2017-05-26 04:48:44 +08:00
|
|
|
/*
|
|
|
|
* ClusterController.actor.cpp
|
|
|
|
*
|
|
|
|
* This source file is part of the FoundationDB open source project
|
|
|
|
*
|
2019-02-16 09:29:52 +08:00
|
|
|
* Copyright 2013-2019 Apple Inc. and the FoundationDB project authors
|
2018-02-22 02:25:11 +08:00
|
|
|
*
|
2017-05-26 04:48:44 +08:00
|
|
|
* Licensed under the Apache License, Version 2.0 (the "License");
|
|
|
|
* you may not use this file except in compliance with the License.
|
|
|
|
* You may obtain a copy of the License at
|
2018-02-22 02:25:11 +08:00
|
|
|
*
|
2017-05-26 04:48:44 +08:00
|
|
|
* http://www.apache.org/licenses/LICENSE-2.0
|
2018-02-22 02:25:11 +08:00
|
|
|
*
|
2017-05-26 04:48:44 +08:00
|
|
|
* Unless required by applicable law or agreed to in writing, software
|
|
|
|
* distributed under the License is distributed on an "AS IS" BASIS,
|
|
|
|
* WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
|
|
|
|
* See the License for the specific language governing permissions and
|
|
|
|
* limitations under the License.
|
|
|
|
*/
|
|
|
|
|
2019-05-16 07:13:04 +08:00
|
|
|
#include <algorithm>
|
|
|
|
#include <iterator>
|
|
|
|
#include <map>
|
2022-01-07 04:15:51 +08:00
|
|
|
#include <memory>
|
2019-05-16 07:13:04 +08:00
|
|
|
#include <set>
|
|
|
|
#include <vector>
|
|
|
|
|
2021-12-08 13:43:58 +08:00
|
|
|
#include "fdbclient/SystemData.h"
|
2017-05-26 04:48:44 +08:00
|
|
|
#include "fdbrpc/FailureMonitor.h"
|
2022-01-14 03:11:01 +08:00
|
|
|
#include "fdbserver/EncryptKeyProxyInterface.h"
|
2017-05-26 04:48:44 +08:00
|
|
|
#include "flow/ActorCollection.h"
|
2021-10-11 11:44:56 +08:00
|
|
|
#include "fdbclient/ClusterConnectionMemoryRecord.h"
|
2019-02-18 07:41:16 +08:00
|
|
|
#include "fdbclient/NativeAPI.actor.h"
|
2022-01-07 04:15:51 +08:00
|
|
|
#include "fdbserver/ApplyMetadataMutation.h"
|
2019-04-25 06:12:37 +08:00
|
|
|
#include "fdbserver/BackupInterface.h"
|
2022-01-07 04:15:51 +08:00
|
|
|
#include "fdbserver/BackupProgress.actor.h"
|
|
|
|
#include "fdbserver/ConfigBroadcaster.h"
|
|
|
|
#include "fdbserver/CoordinatedState.h"
|
|
|
|
#include "fdbserver/CoordinationInterface.h" // copy constructors for ServerCoordinators class
|
|
|
|
#include "fdbserver/ClusterController.actor.h"
|
|
|
|
#include "fdbserver/ClusterRecovery.actor.h"
|
|
|
|
#include "fdbserver/DataDistributorInterface.h"
|
|
|
|
#include "fdbserver/DBCoreState.h"
|
|
|
|
#include "fdbserver/ConfigBroadcaster.h"
|
|
|
|
#include "fdbserver/MoveKeys.actor.h"
|
|
|
|
#include "fdbserver/LeaderElection.h"
|
|
|
|
#include "fdbserver/LogSystem.h"
|
|
|
|
#include "fdbserver/LogSystemConfig.h"
|
|
|
|
#include "fdbserver/LogSystemDiskQueueAdapter.h"
|
|
|
|
#include "fdbserver/WaitFailure.h"
|
|
|
|
#include "fdbserver/RatekeeperInterface.h"
|
|
|
|
#include "fdbserver/BlobManagerInterface.h"
|
|
|
|
#include "fdbserver/ServerDBInfo.h"
|
|
|
|
#include "fdbserver/Status.h"
|
|
|
|
#include "fdbserver/LatencyBandConfig.h"
|
|
|
|
#include "fdbclient/GlobalConfig.actor.h"
|
|
|
|
#include "fdbserver/RecoveryState.h"
|
|
|
|
#include "fdbclient/ReadYourWrites.h"
|
|
|
|
#include "fdbrpc/Replication.h"
|
|
|
|
#include "fdbrpc/ReplicationUtils.h"
|
|
|
|
#include "fdbrpc/sim_validation.h"
|
|
|
|
#include "fdbclient/KeyBackedTypes.h"
|
2022-01-14 03:11:01 +08:00
|
|
|
#include "flow/Trace.h"
|
2022-01-07 04:15:51 +08:00
|
|
|
#include "flow/Util.h"
|
|
|
|
#include "flow/actorcompiler.h" // This must be the last #include.
|
2021-06-29 12:45:02 +08:00
|
|
|
|
2022-01-07 04:15:51 +08:00
|
|
|
void failAfter(Future<Void> trigger, Endpoint e);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2022-01-07 04:15:51 +08:00
|
|
|
// This is used to artificially amplify the used count for processes
|
|
|
|
// occupied by non-singletons. This ultimately makes it less desirable
|
|
|
|
// for singletons to use those processes as well. This constant should
|
|
|
|
// be increased if we ever have more than 100 singletons (unlikely).
|
|
|
|
static const int PID_USED_AMP_FOR_NON_SINGLETON = 100;
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
// Wrapper for singleton interfaces
|
|
|
|
template <class Interface>
|
|
|
|
struct Singleton {
|
|
|
|
const Optional<Interface>& interface;
|
|
|
|
|
|
|
|
Singleton(const Optional<Interface>& interface) : interface(interface) {}
|
|
|
|
|
|
|
|
virtual Role getRole() const = 0;
|
|
|
|
virtual ProcessClass::ClusterRole getClusterRole() const = 0;
|
|
|
|
|
2021-09-16 22:24:15 +08:00
|
|
|
virtual void setInterfaceToDbInfo(ClusterControllerData* cc) const = 0;
|
2021-09-13 21:58:38 +08:00
|
|
|
virtual void halt(ClusterControllerData* cc, Optional<Standalone<StringRef>> pid) const = 0;
|
|
|
|
virtual void recruit(ClusterControllerData* cc) const = 0;
|
|
|
|
};
|
|
|
|
|
|
|
|
struct RatekeeperSingleton : Singleton<RatekeeperInterface> {
|
|
|
|
|
|
|
|
RatekeeperSingleton(const Optional<RatekeeperInterface>& interface) : Singleton(interface) {}
|
|
|
|
|
|
|
|
Role getRole() const { return Role::RATEKEEPER; }
|
|
|
|
ProcessClass::ClusterRole getClusterRole() const { return ProcessClass::Ratekeeper; }
|
|
|
|
|
2021-09-16 22:24:15 +08:00
|
|
|
void setInterfaceToDbInfo(ClusterControllerData* cc) const {
|
2021-09-13 21:58:38 +08:00
|
|
|
if (interface.present()) {
|
2022-01-21 03:27:00 +08:00
|
|
|
TraceEvent("CCRK_SetInf", cc->id).detail("Id", interface.get().id());
|
2021-09-13 21:58:38 +08:00
|
|
|
cc->db.setRatekeeper(interface.get());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
void halt(ClusterControllerData* cc, Optional<Standalone<StringRef>> pid) const {
|
|
|
|
if (interface.present()) {
|
|
|
|
cc->id_worker[pid].haltRatekeeper =
|
|
|
|
brokenPromiseToNever(interface.get().haltRatekeeper.getReply(HaltRatekeeperRequest(cc->id)));
|
|
|
|
}
|
|
|
|
}
|
2021-12-07 05:12:27 +08:00
|
|
|
void recruit(ClusterControllerData* cc) const {
|
|
|
|
cc->lastRecruitTime = now();
|
|
|
|
cc->recruitRatekeeper.set(true);
|
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
};
|
|
|
|
|
|
|
|
struct DataDistributorSingleton : Singleton<DataDistributorInterface> {
|
|
|
|
|
|
|
|
DataDistributorSingleton(const Optional<DataDistributorInterface>& interface) : Singleton(interface) {}
|
|
|
|
|
|
|
|
Role getRole() const { return Role::DATA_DISTRIBUTOR; }
|
|
|
|
ProcessClass::ClusterRole getClusterRole() const { return ProcessClass::DataDistributor; }
|
|
|
|
|
2021-09-16 22:24:15 +08:00
|
|
|
void setInterfaceToDbInfo(ClusterControllerData* cc) const {
|
2021-09-13 21:58:38 +08:00
|
|
|
if (interface.present()) {
|
2022-01-21 03:27:00 +08:00
|
|
|
TraceEvent("CCDD_SetInf", cc->id).detail("Id", interface.get().id());
|
2021-09-13 21:58:38 +08:00
|
|
|
cc->db.setDistributor(interface.get());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
void halt(ClusterControllerData* cc, Optional<Standalone<StringRef>> pid) const {
|
|
|
|
if (interface.present()) {
|
|
|
|
cc->id_worker[pid].haltDistributor =
|
|
|
|
brokenPromiseToNever(interface.get().haltDataDistributor.getReply(HaltDataDistributorRequest(cc->id)));
|
|
|
|
}
|
|
|
|
}
|
2021-12-07 05:12:27 +08:00
|
|
|
void recruit(ClusterControllerData* cc) const {
|
|
|
|
cc->lastRecruitTime = now();
|
|
|
|
cc->recruitDistributor.set(true);
|
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
};
|
|
|
|
|
2021-09-15 23:35:58 +08:00
|
|
|
struct BlobManagerSingleton : Singleton<BlobManagerInterface> {
|
|
|
|
|
|
|
|
BlobManagerSingleton(const Optional<BlobManagerInterface>& interface) : Singleton(interface) {}
|
|
|
|
|
|
|
|
Role getRole() const { return Role::BLOB_MANAGER; }
|
|
|
|
ProcessClass::ClusterRole getClusterRole() const { return ProcessClass::BlobManager; }
|
|
|
|
|
|
|
|
void setInterfaceToDbInfo(ClusterControllerData* cc) const {
|
|
|
|
if (interface.present()) {
|
2022-01-21 03:27:00 +08:00
|
|
|
TraceEvent("CCBM_SetInf", cc->id).detail("Id", interface.get().id());
|
2021-09-15 23:35:58 +08:00
|
|
|
cc->db.setBlobManager(interface.get());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
void halt(ClusterControllerData* cc, Optional<Standalone<StringRef>> pid) const {
|
|
|
|
if (interface.present()) {
|
|
|
|
cc->id_worker[pid].haltBlobManager =
|
|
|
|
brokenPromiseToNever(interface.get().haltBlobManager.getReply(HaltBlobManagerRequest(cc->id)));
|
|
|
|
}
|
|
|
|
}
|
2021-12-07 05:12:27 +08:00
|
|
|
void recruit(ClusterControllerData* cc) const {
|
|
|
|
cc->lastRecruitTime = now();
|
|
|
|
cc->recruitBlobManager.set(true);
|
|
|
|
}
|
2022-02-18 06:50:35 +08:00
|
|
|
|
2021-12-08 13:43:58 +08:00
|
|
|
void haltBlobGranules(ClusterControllerData* cc, Optional<Standalone<StringRef>> pid) const {
|
|
|
|
if (interface.present()) {
|
|
|
|
cc->id_worker[pid].haltBlobManager =
|
|
|
|
brokenPromiseToNever(interface.get().haltBlobGranules.getReply(HaltBlobGranulesRequest(cc->id)));
|
|
|
|
}
|
|
|
|
}
|
2021-09-15 23:35:58 +08:00
|
|
|
};
|
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
struct EncryptKeyProxySingleton : Singleton<EncryptKeyProxyInterface> {
|
|
|
|
|
|
|
|
EncryptKeyProxySingleton(const Optional<EncryptKeyProxyInterface>& interface) : Singleton(interface) {}
|
|
|
|
|
|
|
|
Role getRole() const { return Role::ENCRYPT_KEY_PROXY; }
|
|
|
|
ProcessClass::ClusterRole getClusterRole() const { return ProcessClass::EncryptKeyProxy; }
|
|
|
|
|
|
|
|
void setInterfaceToDbInfo(ClusterControllerData* cc) const {
|
|
|
|
if (interface.present()) {
|
|
|
|
TraceEvent("CCEKP_SetInf", cc->id).detail("Id", interface.get().id());
|
|
|
|
cc->db.setEncryptKeyProxy(interface.get());
|
|
|
|
}
|
|
|
|
}
|
|
|
|
void halt(ClusterControllerData* cc, Optional<Standalone<StringRef>> pid) const {
|
|
|
|
if (interface.present()) {
|
|
|
|
cc->id_worker[pid].haltEncryptKeyProxy =
|
|
|
|
brokenPromiseToNever(interface.get().haltEncryptKeyProxy.getReply(HaltEncryptKeyProxyRequest(cc->id)));
|
|
|
|
}
|
|
|
|
}
|
2022-02-18 06:50:35 +08:00
|
|
|
void recruit(ClusterControllerData* cc) const {
|
|
|
|
cc->lastRecruitTime = now();
|
|
|
|
cc->recruitEncryptKeyProxy.set(true);
|
|
|
|
}
|
2022-01-14 03:11:01 +08:00
|
|
|
};
|
|
|
|
|
2022-01-07 04:15:51 +08:00
|
|
|
ACTOR Future<Void> handleLeaderReplacement(Reference<ClusterRecoveryData> self, Future<Void> leaderFail) {
|
|
|
|
loop choose {
|
|
|
|
when(wait(leaderFail)) {
|
|
|
|
TraceEvent("LeaderReplaced", self->controllerData->id).log();
|
|
|
|
// We are no longer the leader if this has changed.
