summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
Diffstat
-rw-r--r--buildscripts/resmokeconfig/suites/sharding_csrs_continuous_config_stepdown.yml1
-rw-r--r--etc/backports_required_for_multiversion_tests.yml4
-rw-r--r--jstests/sharding/multi_collection_transaction_placement_conflict_workaround.js76
-rw-r--r--src/mongo/db/catalog_raii.cpp5
-rw-r--r--src/mongo/db/db_raii.cpp5
-rw-r--r--src/mongo/db/s/resharding/resharding_coordinator_service.cpp7
-rw-r--r--src/mongo/db/s/resharding/resharding_destined_recipient_test.cpp2
-rw-r--r--src/mongo/db/s/scoped_collection_metadata.h4
-rw-r--r--src/mongo/db/shard_role.cpp91
-rw-r--r--src/mongo/db/shard_role.h10
10 files changed, 191 insertions, 14 deletions
diff --git a/buildscripts/resmokeconfig/suites/sharding_csrs_continuous_config_stepdown.yml b/buildscripts/resmokeconfig/suites/sharding_csrs_continuous_config_stepdown.yml
index 78a4f40e08b..7e4dc129f1a 100644
--- a/buildscripts/resmokeconfig/suites/sharding_csrs_continuous_config_stepdown.yml
+++ b/buildscripts/resmokeconfig/suites/sharding_csrs_continuous_config_stepdown.yml
@@ -104,6 +104,7 @@ selector:
- jstests/sharding/move_chunk_insert_with_write_retryability.js
- jstests/sharding/move_chunk_remove_with_write_retryability.js
- jstests/sharding/move_chunk_update_with_write_retryability.js
+ - jstests/sharding/multi_collection_transaction_placement_conflict_workaround.js # sets a failpoint on the config primary
- jstests/sharding/refine_collection_shard_key_atomic.js # sets a failpoint on the config primary
- jstests/sharding/restart_transactions.js
- jstests/sharding/txn_two_phase_commit_commands_basic_requirements.js
diff --git a/etc/backports_required_for_multiversion_tests.yml b/etc/backports_required_for_multiversion_tests.yml
index 12730f5eae9..936e92585e8 100644
--- a/etc/backports_required_for_multiversion_tests.yml
+++ b/etc/backports_required_for_multiversion_tests.yml
@@ -394,7 +394,7 @@ last-continuous:
- test_file: jstests/sharding/query/sharded_lookup_execution.js
ticket: SERVER-77427
- test_file: jstests/sharding/multi_collection_transaction_placement_conflict_workaround.js
- ticket: SERVER-84723
+ ticket: SERVER-87061
- test_file: jstests/sharding/fsync_lock_unlock.js
ticket: SERVER-73685
- test_file: jstests/sharding/fsync_lock_ddl_lock.js
@@ -1019,7 +1019,7 @@ last-lts:
- test_file: jstests/sharding/query/sharded_lookup_execution.js
ticket: SERVER-77427
- test_file: jstests/sharding/multi_collection_transaction_placement_conflict_workaround.js
- ticket: SERVER-84723
+ ticket: SERVER-87061
- test_file: jstests/sharding/fsync_lock_unlock.js
ticket: SERVER-73685
- test_file: jstests/sharding/fsync_lock_ddl_lock.js
diff --git a/jstests/sharding/multi_collection_transaction_placement_conflict_workaround.js b/jstests/sharding/multi_collection_transaction_placement_conflict_workaround.js
index 3e03f448129..fa507919a89 100644
--- a/jstests/sharding/multi_collection_transaction_placement_conflict_workaround.js
+++ b/jstests/sharding/multi_collection_transaction_placement_conflict_workaround.js
@@ -3,6 +3,7 @@
* TransientTransactionError label) if a collection or database placement changes have occurred
* later than the transaction data snapshot timestamp.
