have been pushed.
     proceed push_ready_fragments() {
         if (_ready) {
-            return push_ready_fragments_with_ready_set();
+            if (push_ready_fragments_with_ready_set() == proceed::no) {
+                return proceed::no;
+            }
         }
 
         if (_out_of_range) {

sstables/sstables.cc

@@ -787,6 +787,11 @@ future<> parse(sstable_version_types v, random_access_reader& in, utils::estimat
         if (length == 0) {
             throw malformed_sstable_exception("Estimated histogram with zero size found. Can't continue!");
         }
+
+        // Arrays are potentially pre-initialized by the estimated_histogram constructor.
+        eh.bucket_offsets.clear();
+        eh.buckets.clear();
+
         eh.bucket_offsets.reserve(length - 1);
         eh.buckets.reserve(length);
 

streaming/stream_manager.cc

@@ -292,7 +292,7 @@ void stream_manager::on_restart(inet_address endpoint, endpoint_state ep_state)
 }
 
 void stream_manager::on_dead(inet_address endpoint, endpoint_state ep_state) {
-    if (has_peer(endpoint) && ep_state.is_shutdown()) {
+    if (has_peer(endpoint)) {
         sslog.info("stream_manager: Close all stream_session with peer = {} in on_dead", endpoint);
         get_stream_manager().invoke_on_all([endpoint] (auto& sm) {
             sm.fail_sessions(endpoint);

streaming/stream_session.cc

@@ -509,8 +509,7 @@ void stream_session::close_session(stream_session_state final_state) {
             _stream_result->handle_session_complete(shared_from_this());
         }
 
-        sslog.debug("[Stream #{}] close_session session={}, state={}, cancel keep_alive timer", plan_id(), this, final_state);
-        _keep_alive.cancel();
+        sslog.debug("[Stream #{}] close_session session={}, state={}", plan_id(), this, final_state);
     }
 }
 
@@ -537,41 +536,6 @@ bool stream_session::is_initialized() const {
 
 void stream_session::init(shared_ptr<stream_result_future> stream_result_) {
     _stream_result = stream_result_;
-    _keep_alive.set_callback([this] {
-        auto plan_id = this->plan_id();
-        auto peer = this->peer;
-        get_local_stream_manager().get_progress_on_all_shards(plan_id, peer).then([this, peer, plan_id] (stream_bytes sbytes) {
-            if (this->_is_aborted) {
-                sslog.info("[Stream #{}] The session {} is closed, keep alive timer will do nothing", plan_id, this);
-                return;
-            }
-            auto now = lowres_clock::now();
-            sslog.debug("[Stream #{}] keep alive timer callback sbytes old: tx={}, rx={} new: tx={} rx={}",
-                    plan_id, this->_last_stream_bytes.bytes_sent, this->_last_stream_bytes.bytes_received,
-                    sbytes.bytes_sent, sbytes.bytes_received);
-            if (sbytes.bytes_sent > this->_last_stream_bytes.bytes_sent ||
-                sbytes.bytes_received > this->_last_stream_bytes.bytes_received) {
-                sslog.debug("[Stream #{}] The session {} made progress with peer {}", plan_id, this, peer);
-                // Progress has been made
-                this->_last_stream_bytes = sbytes;
-                this->_last_stream_progress = now;
-                this->start_keep_alive_timer();
-            } else if (now - this->_last_stream_progress >= this->_keep_alive_timeout) {
-                // Timeout
-                sslog.info("[Stream #{}] The session {} is idle for {} seconds, the peer {} is probably gone, close it",
-                        plan_id, this, this->_keep_alive_timeout.count(), peer);
-                this->on_error();
-            } else {
-                // Start the timer to check again
-                sslog.info("[Stream #{}] The session {} made no progress with peer {}", plan_id, this, peer);
-                this->start_keep_alive_timer();
-            }
-        }).handle_exception([plan_id, peer, session = this->shared_from_this()] (auto ep) {
-           sslog.info("[Stream #{}] keep alive timer callback fails with peer {}: {}", plan_id, peer, ep);
-        });
-    });
-    _last_stream_progress = lowres_clock::now();
-    start_keep_alive_timer();
 }
 
 utils::UUID stream_session::plan_id() {