|
|
|
|
self->controllerData->shouldCommitSuicide = true;
|
|
|
|
throw restart_cluster_controller();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
ACTOR Future<Void> clusterWatchDatabase(ClusterControllerData* cluster,
|
|
|
|
ClusterControllerData::DBInfo* db,
|
|
|
|
ServerCoordinators coordinators,
|
|
|
|
Future<Void> leaderFail) {
|
2017-05-26 04:48:44 +08:00
|
|
|
state MasterInterface iMaster;
|
2022-01-07 04:15:51 +08:00
|
|
|
state Reference<ClusterRecoveryData> recoveryData;
|
|
|
|
state PromiseStream<Future<Void>> addActor;
|
|
|
|
state Future<Void> recoveryCore;
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
// SOMEDAY: If there is already a non-failed master referenced by zkMasterInfo, use that one until it fails
|
|
|
|
// When this someday is implemented, make sure forced failures still cause the master to be recruited again
|
|
|
|
|
|
|
|
loop {
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("CCWDB", cluster->id).log();
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
|
|
|
state double recoveryStart = now();
|
2022-01-07 04:15:51 +08:00
|
|
|
state MasterInterface newMaster;
|
|
|
|
state Future<Void> collection;
|
2018-02-10 08:48:55 +08:00
|
|
|
|
2022-01-07 04:15:51 +08:00
|
|
|
TraceEvent("CCWDB", cluster->id).detail("Recruiting", "Master");
|
|
|
|
wait(recruitNewMaster(cluster, db, std::addressof(newMaster)));
|
|
|
|
|
|
|
|
iMaster = newMaster;
|
|
|
|
|
|
|
|
db->masterRegistrationCount = 0;
|
|
|
|
db->recoveryStalled = false;
|
|
|
|
|
|
|
|
auto dbInfo = ServerDBInfo();
|
|
|
|
dbInfo.master = iMaster;
|
|
|
|
dbInfo.id = deterministicRandom()->randomUniqueID();
|
|
|
|
dbInfo.infoGeneration = ++db->dbInfoCount;
|
|
|
|
dbInfo.masterLifetime = db->serverInfo->get().masterLifetime;
|
|
|
|
++dbInfo.masterLifetime;
|
|
|
|
dbInfo.clusterInterface = db->serverInfo->get().clusterInterface;
|
|
|
|
dbInfo.distributor = db->serverInfo->get().distributor;
|
|
|
|
dbInfo.ratekeeper = db->serverInfo->get().ratekeeper;
|
|
|
|
dbInfo.blobManager = db->serverInfo->get().blobManager;
|
2022-01-14 03:11:01 +08:00
|
|
|
dbInfo.encryptKeyProxy = db->serverInfo->get().encryptKeyProxy;
|
2022-01-07 04:15:51 +08:00
|
|
|
dbInfo.latencyBandConfig = db->serverInfo->get().latencyBandConfig;
|
|
|
|
dbInfo.myLocality = db->serverInfo->get().myLocality;
|
|
|
|
dbInfo.client = ClientDBInfo();
|
2022-02-19 13:22:31 +08:00
|
|
|
dbInfo.client.tenantMode = db->config.tenantMode;
|
2022-01-07 04:15:51 +08:00
|
|
|
|
|
|
|
TraceEvent("CCWDB", cluster->id)
|
|
|
|
.detail("NewMaster", dbInfo.master.id().toString())
|
|
|
|
.detail("Lifetime", dbInfo.masterLifetime.toString())
|
|
|
|
.detail("ChangeID", dbInfo.id);
|
|
|
|
db->serverInfo->set(dbInfo);
|
|
|
|
|
|
|
|
state Future<Void> spinDelay = delay(
|
|
|
|
SERVER_KNOBS
|
2022-01-19 05:54:00 +08:00
|
|
|
->MASTER_SPIN_DELAY); // Don't retry cluster recovery more than once per second, but don't delay
|
2022-01-07 04:15:51 +08:00
|
|
|
// the "first" recovery after more than a second of normal operation
|
|
|
|
|
|
|
|
TraceEvent("CCWDB", cluster->id).detail("Watching", iMaster.id());
|
|
|
|
recoveryData = makeReference<ClusterRecoveryData>(cluster,
|
|
|
|
db->serverInfo,
|
|
|
|
db->serverInfo->get().master,
|
|
|
|
db->serverInfo->get().masterLifetime,
|
|
|
|
coordinators,
|
|
|
|
db->serverInfo->get().clusterInterface,
|
|
|
|
LiteralStringRef(""),
|
|
|
|
addActor,
|
|
|
|
db->forceRecovery);
|
|
|
|
|
|
|
|
collection = actorCollection(recoveryData->addActor.getFuture());
|
|
|
|
recoveryCore = clusterRecoveryCore(recoveryData);
|
|
|
|
|
|
|
|
// Master failure detection is pretty sensitive, but if we are in the middle of a very long recovery we
|
|
|
|
// really don't want to have to start over
|
|
|
|
loop choose {
|
|
|
|
when(wait(recoveryCore)) {}
|
|
|
|
when(wait(waitFailureClient(
|
|
|
|
iMaster.waitFailure,
|
|
|
|
db->masterRegistrationCount
|
|
|
|
? SERVER_KNOBS->MASTER_FAILURE_REACTION_TIME
|
|
|
|
: (now() - recoveryStart) * SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY,
|
|
|
|
db->masterRegistrationCount ? -SERVER_KNOBS->MASTER_FAILURE_REACTION_TIME /
|
|
|
|
SERVER_KNOBS->SECONDS_BEFORE_NO_FAILURE_DELAY
|
|
|
|
: SERVER_KNOBS->MASTER_FAILURE_SLOPE_DURING_RECOVERY) ||
|
|
|
|
db->forceMasterFailure.onTrigger())) {
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
when(wait(db->serverInfo->onChange())) {}
|
|
|
|
when(BackupWorkerDoneRequest req =
|
|
|
|
waitNext(db->serverInfo->get().clusterInterface.notifyBackupWorkerDone.getFuture())) {
|
|
|
|
if (recoveryData->logSystem.isValid() && recoveryData->logSystem->removeBackupWorker(req)) {
|
|
|
|
recoveryData->registrationTrigger.trigger();
|
2021-03-11 02:06:03 +08:00
|
|
|
}
|
2022-01-07 04:15:51 +08:00
|
|
|
++recoveryData->backupWorkerDoneRequests;
|
|
|
|
req.reply.send(Void());
|
|
|
|
TraceEvent(SevDebug, "BackupWorkerDoneRequest", cluster->id).log();
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2022-01-07 04:15:51 +08:00
|
|
|
when(wait(collection)) { throw internal_error(); }
|
|
|
|
when(wait(handleLeaderReplacement(recoveryData, leaderFail))) { throw internal_error(); }
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2022-01-07 04:15:51 +08:00
|
|
|
// failed master (better master exists) could happen while change-coordinators request processing is
|
|
|
|
// in-progress
|
|
|
|
if (cluster->shouldCommitSuicide) {
|
|
|
|
throw restart_cluster_controller();
|
|
|
|
}
|
|
|
|
|
|
|
|
recoveryCore.cancel();
|
|
|
|
wait(cleanupRecoveryActorCollection(recoveryData, /*exThrown=*/false));
|
|
|
|
ASSERT(addActor.isEmpty());
|
|
|
|
|
|
|
|
wait(spinDelay);
|
|
|
|
|
|
|
|
TEST(true); // clusterWatchDatabase() master failed
|
|
|
|
TraceEvent(SevWarn, "DetectedFailedRecovery", cluster->id).detail("OldMaster", iMaster.id());
|
2017-05-26 04:48:44 +08:00
|
|
|
} catch (Error& e) {
|
2022-01-07 04:15:51 +08:00
|
|
|
state Error err = e;
|
2022-02-25 04:25:52 +08:00
|
|
|
TraceEvent("CCWDB", cluster->id).errorUnsuppressed(e).detail("Master", iMaster.id());
|
2022-01-07 04:15:51 +08:00
|
|
|
if (e.code() != error_code_actor_cancelled)
|
|
|
|
wait(delay(0.0));
|
|
|
|
|
|
|
|
recoveryCore.cancel();
|
|
|
|
wait(cleanupRecoveryActorCollection(recoveryData, true /* exThrown */));
|
|
|
|
ASSERT(addActor.isEmpty());
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2022-01-07 04:15:51 +08:00
|
|
|
TEST(err.code() == error_code_tlog_failed); // Terminated due to tLog failure
|
|
|
|
TEST(err.code() == error_code_commit_proxy_failed); // Terminated due to commit proxy failure
|
|
|
|
TEST(err.code() == error_code_grv_proxy_failed); // Terminated due to GRV proxy failure
|
|
|
|
TEST(err.code() == error_code_resolver_failed); // Terminated due to resolver failure
|
|
|
|
TEST(err.code() == error_code_backup_worker_failed); // Terminated due to backup worker failure
|
|
|
|
TEST(err.code() == error_code_operation_failed); // Terminated due to failed operation
|
|
|
|
TEST(err.code() == error_code_restart_cluster_controller); // Terminated due to cluster-controller restart.
|
|
|
|
|
|
|
|
if (cluster->shouldCommitSuicide || err.code() == error_code_coordinators_changed) {
|
2022-02-25 04:25:52 +08:00
|
|
|
TraceEvent("ClusterControllerTerminate", cluster->id).errorUnsuppressed(err);
|
2022-01-07 04:15:51 +08:00
|
|
|
throw restart_cluster_controller();
|
|
|
|
}
|
|
|
|
|
|
|
|
if (isNormalClusterRecoveryError(err)) {
|
|
|
|
TraceEvent(SevWarn, "ClusterRecoveryRetrying", cluster->id).error(err);
|
|
|
|
} else {
|
|
|
|
bool ok = err.code() == error_code_no_more_servers;
|
|
|
|
TraceEvent(ok ? SevWarn : SevError, "ClusterWatchDatabaseRetrying", cluster->id).error(err);
|
|
|
|
if (!ok)
|
|
|
|
throw err;
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(delay(SERVER_KNOBS->ATTEMPT_RECRUITMENT_DELAY));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> clusterGetServerInfo(ClusterControllerData::DBInfo* db,
|
|
|
|
UID knownServerInfoID,
|
2020-04-12 10:30:05 +08:00
|
|
|
ReplyPromise<ServerDBInfo> reply) {
|
2021-03-11 02:06:03 +08:00
|
|
|
while (db->serverInfo->get().id == knownServerInfoID) {
|
2017-05-26 04:48:44 +08:00
|
|
|
choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(yieldedFuture(db->serverInfo->onChange()))) {}
|
|
|
|
when(wait(delayJittered(300))) { break; } // The server might be long gone!
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
reply.send(db->serverInfo->get());
|
2017-05-26 04:48:44 +08:00
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2019-07-26 08:15:31 +08:00
|
|
|
ACTOR Future<Void> clusterOpenDatabase(ClusterControllerData::DBInfo* db, OpenDatabaseRequest req) {
|
|
|
|
db->clientStatus[req.reply.getEndpoint().getPrimaryAddress()] = std::make_pair(now(), req);
|
2021-03-11 02:06:03 +08:00
|
|
|
if (db->clientStatus.size() > 10000) {
|
2019-07-26 08:15:31 +08:00
|
|
|
TraceEvent(SevWarnAlways, "TooManyClientStatusEntries").suppressFor(1.0);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2020-01-08 11:53:09 +08:00
|
|
|
|
2019-07-26 08:15:31 +08:00
|
|
|
while (db->clientInfo->get().id == req.knownClientInfoID) {
|
2017-05-26 04:48:44 +08:00
|
|
|
choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(db->clientInfo->onChange())) {}
|
|
|
|
when(wait(delayJittered(SERVER_KNOBS->COORDINATOR_REGISTER_INTERVAL))) {
|
|
|
|
break;
|
|
|
|
} // The client might be long gone!
