summaryrefslogtreecommitdiff
diff options
context:
space:
mode:
authorGregory Wlodarek <gregory.wlodarek@mongodb.com>2022-08-31 22:11:08 +0000
committerEvergreen Agent <no-reply@evergreen.mongodb.com>2022-08-31 22:59:41 +0000
commit1a5cebdcd68691ce7e3f4e7b1caf367033000e7a (patch)
tree53bb57976704e3d5fc1361b0bae8f4c7e163bcd6
parentd7f979031a7c47deb3a887700b44706792233252 (diff)
downloadmongo-1a5cebdcd68691ce7e3f4e7b1caf367033000e7a.tar.gz
SERVER-68254 Create a function to instantiate collections and index entries from earlier points in time
-rw-r--r--src/mongo/db/catalog/collection.h8
-rw-r--r--src/mongo/db/catalog/collection_catalog.cpp102
-rw-r--r--src/mongo/db/catalog/collection_catalog.h24
-rw-r--r--src/mongo/db/catalog/collection_catalog_test.cpp512
-rw-r--r--src/mongo/db/catalog/collection_impl.cpp120
-rw-r--r--src/mongo/db/catalog/collection_impl.h8
-rw-r--r--src/mongo/db/catalog/index_catalog.h11
-rw-r--r--src/mongo/db/catalog/index_catalog_impl.cpp86
-rw-r--r--src/mongo/db/catalog/index_catalog_impl.h22
-rw-r--r--src/mongo/db/service_context_d_test_fixture.cpp6
-rw-r--r--src/mongo/db/service_context_d_test_fixture.h6
-rw-r--r--src/mongo/db/storage/storage_options.cpp1
-rw-r--r--src/mongo/db/storage/storage_options.h3
-rw-r--r--src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp6
14 files changed, 879 insertions, 36 deletions
diff --git a/src/mongo/db/catalog/collection.h b/src/mongo/db/catalog/collection.h
index 097b115864f..ad129695d7c 100644
--- a/src/mongo/db/catalog/collection.h
+++ b/src/mongo/db/catalog/collection.h
@@ -218,6 +218,14 @@ public:
virtual void init(OperationContext* opCtx) {}
+ /**
+ * Initializes a collection representative at the provided read timestamp using the shared state
+ * from an already existing, later collection.
+ */
+ virtual void initFromExisting(OperationContext* opCtx,
+ std::shared_ptr<Collection> collection,
+ Timestamp readTimestamp) {}
+
virtual bool isCommitted() const {
return true;
}
diff --git a/src/mongo/db/catalog/collection_catalog.cpp b/src/mongo/db/catalog/collection_catalog.cpp
index 1c4fd93cffb..29bb7c6093c 100644
--- a/src/mongo/db/catalog/collection_catalog.cpp
+++ b/src/mongo/db/catalog/collection_catalog.cpp
@@ -38,6 +38,7 @@
#include "mongo/db/multitenancy_gen.h"
#include "mongo/db/server_feature_flags_gen.h"
#include "mongo/db/server_options.h"
+#include "mongo/db/storage/kv/kv_engine.h"
#include "mongo/db/storage/recovery_unit.h"
#include "mongo/db/storage/snapshot_helper.h"
#include "mongo/db/storage/storage_parameters_gen.h"
@@ -61,6 +62,22 @@ std::shared_ptr<CollectionCatalog> batchedCatalogWriteInstance;
const OperationContext::Decoration<std::shared_ptr<const CollectionCatalog>> stashedCatalog =
OperationContext::declareDecoration<std::shared_ptr<const CollectionCatalog>>();
+/**
+ * Returns true if the collection is compatible with the read timestamp.
+ */
+bool isCollectionCompatible(std::shared_ptr<Collection> coll, Timestamp readTimestamp) {
+ if (!coll) {
+ return false;
+ }
+
+ boost::optional<Timestamp> minValidSnapshot = coll->getMinimumValidSnapshot();
+ if (!minValidSnapshot) {
+ // Collection is valid in all snapshots.
+ return true;
+ }
+ return readTimestamp >= *minValidSnapshot;
+}
+
} // namespace
class IgnoreExternalViewChangesForDatabase {
@@ -659,6 +676,91 @@ Status CollectionCatalog::reloadViews(OperationContext* opCtx, const DatabaseNam
return status;
}
+std::shared_ptr<Collection> CollectionCatalog::openCollection(OperationContext* opCtx,
+ const DurableCatalog::Entry& entry,
+ Timestamp readTimestamp) const {
+ if (!feature_flags::gPointInTimeCatalogLookups.isEnabledAndIgnoreFCV()) {
+ return nullptr;
+ }
+
+ auto metadata = DurableCatalog::get(opCtx)->getMetaData(opCtx, entry.catalogId);
+ if (!metadata) {
+ // Treat the collection as non-existent at a point-in-time the same as not-existent at
+ // latest.
+ return nullptr;
+ }
+
+ const auto& collectionOptions = metadata->options;
+
+ // Check if the collection already exists in the catalog and if it's compatible with the read
+ // timestamp.
+ std::shared_ptr<Collection> latestColl = _lookupCollectionByUUID(*collectionOptions.uuid);
+ if (isCollectionCompatible(latestColl, readTimestamp)) {
+ return latestColl;
+ }
+
+ // Check if the collection is drop pending, not expired, and compatible with the read timestamp.
+ std::shared_ptr<Collection> dropPendingColl = [&]() -> std::shared_ptr<Collection> {
+ auto dropPendingIt = _dropPendingCollection.find(entry.ident);
+ if (dropPendingIt == _dropPendingCollection.end()) {
+ return nullptr;
+ }
+
+ return dropPendingIt->second.lock();
+ }();
+
+ if (isCollectionCompatible(dropPendingColl, readTimestamp)) {
+ return dropPendingColl;
+ }
+
+ // Neither the latest collection or drop pending collection exist, or were compatible with the
+ // read timestamp. We'll need to instantiate a new Collection instance.
+ if (latestColl || dropPendingColl) {
+ // If the latest or drop pending collection exists, instantiate a new collection using
+ // their shared state.
+ LOGV2_DEBUG(6825400,
+ 1,
+ "Instantiating a collection using shared state",
+ logAttrs(entry.nss),
+ "ident"_attr = entry.ident,
+ "md"_attr = metadata->toBSON(),
+ "timestamp"_attr = readTimestamp);
+
+ std::shared_ptr<Collection> collToReturn = Collection::Factory::get(opCtx)->make(
+ opCtx, entry.nss, entry.catalogId, metadata, /*rs=*/nullptr);
+ collToReturn->initFromExisting(
+ opCtx, latestColl ? latestColl : dropPendingColl, readTimestamp);
+ return collToReturn;
+ }
+
+ // Instantiate a new collection without any shared state.
