24-3-13-hotfix: Fix bulk operations breaking frozen locks (#12019)

snaury · web-flow · commit 83b186050b7e · 2024-11-27T13:34:32.000+03:00
diff --git a/ydb/core/tx/datashard/datashard__op_rows.cpp b/ydb/core/tx/datashard/datashard__op_rows.cpp
@@ -220,6 +220,11 @@ void TDataShard::Handle(TEvDataShard::TEvUploadRowsRequest::TPtr& ev, const TAct
         UpdateProposeQueueSize();
         return;
     }
+    if (Pipeline.HasProposeDelayers()) {
+        DelayedProposeQueue.emplace_back().Reset(ev.Release());
+        UpdateProposeQueueSize();
+        return;
+    }
     if (IsReplicated()) {
         return Reject<TEvDataShard::TEvUploadRowsResponse>(this, ev, "bulk upsert",
             ERejectReasons::WrongState, "Can't execute bulk upsert at replicated table", &ReadOnly, ctx, TDataShard::ELogThrottlerType::UploadRows_Reject);
@@ -237,6 +242,11 @@ void TDataShard::Handle(TEvDataShard::TEvEraseRowsRequest::TPtr& ev, const TActo
         UpdateProposeQueueSize();
         return;
     }
+    if (Pipeline.HasProposeDelayers()) {
+        DelayedProposeQueue.emplace_back().Reset(ev.Release());
+        UpdateProposeQueueSize();
+        return;
+    }
     if (IsReplicated()) {
         return Reject<TEvDataShard::TEvEraseRowsResponse>(this, ev, "erase",
             ERejectReasons::WrongState, "Can't execute erase at replicated table", &ExecError, ctx, TDataShard::ELogThrottlerType::EraseRows_Reject);
diff --git a/ydb/core/tx/datashard/datashard_ut_snapshot.cpp b/ydb/core/tx/datashard/datashard_ut_snapshot.cpp
@@ -3977,6 +3977,190 @@ Y_UNIT_TEST_SUITE(DataShardSnapshots) {
             "{ items { uint32_value: 4 } items { uint32_value: 40 } }");
     }
 
+    Y_UNIT_TEST(UncommittedWriteRestartDuringCommitThenBulkErase) {
+        NKikimrConfig::TAppConfig app;
+
+        TPortManager pm;
+        TServerSettings serverSettings(pm.GetPort(2134));
+        serverSettings.SetDomainName("Root")
+            .SetUseRealThreads(false)
+            .SetDomainPlanResolution(100)
+            .SetAppConfig(app)
+            // Bug was with non-volatile transactions
+            .SetEnableDataShardVolatileTransactions(false);
+
+        Tests::TServer::TPtr server = new TServer(serverSettings);
+        auto &runtime = *server->GetRuntime();
+        auto sender = runtime.AllocateEdgeActor();
+
+        runtime.SetLogPriority(NKikimrServices::TX_DATASHARD, NLog::PRI_TRACE);
+        runtime.SetLogPriority(NKikimrServices::TX_PROXY, NLog::PRI_DEBUG);
+        runtime.SetLogPriority(NKikimrServices::KQP_EXECUTER, NLog::PRI_TRACE);
+        runtime.SetLogPriority(NKikimrServices::KQP_SESSION, NLog::PRI_TRACE);
+
+        InitRoot(server, sender);
+
+        TDisableDataShardLogBatching disableDataShardLogBatching;
+        UNIT_ASSERT_VALUES_EQUAL(
+            KqpSchemeExec(runtime, R"(
+                CREATE TABLE `/Root/table` (key Uint32, value Uint32, PRIMARY KEY (key))
+                WITH (PARTITION_AT_KEYS = (5));
+            )"),
+            "SUCCESS");
+
+        // Insert some initial data
+        ExecSQL(server, sender, "UPSERT INTO `/Root/table` (key, value) VALUES (1, 10), (5, 50);");
+
+        const auto shards = GetTableShards(server, sender, "/Root/table");
+        const auto tableId = ResolveTableId(server, sender, "/Root/table");
+        UNIT_ASSERT_VALUES_EQUAL(shards.size(), 2u);
+
+        TString sessionId, txId;
+
+        // Start inserting a couple of rows into the table
+        Cerr << "... sending initial upsert" << Endl;
+        UNIT_ASSERT_VALUES_EQUAL(
+            KqpSimpleBegin(runtime, sessionId, txId, R"(
+                SELECT key, value FROM `/Root/table` WHERE key = 1;
+                UPSERT INTO `/Root/table` (key, value) VALUES (2, 20), (6, 60);
+                )"),
+            "{ items { uint32_value: 1 } items { uint32_value: 10 } }");
+
+        // We want to block readsets next
+        std::vector<std::unique_ptr<IEventHandle>> readSets;
+        auto blockReadSets = runtime.AddObserver<TEvTxProcessing::TEvReadSet>([&](TEvTxProcessing::TEvReadSet::TPtr& ev) {
+            readSets.emplace_back(ev.Release());
+        });
+
+        // Start committing an additional read/write
+        // Note: select on the table flushes accumulated changes first
+        Cerr << "... sending commit request" << Endl;