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
req.reply.send(db->clientInfo->get());
|
2017-05-26 04:48:44 +08:00
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
void checkOutstandingRecruitmentRequests(ClusterControllerData* self) {
|
|
|
|
for (int i = 0; i < self->outstandingRecruitmentRequests.size(); i++) {
|
2022-01-07 04:15:51 +08:00
|
|
|
Reference<RecruitWorkersInfo> info = self->outstandingRecruitmentRequests[i];
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
2022-01-07 04:15:51 +08:00
|
|
|
info->rep = self->findWorkersForConfiguration(info->req);
|
|
|
|
if (info->dbgId.present()) {
|
|
|
|
TraceEvent("CheckOutstandingRecruitment", info->dbgId.get())
|
|
|
|
.detail("Request", info->req.configuration.toString());
|
|
|
|
}
|
|
|
|
info->waitForCompletion.trigger();
|
2021-03-11 02:06:03 +08:00
|
|
|
swapAndPop(&self->outstandingRecruitmentRequests, i--);
|
2017-05-26 04:48:44 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
if (e.code() == error_code_no_more_servers || e.code() == error_code_operation_failed) {
|
|
|
|
TraceEvent(SevWarn, "RecruitTLogMatchingSetNotAvailable", self->id).error(e);
|
|
|
|
} else {
|
|
|
|
TraceEvent(SevError, "RecruitTLogsRequestError", self->id).error(e);
|
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
void checkOutstandingRemoteRecruitmentRequests(ClusterControllerData* self) {
|
|
|
|
for (int i = 0; i < self->outstandingRemoteRecruitmentRequests.size(); i++) {
|
2022-01-07 04:15:51 +08:00
|
|
|
Reference<RecruitRemoteWorkersInfo> info = self->outstandingRemoteRecruitmentRequests[i];
|
2017-09-12 08:40:46 +08:00
|
|
|
try {
|
2022-01-07 04:15:51 +08:00
|
|
|
info->rep = self->findRemoteWorkersForConfiguration(info->req);
|
|
|
|
if (info->dbgId.present()) {
|
|
|
|
TraceEvent("CheckOutstandingRemoteRecruitment", info->dbgId.get())
|
|
|
|
.detail("Request", info->req.configuration.toString());
|
|
|
|
}
|
|
|
|
info->waitForCompletion.trigger();
|
2021-03-11 02:06:03 +08:00
|
|
|
swapAndPop(&self->outstandingRemoteRecruitmentRequests, i--);
|
2017-09-12 08:40:46 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
if (e.code() == error_code_no_more_servers || e.code() == error_code_operation_failed) {
|
|
|
|
TraceEvent(SevWarn, "RecruitRemoteTLogMatchingSetNotAvailable", self->id).error(e);
|
|
|
|
} else {
|
|
|
|
TraceEvent(SevError, "RecruitRemoteTLogsRequestError", self->id).error(e);
|
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
void checkOutstandingStorageRequests(ClusterControllerData* self) {
|
|
|
|
for (int i = 0; i < self->outstandingStorageRequests.size(); i++) {
|
2017-05-26 04:48:44 +08:00
|
|
|
auto& req = self->outstandingStorageRequests[i];
|
|
|
|
try {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (req.second < now()) {
|
2017-05-26 04:48:44 +08:00
|
|
|
req.first.reply.sendError(timed_out());
|
2021-03-11 02:06:03 +08:00
|
|
|
swapAndPop(&self->outstandingStorageRequests, i--);
|
2017-05-26 04:48:44 +08:00
|
|
|
} else {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!self->gotProcessClasses && !req.first.criticalRecruitment)
|
2017-05-26 04:48:44 +08:00
|
|
|
throw no_more_servers();
|
|
|
|
|
|
|
|
auto worker = self->getStorageWorker(req.first);
|
|
|
|
RecruitStorageReply rep;
|
2019-03-09 00:25:07 +08:00
|
|
|
rep.worker = worker.interf;
|
|
|
|
rep.processClass = worker.processClass;
|
2021-03-11 02:06:03 +08:00
|
|
|
req.first.reply.send(rep);
|
|
|
|
swapAndPop(&self->outstandingStorageRequests, i--);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
} catch (Error& e) {
|
|
|
|
if (e.code() == error_code_no_more_servers) {
|
2019-08-13 01:08:12 +08:00
|
|
|
TraceEvent(SevWarn, "RecruitStorageNotAvailable", self->id)
|
2022-02-25 04:25:52 +08:00
|
|
|
.errorUnsuppressed(e)
|
2019-08-20 04:47:48 +08:00
|
|
|
.suppressFor(1.0)
|
2019-08-13 01:08:12 +08:00
|
|
|
.detail("OutstandingReq", i)
|
2022-02-25 04:25:52 +08:00
|
|
|
.detail("IsCriticalRecruitment", req.first.criticalRecruitment);
|
2017-05-26 04:48:44 +08:00
|
|
|
} else {
|
|
|
|
TraceEvent(SevError, "RecruitStorageError", self->id).error(e);
|
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-29 07:15:32 +08:00
|
|
|
// When workers aren't available at the time of request, the request
|
|
|
|
// gets added to a list of outstanding reqs. Here, we try to resolve these
|
|
|
|
// outstanding requests.
|
2021-09-21 03:42:20 +08:00
|
|
|
void checkOutstandingBlobWorkerRequests(ClusterControllerData* self) {
|
|
|
|
for (int i = 0; i < self->outstandingBlobWorkerRequests.size(); i++) {
|
|
|
|
auto& req = self->outstandingBlobWorkerRequests[i];
|
|
|
|
try {
|
|
|
|
if (req.second < now()) {
|
|
|
|
req.first.reply.sendError(timed_out());
|
|
|
|
swapAndPop(&self->outstandingBlobWorkerRequests, i--);
|
|
|
|
} else {
|
|
|
|
if (!self->gotProcessClasses)
|
|
|
|
throw no_more_servers();
|
|
|
|
|
|
|
|
auto worker = self->getBlobWorker(req.first);
|
|
|
|
RecruitBlobWorkerReply rep;
|
|
|
|
rep.worker = worker.interf;
|
|
|
|
rep.processClass = worker.processClass;
|
|
|
|
req.first.reply.send(rep);
|
2021-09-29 07:15:32 +08:00
|
|
|
// can remove it once we know the worker was found
|
2021-09-21 03:42:20 +08:00
|
|
|
swapAndPop(&self->outstandingBlobWorkerRequests, i--);
|
|
|
|
}
|
|
|
|
} catch (Error& e) {
|
|
|
|
if (e.code() == error_code_no_more_servers) {
|
|
|
|
TraceEvent(SevWarn, "RecruitBlobWorkerNotAvailable", self->id)
|
2022-02-25 04:25:52 +08:00
|
|
|
.errorUnsuppressed(e)
|
2021-09-21 03:42:20 +08:00
|
|
|
.suppressFor(1.0)
|
2022-02-25 04:25:52 +08:00
|
|
|
.detail("OutstandingReq", i);
|
2021-09-21 03:42:20 +08:00
|
|
|
} else {
|
|
|
|
TraceEvent(SevError, "RecruitBlobWorkerError", self->id).error(e);
|
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
// Finds and returns a new process for role
|
|
|
|
WorkerDetails findNewProcessForSingleton(ClusterControllerData* self,
|
|
|
|
const ProcessClass::ClusterRole role,
|
|
|
|
std::map<Optional<Standalone<StringRef>>, int>& id_used) {
|
|
|
|
// find new process in cluster for role
|
|
|
|
WorkerDetails newWorker =
|
|
|
|
self->getWorkerForRoleInDatacenter(
|
|
|
|
self->clusterControllerDcId, role, ProcessClass::NeverAssign, self->db.config, id_used, {}, true)
|
|
|
|
.worker;
|
|
|
|
|
|
|
|
// check if master's process is actually better suited for role
|
|
|
|
if (self->onMasterIsBetter(newWorker, role)) {
|
|
|
|
newWorker = self->id_worker[self->masterProcessId.get()].details;
|
|
|
|
}
|
|
|
|
|
|
|
|
// acknowledge that the pid is now potentially used by this role as well
|
|
|
|
id_used[newWorker.interf.locality.processId()]++;
|
|
|
|
|
|
|
|
return newWorker;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Return best possible fitness for singleton. Note that lower fitness is better.
|
|
|
|
ProcessClass::Fitness findBestFitnessForSingleton(const ClusterControllerData* self,
|
|
|
|
const WorkerDetails& worker,
|
|
|
|
const ProcessClass::ClusterRole& role) {
|
|
|
|
auto bestFitness = worker.processClass.machineClassFitness(role);
|
|
|
|
// If the process has been marked as excluded, we take the max with ExcludeFit to ensure its fit
|
|
|
|
// is at least as bad as ExcludeFit. This assists with successfully offboarding such processes
|
|
|
|
// and removing them from the cluster.
|
|
|
|
if (self->db.config.isExcludedServer(worker.interf.addresses())) {
|
|
|
|
bestFitness = std::max(bestFitness, ProcessClass::ExcludeFit);
|
|
|
|
}
|
|
|
|
return bestFitness;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Returns true iff the singleton is healthy. "Healthy" here means that
|
|
|
|
// the singleton is stable (see below) and doesn't need to be rerecruited.
|
|
|
|
// Side effects: (possibly) initiates recruitment
|
|
|
|
template <class Interface>
|
2021-09-15 22:44:01 +08:00
|
|
|
bool isHealthySingleton(ClusterControllerData* self,
|
2021-09-13 21:58:38 +08:00
|
|
|
const WorkerDetails& newWorker,
|
|
|
|
const Singleton<Interface>& singleton,
|
|
|
|
const ProcessClass::Fitness& bestFitness,
|
|
|
|
const Optional<UID> recruitingID) {
|
|
|
|
// A singleton is stable if it exists in cluster, has not been killed off of proc and is not being recruited
|
|
|
|
bool isStableSingleton = singleton.interface.present() &&
|
|
|
|
self->id_worker.count(singleton.interface.get().locality.processId()) &&
|
|
|
|
(!recruitingID.present() || (recruitingID.get() == singleton.interface.get().id()));
|
|
|
|
|
|
|
|
if (!isStableSingleton) {
|
|
|
|
return false; // not healthy because unstable
|
|
|
|
}
|
|
|
|
|
|
|
|
auto& currWorker = self->id_worker[singleton.interface.get().locality.processId()];
|
|
|
|
auto currFitness = currWorker.details.processClass.machineClassFitness(singleton.getClusterRole());
|
|
|
|
if (currWorker.priorityInfo.isExcluded) {
|
|
|
|
currFitness = ProcessClass::ExcludeFit;
|
|
|
|
}
|
|
|
|
// If any of the following conditions are met, we will switch the singleton's process:
|
|
|
|
// - if the current proc is used by some non-master, non-singleton role
|
|
|
|
// - if the current fitness is less than optimal (lower fitness is better)
|
|
|
|
// - if currently at peak fitness but on same process as master, and the new worker is on different process
|
|
|
|
bool shouldRerecruit =
|
|
|
|
self->isUsedNotMaster(currWorker.details.interf.locality.processId()) || bestFitness < currFitness ||
|
|
|
|
(currFitness == bestFitness && currWorker.details.interf.locality.processId() == self->masterProcessId &&
|
|
|
|
newWorker.interf.locality.processId() != self->masterProcessId);
|
|
|
|
if (shouldRerecruit) {
|
|
|
|
std::string roleAbbr = singleton.getRole().abbreviation;
|
|
|
|
TraceEvent(("CCHalt" + roleAbbr).c_str(), self->id)
|
|
|
|
.detail(roleAbbr + "ID", singleton.interface.get().id())
|
|
|
|
.detail("Excluded", currWorker.priorityInfo.isExcluded)
|
|
|
|
.detail("Fitness", currFitness)
|
|
|
|
.detail("BestFitness", bestFitness);
|
2021-09-15 22:44:01 +08:00
|
|
|
singleton.recruit(self); // SIDE EFFECT: initiating recruitment
|
2021-09-13 21:58:38 +08:00
|
|
|
return false; // not healthy since needed to be rerecruited
|
|
|
|
} else {
|
|
|
|
return true; // healthy because doesn't need to be rerecruited
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
// Returns a mapping from pid->pidCount for pids
|
2021-09-21 04:41:04 +08:00
|
|
|
std::map<Optional<Standalone<StringRef>>, int> getColocCounts(
|
|
|
|
const std::vector<Optional<Standalone<StringRef>>>& pids) {
|
2021-09-13 21:58:38 +08:00
|
|
|
std::map<Optional<Standalone<StringRef>>, int> counts;
|
|
|
|
for (const auto& pid : pids) {
|
2021-10-19 03:04:22 +08:00
|
|
|
if (pid.present()) {
|
|
|
|
++counts[pid];
|
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
}
|
|
|
|
return counts;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Checks if there exists a better process for each singleton (e.g. DD) compared
|
|
|
|
// to the process it is currently on.
|
2021-10-19 03:04:22 +08:00
|
|
|
// Note: there is a lot of extra logic here to only recruit the blob manager when gate is open.
|
|
|
|
// When adding new singletons, just follow the ratekeeper/data distributor examples.
|
2021-09-13 21:58:38 +08:00
|
|
|
void checkBetterSingletons(ClusterControllerData* self) {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!self->masterProcessId.present() ||
|
|
|
|
self->db.serverInfo->get().recoveryState < RecoveryState::ACCEPTING_COMMITS) {
|
2019-03-20 06:58:25 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
// note: this map doesn't consider pids used by existing singletons
|
2019-11-13 06:22:36 +08:00
|
|
|
std::map<Optional<Standalone<StringRef>>, int> id_used = self->getUsedIds();
|
2021-09-13 21:58:38 +08:00
|
|
|
|
|
|
|
// We prefer spreading out other roles more than separating singletons on their own process
|
|
|
|
// so we artificially amplify the pid count for the processes used by non-singleton roles.
|
|
|
|
// In other words, we make the processes used for other roles less desirable to be used
|
|
|
|
// by singletons as well.
