wbtiger
/
foundationdb
mirror of https://github.com/apple/foundationdb.git
0
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

fix: do not call better master exists until the long lived stateless processes have settled into their desired locations

This commit is contained in:
Evan Tschannen 2021-12-06 13:12:27 -08:00
parent a4fff19320
commit 951bc4acd7
3 changed files with 18 additions and 3 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
fdbclient/ServerKnobs.cpp
View File

@ -509,6 +509,7 @@ void ServerKnobs::initialize(Randomize randomize, ClientKnobs* clientKnobs, IsSi
init( POLICY_GENERATIONS, 100 ); if( randomize && BUGGIFY ) POLICY_GENERATIONS = 10;
init( DBINFO_SEND_AMOUNT, 5 );
init( DBINFO_BATCH_DELAY, 0.1 );
init( SINGLETON_RECRUIT_BME_DELAY, 10.0 );
//Move Keys
init( SHARD_READY_DELAY, 0.25 );

1
fdbclient/ServerKnobs.h
View File

@ -446,6 +446,7 @@ public:
double RECRUITMENT_TIMEOUT;
int DBINFO_SEND_AMOUNT;
double DBINFO_BATCH_DELAY;
double SINGLETON_RECRUIT_BME_DELAY;
// Move Keys
double SHARD_READY_DELAY;

19
fdbserver/ClusterController.actor.cpp
View File

@ -3195,6 +3195,7 @@ public:
// recruitX is used to signal when role X needs to be (re)recruited.
// recruitingXID is used to track the ID of X's interface which is being recruited.
// We use AsyncVars to kill (i.e. halt) singletons that have been replaced.
double lastRecruitTime = 0;
AsyncVar<bool> recruitDistributor;
Optional<UID> recruitingDistributorID;
AsyncVar<bool> recruitRatekeeper;
@ -3299,7 +3300,10 @@ struct RatekeeperSingleton : Singleton<RatekeeperInterface> {
brokenPromiseToNever(interface.get().haltRatekeeper.getReply(HaltRatekeeperRequest(cc->id)));
}
}
void recruit(ClusterControllerData* cc) const { cc->recruitRatekeeper.set(true); }
void recruit(ClusterControllerData* cc) const {
cc->lastRecruitTime = now();
cc->recruitRatekeeper.set(true);
}
};
struct DataDistributorSingleton : Singleton<DataDistributorInterface> {
@ -3320,7 +3324,10 @@ struct DataDistributorSingleton : Singleton<DataDistributorInterface> {
brokenPromiseToNever(interface.get().haltDataDistributor.getReply(HaltDataDistributorRequest(cc->id)));
}
}
void recruit(ClusterControllerData* cc) const { cc->recruitDistributor.set(true); }
void recruit(ClusterControllerData* cc) const {
cc->lastRecruitTime = now();
cc->recruitDistributor.set(true);
}
};
struct BlobManagerSingleton : Singleton<BlobManagerInterface> {
@ -3341,7 +3348,10 @@ struct BlobManagerSingleton : Singleton<BlobManagerInterface> {
brokenPromiseToNever(interface.get().haltBlobManager.getReply(HaltBlobManagerRequest(cc->id)));
}
}
void recruit(ClusterControllerData* cc) const { cc->recruitBlobManager.set(true); }
void recruit(ClusterControllerData* cc) const {
cc->lastRecruitTime = now();
cc->recruitBlobManager.set(true);
}
};
ACTOR Future<Void> clusterWatchDatabase(ClusterControllerData* cluster, ClusterControllerData::DBInfo* db) {
@ -3798,6 +3808,9 @@ ACTOR Future<Void> doCheckOutstandingRequests(ClusterControllerData* self) {
while (!self->goodRecruitmentTime.isReady()) {
wait(self->goodRecruitmentTime);
}
while (now() - self->lastRecruitTime < SERVER_KNOBS->SINGLETON_RECRUIT_BME_DELAY) {
wait(delay(SERVER_KNOBS->SINGLETON_RECRUIT_BME_DELAY + 0.001 - (now() - self->lastRecruitTime)));
}
checkOutstandingRecruitmentRequests(self);
checkOutstandingStorageRequests(self);