SERVER-45743 Add a flag to allow opting out of using the Resumable RangeDeleter

7 files changed, 46 insertions, 27 deletions

diff --git a/src/mongo/db/s/migration_chunk_cloner_source_legacy.cpp b/src/mongo/db/s/migration_chunk_cloner_source_legacy.cpp
index 5ba3a2068e7..b9dd891ed17 100644
--- a/src/mongo/db/s/migration_chunk_cloner_source_legacy.cpp
+++ b/src/mongo/db/s/migration_chunk_cloner_source_legacy.cpp
@@ -44,6 +44,7 @@
 #include "mongo/db/repl/replication_process.h"
 #include "mongo/db/s/collection_sharding_runtime.h"
 #include "mongo/db/s/migration_source_manager.h"
+#include "mongo/db/s/sharding_runtime_d_params_gen.h"
 #include "mongo/db/s/sharding_statistics.h"
 #include "mongo/db/s/start_chunk_clone_request.h"
 #include "mongo/db/service_context.h"
@@ -279,7 +280,8 @@ Status MigrationChunkClonerSourceLegacy::startClone(OperationContext* opCtx,
     BSONObjBuilder cmdBuilder;
 
     auto fcvVersion = serverGlobalParams.featureCompatibility.getVersion();
-    if (fcvVersion == ServerGlobalParams::FeatureCompatibility::Version::kFullyUpgradedTo44) {
+    if (fcvVersion == ServerGlobalParams::FeatureCompatibility::Version::kFullyUpgradedTo44 &&
+        !disableResumableRangeDeleter.load()) {
         StartChunkCloneRequest::appendAsCommand(&cmdBuilder,
                                                 _args.getNss(),
                                                 migrationId,
@@ -292,7 +294,8 @@ Status MigrationChunkClonerSourceLegacy::startClone(OperationContext* opCtx,
                                                 _shardKeyPattern.toBSON(),
                                                 _args.getSecondaryThrottle());
     } else {
-        // TODO (SERVER-44787): Remove this overload after 4.4 is released.
+        // TODO (SERVER-44787): Remove this overload after 4.4 is released AND
+        // disableResumableRangeDeleter has been removed from server parameters.
         StartChunkCloneRequest::appendAsCommand(&cmdBuilder,
                                                 _args.getNss(),
                                                 _sessionId,
diff --git a/src/mongo/db/s/migration_destination_manager.cpp b/src/mongo/db/s/migration_destination_manager.cpp
index ad170299a1a..a44e6bdda24 100644
--- a/src/mongo/db/s/migration_destination_manager.cpp
+++ b/src/mongo/db/s/migration_destination_manager.cpp
@@ -343,14 +343,15 @@ Status MigrationDestinationManager::start(OperationContext* opCtx,
     // Note: It is expected that the FCV cannot change while the node is donating or receiving a
     // chunk. This is guaranteed by the setFCV command serializing with donating and receiving
     // chunks via the ActiveMigrationsRegistry.
-    _useFCV44Protocol =
-        fcvVersion == ServerGlobalParams::FeatureCompatibility::Version::kFullyUpgradedTo44;
+    _enableResumableRangeDeleter =
+        fcvVersion == ServerGlobalParams::FeatureCompatibility::Version::kFullyUpgradedTo44 &&
+        !disableResumableRangeDeleter.load();
 
     _state = READY;
     _stateChangedCV.notify_all();
     _errmsg = "";
 
-    if (_useFCV44Protocol) {
+    if (_enableResumableRangeDeleter) {
         uassert(ErrorCodes::ConflictingOperationInProgress,
                 "Missing migrationId in FCV 4.4",
                 cloneRequest.hasMigrationId());
@@ -767,7 +768,7 @@ void MigrationDestinationManager::_migrateThread() {
         _setStateFail(str::stream() << "migrate failed: " << redact(exceptionToStatus()));
     }
 
-    if (!_useFCV44Protocol) {
+    if (!_enableResumableRangeDeleter) {
         if (getState() != DONE) {
             _forgetPending(opCtx.get(), ChunkRange(_min, _max));
         }
@@ -812,7 +813,7 @@ void MigrationDestinationManager::_migrateDriver(OperationContext* opCtx) {
 
         // 2. Ensure any data which might have been left orphaned in the range being moved has been
         // deleted.
-        if (_useFCV44Protocol) {
+        if (_enableResumableRangeDeleter) {
             while (migrationutil::checkForConflictingDeletions(
                 opCtx, range, donorCollectionOptionsAndIndexes.uuid)) {
                 LOG(0) << "Migration paused because range overlaps with a "
diff --git a/src/mongo/db/s/migration_destination_manager.h b/src/mongo/db/s/migration_destination_manager.h
index 78ea4741e8e..6be51ce2dd8 100644
--- a/src/mongo/db/s/migration_destination_manager.h
+++ b/src/mongo/db/s/migration_destination_manager.h
@@ -202,7 +202,9 @@ private:
 
     stdx::thread _migrateThreadHandle;
 