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto& it : id_used) {
|
2021-09-13 21:58:38 +08:00
|
|
|
it.second *= PID_USED_AMP_FOR_NON_SINGLETON;
|
2019-11-13 06:22:36 +08:00
|
|
|
}
|
2019-03-20 06:58:25 +08:00
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
// Try to find a new process for each singleton.
|
|
|
|
WorkerDetails newRKWorker = findNewProcessForSingleton(self, ProcessClass::Ratekeeper, id_used);
|
|
|
|
WorkerDetails newDDWorker = findNewProcessForSingleton(self, ProcessClass::DataDistributor, id_used);
|
2021-10-19 03:04:22 +08:00
|
|
|
|
|
|
|
WorkerDetails newBMWorker;
|
2021-12-10 05:22:11 +08:00
|
|
|
if (self->db.blobGranulesEnabled.get()) {
|
2021-10-19 03:04:22 +08:00
|
|
|
newBMWorker = findNewProcessForSingleton(self, ProcessClass::BlobManager, id_used);
|
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
WorkerDetails newEKPWorker;
|
|
|
|
if (SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY) {
|
|
|
|
newEKPWorker = findNewProcessForSingleton(self, ProcessClass::EncryptKeyProxy, id_used);
|
|
|
|
}
|
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
// Find best possible fitnesses for each singleton.
|
|
|
|
auto bestFitnessForRK = findBestFitnessForSingleton(self, newRKWorker, ProcessClass::Ratekeeper);
|
|
|
|
auto bestFitnessForDD = findBestFitnessForSingleton(self, newDDWorker, ProcessClass::DataDistributor);
|
2021-10-19 03:04:22 +08:00
|
|
|
|
|
|
|
ProcessClass::Fitness bestFitnessForBM;
|
2021-12-10 05:22:11 +08:00
|
|
|
if (self->db.blobGranulesEnabled.get()) {
|
2021-10-19 03:04:22 +08:00
|
|
|
bestFitnessForBM = findBestFitnessForSingleton(self, newBMWorker, ProcessClass::BlobManager);
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
ProcessClass::Fitness bestFitnessForEKP;
|
|
|
|
if (SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY) {
|
|
|
|
bestFitnessForEKP = findBestFitnessForSingleton(self, newEKPWorker, ProcessClass::EncryptKeyProxy);
|
|
|
|
}
|
|
|
|
|
2020-04-12 10:30:05 +08:00
|
|
|
auto& db = self->db.serverInfo->get();
|
2021-09-13 21:58:38 +08:00
|
|
|
auto rkSingleton = RatekeeperSingleton(db.ratekeeper);
|
|
|
|
auto ddSingleton = DataDistributorSingleton(db.distributor);
|
2021-10-19 03:04:22 +08:00
|
|
|
BlobManagerSingleton bmSingleton(db.blobManager);
|
2022-01-14 03:11:01 +08:00
|
|
|
EncryptKeyProxySingleton ekpSingleton(db.encryptKeyProxy);
|
2021-09-13 21:58:38 +08:00
|
|
|
|
2021-09-15 22:44:01 +08:00
|
|
|
// Check if the singletons are healthy.
|
|
|
|
// side effect: try to rerecruit the singletons to more optimal processes
|
|
|
|
bool rkHealthy = isHealthySingleton<RatekeeperInterface>(
|
2021-09-13 21:58:38 +08:00
|
|
|
self, newRKWorker, rkSingleton, bestFitnessForRK, self->recruitingRatekeeperID);
|
|
|
|
|
2021-09-15 22:44:01 +08:00
|
|
|
bool ddHealthy = isHealthySingleton<DataDistributorInterface>(
|
2021-09-13 21:58:38 +08:00
|
|
|
self, newDDWorker, ddSingleton, bestFitnessForDD, self->recruitingDistributorID);
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
bool bmHealthy = true;
|
2021-12-10 05:22:11 +08:00
|
|
|
if (self->db.blobGranulesEnabled.get()) {
|
2021-10-19 03:04:22 +08:00
|
|
|
bmHealthy = isHealthySingleton<BlobManagerInterface>(
|
|
|
|
self, newBMWorker, bmSingleton, bestFitnessForBM, self->recruitingBlobManagerID);
|
|
|
|
}
|
2021-09-15 23:35:58 +08:00
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
bool ekpHealthy = true;
|
|
|
|
if (SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY) {
|
|
|
|
ekpHealthy = isHealthySingleton<EncryptKeyProxyInterface>(
|
|
|
|
self, newEKPWorker, ekpSingleton, bestFitnessForEKP, self->recruitingEncryptKeyProxyID);
|
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
// if any of the singletons are unhealthy (rerecruited or not stable), then do not
|
|
|
|
// consider any further re-recruitments
|
2022-01-14 03:11:01 +08:00
|
|
|
if (!(rkHealthy && ddHealthy && bmHealthy && ekpHealthy)) {
|
2021-09-13 21:58:38 +08:00
|
|
|
return;
|
2019-03-20 06:58:25 +08:00
|
|
|
}
|
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
// if we reach here, we know that the singletons are healthy so let's
|
|
|
|
// check if we can colocate the singletons in a more optimal way
|
2021-09-16 01:41:59 +08:00
|
|
|
Optional<Standalone<StringRef>> currRKProcessId = rkSingleton.interface.get().locality.processId();
|
|
|
|
Optional<Standalone<StringRef>> currDDProcessId = ddSingleton.interface.get().locality.processId();
|
2021-09-13 21:58:38 +08:00
|
|
|
Optional<Standalone<StringRef>> newRKProcessId = newRKWorker.interf.locality.processId();
|
2021-09-16 10:17:27 +08:00
|
|
|
Optional<Standalone<StringRef>> newDDProcessId = newDDWorker.interf.locality.processId();
|
2021-09-13 21:58:38 +08:00
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
Optional<Standalone<StringRef>> currBMProcessId, newBMProcessId;
|
2021-12-10 05:22:11 +08:00
|
|
|
if (self->db.blobGranulesEnabled.get()) {
|
2021-10-19 03:04:22 +08:00
|
|
|
currBMProcessId = bmSingleton.interface.get().locality.processId();
|
|
|
|
newBMProcessId = newBMWorker.interf.locality.processId();
|
|
|
|
}
|
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
Optional<Standalone<StringRef>> currEKPProcessId, newEKPProcessId;
|
|
|
|
if (SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY) {
|
|
|
|
currEKPProcessId = ekpSingleton.interface.get().locality.processId();
|
|
|
|
newEKPProcessId = newEKPWorker.interf.locality.processId();
|
|
|
|
}
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
std::vector<Optional<Standalone<StringRef>>> currPids = { currRKProcessId, currDDProcessId };
|
|
|
|
std::vector<Optional<Standalone<StringRef>>> newPids = { newRKProcessId, newDDProcessId };
|
2021-12-10 05:22:11 +08:00
|
|
|
if (self->db.blobGranulesEnabled.get()) {
|
2021-10-19 03:04:22 +08:00
|
|
|
currPids.emplace_back(currBMProcessId);
|
|
|
|
newPids.emplace_back(newBMProcessId);
|
|
|
|
}
|
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
if (SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY) {
|
|
|
|
currPids.emplace_back(currEKPProcessId);
|
|
|
|
newPids.emplace_back(newEKPProcessId);
|
|
|
|
}
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
auto currColocMap = getColocCounts(currPids);
|
|
|
|
auto newColocMap = getColocCounts(newPids);
|
|
|
|
|
|
|
|
// if the knob is disabled, the BM coloc counts should have no affect on the coloc counts check below
|
2021-12-10 05:22:11 +08:00
|
|
|
if (!self->db.blobGranulesEnabled.get()) {
|
2021-10-19 03:04:22 +08:00
|
|
|
ASSERT(currColocMap[currBMProcessId] == 0);
|
|
|
|
ASSERT(newColocMap[newBMProcessId] == 0);
|
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
// if the knob is disabled, the EKP coloc counts should have no affect on the coloc counts check below
|
|
|
|
if (!SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY) {
|
|
|
|
ASSERT(currColocMap[currEKPProcessId] == 0);
|
|
|
|
ASSERT(newColocMap[newEKPProcessId] == 0);
|
|
|
|
}
|
|
|
|
|
2021-09-15 23:35:58 +08:00
|
|
|
// if the new coloc counts are collectively better (i.e. each singleton's coloc count has not increased)
|
2021-09-16 01:41:59 +08:00
|
|
|
if (newColocMap[newRKProcessId] <= currColocMap[currRKProcessId] &&
|
2021-09-15 23:35:58 +08:00
|
|
|
newColocMap[newDDProcessId] <= currColocMap[currDDProcessId] &&
|
2022-01-14 03:11:01 +08:00
|
|
|
newColocMap[newBMProcessId] <= currColocMap[currBMProcessId] &&
|
|
|
|
newColocMap[newEKPProcessId] <= currColocMap[currEKPProcessId]) {
|
2021-09-13 21:58:38 +08:00
|
|
|
// rerecruit the singleton for which we have found a better process, if any
|
2021-09-16 01:41:59 +08:00
|
|
|
if (newColocMap[newRKProcessId] < currColocMap[currRKProcessId]) {
|
2021-09-13 21:58:38 +08:00
|
|
|
rkSingleton.recruit(self);
|
2021-09-16 01:41:59 +08:00
|
|
|
} else if (newColocMap[newDDProcessId] < currColocMap[currDDProcessId]) {
|
2021-09-13 21:58:38 +08:00
|
|
|
ddSingleton.recruit(self);
|
2021-12-10 05:22:11 +08:00
|
|
|
} else if (self->db.blobGranulesEnabled.get() && newColocMap[newBMProcessId] < currColocMap[currBMProcessId]) {
|
2021-09-15 23:35:58 +08:00
|
|
|
bmSingleton.recruit(self);
|
2022-01-14 03:11:01 +08:00
|
|
|
} else if (SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY &&
|
|
|
|
newColocMap[newEKPProcessId] < currColocMap[currEKPProcessId]) {
|
|
|
|
ekpSingleton.recruit(self);
|
2019-03-20 06:58:25 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> doCheckOutstandingRequests(ClusterControllerData* self) {
|
2018-06-29 14:15:32 +08:00
|
|
|
try {
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(delay(SERVER_KNOBS->CHECK_OUTSTANDING_INTERVAL));
|
2021-12-07 05:50:31 +08:00
|
|
|
while (now() - self->lastRecruitTime < SERVER_KNOBS->SINGLETON_RECRUIT_BME_DELAY ||
|
|
|
|
!self->goodRecruitmentTime.isReady()) {
|
|
|
|
if (now() - self->lastRecruitTime < SERVER_KNOBS->SINGLETON_RECRUIT_BME_DELAY) {
|
|
|
|
wait(delay(SERVER_KNOBS->SINGLETON_RECRUIT_BME_DELAY + 0.001 - (now() - self->lastRecruitTime)));
|
|
|
|
}
|
|
|
|
if (!self->goodRecruitmentTime.isReady()) {
|
|
|
|
wait(self->goodRecruitmentTime);
|
|
|
|
}
|
2021-12-07 05:12:27 +08:00
|
|
|
}
|
2018-06-29 14:15:32 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
checkOutstandingRecruitmentRequests(self);
|
|
|
|
checkOutstandingStorageRequests(self);
|
2021-10-19 03:04:22 +08:00
|
|
|
|
2021-12-10 05:22:11 +08:00
|
|
|
if (self->db.blobGranulesEnabled.get()) {
|
2021-10-19 03:04:22 +08:00
|
|
|
checkOutstandingBlobWorkerRequests(self);
|
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
checkBetterSingletons(self);
|
2018-06-29 14:15:32 +08:00
|
|
|
|
|
|
|
self->checkRecoveryStalled();
|
|
|
|
if (self->betterMasterExists()) {
|
2019-02-19 06:54:28 +08:00
|
|
|
self->db.forceMasterFailure.trigger();
|
2020-04-12 10:30:05 +08:00
|
|
|
TraceEvent("MasterRegistrationKill", self->id).detail("MasterId", self->db.serverInfo->get().master.id());
|
2018-06-29 14:15:32 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
if (e.code() != error_code_no_more_servers) {
|
2020-02-20 08:48:30 +08:00
|
|
|
TraceEvent(SevError, "CheckOutstandingError").error(e);
|
|
|
|
}
|
|
|
|
}
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> doCheckOutstandingRemoteRequests(ClusterControllerData* self) {
|
2020-02-20 08:48:30 +08:00
|
|
|
try {
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(delay(SERVER_KNOBS->CHECK_OUTSTANDING_INTERVAL));
|
|
|
|
while (!self->goodRemoteRecruitmentTime.isReady()) {
|
2020-02-20 08:48:30 +08:00
|
|
|
wait(self->goodRemoteRecruitmentTime);
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
checkOutstandingRemoteRecruitmentRequests(self);
|
|
|
|
} catch (Error& e) {
|
|
|
|
if (e.code() != error_code_no_more_servers) {
|
2018-06-29 14:15:32 +08:00
|
|
|
TraceEvent(SevError, "CheckOutstandingError").error(e);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
void checkOutstandingRequests(ClusterControllerData* self) {
|
|
|
|
if (self->outstandingRemoteRequestChecker.isReady()) {
|
2020-02-20 08:48:30 +08:00
|
|
|
self->outstandingRemoteRequestChecker = doCheckOutstandingRemoteRequests(self);
|
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (self->outstandingRequestChecker.isReady()) {
|
2020-02-20 08:48:30 +08:00
|
|
|
self->outstandingRequestChecker = doCheckOutstandingRequests(self);
|
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> rebootAndCheck(ClusterControllerData* cluster, Optional<Standalone<StringRef>> processID) {
|
2019-02-18 10:46:59 +08:00
|
|
|
{
|
|
|
|
auto watcher = cluster->id_worker.find(processID);
|
|
|
|
ASSERT(watcher != cluster->id_worker.end());
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2019-02-18 10:46:59 +08:00
|
|
|
watcher->second.reboots++;
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(delay(g_network->isSimulated() ? SERVER_KNOBS->SIM_SHUTDOWN_TIMEOUT : SERVER_KNOBS->SHUTDOWN_TIMEOUT));
|
2019-02-18 10:46:59 +08:00
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2019-02-18 10:46:59 +08:00
|
|
|
{
|
|
|
|
auto watcher = cluster->id_worker.find(processID);
|
2021-03-11 02:06:03 +08:00
|
|
|
if (watcher != cluster->id_worker.end()) {
|
2019-02-18 10:46:59 +08:00
|
|
|
watcher->second.reboots--;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (watcher->second.reboots < 2)
|
|
|
|
checkOutstandingRequests(cluster);
|
2019-02-18 10:46:59 +08:00
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> workerAvailabilityWatch(WorkerInterface worker,
|
|
|
|
ProcessClass startingClass,
|
|
|
|
ClusterControllerData* cluster) {
|
2019-05-30 04:43:21 +08:00
|
|
|
state Future<Void> failed =
|
|
|
|
(worker.address() == g_network->getLocalAddress() || startingClass.classType() == ProcessClass::TesterClass)
|
|
|
|
? Never()
|
|
|
|
: waitFailureClient(worker.waitFailure, SERVER_KNOBS->WORKER_FAILURE_TIME);
|
2021-03-11 02:06:03 +08:00
|
|
|
cluster->updateWorkerList.set(worker.locality.processId(),
|
|
|
|
ProcessData(worker.locality, startingClass, worker.stableAddress()));
|
|
|
|
// This switching avoids a race where the worker can be added to id_worker map after the workerAvailabilityWatch
|
|
|
|
// fails for the worker.