+ LOGV2_DEBUG(6825401,
+ 1,
+ "Instantiating a new collection",
+ logAttrs(entry.nss),
+ "ident"_attr = entry.ident,
+ "md"_attr = metadata->toBSON(),
+ "timestamp"_attr = readTimestamp);
+
+ std::unique_ptr<RecordStore> rs =
+ opCtx->getServiceContext()->getStorageEngine()->getEngine()->getRecordStore(
+ opCtx, entry.nss, entry.ident, collectionOptions);
+ std::shared_ptr<Collection> collToReturn = Collection::Factory::get(opCtx)->make(
+ opCtx, entry.nss, entry.catalogId, metadata, std::move(rs));
+ collToReturn->init(opCtx);
+ return collToReturn;
+}
+
+std::shared_ptr<IndexCatalogEntry> CollectionCatalog::findDropPendingIndex(
+ const std::string& ident) const {
+ auto it = _dropPendingIndex.find(ident);
+ if (it == _dropPendingIndex.end()) {
+ return nullptr;
+ }
+
+ return it->second.lock();
+}
+
void CollectionCatalog::onCreateCollection(OperationContext* opCtx,
std::shared_ptr<Collection> coll) const {
invariant(coll);
diff --git a/src/mongo/db/catalog/collection_catalog.h b/src/mongo/db/catalog/collection_catalog.h
index f38f85c5841..e5d2845a527 100644
--- a/src/mongo/db/catalog/collection_catalog.h
+++ b/src/mongo/db/catalog/collection_catalog.h
@@ -38,6 +38,8 @@
#include "mongo/db/database_name.h"
#include "mongo/db/profile_filter.h"
#include "mongo/db/service_context.h"
+// TODO SERVER-68265: remove include.
+#include "mongo/db/storage/durable_catalog.h"
#include "mongo/db/views/view.h"
#include "mongo/stdx/unordered_map.h"
#include "mongo/util/uuid.h"
@@ -206,6 +208,28 @@ public:
Status reloadViews(OperationContext* opCtx, const DatabaseName& dbName) const;
/**
+ * Returns the collection instance representative of 'entry' at the provided read timestamp.
+ *
+ * TODO SERVER-68571:
+ * - If the data files have already been removed, return nullptr.
+ * - Update the drop pending map ident token when not initialized from shared state.
+ *
+ * TODO SERVER-68265:
+ * - Use NamespaceString instead of DurableCatalog::Entry.
+ * - Remove DurableCatalog dependency.
+ *
+ * Returns nullptr when reading from a point-in-time where the collection did not exist.
+ */
+ std::shared_ptr<Collection> openCollection(OperationContext* opCtx,
+ const DurableCatalog::Entry& entry,
+ Timestamp readTimestamp) const;
+
+ /**
+ * Returns a shared_ptr to a drop pending index if it's found and not expired.
+ */
+ std::shared_ptr<IndexCatalogEntry> findDropPendingIndex(const std::string& ident) const;
+
+ /**
* Handles committing a collection to the catalog within a WriteUnitOfWork.
*
* Must be called within a WriteUnitOfWork.
diff --git a/src/mongo/db/catalog/collection_catalog_test.cpp b/src/mongo/db/catalog/collection_catalog_test.cpp
index 1d7632709ec..43a089fea2b 100644
--- a/src/mongo/db/catalog/collection_catalog_test.cpp
+++ b/src/mongo/db/catalog/collection_catalog_test.cpp
@@ -34,8 +34,11 @@
#include "mongo/db/catalog/catalog_test_fixture.h"
#include "mongo/db/catalog/collection_catalog_helper.h"
#include "mongo/db/catalog/collection_mock.h"
+#include "mongo/db/catalog_raii.h"
#include "mongo/db/concurrency/lock_manager_defs.h"
+#include "mongo/db/index_builds_coordinator.h"
#include "mongo/db/service_context_d_test_fixture.h"
+#include "mongo/idl/server_parameter_test_util.h"
#include "mongo/unittest/death_test.h"
#include "mongo/unittest/unittest.h"
@@ -857,5 +860,514 @@ TEST_F(ForEachCollectionFromDbTest, ForEachCollectionFromDbWithPredicate) {
}
}
+/**
+ * RAII type for operating at a timestamp. Will remove any timestamping when the object destructs.
+ */
+class OneOffRead {
+public:
+ OneOffRead(OperationContext* opCtx, const Timestamp& ts) : _opCtx(opCtx) {
+ _opCtx->recoveryUnit()->abandonSnapshot();
+ if (ts.isNull()) {
+ _opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp);
+ } else {
+ _opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kProvided, ts);
+ }
+ }
+
+ ~OneOffRead() {
+ _opCtx->recoveryUnit()->abandonSnapshot();
+ _opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp);
+ }
+
+private:
+ OperationContext* _opCtx;
+};
+
+class CollectionCatalogTimestampTest : public ServiceContextMongoDTest {
+public:
+ // Disable table logging. When table logging is enabled, timestamps are discarded by WiredTiger.