+        auto commitFuture = SendRequest(runtime, MakeSimpleRequestRPC(R"(
+            SELECT key, value FROM `/Root/table` ORDER BY key;
+            )", sessionId, txId, /* commitTx */ true));
+
+        WaitFor(runtime, [&]{ return readSets.size() >= 2; }, "readset exchange");
+        UNIT_ASSERT_VALUES_EQUAL(readSets.size(), 2u);
+
+        // We want to make sure we block the first progress message when shards reboot
+        std::vector<TActorId> shardActors(shards.size());
+        UNIT_ASSERT_VALUES_EQUAL(shardActors.size(), 2u);
+        std::vector<std::unique_ptr<IEventHandle>> blockedProgress;
+        auto blockProgressQueue = runtime.AddObserver([&](TAutoPtr<IEventHandle>& ev) noexcept {
+            switch (ev->GetTypeRewrite()) {
+                case TEvTablet::TEvBoot::EventType: {
+                    auto* msg = ev->Get<TEvTablet::TEvBoot>();
+                    Cerr << "... observed TEvBoot for " << msg->TabletID << " at " << ev->GetRecipientRewrite() << Endl;
+                    auto it = std::find(shards.begin(), shards.end(), msg->TabletID);
+                    if (it != shards.end()) {
+                        shardActors.at(it - shards.begin()) = ev->GetRecipientRewrite();
+                    }
+                    break;
+                }
+                case EventSpaceBegin(TKikimrEvents::ES_PRIVATE) + 0 /* EvProgressTransaction */: {
+                    auto it = std::find(shardActors.begin(), shardActors.end(), ev->GetRecipientRewrite());
+                    if (it != shardActors.end()) {
+                        ui64 shardId = shards.at(it - shardActors.begin());
+                        Cerr << "... blocking TEvProgressTranasction at " << ev->GetRecipientRewrite() << " shard " << shardId << Endl;
+                        blockedProgress.emplace_back(ev.Release());
+                        return;
+                    }
+                    break;
+                }
+            }
+        });
+
+        // Clear old readsets and reboot both shards with TEvPoison
+        // This way shards don't have a chance to reply causing an UNDETERMINED error
+        readSets.clear();
+        for (ui64 shardId : shards) {
+            Cerr << "... sending TEvPoison to " << shardId << Endl;
+            ForwardToTablet(runtime, shardId, sender, new TEvents::TEvPoison);
+        }
+
+        // Note: we cannot wait for the commit result, since KQP is blocked trying to abort
+
+        // Sleep a little to make sure everything settles
+        Cerr << "... sleeping for 1 second" << Endl;
+        runtime.SimulateSleep(TDuration::Seconds(1));
+
+        UNIT_ASSERT_VALUES_EQUAL(readSets.size(), 2u);
+        UNIT_ASSERT_VALUES_EQUAL(blockedProgress.size(), 2u);
+
+        // Send an erase rows request before the progress queue resumes
+        {
+            Cerr << "... sending TEvEraseRowsRequest to shard 1 for key 1" << Endl;
+            auto req = std::make_unique<TEvDataShard::TEvEraseRowsRequest>();
+            req->Record.SetTableId(tableId.PathId.LocalPathId);
+            req->Record.SetSchemaVersion(tableId.SchemaVersion);
+            req->Record.AddKeyColumnIds(1);
+            ui32 key = 1;
+            TCell keyCell = TCell::Make(key);
+            req->Record.AddKeyColumns(TSerializedCellVec::Serialize(TArrayRef<const TCell>(&keyCell, 1)));
+            runtime.Send(new IEventHandle(shardActors.at(0), sender, req.release()), 0, true);
+            // Give shard 1 a chance to process this request incorrectly
+            Cerr << "... sleeping for 1 second" << Endl;
+            runtime.SimulateSleep(TDuration::Seconds(1));
+        }
+
+        // Unblock progress queue and resend blocked messages
+        Cerr << "... resending progress queue" << Endl;
+        blockProgressQueue.Remove();
+        for (auto& ev : blockedProgress) {
+            runtime.Send(ev.release(), 0, true);
+        }
+        blockedProgress.clear();
+
+        // This insert must run after the currently committing transaction, so it must fail: either read happens before
+        // the commit and is broken later by the commit, or the read finds a duplicate row and insert fails. Due to a
+        // bug the commit lock might already be broken, causing conflicts not to work properly, and allowing the insert
+        // to overwrite key = 2.