*/
+import {configureFailPoint} from "jstests/libs/fail_point_util.js";
const st = new ShardingTest({mongos: 1, shards: 2});
@@ -244,6 +245,81 @@ const st = new ShardingTest({mongos: 1, shards: 2});
runTest(readConcern, command);
}));
}
+
+ // Test transaction concurrent with reshardCollection.
+ {
+ function runTest(readConcernLevel, command) {
+ jsTest.log("Running transaction + resharding test with read concern " +
+ readConcernLevel + " and command " + command);
+
+ // Setup initial state:
+ assert.commandWorked(st.s.getDB(dbName).dropDatabase());
+ st.adminCommand({enableSharding: dbName, primaryShard: st.shard0.shardName});
+
+ st.adminCommand({shardCollection: ns1, key: {x: 1}});
+ assert.commandWorked(st.splitAt(ns1, {x: 0}));
+ assert.commandWorked(st.moveChunk(ns1, {x: -1}, st.shard0.shardName));
+ assert.commandWorked(st.moveChunk(ns1, {x: 1}, st.shard1.shardName));
+
+ assert.commandWorked(coll1.insertMany([{x: -1, y: 0}, {x: 1, y: 0}]));
+
+ assert.commandWorked(coll2.insertOne({a: 1}));
+
+ // Set fp to block resharding after commit on configsvr but before commit on shards.
+ // We seek to test an interleaving where the transaction executes at a logical timestamp
+ // that falls between the timestamp at which resharding commits on the configsvr and the
+ // timestamp at which the temporary resharding collections are renamed to the original
+ // nss on the shards.
+ const fp = configureFailPoint(st.configRS.getPrimary(),
+ "reshardingPauseBeforeTellingParticipantsToCommit");
+
+ // On parallel shell, start resharding
+ const joinResharding = startParallelShell(() => {
+ assert.commandWorked(db.adminCommand({reshardCollection: 'test.foo', key: {y: 1}}));
+ }, st.s.port);
+
+ // Await configsvr to have done its part of the commit.
+ fp.wait();
+
+ const session = st.s.startSession();
+ const sessionDB = session.getDatabase(dbName);
+ const sessionColl1 = sessionDB.getCollection(collName1);
+ const sessionColl2 = sessionDB.getCollection(collName2);
+
+ // Make sure the session knows of a clusterTime inclusive of the resharding operation up
+ // to the commit on the configsvr.
+ assert.commandWorked(sessionColl2.insert({a: 2}));
+
+ // Start txn.
+ session.startTransaction({readConcern: {level: readConcernLevel}});
+ assert.eq(2, sessionColl2.find().itcount());
+
+ // Unset fp and wait for resharding to finish.
+ fp.off();
+ joinResharding();
+
+ // Make sure the router is aware of the new (post-resharding) routing table for test.foo
+ assert.eq(2, coll1.find({y: 0}).itcount());
+
+ // Now operate on coll1 within the transaction and expect to get a conflict.