+ CollectionCatalogTimestampTest()
+ : ServiceContextMongoDTest(Options{}.forceDisableTableLogging()) {}
+
+ void setUp() override {
+ ServiceContextMongoDTest::setUp();
+ opCtx = makeOperationContext();
+ }
+
+ void createCollection(OperationContext* opCtx,
+ const NamespaceString& nss,
+ Timestamp timestamp) {
+ opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp);
+ opCtx->recoveryUnit()->abandonSnapshot();
+
+ if (!opCtx->recoveryUnit()->getCommitTimestamp().isNull()) {
+ opCtx->recoveryUnit()->clearCommitTimestamp();
+ }
+ opCtx->recoveryUnit()->setCommitTimestamp(timestamp);
+
+ AutoGetDb databaseWriteGuard(opCtx, nss.dbName(), MODE_IX);
+ auto db = databaseWriteGuard.ensureDbExists(opCtx);
+ ASSERT(db);
+
+ Lock::CollectionLock lk(opCtx, nss, MODE_IX);
+
+ WriteUnitOfWork wuow(opCtx);
+ CollectionOptions options;
+ options.uuid.emplace(UUID::gen());
+
+ auto storageEngine = opCtx->getServiceContext()->getStorageEngine();
+ std::pair<RecordId, std::unique_ptr<RecordStore>> catalogIdRecordStorePair =
+ uassertStatusOK(storageEngine->getCatalog()->createCollection(
+ opCtx, nss, options, /*allocateDefaultSpace=*/true));
+ auto& catalogId = catalogIdRecordStorePair.first;
+ std::shared_ptr<Collection> ownedCollection = Collection::Factory::get(opCtx)->make(
+ opCtx, nss, catalogId, options, std::move(catalogIdRecordStorePair.second));
+ ownedCollection->init(opCtx);
+ ownedCollection->setCommitted(false);
+
+ CollectionCatalog::get(opCtx)->onCreateCollection(opCtx, std::move(ownedCollection));
+
+ wuow.commit();
+ }
+
+ void dropCollection(OperationContext* opCtx, const NamespaceString& nss, Timestamp timestamp) {
+ opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp);
+ opCtx->recoveryUnit()->abandonSnapshot();
+
+ if (!opCtx->recoveryUnit()->getCommitTimestamp().isNull()) {
+ opCtx->recoveryUnit()->clearCommitTimestamp();
+ }
+ opCtx->recoveryUnit()->setCommitTimestamp(timestamp);
+
+ Lock::DBLock dbLk(opCtx, nss.db(), MODE_IX);
+ Lock::CollectionLock collLk(opCtx, nss, MODE_X);
+ CollectionWriter collection(opCtx, nss);
+
+ WriteUnitOfWork wuow(opCtx);
+ CollectionCatalog::get(opCtx)->dropCollection(
+ opCtx, collection.getWritableCollection(opCtx), /*isDropPending=*/true);
+ wuow.commit();
+ }
+
+ void createIndex(OperationContext* opCtx,
+ const NamespaceString& nss,
+ BSONObj indexSpec,
+ Timestamp timestamp) {
+ opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp);
+ opCtx->recoveryUnit()->abandonSnapshot();
+
+ if (!opCtx->recoveryUnit()->getCommitTimestamp().isNull()) {
+ opCtx->recoveryUnit()->clearCommitTimestamp();
+ }
+ opCtx->recoveryUnit()->setCommitTimestamp(timestamp);
+
+ AutoGetCollection autoColl(opCtx, nss, MODE_X);
+
+ WriteUnitOfWork wuow(opCtx);
+ CollectionWriter collection(opCtx, nss);
+
+ IndexBuildsCoordinator::get(opCtx)->createIndexesOnEmptyCollection(
+ opCtx, collection, {indexSpec}, /*fromMigrate=*/false);
+
+ wuow.commit();
+ }
+
+ void dropIndex(OperationContext* opCtx,
+ const NamespaceString& nss,
+ const std::string& indexName,
+ Timestamp timestamp) {
+ opCtx->recoveryUnit()->setTimestampReadSource(RecoveryUnit::ReadSource::kNoTimestamp);
+ opCtx->recoveryUnit()->abandonSnapshot();
+
+ if (!opCtx->recoveryUnit()->getCommitTimestamp().isNull()) {
+ opCtx->recoveryUnit()->clearCommitTimestamp();
+ }
+ opCtx->recoveryUnit()->setCommitTimestamp(timestamp);
+
+ AutoGetCollection autoColl(opCtx, nss, MODE_X);
+
+ WriteUnitOfWork wuow(opCtx);
+ CollectionWriter collection(opCtx, nss);
+
+ Collection* writableCollection = collection.getWritableCollection(opCtx);
+
+ IndexCatalog* indexCatalog = writableCollection->getIndexCatalog();
+ auto indexDescriptor =
+ indexCatalog->findIndexByName(opCtx, indexName, IndexCatalog::InclusionPolicy::kReady);
+ ASSERT_OK(indexCatalog->dropIndex(opCtx, writableCollection, indexDescriptor));
+
+ wuow.commit();
+ }
+
+ DurableCatalog::Entry getEntry(OperationContext* opCtx, const NamespaceString& nss) {
+ AutoGetCollection autoColl(opCtx, nss, LockMode::MODE_IS);
+ auto durableCatalog = DurableCatalog::get(opCtx);
+ return durableCatalog->getEntry(autoColl->getCatalogId());
+ }
+
+ std::shared_ptr<Collection> openCollection(OperationContext* opCtx,
+ const DurableCatalog::Entry& entry,
+ Timestamp readTimestamp) {
+ Lock::GlobalLock globalLock(opCtx, MODE_IS);
+ auto coll = CollectionCatalog::get(opCtx)->openCollection(opCtx, entry, readTimestamp);
+ ASSERT(coll);
+ return coll;
+ }
+
+protected:
+ ServiceContext::UniqueOperationContext opCtx;
+};
+
+TEST_F(CollectionCatalogTimestampTest, MinimumValidSnapshot) {
+ const NamespaceString nss("a.b");
+ const Timestamp createCollectionTs = Timestamp(10, 10);
+ const Timestamp createXIndexTs = Timestamp(20, 20);
+ const Timestamp createYIndexTs = Timestamp(30, 30);
+ const Timestamp dropIndexTs = Timestamp(40, 40);
+
+ createCollection(opCtx.get(), nss, createCollectionTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "x_1"
+ << "key" << BSON("x" << 1)),
+ createXIndexTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "y_1"
+ << "key" << BSON("y" << 1)),
+ createYIndexTs);
+
+ auto coll = CollectionCatalog::get(opCtx.get())->lookupCollectionByNamespace(opCtx.get(), nss);
+ ASSERT(coll);
+ ASSERT_EQ(coll->getMinimumVisibleSnapshot(), createCollectionTs);
+ ASSERT_EQ(coll->getMinimumValidSnapshot(), createYIndexTs);
+
+ const IndexDescriptor* desc = coll->getIndexCatalog()->findIndexByName(opCtx.get(), "x_1");
+ const IndexCatalogEntry* entry = coll->getIndexCatalog()->getEntry(desc);
+ ASSERT_EQ(entry->getMinimumVisibleSnapshot(), createXIndexTs);
+
+ desc = coll->getIndexCatalog()->findIndexByName(opCtx.get(), "y_1");
+ entry = coll->getIndexCatalog()->getEntry(desc);
+ ASSERT_EQ(entry->getMinimumVisibleSnapshot(), createYIndexTs);
+
+ dropIndex(opCtx.get(), nss, "x_1", dropIndexTs);
+ dropIndex(opCtx.get(), nss, "y_1", dropIndexTs);
+
+ // Fetch the latest collection instance without the indexes.
+ coll = CollectionCatalog::get(opCtx.get())->lookupCollectionByNamespace(opCtx.get(), nss);
+ ASSERT(coll);
+ ASSERT_EQ(coll->getMinimumVisibleSnapshot(), createCollectionTs);
+ ASSERT_EQ(coll->getMinimumValidSnapshot(), dropIndexTs);
+}
+
+TEST_F(CollectionCatalogTimestampTest, OpenCollectionBeforeCreateTimestamp) {
+ RAIIServerParameterControllerForTest featureFlagController(
+ "featureFlagPointInTimeCatalogLookups", true);
+
+ const NamespaceString nss("a.b");
+ const Timestamp createCollectionTs = Timestamp(10, 10);
+
+ createCollection(opCtx.get(), nss, createCollectionTs);
+ DurableCatalog::Entry entry = getEntry(opCtx.get(), nss);
+
+ // Try to open the collection before it was created.