-    bool _useFCV44Protocol{false};
+    // Whether to use the resumable range deleter. This decision is based on whether the FCV 4.2 or
+    // FCV 4.4 protocol are in use and the disableResumableRangeDeleter option is off.
+    bool _enableResumableRangeDeleter{true};
 
     UUID _migrationId;
     NamespaceString _nss;
diff --git a/src/mongo/db/s/migration_source_manager.cpp b/src/mongo/db/s/migration_source_manager.cpp
index 728366a814c..1f779820e75 100644
--- a/src/mongo/db/s/migration_source_manager.cpp
+++ b/src/mongo/db/s/migration_source_manager.cpp
@@ -50,6 +50,7 @@
 #include "mongo/db/s/shard_filtering_metadata_refresh.h"
 #include "mongo/db/s/shard_metadata_util.h"
 #include "mongo/db/s/sharding_logging.h"
+#include "mongo/db/s/sharding_runtime_d_params_gen.h"
 #include "mongo/db/s/sharding_state.h"
 #include "mongo/db/s/sharding_state_recovery.h"
 #include "mongo/db/s/sharding_statistics.h"
@@ -116,6 +117,12 @@ void refreshRecipientRoutingTable(OperationContext* opCtx,
     executor->scheduleRemoteCommand(request, noOp).getStatus().ignore();
 }
 
+bool isFCVLatest() {
+    auto fcvVersion = serverGlobalParams.featureCompatibility.getVersion();
+
+    return fcvVersion == ServerGlobalParams::FeatureCompatibility::Version::kFullyUpgradedTo44;
+}
+
 }  // namespace
 
 MONGO_FAIL_POINT_DEFINE(doNotRefreshRecipientAfterCommit);
@@ -143,12 +150,13 @@ MigrationSourceManager::MigrationSourceManager(OperationContext* opCtx,
     // chunk. This is guaranteed by the setFCV command serializing with donating and receiving
     // chunks via the ActiveMigrationsRegistry.
     auto fcvVersion = serverGlobalParams.featureCompatibility.getVersion();
+
     uassert(ErrorCodes::ConflictingOperationInProgress,
             "Can't donate chunk while FCV is upgrading/downgrading",
             fcvVersion != ServerGlobalParams::FeatureCompatibility::Version::kUpgradingTo44 &&
                 fcvVersion != ServerGlobalParams::FeatureCompatibility::Version::kDowngradingTo42);
-    _useFCV44Protocol =
-        fcvVersion == ServerGlobalParams::FeatureCompatibility::Version::kFullyUpgradedTo44;
+
+    _enableResumableRangeDeleter = isFCVLatest() && !disableResumableRangeDeleter.load();
 
     // Disallow moving a chunk to ourselves
     uassert(ErrorCodes::InvalidOptions,
@@ -283,7 +291,7 @@ Status MigrationSourceManager::startClone() {
         auto lockedCsr = CollectionShardingRuntime::CSRLock::lockExclusive(_opCtx, csr);
         invariant(nullptr == std::exchange(msmForCsr(csr), this));
 
-        if (_useFCV44Protocol) {
+        if (_enableResumableRangeDeleter) {
             _coordinator = std::make_unique<migrationutil::MigrationCoordinator>(
                 migrationId,
                 _args.getFromShardId(),
@@ -306,7 +314,7 @@ Status MigrationSourceManager::startClone() {
             _opCtx, readConcernArgs, PrepareConflictBehavior::kEnforce);
     }
 
-    if (_useFCV44Protocol) {
+    if (_enableResumableRangeDeleter) {
         _coordinator->startMigration(_opCtx, _args.getWaitForDelete());
     }
 
@@ -426,7 +434,7 @@ Status MigrationSourceManager::commitChunkMetadataOnConfig() {
         ChunkType migratedChunkType;
         migratedChunkType.setMin(_args.getMinKey());
         migratedChunkType.setMax(_args.getMaxKey());
-        if (_useFCV44Protocol) {
+        if (isFCVLatest()) {
             migratedChunkType.setVersion(_chunkVersion);
         }
 
@@ -467,7 +475,7 @@ Status MigrationSourceManager::commitChunkMetadataOnConfig() {
         Shard::CommandResponse::getEffectiveStatus(commitChunkMigrationResponse);
 
     if (!migrationCommitStatus.isOK()) {
-        if (_useFCV44Protocol) {
+        if (isFCVLatest()) {
             migrationutil::ensureChunkVersionIsGreaterThan(_opCtx, _args.getRange(), _chunkVersion);
         } else {
             // This is the FCV 4.2 and below protocol.
@@ -542,7 +550,7 @@ Status MigrationSourceManager::commitChunkMetadataOnConfig() {
                     "have re-received the chunk"};
         }
 