+ let err = assert.throwsWithCode(() => {
+ if (command === 'find') {
+ sessionColl1.find().itcount();
+ } else if (command === 'aggregate') {
+ sessionColl1.aggregate().itcount();
+ } else if (command === 'update') {
+ assert.commandWorked(sessionColl1.updateMany({}, {$set: {c: 1}}));
+ }
+ }, [ErrorCodes.WriteConflict, ErrorCodes.SnapshotUnavailable]);
+ assert.contains("TransientTransactionError", err.errorLabels, tojson(err));
+
+ session.abortTransaction();
+ }
+
+ readConcerns.forEach((readConcern) => commands.forEach((command) => {
+ runTest(readConcern, command);
+ }));
+ }
}
st.stop();
diff --git a/src/mongo/db/catalog_raii.cpp b/src/mongo/db/catalog_raii.cpp
index d226e8f628c..ab7ce48f8d2 100644
--- a/src/mongo/db/catalog_raii.cpp
+++ b/src/mongo/db/catalog_raii.cpp
@@ -379,6 +379,11 @@ AutoGetCollection::AutoGetCollection(OperationContext* opCtx,
opCtx, *catalog, _coll, _resolvedNss);
}
+ if (receivedShardVersion) {
+ shard_role_details::checkShardingAndLocalCatalogCollectionUUIDMatch(
+ opCtx, _resolvedNss, *receivedShardVersion, collDesc, _coll);
+ }
+
return;
}
diff --git a/src/mongo/db/db_raii.cpp b/src/mongo/db/db_raii.cpp
index 3fedd36b5b5..b774e2a3f35 100644
--- a/src/mongo/db/db_raii.cpp
+++ b/src/mongo/db/db_raii.cpp
@@ -452,6 +452,11 @@ AutoGetCollectionForRead::AutoGetCollectionForRead(OperationContext* opCtx,
checkCollectionUUIDMismatch(opCtx, *catalog, _resolvedNss, _coll, options._expectedUUID);
+ if (receivedShardVersion) {
+ shard_role_details::checkShardingAndLocalCatalogCollectionUUIDMatch(
+ opCtx, _resolvedNss, *receivedShardVersion, collDesc, _coll);
+ }
+
return;
}
diff --git a/src/mongo/db/s/resharding/resharding_coordinator_service.cpp b/src/mongo/db/s/resharding/resharding_coordinator_service.cpp
index 733fbe3d655..d1c59f0438e 100644
--- a/src/mongo/db/s/resharding/resharding_coordinator_service.cpp
+++ b/src/mongo/db/s/resharding/resharding_coordinator_service.cpp
@@ -140,6 +140,7 @@ MONGO_FAIL_POINT_DEFINE(reshardingPauseCoordinatorBeforeInitializing);
MONGO_FAIL_POINT_DEFINE(reshardingPauseCoordinatorBeforeCloning);
MONGO_FAIL_POINT_DEFINE(reshardingPauseCoordinatorBeforeBlockingWrites);
MONGO_FAIL_POINT_DEFINE(reshardingPauseCoordinatorBeforeDecisionPersisted);
+MONGO_FAIL_POINT_DEFINE(reshardingPauseBeforeTellingParticipantsToCommit);
MONGO_FAIL_POINT_DEFINE(reshardingPauseCoordinatorBeforeRemovingStateDoc);
MONGO_FAIL_POINT_DEFINE(reshardingPauseCoordinatorBeforeCompletion);
MONGO_FAIL_POINT_DEFINE(reshardingPauseCoordinatorBeforeStartingErrorFlow);
@@ -2754,6 +2755,12 @@ void ReshardingCoordinator::_tellAllDonorsToRefresh(
void ReshardingCoordinator::_tellAllParticipantsToCommit(
const NamespaceString& nss, const std::shared_ptr<executor::ScopedTaskExecutor>& executor) {
+ {
+ auto opCtx = _cancelableOpCtxFactory->makeOperationContext(&cc());
+ reshardingPauseBeforeTellingParticipantsToCommit.pauseWhileSetAndNotCanceled(
+ opCtx.get(), _ctHolder->getAbortToken());
+ }
+
auto opts = createShardsvrCommitReshardCollectionOptions(
nss, _coordinatorDoc.getReshardingUUID(), **executor, _ctHolder->getStepdownToken(), {});
opts->cmd.setDbName(DatabaseName::kAdmin);
diff --git a/src/mongo/db/s/resharding/resharding_destined_recipient_test.cpp b/src/mongo/db/s/resharding/resharding_destined_recipient_test.cpp
index 52bac27be96..cfce85e7a30 100644
--- a/src/mongo/db/s/resharding/resharding_destined_recipient_test.cpp
+++ b/src/mongo/db/s/resharding/resharding_destined_recipient_test.cpp
@@ -279,7 +279,7 @@ protected:
env.version.placementVersion().epoch(),
env.version.placementVersion().getTimestamp(),
Date_t::now(),
- UUID::gen(),
+ env.sourceUuid,