+ const Timestamp readTimestamp(5, 5);
+ Lock::GlobalLock globalLock(opCtx.get(), MODE_IS);
+ OneOffRead oor(opCtx.get(), readTimestamp);
+ auto coll =
+ CollectionCatalog::get(opCtx.get())->openCollection(opCtx.get(), entry, readTimestamp);
+ ASSERT(!coll);
+}
+
+TEST_F(CollectionCatalogTimestampTest, OpenEarlierCollection) {
+ RAIIServerParameterControllerForTest featureFlagController(
+ "featureFlagPointInTimeCatalogLookups", true);
+
+ const NamespaceString nss("a.b");
+ const Timestamp createCollectionTs = Timestamp(10, 10);
+ const Timestamp createIndexTs = Timestamp(20, 20);
+
+ createCollection(opCtx.get(), nss, createCollectionTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "x_1"
+ << "key" << BSON("x" << 1)),
+ createIndexTs);
+
+ DurableCatalog::Entry entry = getEntry(opCtx.get(), nss);
+
+ // Open an instance of the collection before the index was created.
+ const Timestamp readTimestamp(15, 15);
+ OneOffRead oor(opCtx.get(), readTimestamp);
+ std::shared_ptr<Collection> coll = openCollection(opCtx.get(), entry, readTimestamp);
+ ASSERT(coll);
+ ASSERT_EQ(0, coll->getIndexCatalog()->numIndexesTotal(opCtx.get()));
+
+ // Verify that the CollectionCatalog returns the latest collection with the index present.
+ auto latestColl = CollectionCatalog::get(opCtx.get())
+ ->lookupCollectionByNamespaceForRead(opCtx.get(), entry.nss);
+ ASSERT(latestColl);
+ ASSERT_EQ(1, latestColl->getIndexCatalog()->numIndexesTotal(opCtx.get()));
+
+ // Ensure the idents are shared between the collection instances.
+ ASSERT_NE(coll, latestColl);
+ ASSERT_EQ(coll->getSharedIdent(), latestColl->getSharedIdent());
+}
+
+TEST_F(CollectionCatalogTimestampTest, OpenEarlierCollectionWithIndex) {
+ RAIIServerParameterControllerForTest featureFlagController(
+ "featureFlagPointInTimeCatalogLookups", true);
+
+ const NamespaceString nss("a.b");
+ const Timestamp createCollectionTs = Timestamp(10, 10);
+ const Timestamp createXIndexTs = Timestamp(20, 20);
+ const Timestamp createYIndexTs = Timestamp(30, 30);
+
+ createCollection(opCtx.get(), nss, createCollectionTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "x_1"
+ << "key" << BSON("x" << 1)),
+ createXIndexTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "y_1"
+ << "key" << BSON("y" << 1)),
+ createYIndexTs);
+
+ DurableCatalog::Entry entry = getEntry(opCtx.get(), nss);
+
+ // Open an instance of the collection when only one of the two indexes were present.
+ const Timestamp readTimestamp(25, 25);
+ OneOffRead oor(opCtx.get(), readTimestamp);
+ std::shared_ptr<Collection> coll = openCollection(opCtx.get(), entry, readTimestamp);
+ ASSERT(coll);
+ ASSERT_EQ(1, coll->getIndexCatalog()->numIndexesTotal(opCtx.get()));
+
+ // Verify that the CollectionCatalog returns the latest collection.
+ auto latestColl = CollectionCatalog::get(opCtx.get())
+ ->lookupCollectionByNamespaceForRead(opCtx.get(), entry.nss);
+ ASSERT(latestColl);
+ ASSERT_EQ(2, latestColl->getIndexCatalog()->numIndexesTotal(opCtx.get()));
+
+ // Ensure the idents are shared between the collection and index instances.
+ ASSERT_NE(coll, latestColl);
+ ASSERT_EQ(coll->getSharedIdent(), latestColl->getSharedIdent());
+
+ auto indexDescPast = coll->getIndexCatalog()->findIndexByName(opCtx.get(), "x_1");
+ auto indexDescLatest = latestColl->getIndexCatalog()->findIndexByName(opCtx.get(), "x_1");
+ ASSERT_BSONOBJ_EQ(indexDescPast->infoObj(), indexDescLatest->infoObj());
+ ASSERT_EQ(coll->getIndexCatalog()->getEntryShared(indexDescPast),
+ latestColl->getIndexCatalog()->getEntryShared(indexDescLatest));
+}
+
+TEST_F(CollectionCatalogTimestampTest, OpenLatestCollectionWithIndex) {
+ RAIIServerParameterControllerForTest featureFlagController(
+ "featureFlagPointInTimeCatalogLookups", true);
+
+ const NamespaceString nss("a.b");
+ const Timestamp createCollectionTs = Timestamp(10, 10);
+ const Timestamp createXIndexTs = Timestamp(20, 20);
+
+ createCollection(opCtx.get(), nss, createCollectionTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "x_1"
+ << "key" << BSON("x" << 1)),
+ createXIndexTs);
+
+ DurableCatalog::Entry entry = getEntry(opCtx.get(), nss);
+
+ // Setting the read timestamp to the last DDL operation on the collection returns the latest
+ // collection.
+ const Timestamp readTimestamp(20, 20);
+ OneOffRead oor(opCtx.get(), readTimestamp);
+ std::shared_ptr<Collection> coll = openCollection(opCtx.get(), entry, readTimestamp);
+ ASSERT(coll);
+
+ // Verify that the CollectionCatalog returns the latest collection.
+ auto currentColl = CollectionCatalog::get(opCtx.get())
+ ->lookupCollectionByNamespaceForRead(opCtx.get(), entry.nss);
+ ASSERT_EQ(coll, currentColl);
+
+ // Ensure the idents are shared between the collection and index instances.
+ ASSERT_EQ(coll->getSharedIdent(), currentColl->getSharedIdent());
+
+ auto indexDesc = coll->getIndexCatalog()->findIndexByName(opCtx.get(), "x_1");
+ auto indexDescCurrent = currentColl->getIndexCatalog()->findIndexByName(opCtx.get(), "x_1");
+ ASSERT_BSONOBJ_EQ(indexDesc->infoObj(), indexDescCurrent->infoObj());
+ ASSERT_EQ(coll->getIndexCatalog()->getEntryShared(indexDesc),
+ currentColl->getIndexCatalog()->getEntryShared(indexDescCurrent));
+}
+
+TEST_F(CollectionCatalogTimestampTest, OpenEarlierCollectionWithDropPendingIndex) {
+ RAIIServerParameterControllerForTest featureFlagController(
+ "featureFlagPointInTimeCatalogLookups", true);
+
+ const NamespaceString nss("a.b");
+ const Timestamp createCollectionTs = Timestamp(10, 10);
+ const Timestamp createIndexTs = Timestamp(20, 20);
+ const Timestamp dropIndexTs = Timestamp(30, 30);
+
+ createCollection(opCtx.get(), nss, createCollectionTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "x_1"
+ << "key" << BSON("x" << 1)),
+ createIndexTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "y_1"
+ << "key" << BSON("y" << 1)),
+ createIndexTs);
+
+ DurableCatalog::Entry entry = getEntry(opCtx.get(), nss);
+
+ // Maintain a shared_ptr to "x_1", so it's not expired in drop pending map, but not for "y_1".
