[release-7.3] Pause perpetual storage wiggle when TSS count target is…

… met. (#11824) * TSS pause * Add condition
apple · Dec 16, 2024 · 9fb460b · 9fb460b
1 parent a97b392
commit 9fb460b
Show file tree

Hide file tree

Showing 3 changed files with 21 additions and 10 deletions.
diff --git a/fdbclient/ServerKnobs.cpp b/fdbclient/ServerKnobs.cpp
@@ -286,6 +286,7 @@ void ServerKnobs::initialize(Randomize randomize, ClientKnobs* clientKnobs, IsSi
 	init( PERPETUAL_WIGGLE_MIN_BYTES_BALANCE_RATIO,             0.85 );
 	init( PW_MAX_SS_LESSTHAN_MIN_BYTES_BALANCE_RATIO,              8 );
 	init( PERPETUAL_WIGGLE_DISABLE_REMOVER,                     true );
+	init( PERPETUAL_WIGGLE_PAUSE_AFTER_TSS_TARGET_MET,         false ); if (isSimulated)PERPETUAL_WIGGLE_PAUSE_AFTER_TSS_TARGET_MET = deterministicRandom()->coinflip();
 	init( LOG_ON_COMPLETION_DELAY,         DD_QUEUE_LOGGING_INTERVAL );
 	init( BEST_TEAM_MAX_TEAM_TRIES,                               10 );
 	init( BEST_TEAM_OPTION_COUNT,                                  4 );

diff --git a/fdbclient/include/fdbclient/ServerKnobs.h b/fdbclient/include/fdbclient/ServerKnobs.h
@@ -269,6 +269,7 @@ class ServerKnobs : public KnobsImpl<ServerKnobs> {
 	                                                // balanced/filledup before starting the next wiggle.
 	double PERPETUAL_WIGGLE_DELAY; // The max interval between the last wiggle finish and the next wiggle start
 	bool PERPETUAL_WIGGLE_DISABLE_REMOVER; // Whether the start of perpetual wiggle replace team remover
+	bool PERPETUAL_WIGGLE_PAUSE_AFTER_TSS_TARGET_MET;
 	double LOG_ON_COMPLETION_DELAY;
 	int BEST_TEAM_MAX_TEAM_TRIES;
 	int BEST_TEAM_OPTION_COUNT;

diff --git a/fdbserver/DDTeamCollection.actor.cpp b/fdbserver/DDTeamCollection.actor.cpp
@@ -2234,19 +2234,28 @@ class DDTeamCollectionImpl {
 			takeRest = self->server_info.size() <= self->configuration.storageTeamSize ||
 			           self->machine_info.size() < self->configuration.storageTeamSize || imbalance;
 
+			if (SERVER_KNOBS->PERPETUAL_WIGGLE_PAUSE_AFTER_TSS_TARGET_MET &&
+			    self->configuration.storageMigrationType == StorageMigrationType::DEFAULT) {
+				takeRest = takeRest || (self->getTargetTSSInDC() > 0 && self->reachTSSPairTarget());
+			}
+
 			// log the extra delay and change the wiggler state
 			if (takeRest) {
 				self->storageWiggler->setWiggleState(StorageWiggler::PAUSE);
-				if (self->configuration.storageMigrationType == StorageMigrationType::GRADUAL) {
-					TraceEvent(SevWarn, "PerpetualStorageWiggleSleep", self->distributorId)
-					    .suppressFor(SERVER_KNOBS->PERPETUAL_WIGGLE_DELAY * 4)
-					    .detail("ImbalanceFactor",
-					            SERVER_KNOBS->PW_MAX_SS_LESSTHAN_MIN_BYTES_BALANCE_RATIO ? numSSToBeLoadBytesBalanced
-					                                                                     : ratio)
-					    .detail("ServerSize", self->server_info.size())
-					    .detail("MachineSize", self->machine_info.size())
-					    .detail("StorageTeamSize", self->configuration.storageTeamSize);
-				}
+				Severity sev =
+				    self->configuration.storageMigrationType == StorageMigrationType::GRADUAL ? SevWarn : SevInfo;
+				TraceEvent(sev, "PerpetualStorageWiggleSleep", self->distributorId)
+				    .suppressFor(SERVER_KNOBS->PERPETUAL_WIGGLE_DELAY * 4)
+				    .detail("Primary", self->primary)
+				    .detail("ImbalanceFactor",
+				            SERVER_KNOBS->PW_MAX_SS_LESSTHAN_MIN_BYTES_BALANCE_RATIO ? numSSToBeLoadBytesBalanced
+				                                                                     : ratio)
+				    .detail("ServerSize", self->server_info.size())
+				    .detail("MachineSize", self->machine_info.size())
+				    .detail("StorageTeamSize", self->configuration.storageTeamSize)
+				    .detail("TargetTSSInDC", self->getTargetTSSInDC())
+				    .detail("ReachTSSPairTarget", self->reachTSSPairTarget())
+				    .detail("MigrationType", self->configuration.storageMigrationType.toString());
 			}
 		}
 		return Void();