+        Cerr << "... sending an insert" << Endl;
+        auto insertFuture = KqpSimpleSend(runtime, R"(
+            INSERT INTO `/Root/table` (key, value) VALUES (2, 22);
+        )");
+
+        // Sleep a little to make sure everything settles
+        Cerr << "... sleeping for 1 second" << Endl;
+        runtime.SimulateSleep(TDuration::Seconds(1));
+
+        // Unblock readsets letting transaction to complete
+        Cerr << "... resending readsets" << Endl;
+        blockReadSets.Remove();
+        for (auto& ev : readSets) {
+            runtime.Send(ev.release(), 0, true);
+        }
+        readSets.clear();
+
+        // Sleep a little to make sure everything settles
+        Cerr << "... sleeping for 1 second" << Endl;
+        runtime.SimulateSleep(TDuration::Seconds(1));
+
+        // We expect erase to succeed by this point
+        Cerr << "... checking the erase result" << Endl;
+        {
+            auto ev = runtime.GrabEdgeEventRethrow<TEvDataShard::TEvEraseRowsResponse>(sender);
+            UNIT_ASSERT_VALUES_EQUAL(ev->Get()->Record.GetStatus(), NKikimrTxDataShard::TEvEraseRowsResponse::OK);
+        }
+
+        // We expect commit to fail with an UNDETERMINED error
+        Cerr << "... checking the commit result" << Endl;
+        UNIT_ASSERT_VALUES_EQUAL(
+            FormatResult(AwaitResponse(runtime, std::move(commitFuture))),
+            "ERROR: UNDETERMINED");
+
+        // Now make a read query, we must not observe any partial commits
+        Cerr << "... checking final table state" << Endl;
+        UNIT_ASSERT_VALUES_EQUAL(
+            KqpSimpleExec(runtime, R"(
+                SELECT key, value FROM `/Root/table`
+                ORDER BY key;
+                )"),
+            "{ items { uint32_value: 2 } items { uint32_value: 20 } }, "
+            "{ items { uint32_value: 5 } items { uint32_value: 50 } }, "
+            "{ items { uint32_value: 6 } items { uint32_value: 60 } }");
+    }
+
     /**
      * This observer forces newly created nodes to start on particular nodes
      */
diff --git a/ydb/core/tx/locks/locks.h b/ydb/core/tx/locks/locks.h
@@ -239,7 +239,7 @@ enum class ELockRangeFlags : ui8 {
 using ELockRangeFlagsRaw = std::underlying_type<ELockRangeFlags>::type;
 
 inline ELockRangeFlags operator|(ELockRangeFlags a, ELockRangeFlags b) { return ELockRangeFlags(ELockRangeFlagsRaw(a) | ELockRangeFlagsRaw(b)); }
-inline ELockRangeFlags operator&(ELockRangeFlags a, ELockRangeFlags b) { return ELockRangeFlags(ELockRangeFlagsRaw(a) | ELockRangeFlagsRaw(b)); }
+inline ELockRangeFlags operator&(ELockRangeFlags a, ELockRangeFlags b) { return ELockRangeFlags(ELockRangeFlagsRaw(a) & ELockRangeFlagsRaw(b)); }
 inline ELockRangeFlags& operator|=(ELockRangeFlags& a, ELockRangeFlags b) { return a = a | b; }
 inline ELockRangeFlags& operator&=(ELockRangeFlags& a, ELockRangeFlags b) { return a = a & b; }
 inline bool operator!(ELockRangeFlags c) { return ELockRangeFlagsRaw(c) == 0; }