+ std::shared_ptr<const IndexCatalogEntry> index;
+ {
+ auto latestColl = CollectionCatalog::get(opCtx.get())
+ ->lookupCollectionByNamespaceForRead(opCtx.get(), entry.nss);
+ auto desc = latestColl->getIndexCatalog()->findIndexByName(opCtx.get(), "x_1");
+ index = latestColl->getIndexCatalog()->getEntryShared(desc);
+ }
+
+ dropIndex(opCtx.get(), nss, "x_1", dropIndexTs);
+ dropIndex(opCtx.get(), nss, "y_1", dropIndexTs);
+
+ // Open the collection while both indexes were present.
+ const Timestamp readTimestamp(20, 20);
+ OneOffRead oor(opCtx.get(), readTimestamp);
+ std::shared_ptr<Collection> coll = openCollection(opCtx.get(), entry, readTimestamp);
+ ASSERT(coll);
+
+ // Collection is not shared from the latest instance.
+ auto latestColl = CollectionCatalog::get(opCtx.get())
+ ->lookupCollectionByNamespaceForRead(opCtx.get(), entry.nss);
+ ASSERT_NE(coll, latestColl);
+
+ auto indexDescX = coll->getIndexCatalog()->findIndexByName(opCtx.get(), "x_1");
+ auto indexDescY = coll->getIndexCatalog()->findIndexByName(opCtx.get(), "y_1");
+
+ auto indexEntryX = coll->getIndexCatalog()->getEntryShared(indexDescX);
+ auto indexEntryY = coll->getIndexCatalog()->getEntryShared(indexDescY);
+
+ // Check use_count(). 2 in the unit test, 1 in the opened collection.
+ ASSERT_EQ(3, indexEntryX.use_count());
+
+ // Check use_count(). 1 in the unit test, 1 in the opened collection.
+ ASSERT_EQ(2, indexEntryY.use_count());
+
+ // Verify that "x_1" was retrieved from the drop pending map for the opened collection.
+ ASSERT_EQ(index, indexEntryX);
+}
+
+TEST_F(CollectionCatalogTimestampTest, OpenEarlierAlreadyDropPendingCollection) {
+ RAIIServerParameterControllerForTest featureFlagController(
+ "featureFlagPointInTimeCatalogLookups", true);
+
+ const NamespaceString firstNss("a.b");
+ const NamespaceString secondNss("c.d");
+ const Timestamp createCollectionTs = Timestamp(10, 10);
+ const Timestamp dropCollectionTs = Timestamp(30, 30);
+
+ createCollection(opCtx.get(), firstNss, createCollectionTs);
+ createCollection(opCtx.get(), secondNss, createCollectionTs);
+
+ DurableCatalog::Entry firstEntry = getEntry(opCtx.get(), firstNss);
+ DurableCatalog::Entry secondEntry = getEntry(opCtx.get(), secondNss);
+
+ // Maintain a shared_ptr to "a.b" so that it isn't expired in the drop pending map after we drop
+ // the collections.
+ std::shared_ptr<const Collection> coll =
+ CollectionCatalog::get(opCtx.get())
+ ->lookupCollectionByNamespaceForRead(opCtx.get(), firstNss);
+ ASSERT(coll);
+
+ // Make the collections drop pending.
+ dropCollection(opCtx.get(), firstNss, dropCollectionTs);
+ dropCollection(opCtx.get(), secondNss, dropCollectionTs);
+
+ // Set the read timestamp to be before the drop timestamp.
+ const Timestamp readTimestamp(20, 20);
+ OneOffRead oor(opCtx.get(), readTimestamp);
+
+ {
+ // Open "a.b", which is not expired in the drop pending map.
+ std::shared_ptr<Collection> openedColl =
+ openCollection(opCtx.get(), firstEntry, readTimestamp);
+ ASSERT(openedColl);
+ ASSERT_EQ(coll, openedColl);
+
+ // Check use_count(). 2 in the unit test.
+ ASSERT_EQ(2, openedColl.use_count());
+ }
+
+ {
+ // Open "c.d" which is expired in the drop pending map.
+ std::shared_ptr<Collection> openedColl =
+ openCollection(opCtx.get(), secondEntry, readTimestamp);
+ ASSERT(openedColl);
+ ASSERT_NE(coll, openedColl);
+
+ // Check use_count(). 1 in the unit test.
+ ASSERT_EQ(1, openedColl.use_count());
+ }
+}
+
+TEST_F(CollectionCatalogTimestampTest, OpenNewCollectionUsingDropPendingCollectionSharedState) {
+ RAIIServerParameterControllerForTest featureFlagController(
+ "featureFlagPointInTimeCatalogLookups", true);
+
+ const NamespaceString nss("a.b");
+ const Timestamp createCollectionTs = Timestamp(10, 10);
+ const Timestamp createIndexTs = Timestamp(20, 20);
+ const Timestamp dropCollectionTs = Timestamp(30, 30);
+
+ createCollection(opCtx.get(), nss, createCollectionTs);
+ createIndex(opCtx.get(),
+ nss,
+ BSON("v" << 2 << "name"
+ << "x_1"
+ << "key" << BSON("x" << 1)),
+ createIndexTs);
+
+ DurableCatalog::Entry entry = getEntry(opCtx.get(), nss);
+
+ // Maintain a shared_ptr to "a.b" so that it isn't expired in the drop pending map after we drop
+ // it.
+ std::shared_ptr<const Collection> coll =
+ CollectionCatalog::get(opCtx.get())->lookupCollectionByNamespaceForRead(opCtx.get(), nss);
+ ASSERT(coll);
+ ASSERT_EQ(*coll->getMinimumValidSnapshot(), createIndexTs);
+
+ // Make the collection drop pending.
+ dropCollection(opCtx.get(), nss, dropCollectionTs);
+
+ // Open the collection before the index was created. The drop pending collection is incompatible
+ // as it has an index entry. But we can still use the drop pending collections shared state to
+ // instantiate a new collection.
+ const Timestamp readTimestamp(10, 10);
+ OneOffRead oor(opCtx.get(), readTimestamp);
+
+ std::shared_ptr<Collection> openedColl = openCollection(opCtx.get(), entry, readTimestamp);
+ ASSERT(openedColl);
+ ASSERT_NE(coll, openedColl);
+
+ // Check use_count(). 1 in the unit test.
+ ASSERT_EQ(1, openedColl.use_count());
+
+ // Ensure the idents are shared between the opened collection and the drop pending collection.