-        if (_useFCV44Protocol) {
+        if (_enableResumableRangeDeleter) {
             _coordinator->setMigrationDecision(
                 migrationutil::MigrationCoordinator::Decision::kAborted);
         }
@@ -555,7 +563,7 @@ Status MigrationSourceManager::commitChunkMetadataOnConfig() {
     LOG(0) << "Migration succeeded and updated collection version to "
            << refreshedMetadata->getCollVersion();
 
-    if (_useFCV44Protocol) {
+    if (_enableResumableRangeDeleter) {
         _coordinator->setMigrationDecision(
             migrationutil::MigrationCoordinator::Decision::kCommitted);
     }
@@ -595,7 +603,7 @@ Status MigrationSourceManager::commitChunkMetadataOnConfig() {
         << "Moved chunks successfully but failed to clean up " << getNss().ns() << " range "
         << redact(range.toString()) << " due to: ";
 
-    if (_useFCV44Protocol) {
+    if (_enableResumableRangeDeleter) {
         if (_args.getWaitForDelete()) {
             log() << "Waiting for cleanup of " << getNss().ns() << " range "
                   << redact(range.toString());
@@ -778,7 +786,7 @@ void MigrationSourceManager::_cleanup() {
         ShardingStateRecovery::endMetadataOp(_opCtx);
     }
 
-    if (_useFCV44Protocol) {
+    if (_enableResumableRangeDeleter) {
         if (_state >= kCloning && _state < kCommittingOnConfig) {
             invariant(_coordinator);
             _coordinator->setMigrationDecision(
diff --git a/src/mongo/db/s/migration_source_manager.h b/src/mongo/db/s/migration_source_manager.h
index e833e2a641c..b0aef2c7327 100644
--- a/src/mongo/db/s/migration_source_manager.h
+++ b/src/mongo/db/s/migration_source_manager.h
@@ -252,9 +252,9 @@ private:
     // collection doesn't have UUID.
     boost::optional<UUID> _collectionUuid;
 
-    // Whether to use the FCV 4.2 or FCV 4.4 protocol. After branching for v4.4, this should be
-    // removed and the FCV 4.4 protocol should always be used.
-    bool _useFCV44Protocol;
+    // Whether to use the resumable range deleter. This decision is based on whether the FCV 4.2 or
+    // FCV 4.4 protocol are in use and the disableResumableRangeDeleter option is off.
+    bool _enableResumableRangeDeleter;
 
     // Contains logic for ensuring the donor's and recipient's config.rangeDeletions entries are
     // correctly updated based on whether the migration committed or aborted.
diff --git a/src/mongo/db/s/sharding_runtime_d_params.idl b/src/mongo/db/s/sharding_runtime_d_params.idl
index 44f808d412a..ddb8230386c 100644
--- a/src/mongo/db/s/sharding_runtime_d_params.idl
+++ b/src/mongo/db/s/sharding_runtime_d_params.idl
@@ -90,3 +90,10 @@ server_parameters:
         cpp_vartype: AtomicWord<int>
         cpp_varname: orphanCleanupDelaySecs
         default: 900
+
+    disableResumableRangeDeleter:
+        description: 'Disable the resumable range deleter and revert to prior behavior.'
+        set_at: [startup, runtime]
+        cpp_vartype: AtomicWord<bool> 
+        cpp_varname : disableResumableRangeDeleter
+        default: false
diff --git a/src/mongo/db/s/start_chunk_clone_request.cpp b/src/mongo/db/s/start_chunk_clone_request.cpp
index 30d2813a428..883bca6b9f7 100644
--- a/src/mongo/db/s/start_chunk_clone_request.cpp
+++ b/src/mongo/db/s/start_chunk_clone_request.cpp
@@ -74,12 +74,10 @@ StatusWith<StartChunkCloneRequest> StartChunkCloneRequest::createFromCommand(Nam
                                    std::move(sessionIdStatus.getValue()),
                                    std::move(secondaryThrottleStatus.getValue()));
 
-    // TODO (SERVER-44787): Remove this FCV check after 4.4 is released.
-    if (serverGlobalParams.featureCompatibility.getVersion() ==
-        ServerGlobalParams::FeatureCompatibility::Version::kFullyUpgradedTo44) {
-        if (obj.getField("uuid"))
-            request._migrationId = UUID::parse(obj);
-    }
+    // TODO (SERVER-44787): Remove this existence check after 4.4 is released and the
+    // disableResumableRangeDeleter option is removed.
+    if (obj.getField("uuid"))
+        request._migrationId = UUID::parse(obj);
 
     {
         std::string fromShardConnectionString;