BSON(kShardKey << 1));
coll.setAllowMigrations(false);
diff --git a/src/mongo/db/s/scoped_collection_metadata.h b/src/mongo/db/s/scoped_collection_metadata.h
index 11b460bd279..2e0d512ca46 100644
--- a/src/mongo/db/s/scoped_collection_metadata.h
+++ b/src/mongo/db/s/scoped_collection_metadata.h
@@ -102,6 +102,10 @@ public:
return _impl->get().uuidMatches(uuid);
}
+ const UUID& getUUID() const {
+ return _impl->get().getUUID();
+ }
+
const boost::optional<TypeCollectionReshardingFields>& getReshardingFields() const {
return _impl->get().getReshardingFields();
}
diff --git a/src/mongo/db/shard_role.cpp b/src/mongo/db/shard_role.cpp
index cad506c8d74..8f5b36d01d2 100644
--- a/src/mongo/db/shard_role.cpp
+++ b/src/mongo/db/shard_role.cpp
@@ -35,6 +35,7 @@
#include <boost/optional.hpp>
#include <boost/optional/optional.hpp>
#include <boost/utility/in_place_factory.hpp> // IWYU pragma: keep
+#include <fmt/format.h>
#include <iterator>
#include <list>
#include <map>
@@ -86,6 +87,8 @@
namespace mongo {
+using namespace fmt::literals;
+
using TransactionResources = shard_role_details::TransactionResources;
namespace {
@@ -356,21 +359,32 @@ CollectionOrViewAcquisitions acquireResolvedCollectionsOrViewsWithoutTakingLocks
const bool isCollection =
holds_alternative<CollectionPtr>(snapshotedServices.collectionPtrOrView);
- if (holds_alternative<PlacementConcern>(prerequisites.placementConcern)) {
- const auto& placementConcern = get<PlacementConcern>(prerequisites.placementConcern);
-
- if (placementConcern.shardVersion == ShardVersion::UNSHARDED()) {
- shard_role_details::checkLocalCatalogIsValidForUnshardedShardVersion(
- opCtx,
- catalog,
- isCollection ? get<CollectionPtr>(snapshotedServices.collectionPtrOrView)
- : CollectionPtr::null,
- prerequisites.nss);
- }
+ const boost::optional<ShardVersion> placementConcernShardVersion =
+ holds_alternative<PlacementConcern>(prerequisites.placementConcern)
+ ? get<PlacementConcern>(prerequisites.placementConcern).shardVersion
+ : boost::none;
+
+ if (placementConcernShardVersion == ShardVersion::UNSHARDED()) {
+ shard_role_details::checkLocalCatalogIsValidForUnshardedShardVersion(
+ opCtx,
+ catalog,
+ isCollection ? get<CollectionPtr>(snapshotedServices.collectionPtrOrView)
+ : CollectionPtr::null,
+ prerequisites.nss);
}
if (isCollection) {
const auto& collectionPtr = get<CollectionPtr>(snapshotedServices.collectionPtrOrView);
+
+ if (placementConcernShardVersion && snapshotedServices.collectionDescription) {
+ shard_role_details::checkShardingAndLocalCatalogCollectionUUIDMatch(
+ opCtx,
+ prerequisites.nss,
+ *placementConcernShardVersion,
+ *snapshotedServices.collectionDescription,
+ collectionPtr);
+ }
+
invariant(!prerequisites.uuid || prerequisites.uuid == collectionPtr->uuid());
if (!prerequisites.uuid && collectionPtr) {
// If the uuid wasn't originally set on the AcquisitionRequest, set it now on the
@@ -1665,4 +1679,59 @@ void shard_role_details::checkLocalCatalogIsValidForUnshardedShardVersion(
}
}
+void shard_role_details::checkShardingAndLocalCatalogCollectionUUIDMatch(
+ OperationContext* opCtx,
+ const NamespaceString& nss,
+ const ShardVersion& requestedShardVersion,
+ const ScopedCollectionDescription& shardingCollectionDescription,
+ const CollectionPtr& collectionPtr) {
+ // Skip the check if the requested shard version corresponds to an untracked collection or
+ // corresponds to this shard not own any chunk. Also skip the check if the router attached
+ // ShardVersion::IGNORED, since in this case the router broadcasts request to shards that may
+ // not even own the collection at all (so they won't have any uuid on their local catalog).