|
2019-01-11 02:28:32 +08:00
|
|
|
wait(delay(0));
|
|
|
|
|
2017-05-26 04:48:44 +08:00
|
|
|
loop {
|
|
|
|
choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(IFailureMonitor::failureMonitor().onStateEqual(
|
|
|
|
worker.storage.getEndpoint(),
|
|
|
|
FailureStatus(
|
|
|
|
IFailureMonitor::failureMonitor().getState(worker.storage.getEndpoint()).isAvailable())))) {
|
|
|
|
if (IFailureMonitor::failureMonitor().getState(worker.storage.getEndpoint()).isAvailable()) {
|
|
|
|
cluster->ac.add(rebootAndCheck(cluster, worker.locality.processId()));
|
|
|
|
checkOutstandingRequests(cluster);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(failed)) { // remove workers that have failed
|
|
|
|
WorkerInfo& failedWorkerInfo = cluster->id_worker[worker.locality.processId()];
|
2019-12-07 05:28:44 +08:00
|
|
|
|
2017-10-05 08:11:12 +08:00
|
|
|
if (!failedWorkerInfo.reply.isSet()) {
|
2021-03-11 02:06:03 +08:00
|
|
|
failedWorkerInfo.reply.send(
|
|
|
|
RegisterWorkerReply(failedWorkerInfo.details.processClass, failedWorkerInfo.priorityInfo));
|
2017-10-05 08:11:12 +08:00
|
|
|
}
|
2019-03-21 07:03:36 +08:00
|
|
|
if (worker.locality.processId() == cluster->masterProcessId) {
|
|
|
|
cluster->masterProcessId = Optional<Key>();
|
|
|
|
}
|
2021-04-30 16:42:54 +08:00
|
|
|
TraceEvent("ClusterControllerWorkerFailed", cluster->id)
|
2021-05-14 03:20:31 +08:00
|
|
|
.detail("ProcessId", worker.locality.processId())
|
|
|
|
.detail("ProcessClass", failedWorkerInfo.details.processClass.toString())
|
|
|
|
.detail("Address", worker.address());
|
2020-04-06 14:09:36 +08:00
|
|
|
cluster->removedDBInfoEndpoints.insert(worker.updateServerDBInfo.getEndpoint());
|
2021-03-11 02:06:03 +08:00
|
|
|
cluster->id_worker.erase(worker.locality.processId());
|
|
|
|
cluster->updateWorkerList.set(worker.locality.processId(), Optional<ProcessData>());
|
2017-05-26 04:48:44 +08:00
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
struct FailureStatusInfo {
|
|
|
|
FailureStatus status;
|
|
|
|
double lastRequestTime;
|
|
|
|
double penultimateRequestTime;
|
|
|
|
|
|
|
|
FailureStatusInfo() : lastRequestTime(0), penultimateRequestTime(0) {}
|
|
|
|
|
|
|
|
void insertRequest(double now) {
|
|
|
|
penultimateRequestTime = lastRequestTime;
|
|
|
|
lastRequestTime = now;
|
|
|
|
}
|
|
|
|
|
|
|
|
double latency(double now) const {
|
2021-03-11 02:06:03 +08:00
|
|
|
return std::max(now - lastRequestTime, lastRequestTime - penultimateRequestTime);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
};
|
|
|
|
|
2021-09-17 08:42:34 +08:00
|
|
|
ACTOR Future<std::vector<TLogInterface>> requireAll(std::vector<Future<Optional<std::vector<TLogInterface>>>> in) {
|
|
|
|
state std::vector<TLogInterface> out;
|
2017-05-26 04:48:44 +08:00
|
|
|
state int i;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (i = 0; i < in.size(); i++) {
|
2021-09-17 08:42:34 +08:00
|
|
|
Optional<std::vector<TLogInterface>> x = wait(in[i]);
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!x.present())
|
|
|
|
throw recruitment_failed();
|
2017-05-26 04:48:44 +08:00
|
|
|
out.insert(out.end(), x.get().begin(), x.get().end());
|
|
|
|
}
|
|
|
|
return out;
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
void clusterRecruitStorage(ClusterControllerData* self, RecruitStorageRequest req) {
|
2017-05-26 04:48:44 +08:00
|
|
|
try {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!self->gotProcessClasses && !req.criticalRecruitment)
|
2017-05-26 04:48:44 +08:00
|
|
|
throw no_more_servers();
|
|
|
|
auto worker = self->getStorageWorker(req);
|
|
|
|
RecruitStorageReply rep;
|
2019-03-09 00:25:07 +08:00
|
|
|
rep.worker = worker.interf;
|
|
|
|
rep.processClass = worker.processClass;
|
2021-03-11 02:06:03 +08:00
|
|
|
req.reply.send(rep);
|
|
|
|
} catch (Error& e) {
|
2017-05-26 04:48:44 +08:00
|
|
|
if (e.code() == error_code_no_more_servers) {
|
2021-12-02 14:46:35 +08:00
|
|
|
self->outstandingStorageRequests.emplace_back(req, now() + SERVER_KNOBS->RECRUITMENT_TIMEOUT);
|
2019-08-13 01:08:12 +08:00
|
|
|
TraceEvent(SevWarn, "RecruitStorageNotAvailable", self->id)
|
2022-02-25 04:25:52 +08:00
|
|
|
.error(e)
|
|
|
|
.detail("IsCriticalRecruitment", req.criticalRecruitment);
|
2017-05-26 04:48:44 +08:00
|
|
|
} else {
|
|
|
|
TraceEvent(SevError, "RecruitStorageError", self->id).error(e);
|
2021-03-11 02:06:03 +08:00
|
|
|
throw; // Any other error will bring down the cluster controller
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-29 07:15:32 +08:00
|
|
|
// Trys to send a reply to req with a worker (process) that a blob worker can be recruited on
|
|
|
|
// Otherwise, add the req to a list of outstanding reqs that will eventually be dealt with
|
2021-09-21 03:42:20 +08:00
|
|
|
void clusterRecruitBlobWorker(ClusterControllerData* self, RecruitBlobWorkerRequest req) {
|
|
|
|
try {
|
|
|
|
if (!self->gotProcessClasses)
|
|
|
|
throw no_more_servers();
|
|
|
|
auto worker = self->getBlobWorker(req);
|
|
|
|
RecruitBlobWorkerReply rep;
|
|
|
|
rep.worker = worker.interf;
|
|
|
|
rep.processClass = worker.processClass;
|
|
|
|
req.reply.send(rep);
|
|
|
|
} catch (Error& e) {
|
|
|
|
if (e.code() == error_code_no_more_servers) {
|
2021-12-02 14:46:35 +08:00
|
|
|
self->outstandingBlobWorkerRequests.emplace_back(req, now() + SERVER_KNOBS->RECRUITMENT_TIMEOUT);
|
2021-09-21 03:42:20 +08:00
|
|
|
TraceEvent(SevWarn, "RecruitBlobWorkerNotAvailable", self->id).error(e);
|
|
|
|
} else {
|
|
|
|
TraceEvent(SevError, "RecruitBlobWorkerError", self->id).error(e);
|
|
|
|
throw; // Any other error will bring down the cluster controller
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
void clusterRegisterMaster(ClusterControllerData* self, RegisterMasterRequest const& req) {
|
|
|
|
req.reply.send(Void());
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2020-03-18 05:45:07 +08:00
|
|
|
TraceEvent("MasterRegistrationReceived", self->id)
|
|
|
|
.detail("MasterId", req.id)
|
|
|
|
.detail("Master", req.mi.toString())
|
|
|
|
.detail("Tlogs", describe(req.logSystemConfig.tLogs))
|
|
|
|
.detail("Resolvers", req.resolvers.size())
|
|
|
|
.detail("RecoveryState", (int)req.recoveryState)
|
|
|
|
.detail("RegistrationCount", req.registrationCount)
|
2020-09-11 08:44:15 +08:00
|
|
|
.detail("CommitProxies", req.commitProxies.size())
|
2020-08-12 09:54:54 +08:00
|
|
|
.detail("GrvProxies", req.grvProxies.size())
|
2020-03-18 05:45:07 +08:00
|
|
|
.detail("RecoveryCount", req.recoveryCount)
|
|
|
|
.detail("Stalled", req.recoveryStalled)
|
2021-10-14 13:22:49 +08:00
|
|
|
.detail("OldestBackupEpoch", req.logSystemConfig.oldestBackupEpoch)
|
|
|
|
.detail("ClusterId", req.clusterId);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// make sure the request comes from an active database
|
2017-05-26 04:48:44 +08:00
|
|
|
auto db = &self->db;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (db->serverInfo->get().master.id() != req.id || req.registrationCount <= db->masterRegistrationCount) {
|
|
|
|
TraceEvent("MasterRegistrationNotFound", self->id)
|
|
|
|
.detail("MasterId", req.id)
|
|
|
|
.detail("ExistingId", db->serverInfo->get().master.id())
|
|
|
|
.detail("RegCount", req.registrationCount)
|
|
|
|
.detail("ExistingRegCount", db->masterRegistrationCount);
|
2017-05-26 04:48:44 +08:00
|
|
|
return;
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (req.recoveryState == RecoveryState::FULLY_RECOVERED) {
|
2018-09-01 01:51:55 +08:00
|
|
|
self->db.unfinishedRecoveries = 0;
|
|
|
|
self->db.logGenerations = 0;
|
2021-03-11 02:06:03 +08:00
|
|
|
ASSERT(!req.logSystemConfig.oldTLogs.size());
|
2018-09-01 01:51:55 +08:00
|
|
|
} else {
|
|
|
|
self->db.logGenerations = std::max<int>(self->db.logGenerations, req.logSystemConfig.oldTLogs.size());
|
|
|
|
}
|
|
|
|
|
2017-05-26 04:48:44 +08:00
|
|
|
db->masterRegistrationCount = req.registrationCount;
|
2018-06-14 09:14:14 +08:00
|
|
|
db->recoveryStalled = req.recoveryStalled;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (req.configuration.present()) {
|
2017-11-15 05:57:37 +08:00
|
|
|
db->config = req.configuration.get();
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (req.recoveryState >= RecoveryState::ACCEPTING_COMMITS) {
|
2017-11-16 09:15:24 +08:00
|
|
|
self->gotFullyRecoveredConfig = true;
|
|
|
|
db->fullyRecoveredConfig = req.configuration.get();
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto& it : self->id_worker) {
|
|
|
|
bool isExcludedFromConfig =
|
|
|
|
db->fullyRecoveredConfig.isExcludedServer(it.second.details.interf.addresses());
|
|
|
|
if (it.second.priorityInfo.isExcluded != isExcludedFromConfig) {
|
2018-02-10 08:48:55 +08:00
|
|
|
it.second.priorityInfo.isExcluded = isExcludedFromConfig;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!it.second.reply.isSet()) {
|
|
|
|
it.second.reply.send(
|
|
|
|
RegisterWorkerReply(it.second.details.processClass, it.second.priorityInfo));
|
2018-02-10 08:48:55 +08:00
|
|
|
}
|
2017-11-16 09:15:24 +08:00
|
|
|
}
|
2017-11-15 05:57:37 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
bool isChanged = false;
|
2020-04-12 10:30:05 +08:00
|
|
|
auto dbInfo = self->db.serverInfo->get();
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
if (dbInfo.recoveryState != req.recoveryState) {
|
|
|
|
dbInfo.recoveryState = req.recoveryState;
|
|
|
|
isChanged = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
if (dbInfo.priorCommittedLogServers != req.priorCommittedLogServers) {
|
|
|
|
dbInfo.priorCommittedLogServers = req.priorCommittedLogServers;
|
|
|
|
isChanged = true;
|
|
|
|
}
|
|
|
|
|
|
|
|
// Construct the client information
|
2020-09-11 08:44:15 +08:00
|
|
|
if (db->clientInfo->get().commitProxies != req.commitProxies ||
|
2022-02-19 13:22:31 +08:00
|
|
|
db->clientInfo->get().grvProxies != req.grvProxies ||
|
|
|
|
db->clientInfo->get().tenantMode != db->config.tenantMode) {
|
2022-01-07 04:15:51 +08:00
|
|
|
TraceEvent("PublishNewClientInfo", self->id)
|
|
|
|
.detail("Master", dbInfo.master.id())
|
|
|
|
.detail("GrvProxies", db->clientInfo->get().grvProxies)
|
|
|
|
.detail("ReqGrvProxies", req.grvProxies)
|
|
|
|
.detail("CommitProxies", db->clientInfo->get().commitProxies)
|
2022-02-19 13:22:31 +08:00
|
|
|
.detail("ReqCPs", req.commitProxies)
|
|
|
|
.detail("TenantMode", db->clientInfo->get().tenantMode.toString())
|
|
|
|
.detail("ReqTenantMode", db->config.tenantMode.toString());
|
2017-05-26 04:48:44 +08:00
|
|
|
isChanged = true;
|
2021-03-06 03:28:15 +08:00
|
|
|
// TODO why construct a new one and not just copy the old one and change proxies + id?