+ ASSERT_EQ(coll->getSharedIdent(), openedColl->getSharedIdent());
+}
+
} // namespace
} // namespace mongo
diff --git a/src/mongo/db/catalog/collection_impl.cpp b/src/mongo/db/catalog/collection_impl.cpp
index 1f37bb989a0..d61a693869f 100644
--- a/src/mongo/db/catalog/collection_impl.cpp
+++ b/src/mongo/db/catalog/collection_impl.cpp
@@ -366,10 +366,107 @@ void CollectionImpl::init(OperationContext* opCtx) {
const auto& collectionOptions = _metadata->options;
_shared->_collator = parseCollation(opCtx, _ns, collectionOptions.collation);
+
+ _initCommon(opCtx);
+
+ if (collectionOptions.clusteredIndex) {
+ if (collectionOptions.expireAfterSeconds) {
+ // If this collection has been newly created, we need to register with the TTL cache at
+ // commit time, otherwise it is startup and we can register immediately.
+ auto svcCtx = opCtx->getClient()->getServiceContext();
+ auto uuid = *collectionOptions.uuid;
+ if (opCtx->lockState()->inAWriteUnitOfWork()) {
+ opCtx->recoveryUnit()->onCommit([svcCtx, uuid](auto ts) {
+ TTLCollectionCache::get(svcCtx).registerTTLInfo(
+ uuid, TTLCollectionCache::Info{TTLCollectionCache::ClusteredId{}});
+ });
+ } else {
+ TTLCollectionCache::get(svcCtx).registerTTLInfo(
+ uuid, TTLCollectionCache::Info{TTLCollectionCache::ClusteredId{}});
+ }
+ }
+ }
+
+ uassertStatusOK(getIndexCatalog()->init(opCtx, this));
+ _initialized = true;
+}
+
+void CollectionImpl::initFromExisting(OperationContext* opCtx,
+ std::shared_ptr<Collection> collection,
+ Timestamp readTimestamp) {
+ LOGV2_DEBUG(
+ 6825402, 1, "Initializing collection using shared state", logAttrs(collection->ns()));
+
+ // We are per definition committed if we initialize from an existing collection.
+ _cachedCommitted = true;
+
+ // Use the shared state from the existing collection.
+ _shared = static_cast<CollectionImpl*>(collection.get())->_shared;
+
+ // When initializing a collection from an earlier point-in-time, we don't know when the last DDL
+ // operation took place at that point-in-time. We conservatively set the minimum valid snapshot
+ // to the read point-in-time.
+ _minVisibleSnapshot = readTimestamp;
+ _minValidSnapshot = readTimestamp;
+
+ _initCommon(opCtx);
+
+ // Determine which indexes from the existing collection can be shared with this newly
+ // initialized collection. The remaining indexes will be initialized by the IndexCatalog.
+ IndexCatalogEntryContainer preexistingIndexes;
+ for (const auto& index : _metadata->indexes) {
+ // First check the index catalog of the existing collection for the index entry.
+ std::shared_ptr<IndexCatalogEntry> entry = [&]() -> std::shared_ptr<IndexCatalogEntry> {
+ auto desc =
+ collection->getIndexCatalog()->findIndexByName(opCtx, index.nameStringData());
+ if (!desc)
+ return nullptr;
+
+ auto entry = collection->getIndexCatalog()->getEntryShared(desc);
+ if (!entry->getMinimumVisibleSnapshot())
+ // Index is valid in all snapshots.
+ return entry;
+ return readTimestamp < *entry->getMinimumVisibleSnapshot() ? nullptr : entry;
+ }();
+
+ if (entry) {
+ preexistingIndexes.add(std::move(entry));
+ continue;
+ }
+
+ // Next check the CollectionCatalog for a compatible drop pending index.
+ entry = [&]() -> std::shared_ptr<IndexCatalogEntry> {
+ const std::string ident = DurableCatalog::get(opCtx)->getIndexIdent(
+ opCtx, _catalogId, index.nameStringData());
+ auto entry = CollectionCatalog::get(opCtx)->findDropPendingIndex(ident);
+ if (!entry)
+ return nullptr;
+
+ if (!entry->getMinimumVisibleSnapshot())
+ // Index is valid in all snapshots.
+ return entry;
+ return readTimestamp < *entry->getMinimumVisibleSnapshot() ? nullptr : entry;
+ }();
+
+ if (entry) {
+ preexistingIndexes.add(std::move(entry));
+ continue;
+ }
+ }
+
+ uassertStatusOK(
+ getIndexCatalog()->initFromExisting(opCtx, this, preexistingIndexes, readTimestamp));
+ _initialized = true;
+}
+
+void CollectionImpl::_initCommon(OperationContext* opCtx) {
+ invariant(!_initialized);
+
+ const auto& collectionOptions = _metadata->options;
auto validatorDoc = collectionOptions.validator.getOwned();
// Enforce that the validator can be used on this namespace.
- uassertStatusOK(checkValidatorCanBeUsedOnNs(validatorDoc, ns(), _uuid));
+ uassertStatusOK(checkValidatorCanBeUsedOnNs(validatorDoc, _ns, _uuid));
// Make sure validationAction and validationLevel are allowed on this collection
uassertStatusOK(checkValidationOptionsCanBeUsed(
@@ -399,27 +496,6 @@ void CollectionImpl::init(OperationContext* opCtx) {
logAttrs(_ns),
"validatorStatus"_attr = _validator.getStatus());
}
-
- if (collectionOptions.clusteredIndex) {
- if (collectionOptions.expireAfterSeconds) {
- // If this collection has been newly created, we need to register with the TTL cache at
- // commit time, otherwise it is startup and we can register immediately.
- auto svcCtx = opCtx->getClient()->getServiceContext();
- auto uuid = *collectionOptions.uuid;
- if (opCtx->lockState()->inAWriteUnitOfWork()) {
- opCtx->recoveryUnit()->onCommit([svcCtx, uuid](auto ts) {
- TTLCollectionCache::get(svcCtx).registerTTLInfo(
- uuid, TTLCollectionCache::Info{TTLCollectionCache::ClusteredId{}});
- });
- } else {
- TTLCollectionCache::get(svcCtx).registerTTLInfo(
- uuid, TTLCollectionCache::Info{TTLCollectionCache::ClusteredId{}});
- }
- }
- }
-
- getIndexCatalog()->init(opCtx, this).transitional_ignore();
- _initialized = true;
}
bool CollectionImpl::isInitialized() const {
diff --git a/src/mongo/db/catalog/collection_impl.h b/src/mongo/db/catalog/collection_impl.h
index cf7cfaaf805..4300a207635 100644
--- a/src/mongo/db/catalog/collection_impl.h
+++ b/src/mongo/db/catalog/collection_impl.h
@@ -73,6 +73,9 @@ public:
SharedCollectionDecorations* getSharedDecorations() const final;
void init(OperationContext* opCtx) final;
+ void initFromExisting(OperationContext* opCtx,
+ std::shared_ptr<Collection> collection,
+ Timestamp readTimestamp) final;
bool isInitialized() const final;
bool isCommitted() const final;
void setCommitted(bool val) final;
@@ -445,6 +448,11 @@ private:
void _writeMetadata(OperationContext* opCtx, Func func);
/**
+ * Helper for init() and initFromExisting() to initialize common state.