+ if (requestedShardVersion == ShardVersion::UNSHARDED() ||
+ !requestedShardVersion.placementVersion().isSet() ||
+ ShardVersion::isPlacementVersionIgnored(requestedShardVersion)) {
+ return;
+ }
+
+ // Skip checking resharding temporary collections. The reason is that resharding registers the
+ // temporary collections on the sharding catalog before creating them on the shards, without
+ // holding any critical section.
+ // TODO: SERVER-87235 Remove this when resharding creates the temporary collections under a
+ // critical section.
+ if (nss.isTemporaryReshardingCollection()) {
+ return;
+ }
+
+ // Check that the collection uuid in the sharding catalog and the one on the local catalog
+ // match.
+ if (shardingCollectionDescription.hasRoutingTable() &&
+ (!collectionPtr || !shardingCollectionDescription.uuidMatches(collectionPtr->uuid()))) {
+ if ((opCtx->inMultiDocumentTransaction() ||
+ repl::ReadConcernArgs::get(opCtx).getArgsAtClusterTime())) {
+ // If in multi-document transaction or snapshot read, throw SnapshotUnavailable so that
+ // the transaction can be retried. This situation is known to be possible when a
+ // collection undergoes resharding, due to the resharding commit protocol. See
+ // SERVER-87061.
+ // TODO: SERVER-87235: Remove this condition and leave only the tassert below also for
+ // transaction and snapshot reads.
+ uasserted(
+ ErrorCodes::SnapshotUnavailable,
+ "Sharding catalog and local catalog collection uuid do not match. Nss: '{}', sharding uuid: '{}', local uuid: '{}'"_format(
+ nss.toStringForErrorMsg(),
+ shardingCollectionDescription.getUUID().toString(),
+ collectionPtr ? collectionPtr->uuid().toString() : ""));
+
+ } else {
+ tasserted(
+ 8706100,
+ "Sharding catalog and local catalog collection uuid do not match. Nss: '{}', sharding uuid: '{}', local uuid: '{}'"_format(
+ nss.toStringForErrorMsg(),
+ shardingCollectionDescription.getUUID().toString(),
+ collectionPtr ? collectionPtr->uuid().toString() : ""));
+ }
+ }
+}
+
} // namespace mongo
diff --git a/src/mongo/db/shard_role.h b/src/mongo/db/shard_role.h
index 2b79d39c8a4..b83e79a87a9 100644
--- a/src/mongo/db/shard_role.h
+++ b/src/mongo/db/shard_role.h
@@ -607,5 +607,15 @@ void checkLocalCatalogIsValidForUnshardedShardVersion(OperationContext* opCtx,
const CollectionPtr& collectionPtr,
const NamespaceString& nss);
+/*
+ * Check that the collection uuid on the sharding catalog and the local catalog match.
+ */
+void checkShardingAndLocalCatalogCollectionUUIDMatch(
+ OperationContext* opCtx,
+ const NamespaceString& nss,
+ const ShardVersion& requestedShardVersion,
+ const ScopedCollectionDescription& shardingCollectionDescription,
+ const CollectionPtr& collectionPtr);
+
} // namespace shard_role_details
} // namespace mongo
generated by cgit v1.2.3 (git 2.39.1) at 2026-09-11 11:17:58 +0000