|
2017-05-26 04:48:44 +08:00
|
|
|
ClientDBInfo clientInfo;
|
2019-05-11 05:01:52 +08:00
|
|
|
clientInfo.id = deterministicRandom()->randomUniqueID();
|
2020-09-11 08:44:15 +08:00
|
|
|
clientInfo.commitProxies = req.commitProxies;
|
2020-07-15 15:37:41 +08:00
|
|
|
clientInfo.grvProxies = req.grvProxies;
|
2022-02-19 13:22:31 +08:00
|
|
|
clientInfo.tenantMode = db->config.tenantMode;
|
2021-03-11 02:06:03 +08:00
|
|
|
db->clientInfo->set(clientInfo);
|
2017-05-26 04:48:44 +08:00
|
|
|
dbInfo.client = db->clientInfo->get();
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!dbInfo.logSystemConfig.isEqual(req.logSystemConfig)) {
|
2017-05-26 04:48:44 +08:00
|
|
|
isChanged = true;
|
|
|
|
dbInfo.logSystemConfig = req.logSystemConfig;
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (dbInfo.resolvers != req.resolvers) {
|
2017-05-26 04:48:44 +08:00
|
|
|
isChanged = true;
|
|
|
|
dbInfo.resolvers = req.resolvers;
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (dbInfo.recoveryCount != req.recoveryCount) {
|
2017-05-26 04:48:44 +08:00
|
|
|
isChanged = true;
|
|
|
|
dbInfo.recoveryCount = req.recoveryCount;
|
|
|
|
}
|
|
|
|
|
2021-10-14 13:22:49 +08:00
|
|
|
if (dbInfo.clusterId != req.clusterId) {
|
|
|
|
isChanged = true;
|
|
|
|
dbInfo.clusterId = req.clusterId;
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (isChanged) {
|
2019-05-11 05:01:52 +08:00
|
|
|
dbInfo.id = deterministicRandom()->randomUniqueID();
|
2020-04-06 14:09:36 +08:00
|
|
|
dbInfo.infoGeneration = ++self->db.dbInfoCount;
|
2021-03-11 02:06:03 +08:00
|
|
|
self->db.serverInfo->set(dbInfo);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2018-06-28 14:02:08 +08:00
|
|
|
checkOutstandingRequests(self);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
// Halts the registering (i.e. requesting) singleton if one is already in the process of being recruited
|
|
|
|
// or, halts the existing singleton in favour of the requesting one
|
|
|
|
template <class Interface>
|
|
|
|
void haltRegisteringOrCurrentSingleton(ClusterControllerData* self,
|
|
|
|
const WorkerInterface& worker,
|
|
|
|
const Singleton<Interface>& currSingleton,
|
|
|
|
const Singleton<Interface>& registeringSingleton,
|
|
|
|
const Optional<UID> recruitingID) {
|
|
|
|
ASSERT(currSingleton.getRole() == registeringSingleton.getRole());
|
|
|
|
const UID registeringID = registeringSingleton.interface.get().id();
|
|
|
|
const std::string roleName = currSingleton.getRole().roleName;
|
|
|
|
const std::string roleAbbr = currSingleton.getRole().abbreviation;
|
|
|
|
|
|
|
|
// halt the requesting singleton if it isn't the one currently being recruited
|
|
|
|
if ((recruitingID.present() && recruitingID.get() != registeringID) ||
|
|
|
|
self->clusterControllerDcId != worker.locality.dcId()) {
|
|
|
|
TraceEvent(("CCHaltRegistering" + roleName).c_str(), self->id)
|
|
|
|
.detail(roleAbbr + "ID", registeringID)
|
|
|
|
.detail("DcID", printable(self->clusterControllerDcId))
|
|
|
|
.detail("ReqDcID", printable(worker.locality.dcId()))
|
|
|
|
.detail("Recruiting" + roleAbbr + "ID", recruitingID.present() ? recruitingID.get() : UID());
|
|
|
|
registeringSingleton.halt(self, worker.locality.processId());
|
|
|
|
} else if (!recruitingID.present()) {
|
|
|
|
// if not currently recruiting, then halt previous one in favour of requesting one
|
|
|
|
TraceEvent(("CCRegister" + roleName).c_str(), self->id).detail(roleAbbr + "ID", registeringID);
|
|
|
|
if (currSingleton.interface.present() && currSingleton.interface.get().id() != registeringID &&
|
|
|
|
self->id_worker.count(currSingleton.interface.get().locality.processId())) {
|
|
|
|
TraceEvent(("CCHaltPrevious" + roleName).c_str(), self->id)
|
|
|
|
.detail(roleAbbr + "ID", currSingleton.interface.get().id())
|
|
|
|
.detail("DcID", printable(self->clusterControllerDcId))
|
|
|
|
.detail("ReqDcID", printable(worker.locality.dcId()))
|
|
|
|
.detail("Recruiting" + roleAbbr + "ID", recruitingID.present() ? recruitingID.get() : UID());
|
|
|
|
currSingleton.halt(self, currSingleton.interface.get().locality.processId());
|
|
|
|
}
|
|
|
|
// set the curr singleton if it doesn't exist or its different from the requesting one
|
|
|
|
if (!currSingleton.interface.present() || currSingleton.interface.get().id() != registeringID) {
|
2021-09-16 22:24:15 +08:00
|
|
|
registeringSingleton.setInterfaceToDbInfo(self);
|
2021-09-13 21:58:38 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-02-02 14:27:12 +08:00
|
|
|
void registerWorker(RegisterWorkerRequest req,
|
|
|
|
ClusterControllerData* self,
|
|
|
|
ServerCoordinators coordinators,
|
|
|
|
ConfigBroadcaster* configBroadcaster) {
|
2019-03-24 11:55:03 +08:00
|
|
|
const WorkerInterface& w = req.wi;
|
2017-10-13 08:11:58 +08:00
|
|
|
ProcessClass newProcessClass = req.processClass;
|
2021-03-11 02:06:03 +08:00
|
|
|
auto info = self->id_worker.find(w.locality.processId());
|
2018-02-10 08:48:55 +08:00
|
|
|
ClusterControllerPriorityInfo newPriorityInfo = req.priorityInfo;
|
2019-03-22 08:56:04 +08:00
|
|
|
newPriorityInfo.processClassFitness = newProcessClass.machineClassFitness(ProcessClass::ClusterController);
|
2022-02-02 14:27:12 +08:00
|
|
|
Optional<ConfigFollowerInterface> cfi;
|
|
|
|
bool isCoordinator =
|
|
|
|
std::find_if(coordinators.configServers.begin(),
|
|
|
|
coordinators.configServers.end(),
|
|
|
|
[&req](const ConfigFollowerInterface& cfi) {
|
|
|
|
return cfi.address() == req.wi.address() || (req.wi.secondaryAddress().present() &&
|
|
|
|
cfi.address() == req.wi.secondaryAddress().get());
|
|
|
|
}) != coordinators.configServers.end();
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto it : req.incompatiblePeers) {
|
2020-04-07 11:58:43 +08:00
|
|
|
self->db.incompatibleConnections[it] = now() + SERVER_KNOBS->INCOMPATIBLE_PEERS_LOGGING_INTERVAL;
|
2020-04-06 14:09:36 +08:00
|
|
|
}
|
2020-04-11 04:45:16 +08:00
|
|
|
self->removedDBInfoEndpoints.erase(w.updateServerDBInfo.getEndpoint());
|
2020-04-06 14:09:36 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (info == self->id_worker.end()) {
|
|
|
|
TraceEvent("ClusterControllerActualWorkers", self->id)
|
|
|
|
.detail("WorkerId", w.id())
|
|
|
|
.detail("ProcessId", w.locality.processId())
|
|
|
|
.detail("ZoneId", w.locality.zoneId())
|
|
|
|
.detail("DataHall", w.locality.dataHallId())
|
|
|
|
.detail("PClass", req.processClass.toString())
|
|
|
|
.detail("Workers", self->id_worker.size());
|
2020-02-20 08:48:30 +08:00
|
|
|
self->goodRecruitmentTime = lowPriorityDelay(SERVER_KNOBS->WAIT_FOR_GOOD_RECRUITMENT_DELAY);
|
2020-02-21 05:46:22 +08:00
|
|
|
self->goodRemoteRecruitmentTime = lowPriorityDelay(SERVER_KNOBS->WAIT_FOR_GOOD_REMOTE_RECRUITMENT_DELAY);
|
2018-11-05 15:07:56 +08:00
|
|
|
} else {
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEvent("ClusterControllerWorkerAlreadyRegistered", self->id)
|
|
|
|
.suppressFor(1.0)
|
|
|
|
.detail("WorkerId", w.id())
|
|
|
|
.detail("ProcessId", w.locality.processId())
|
|
|
|
.detail("ZoneId", w.locality.zoneId())
|
|
|
|
.detail("DataHall", w.locality.dataHallId())
|
|
|
|
.detail("PClass", req.processClass.toString())
|
2021-04-21 15:22:33 +08:00
|
|
|
.detail("Workers", self->id_worker.size())
|
|
|
|
.detail("Degraded", req.degraded);
|
2021-03-11 02:06:03 +08:00
|
|
|
}
|
|
|
|
if (w.address() == g_network->getLocalAddress()) {
|
|
|
|
if (self->changingDcIds.get().first) {
|
|
|
|
if (self->changingDcIds.get().second.present()) {
|
|
|
|
newPriorityInfo.dcFitness = ClusterControllerPriorityInfo::calculateDCFitness(
|
|
|
|
w.locality.dcId(), self->changingDcIds.get().second.get());
|
|
|
|
}
|
|
|
|
} else if (self->changedDcIds.get().second.present()) {
|
|
|
|
newPriorityInfo.dcFitness = ClusterControllerPriorityInfo::calculateDCFitness(
|
|
|
|
w.locality.dcId(), self->changedDcIds.get().second.get());
|
2018-06-29 15:10:29 +08:00
|
|
|
}
|
|
|
|
} else {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!self->changingDcIds.get().first) {
|
|
|
|
if (self->changingDcIds.get().second.present()) {
|
|
|
|
newPriorityInfo.dcFitness = ClusterControllerPriorityInfo::calculateDCFitness(
|
|
|
|
w.locality.dcId(), self->changingDcIds.get().second.get());
|
2018-06-29 15:10:29 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
} else if (self->changedDcIds.get().second.present()) {
|
|
|
|
newPriorityInfo.dcFitness = ClusterControllerPriorityInfo::calculateDCFitness(
|
|
|
|
w.locality.dcId(), self->changedDcIds.get().second.get());
|
2018-02-10 08:48:55 +08:00
|
|
|
}
|
2017-10-26 02:35:29 +08:00
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2017-11-15 05:57:37 +08:00
|
|
|
// Check process class and exclusive property
|
2021-03-11 02:06:03 +08:00
|
|
|
if (info == self->id_worker.end() || info->second.details.interf.id() != w.id() ||
|
|
|
|
req.generation >= info->second.gen) {
|
|
|
|
if (self->gotProcessClasses) {
|
2017-11-15 05:57:37 +08:00
|
|
|
auto classIter = self->id_class.find(w.locality.processId());
|
2019-02-01 10:20:14 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (classIter != self->id_class.end() && (classIter->second.classSource() == ProcessClass::DBSource ||
|
|
|
|
req.initialClass.classType() == ProcessClass::UnsetClass)) {
|
2017-11-15 05:57:37 +08:00
|
|
|
newProcessClass = classIter->second;
|
|
|
|
} else {
|
|
|
|
newProcessClass = req.initialClass;
|
|
|
|
}
|
2018-02-10 08:48:55 +08:00
|
|
|
newPriorityInfo.processClassFitness = newProcessClass.machineClassFitness(ProcessClass::ClusterController);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (self->gotFullyRecoveredConfig) {
|
2020-04-11 04:45:16 +08:00
|
|
|
newPriorityInfo.isExcluded = self->db.fullyRecoveredConfig.isExcludedServer(w.addresses());
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2017-10-13 08:11:58 +08:00
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (info == self->id_worker.end()) {
|
|
|
|
self->id_worker[w.locality.processId()] = WorkerInfo(workerAvailabilityWatch(w, newProcessClass, self),
|
|
|
|
req.reply,
|
|
|
|
req.generation,
|
|
|
|
w,
|
|
|
|
req.initialClass,
|
|
|
|
newProcessClass,
|
|
|
|
newPriorityInfo,
|
|
|
|
req.degraded,
|
|
|
|
req.issues);
|
|
|
|
if (!self->masterProcessId.present() &&
|
|
|
|
w.locality.processId() == self->db.serverInfo->get().master.locality.processId()) {