+ */
+ void _initCommon(OperationContext* opCtx);
+
+ /**
* Holder of shared state between CollectionImpl clones
*/
struct SharedState {
diff --git a/src/mongo/db/catalog/index_catalog.h b/src/mongo/db/catalog/index_catalog.h
index 625a7bc56cc..e02733eaace 100644
--- a/src/mongo/db/catalog/index_catalog.h
+++ b/src/mongo/db/catalog/index_catalog.h
@@ -217,6 +217,17 @@ public:
// Must be called before used.
virtual Status init(OperationContext* opCtx, Collection* collection) = 0;
+ /**
+ * Must be called before used.
+ *
+ * When initializing an index that exists in 'preexistingIndexes', the IndexCatalogEntry will be
+ * taken from there instead of initializing a new IndexCatalogEntry.
+ */
+ virtual Status initFromExisting(OperationContext* opCtx,
+ Collection* collection,
+ const IndexCatalogEntryContainer& preexistingIndexes,
+ boost::optional<Timestamp> readTimestamp) = 0;
+
// ---- accessors -----
virtual bool haveAnyIndexes() const = 0;
diff --git a/src/mongo/db/catalog/index_catalog_impl.cpp b/src/mongo/db/catalog/index_catalog_impl.cpp
index 3b0bc930e25..59f7924b27e 100644
--- a/src/mongo/db/catalog/index_catalog_impl.cpp
+++ b/src/mongo/db/catalog/index_catalog_impl.cpp
@@ -46,6 +46,7 @@
#include "mongo/db/curop.h"
#include "mongo/db/field_ref.h"
#include "mongo/db/fts/fts_spec.h"
+#include "mongo/db/global_settings.h"
#include "mongo/db/index/index_access_method.h"
#include "mongo/db/index/index_descriptor.h"
#include "mongo/db/index/s2_access_method.h"
@@ -184,6 +185,20 @@ std::unique_ptr<IndexCatalog> IndexCatalogImpl::clone() const {
}
Status IndexCatalogImpl::init(OperationContext* opCtx, Collection* collection) {
+ return _init(opCtx, collection, /*preexistingIndexes=*/{}, /*readTimestamp=*/boost::none);
+};
+
+Status IndexCatalogImpl::initFromExisting(OperationContext* opCtx,
+ Collection* collection,
+ const IndexCatalogEntryContainer& preexistingIndexes,
+ boost::optional<Timestamp> readTimestamp) {
+ return _init(opCtx, collection, preexistingIndexes, readTimestamp);
+};
+
+Status IndexCatalogImpl::_init(OperationContext* opCtx,
+ Collection* collection,
+ const IndexCatalogEntryContainer& preexistingIndexes,
+ boost::optional<Timestamp> readTimestamp) {
vector<string> indexNames;
collection->getAllIndexes(&indexNames);
const bool replSetMemberInStandaloneMode =
@@ -200,6 +215,32 @@ Status IndexCatalogImpl::init(OperationContext* opCtx, Collection* collection) {
BSONObj spec = collection->getIndexSpec(indexName).getOwned();
BSONObj keyPattern = spec.getObjectField("key");
+ if (readTimestamp) {
+ auto preexistingIt = std::find_if(
+ preexistingIndexes.begin(),
+ preexistingIndexes.end(),
+ [&](const std::shared_ptr<IndexCatalogEntry>& preexistingEntry) {
+ return spec.woCompare(preexistingEntry->descriptor()->infoObj()) == 0;
+ });
+ if (preexistingIt != preexistingIndexes.end()) {
+ std::shared_ptr<IndexCatalogEntry> existingEntry = *preexistingIt;
+ LOGV2_DEBUG(6825404,
+ 1,
+ "Adding preexisting index entry",
+ logAttrs(collection->ns()),
+ "index"_attr = existingEntry->descriptor()->infoObj());
+
+ if (existingEntry->isReady(opCtx)) {
+ _readyIndexes.add(std::move(existingEntry));
+ } else if (existingEntry->isFrozen()) {
+ _frozenIndexes.add(std::move(existingEntry));
+ } else {
+ _buildingIndexes.add(std::move(existingEntry));
+ }
+ continue;
+ }
+ }
+
auto descriptor = std::make_unique<IndexDescriptor>(_getAccessMethodName(keyPattern), spec);
// TTL indexes with NaN 'expireAfterSeconds' cause problems in multiversion settings.
@@ -257,16 +298,27 @@ Status IndexCatalogImpl::init(OperationContext* opCtx, Collection* collection) {
createIndexEntry(opCtx, collection, std::move(descriptor), flags);
fassert(17340, entry->isReady(opCtx));
- // When initializing indexes from disk, we conservatively set the minimumVisibleSnapshot
- // to non _id indexes to the recovery timestamp. The _id index is left visible. It's
- // assumed if the collection is visible, it's _id is valid to be used.
- if (recoveryTs && !entry->descriptor()->isIdIndex()) {
+ if (readTimestamp) {
+ // When initializing indexes from an earlier point-in-time, we conservatively set
+ // the minimum valid snapshot to the read point-in-time as we don't know when the
+ // last DDL operation took place at that point-in-time.
+ entry->setMinimumVisibleSnapshot(*readTimestamp);
+ } else if (recoveryTs && !entry->descriptor()->isIdIndex()) {
+ // When initializing indexes from disk, we conservatively set the
+ // minimumVisibleSnapshot to non _id indexes to the recovery timestamp. The _id
+ // index is left visible. It's assumed if the collection is visible, it's _id is
+ // valid to be used.
entry->setMinimumVisibleSnapshot(recoveryTs.value());
}
}
}
- CollectionQueryInfo::get(collection).init(opCtx, collection);
+ if (!readTimestamp) {
+ // Only do this when we're not initializing an earlier collection from the shared state of
+ // an existing collection.
+ CollectionQueryInfo::get(collection).init(opCtx, collection);
+ }
+
return Status::OK();
}
@@ -349,15 +401,21 @@ Status IndexCatalogImpl::_isNonIDIndexAndNotAllowedToBuild(OperationContext* opC
const BSONObj& spec) const {
const BSONObj key = spec.getObjectField("key");
invariant(!key.isEmpty());
- if (!IndexDescriptor::isIdIndexPattern(key)) {
- // Check whether the replica set member's config has {buildIndexes:false} set, which means
- // we are not allowed to build non-_id indexes on this server.