|
2019-03-24 11:55:03 +08:00
|
|
|
self->masterProcessId = w.locality.processId();
|
|
|
|
}
|
2022-02-02 14:27:12 +08:00
|
|
|
if (configBroadcaster != nullptr && isCoordinator) {
|
|
|
|
self->addActor.send(configBroadcaster->registerNode(
|
|
|
|
w,
|
2021-08-07 03:42:07 +08:00
|
|
|
req.lastSeenKnobVersion,
|
|
|
|
req.knobConfigClassSet,
|
|
|
|
self->id_worker[w.locality.processId()].watcher,
|
2021-08-10 02:49:57 +08:00
|
|
|
self->id_worker[w.locality.processId()].details.interf.configBroadcastInterface));
|
2021-08-06 09:50:11 +08:00
|
|
|
}
|
2021-11-09 01:43:02 +08:00
|
|
|
self->updateDBInfoEndpoints.insert(w.updateServerDBInfo.getEndpoint());
|
|
|
|
self->updateDBInfo.trigger();
|
2021-03-11 02:06:03 +08:00
|
|
|
checkOutstandingRequests(self);
|
|
|
|
} else if (info->second.details.interf.id() != w.id() || req.generation >= info->second.gen) {
|
2017-09-26 01:36:03 +08:00
|
|
|
if (!info->second.reply.isSet()) {
|
2021-03-11 02:06:03 +08:00
|
|
|
info->second.reply.send(Never());
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
info->second.reply = req.reply;
|
2019-03-09 00:25:07 +08:00
|
|
|
info->second.details.processClass = newProcessClass;
|
2018-02-10 08:48:55 +08:00
|
|
|
info->second.priorityInfo = newPriorityInfo;
|
2017-11-15 05:57:37 +08:00
|
|
|
info->second.initialClass = req.initialClass;
|
2019-03-09 03:40:00 +08:00
|
|
|
info->second.details.degraded = req.degraded;
|
2017-05-26 04:48:44 +08:00
|
|
|
info->second.gen = req.generation;
|
2020-04-06 14:09:36 +08:00
|
|
|
info->second.issues = req.issues;
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (info->second.details.interf.id() != w.id()) {
|
2020-04-06 14:09:36 +08:00
|
|
|
self->removedDBInfoEndpoints.insert(info->second.details.interf.updateServerDBInfo.getEndpoint());
|
2019-03-09 00:25:07 +08:00
|
|
|
info->second.details.interf = w;
|
2021-03-11 02:06:03 +08:00
|
|
|
info->second.watcher = workerAvailabilityWatch(w, newProcessClass, self);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-11-04 03:51:21 +08:00
|
|
|
if (req.requestDbInfo) {
|
|
|
|
self->updateDBInfoEndpoints.insert(w.updateServerDBInfo.getEndpoint());
|
|
|
|
self->updateDBInfo.trigger();
|
|
|
|
}
|
2022-02-02 14:27:12 +08:00
|
|
|
if (configBroadcaster != nullptr && isCoordinator) {
|
|
|
|
self->addActor.send(configBroadcaster->registerNode(w,
|
|
|
|
req.lastSeenKnobVersion,
|
|
|
|
req.knobConfigClassSet,
|
|
|
|
info->second.watcher,
|
|
|
|
info->second.details.interf.configBroadcastInterface));
|
2021-08-06 09:50:11 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
checkOutstandingRequests(self);
|
2019-03-23 08:08:54 +08:00
|
|
|
} else {
|
|
|
|
TEST(true); // Received an old worker registration request.
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
// For each singleton
|
|
|
|
// - if the registering singleton conflicts with the singleton being recruited, kill the registering one
|
|
|
|
// - if the singleton is not being recruited, kill the existing one in favour of the registering one
|
|
|
|
if (req.distributorInterf.present()) {
|
|
|
|
auto currSingleton = DataDistributorSingleton(self->db.serverInfo->get().distributor);
|
|
|
|
auto registeringSingleton = DataDistributorSingleton(req.distributorInterf);
|
|
|
|
haltRegisteringOrCurrentSingleton<DataDistributorInterface>(
|
|
|
|
self, w, currSingleton, registeringSingleton, self->recruitingDistributorID);
|
2019-01-29 03:29:39 +08:00
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
|
2019-03-20 02:29:19 +08:00
|
|
|
if (req.ratekeeperInterf.present()) {
|
2021-09-13 21:58:38 +08:00
|
|
|
auto currSingleton = RatekeeperSingleton(self->db.serverInfo->get().ratekeeper);
|
|
|
|
auto registeringSingleton = RatekeeperSingleton(req.ratekeeperInterf);
|
|
|
|
haltRegisteringOrCurrentSingleton<RatekeeperInterface>(
|
|
|
|
self, w, currSingleton, registeringSingleton, self->recruitingRatekeeperID);
|
2019-01-29 03:29:39 +08:00
|
|
|
}
|
2019-11-13 05:01:29 +08:00
|
|
|
|
2021-12-10 05:22:11 +08:00
|
|
|
if (self->db.blobGranulesEnabled.get() && req.blobManagerInterf.present()) {
|
2021-09-15 23:35:58 +08:00
|
|
|
auto currSingleton = BlobManagerSingleton(self->db.serverInfo->get().blobManager);
|
|
|
|
auto registeringSingleton = BlobManagerSingleton(req.blobManagerInterf);
|
|
|
|
haltRegisteringOrCurrentSingleton<BlobManagerInterface>(
|
|
|
|
self, w, currSingleton, registeringSingleton, self->recruitingBlobManagerID);
|
|
|
|
}
|
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
if (SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY && req.encryptKeyProxyInterf.present()) {
|
|
|
|
auto currSingleton = EncryptKeyProxySingleton(self->db.serverInfo->get().encryptKeyProxy);
|
|
|
|
auto registeringSingleton = EncryptKeyProxySingleton(req.encryptKeyProxyInterf);
|
|
|
|
haltRegisteringOrCurrentSingleton<EncryptKeyProxyInterface>(
|
|
|
|
self, w, currSingleton, registeringSingleton, self->recruitingEncryptKeyProxyID);
|
|
|
|
}
|
|
|
|
|
2019-11-13 05:01:29 +08:00
|
|
|
// Notify the worker to register again with new process class/exclusive property
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!req.reply.isSet() && newPriorityInfo != req.priorityInfo) {
|
|
|
|
req.reply.send(RegisterWorkerReply(newProcessClass, newPriorityInfo));
|
2019-11-13 05:01:29 +08:00
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2017-09-28 07:31:38 +08:00
|
|
|
#define TIME_KEEPER_VERSION LiteralStringRef("1")
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> timeKeeperSetVersion(ClusterControllerData* self) {
|
2020-11-07 15:50:55 +08:00
|
|
|
state Reference<ReadYourWritesTransaction> tr = makeReference<ReadYourWritesTransaction>(self->cx);
|
2017-10-19 05:31:31 +08:00
|
|
|
loop {
|
|
|
|
try {
|
|
|
|
tr->setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
tr->setOption(FDBTransactionOptions::LOCK_AWARE);
|
2018-01-09 10:21:00 +08:00
|
|
|
tr->setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
2017-10-19 05:31:31 +08:00
|
|
|
tr->set(timeKeeperVersionKey, TIME_KEEPER_VERSION);
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(tr->commit());
|
2017-10-19 05:31:31 +08:00
|
|
|
break;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(tr->onError(e));
|
2017-09-28 07:31:38 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
// This actor periodically gets read version and writes it to cluster with current timestamp as key. To avoid
|
|
|
|
// running out of space, it limits the max number of entries and clears old entries on each update. This mapping is
|
|
|
|
// used from backup and restore to get the version information for a timestamp.
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> timeKeeper(ClusterControllerData* self) {
|
2017-09-26 03:40:24 +08:00
|
|
|
state KeyBackedMap<int64_t, Version> versionMap(timeKeeperPrefixRange.begin);
|
|
|
|
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("TimeKeeperStarted").log();
|
2017-09-28 07:31:38 +08:00
|
|
|
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(timeKeeperSetVersion(self));
|
2017-09-28 07:31:38 +08:00
|
|
|
|
2017-09-26 03:40:24 +08:00
|
|
|
loop {
|
2020-11-07 15:50:55 +08:00
|
|
|
state Reference<ReadYourWritesTransaction> tr = makeReference<ReadYourWritesTransaction>(self->cx);
|
2017-10-19 05:31:31 +08:00
|
|
|
loop {
|
|
|
|
try {
|
2021-08-19 06:22:58 +08:00
|
|
|
state UID debugID = deterministicRandom()->randomUniqueID();
|
|
|
|
if (!g_network->isSimulated()) {
|
|
|
|
// This is done to provide an arbitrary logged transaction every ~10s.
|
|
|
|
// FIXME: replace or augment this with logging on the proxy which tracks
|
|
|
|
// how long it is taking to hear responses from each other component.
|
|
|
|
tr->debugTransaction(debugID);
|
|
|
|
}
|
2017-10-19 05:31:31 +08:00
|
|
|
tr->setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
tr->setOption(FDBTransactionOptions::LOCK_AWARE);
|
2018-01-09 10:21:00 +08:00
|
|
|
tr->setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
2017-09-26 03:40:24 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
Optional<Value> disableValue = wait(tr->get(timeKeeperDisableKey));
|
|
|
|
if (disableValue.present()) {
|
2017-10-19 05:31:31 +08:00
|
|
|
break;
|
|
|
|
}
|
2017-09-26 03:40:24 +08:00
|
|
|
|
2017-10-19 05:31:31 +08:00
|
|
|
Version v = tr->getReadVersion().get();
|
|
|
|
int64_t currentTime = (int64_t)now();
|
|
|
|
versionMap.set(tr, currentTime, v);
|
2021-08-18 16:26:55 +08:00
|
|
|
if (!g_network->isSimulated()) {
|
2021-08-19 06:29:13 +08:00
|
|
|
TraceEvent("TimeKeeperCommit", debugID).detail("Version", v);
|
2021-08-18 16:26:55 +08:00
|
|
|
}
|
2017-10-19 05:31:31 +08:00
|
|
|
int64_t ttl = currentTime - SERVER_KNOBS->TIME_KEEPER_DELAY * SERVER_KNOBS->TIME_KEEPER_MAX_ENTRIES;
|
|
|
|
if (ttl > 0) {
|
|
|
|
versionMap.erase(tr, 0, ttl);
|
2017-09-29 04:13:24 +08:00
|
|
|
}
|
2017-10-19 05:31:31 +08:00
|
|
|
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(tr->commit());
|
2017-10-19 05:31:31 +08:00
|
|
|
break;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(tr->onError(e));
|
2017-09-26 03:40:24 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(delay(SERVER_KNOBS->TIME_KEEPER_DELAY));
|
2017-09-26 03:40:24 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> statusServer(FutureStream<StatusRequest> requests,
|
|
|
|
ClusterControllerData* self,
|
2021-05-19 01:47:16 +08:00
|
|
|
ServerCoordinators coordinators,
|
|
|
|
ConfigBroadcaster const* configBroadcaster) {
|
2017-05-26 04:48:44 +08:00
|
|
|
// Seconds since the END of the last GetStatus executed
|
|
|
|
state double last_request_time = 0.0;
|
|
|
|
|
|
|
|
// Place to accumulate a batch of requests to respond to
|
|
|
|
state std::vector<StatusRequest> requests_batch;
|
|
|
|
|
|
|
|
loop {
|
|
|
|
try {
|
|
|
|
// Wait til first request is ready
|
|
|
|
StatusRequest req = waitNext(requests);
|
2019-10-04 06:29:11 +08:00
|
|
|
++self->statusRequests;
|
2017-05-26 04:48:44 +08:00
|
|
|
requests_batch.push_back(req);
|
|
|
|
|
|
|
|
// Earliest time at which we may begin a new request
|
|
|
|
double next_allowed_request_time = last_request_time + SERVER_KNOBS->STATUS_MIN_TIME_BETWEEN_REQUESTS;
|
|
|
|
|
|
|
|
// Wait if needed to satisfy min_time knob, also allows more requets to queue up.