- if (!repl::ReplicationCoordinator::get(opCtx)->buildsIndexes()) {
- // We return an IndexAlreadyExists error so that the caller can catch it and silently
- // skip building it.
- return Status(ErrorCodes::IndexAlreadyExists,
- "this replica set member's 'buildIndexes' setting is set to false");
- }
+ if (IndexDescriptor::isIdIndexPattern(key)) {
+ return Status::OK();
+ }
+
+ if (!getGlobalReplSettings().usingReplSets()) {
+ return Status::OK();
+ }
+
+ // Check whether the replica set member's config has {buildIndexes:false} set, which means
+ // we are not allowed to build non-_id indexes on this server.
+ if (!repl::ReplicationCoordinator::get(opCtx)->buildsIndexes()) {
+ // We return an IndexAlreadyExists error so that the caller can catch it and silently
+ // skip building it.
+ return Status(ErrorCodes::IndexAlreadyExists,
+ "this replica set member's 'buildIndexes' setting is set to false");
}
return Status::OK();
diff --git a/src/mongo/db/catalog/index_catalog_impl.h b/src/mongo/db/catalog/index_catalog_impl.h
index 655c43e6858..4576dddaa3c 100644
--- a/src/mongo/db/catalog/index_catalog_impl.h
+++ b/src/mongo/db/catalog/index_catalog_impl.h
@@ -67,6 +67,17 @@ public:
// must be called before used
Status init(OperationContext* opCtx, Collection* collection) override;
+ /**
+ * Must be called before used.
+ *
+ * When initializing an index that exists in 'preexistingIndexes', the IndexCatalogEntry will be
+ * taken from there instead of initializing a new IndexCatalogEntry.
+ */
+ Status initFromExisting(OperationContext* opCtx,
+ Collection* collection,
+ const IndexCatalogEntryContainer& preexistingIndexes,
+ boost::optional<Timestamp> readTimestamp) override;
+
// ---- accessors -----
bool haveAnyIndexes() const override;
@@ -300,6 +311,17 @@ private:
static const BSONObj _idObj; // { _id : 1 }
/**
+ * Helper for init() and initFromExisting().
+ *
+ * Passing boost::none for 'preexistingIndexes' indicates that the IndexCatalog is not being
+ * initialized at an earlier point-in-time.
+ */
+ Status _init(OperationContext* opCtx,
+ Collection* collection,
+ const IndexCatalogEntryContainer& preexistingIndexes,
+ boost::optional<Timestamp> readTimestamp);
+
+ /**
* In addition to IndexNames::findPluginName, validates that it is a known index type.
* If all you need is to check for a certain type, just use IndexNames::findPluginName.
*
diff --git a/src/mongo/db/service_context_d_test_fixture.cpp b/src/mongo/db/service_context_d_test_fixture.cpp
index caef9950914..cad9b4dfa62 100644
--- a/src/mongo/db/service_context_d_test_fixture.cpp
+++ b/src/mongo/db/service_context_d_test_fixture.cpp
@@ -65,6 +65,10 @@ ServiceContextMongoDTest::ServiceContextMongoDTest(Options options)
: _journalListener(std::move(options._journalListener)),
_tempDir("service_context_d_test_fixture") {
+ if (options._forceDisableTableLogging) {
+ storageGlobalParams.forceDisableTableLogging = true;
+ }
+
if (options._useReplSettings) {
repl::ReplSettings replSettings;
replSettings.setOplogSizeBytes(10 * 1024 * 1024);
@@ -174,6 +178,8 @@ ServiceContextMongoDTest::~ServiceContextMongoDTest() {
std::swap(storageGlobalParams.repair, _stashedStorageParams.repair);
std::swap(serverGlobalParams.enableMajorityReadConcern,
_stashedServerParams.enableMajorityReadConcern);
+
+ storageGlobalParams.reset();
}
void ServiceContextMongoDTest::tearDown() {
diff --git a/src/mongo/db/service_context_d_test_fixture.h b/src/mongo/db/service_context_d_test_fixture.h
index 3973a344ffd..763038eb95e 100644
--- a/src/mongo/db/service_context_d_test_fixture.h
+++ b/src/mongo/db/service_context_d_test_fixture.h
@@ -87,6 +87,11 @@ protected:
return std::move(*this);
}
+ Options forceDisableTableLogging() {
+ _forceDisableTableLogging = true;
+ return std::move(*this);
+ }
+
private:
std::string _engine = "wiredTiger";
// We use ephemeral instances by default to advise Storage Engines (in particular
@@ -99,6 +104,7 @@ protected:
Milliseconds _autoAdvancingMockClockIncrement{0};
std::unique_ptr<TickSource> _mockTickSource;
std::unique_ptr<JournalListener> _journalListener;
+ bool _forceDisableTableLogging = false;
friend class ServiceContextMongoDTest;
};
diff --git a/src/mongo/db/storage/storage_options.cpp b/src/mongo/db/storage/storage_options.cpp
index 5df2299e0ab..f31217b89ae 100644
--- a/src/mongo/db/storage/storage_options.cpp
+++ b/src/mongo/db/storage/storage_options.cpp
@@ -58,6 +58,7 @@ void StorageGlobalParams::reset() {
allowOplogTruncation = true;
disableLockFreeReads = false;
checkpointDelaySecs = 0;
+ forceDisableTableLogging = false;
}
StorageGlobalParams storageGlobalParams;
diff --git a/src/mongo/db/storage/storage_options.h b/src/mongo/db/storage/storage_options.h
index c10b71a25aa..386e13efde5 100644
--- a/src/mongo/db/storage/storage_options.h
+++ b/src/mongo/db/storage/storage_options.h
@@ -132,6 +132,9 @@ struct StorageGlobalParams {
// Delay in seconds between triggering the next checkpoint after the completion of the previous
// one. A value of 0 indicates that checkpointing will be skipped.
size_t checkpointDelaySecs;
+
+ // Test-only option. Disables table logging.
+ bool forceDisableTableLogging = false;
};
extern StorageGlobalParams storageGlobalParams;
diff --git a/src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp b/src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp
index f6087cb0d98..a5a6986bf7c 100644
--- a/src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp
+++ b/src/mongo/db/storage/wiredtiger/wiredtiger_util.cpp
@@ -795,6 +795,12 @@ void WiredTigerUtil::notifyStartupComplete() {
}
bool WiredTigerUtil::useTableLogging(const NamespaceString& nss) {
+ if (storageGlobalParams.forceDisableTableLogging) {
+ invariant(TestingProctor::instance().isEnabled());
+ LOGV2(6825405, "Table logging disabled", logAttrs(nss));
+ return false;
+ }
+
// We only turn off logging in the case of:
// 1) Replication is enabled (the typical deployment), or
// 2) We're running as a standalone with recoverFromOplogAsStandalone=true