|
|
|
|
double minwait = std::max(next_allowed_request_time - now(), 0.0);
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(delay(minwait));
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
// Get all requests that are ready right *now*, before GetStatus() begins.
|
|
|
|
// All of these requests will be responded to with the next GetStatus() result.
|
2019-01-29 07:37:30 +08:00
|
|
|
// If requests are batched, do not respond to more than MAX_STATUS_REQUESTS_PER_SECOND
|
|
|
|
// requests per second
|
|
|
|
while (requests.isReady()) {
|
|
|
|
auto req = requests.pop();
|
|
|
|
if (SERVER_KNOBS->STATUS_MIN_TIME_BETWEEN_REQUESTS > 0.0 &&
|
2021-03-11 02:06:03 +08:00
|
|
|
requests_batch.size() + 1 >
|
|
|
|
SERVER_KNOBS->STATUS_MIN_TIME_BETWEEN_REQUESTS * SERVER_KNOBS->MAX_STATUS_REQUESTS_PER_SECOND) {
|
|
|
|
TraceEvent(SevWarnAlways, "TooManyStatusRequests")
|
|
|
|
.suppressFor(1.0)
|
|
|
|
.detail("BatchSize", requests_batch.size());
|
2019-01-29 07:37:30 +08:00
|
|
|
req.reply.sendError(server_overloaded());
|
|
|
|
} else {
|
|
|
|
requests_batch.push_back(req);
|
|
|
|
}
|
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
// Get status but trap errors to send back to client.
|
2021-09-17 08:42:34 +08:00
|
|
|
std::vector<WorkerDetails> workers;
|
2020-04-18 06:05:01 +08:00
|
|
|
std::vector<ProcessIssues> workerIssues;
|
2020-04-06 14:09:36 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto& it : self->id_worker) {
|
2019-03-09 00:25:07 +08:00
|
|
|
workers.push_back(it.second.details);
|
2021-03-11 02:06:03 +08:00
|
|
|
if (it.second.issues.size()) {
|
2021-12-02 14:46:35 +08:00
|
|
|
workerIssues.emplace_back(it.second.details.interf.address(), it.second.issues);
|
2020-04-06 14:09:36 +08:00
|
|
|
}
|
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
std::vector<NetworkAddress> incompatibleConnections;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto it = self->db.incompatibleConnections.begin(); it != self->db.incompatibleConnections.end();) {
|
|
|
|
if (it->second < now()) {
|
2017-05-26 04:48:44 +08:00
|
|
|
it = self->db.incompatibleConnections.erase(it);
|
|
|
|
} else {
|
|
|
|
incompatibleConnections.push_back(it->first);
|
|
|
|
it++;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
state ErrorOr<StatusReply> result = wait(errorOr(clusterGetStatus(self->db.serverInfo,
|
|
|
|
self->cx,
|
|
|
|
workers,
|
|
|
|
workerIssues,
|
|
|
|
&self->db.clientStatus,
|
|
|
|
coordinators,
|
|
|
|
incompatibleConnections,
|
2021-05-19 01:47:16 +08:00
|
|
|
self->datacenterVersionDifference,
|
|
|
|
configBroadcaster)));
|
2019-02-27 08:20:05 +08:00
|
|
|
|
2017-05-26 04:48:44 +08:00
|
|
|
if (result.isError() && result.getError().code() == error_code_actor_cancelled)
|
|
|
|
throw result.getError();
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// Update last_request_time now because GetStatus is finished and the delay is to be measured between
|
|
|
|
// requests
|
2017-05-26 04:48:44 +08:00
|
|
|
last_request_time = now();
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
while (!requests_batch.empty()) {
|
2017-05-26 04:48:44 +08:00
|
|
|
if (result.isError())
|
|
|
|
requests_batch.back().reply.sendError(result.getError());
|
|
|
|
else
|
|
|
|
requests_batch.back().reply.send(result.get());
|
|
|
|
requests_batch.pop_back();
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(yield());
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2017-05-26 04:48:44 +08:00
|
|
|
TraceEvent(SevError, "StatusServerError").error(e);
|
|
|
|
throw e;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> monitorProcessClasses(ClusterControllerData* self) {
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
state ReadYourWritesTransaction trVer(self->db.db);
|
2017-05-26 04:48:44 +08:00
|
|
|
loop {
|
|
|
|
try {
|
2021-03-11 02:06:03 +08:00
|
|
|
trVer.setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
trVer.setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
Optional<Value> val = wait(trVer.get(processClassVersionKey));
|
|
|
|
|
|
|
|
if (val.present())
|
|
|
|
break;
|
|
|
|
|
2021-05-04 04:14:16 +08:00
|
|
|
RangeResult processClasses = wait(trVer.getRange(processClassKeys, CLIENT_KNOBS->TOO_MANY));
|
2021-03-11 02:06:03 +08:00
|
|
|
ASSERT(!processClasses.more && processClasses.size() < CLIENT_KNOBS->TOO_MANY);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
trVer.clear(processClassKeys);
|
|
|
|
trVer.set(processClassVersionKey, processClassVersionValue);
|
|
|
|
for (auto it : processClasses) {
|
|
|
|
UID processUid = decodeProcessClassKeyOld(it.key);
|
|
|
|
trVer.set(processClassKeyFor(processUid.toString()), it.value);
|
|
|
|
}
|
|
|
|
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(trVer.commit());
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("ProcessClassUpgrade").log();
|
2017-05-26 04:48:44 +08:00
|
|
|
break;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
wait(trVer.onError(e));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
loop {
|
2021-03-11 02:06:03 +08:00
|
|
|
state ReadYourWritesTransaction tr(self->db.db);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
loop {
|
|
|
|
try {
|
2021-03-11 02:06:03 +08:00
|
|
|
tr.setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
tr.setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
2021-05-04 04:14:16 +08:00
|
|
|
RangeResult processClasses = wait(tr.getRange(processClassKeys, CLIENT_KNOBS->TOO_MANY));
|
2021-03-11 02:06:03 +08:00
|
|
|
ASSERT(!processClasses.more && processClasses.size() < CLIENT_KNOBS->TOO_MANY);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (processClasses != self->lastProcessClasses || !self->gotProcessClasses) {
|
2017-05-26 04:48:44 +08:00
|
|
|
self->id_class.clear();
|
2021-03-11 02:06:03 +08:00
|
|
|
for (int i = 0; i < processClasses.size(); i++) {
|
|
|
|
auto c = decodeProcessClassValue(processClasses[i].value);
|
|
|
|
ASSERT(c.classSource() != ProcessClass::CommandLineSource);
|
|
|
|
self->id_class[decodeProcessClassKey(processClasses[i].key)] = c;
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto& w : self->id_worker) {
|
2017-05-26 04:48:44 +08:00
|
|
|
auto classIter = self->id_class.find(w.first);
|
2017-09-26 01:36:03 +08:00
|
|
|
ProcessClass newProcessClass;
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (classIter != self->id_class.end() &&
|
|
|
|
(classIter->second.classSource() == ProcessClass::DBSource ||
|
|
|
|
w.second.initialClass.classType() == ProcessClass::UnsetClass)) {
|
2017-09-26 01:36:03 +08:00
|
|
|
newProcessClass = classIter->second;
|
2017-05-26 04:48:44 +08:00
|
|
|
} else {
|
2017-09-26 01:36:03 +08:00
|
|
|
newProcessClass = w.second.initialClass;
|
|
|
|
}
|
|
|
|
|
2019-03-09 00:25:07 +08:00
|
|
|
if (newProcessClass != w.second.details.processClass) {
|
|
|
|
w.second.details.processClass = newProcessClass;
|
2021-03-11 02:06:03 +08:00
|
|
|
w.second.priorityInfo.processClassFitness =
|
|
|
|
newProcessClass.machineClassFitness(ProcessClass::ClusterController);
|
2017-10-05 06:48:55 +08:00
|
|
|
if (!w.second.reply.isSet()) {
|
2021-03-11 02:06:03 +08:00
|
|
|
w.second.reply.send(
|
|
|
|
RegisterWorkerReply(w.second.details.processClass, w.second.priorityInfo));
|
2017-10-05 06:48:55 +08:00
|
|
|
}
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
self->lastProcessClasses = processClasses;
|
|
|
|
self->gotProcessClasses = true;
|
2021-03-11 02:06:03 +08:00
|
|
|
checkOutstandingRequests(self);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
state Future<Void> watchFuture = tr.watch(processClassChangeKey);
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(tr.commit());
|
|
|
|
wait(watchFuture);
|
2017-05-26 04:48:44 +08:00
|
|
|
break;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
wait(tr.onError(e));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-01-19 08:18:34 +08:00
|
|
|
ACTOR Future<Void> monitorServerInfoConfig(ClusterControllerData::DBInfo* db) {
|
|
|
|
loop {
|
|
|
|
state ReadYourWritesTransaction tr(db->db);
|
|
|
|
loop {
|
|
|
|
try {
|
|
|
|
tr.setOption(FDBTransactionOptions::READ_SYSTEM_KEYS);
|
|
|
|
tr.setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
|
|
|
tr.setOption(FDBTransactionOptions::READ_LOCK_AWARE);
|
|
|
|
|
|
|
|
Optional<Value> configVal = wait(tr.get(latencyBandConfigKey));
|
|
|
|
Optional<LatencyBandConfig> config;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (configVal.present()) {
|
2019-01-19 08:18:34 +08:00
|
|
|
config = LatencyBandConfig::parse(configVal.get());
|
|
|
|
}
|
|
|
|
|
2020-04-12 10:30:05 +08:00
|
|
|
auto serverInfo = db->serverInfo->get();
|
2021-03-11 02:06:03 +08:00
|
|
|
if (config != serverInfo.latencyBandConfig) {
|
2019-01-19 08:18:34 +08:00
|
|
|
TraceEvent("LatencyBandConfigChanged").detail("Present", config.present());
|
2019-05-11 05:01:52 +08:00
|
|
|
serverInfo.id = deterministicRandom()->randomUniqueID();
|
2020-04-07 11:58:43 +08:00
|
|
|
serverInfo.infoGeneration = ++db->dbInfoCount;
|
2019-01-19 08:18:34 +08:00
|
|
|
serverInfo.latencyBandConfig = config;
|
2020-04-12 10:30:05 +08:00
|
|
|
db->serverInfo->set(serverInfo);
|
2019-01-19 08:18:34 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
state Future<Void> configChangeFuture = tr.watch(latencyBandConfigKey);
|
2019-01-25 03:43:26 +08:00
|
|
|
|
|
|
|
wait(tr.commit());
|
|
|
|
wait(configChangeFuture);
|
2019-01-19 08:18:34 +08:00
|
|
|
|
|
|
|
break;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2020-01-08 11:53:09 +08:00
|
|
|
wait(tr.onError(e));
|
2019-01-19 08:18:34 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-20 08:37:01 +08:00
|
|
|
// Monitors the global configuration version key for changes. When changes are
|
|
|
|
// made, the global configuration history is read and any updates are sent to
|
|
|
|
// all processes in the system by updating the ClientDBInfo object. The
|
|
|
|
// GlobalConfig actor class contains the functionality to read the latest
|
|
|
|
// history and update the processes local view.
|
2021-03-16 09:03:54 +08:00
|
|
|
ACTOR Future<Void> monitorGlobalConfig(ClusterControllerData::DBInfo* db) {
|
2017-05-26 04:48:44 +08:00
|
|
|
loop {
|
|
|
|
state ReadYourWritesTransaction tr(db->db);
|
|
|
|
loop {
|
|
|
|
try {
|
|
|
|
tr.setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
tr.setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
2021-02-13 10:55:01 +08:00
|
|
|
state Optional<Value> globalConfigVersion = wait(tr.get(globalConfigVersionKey));
|
2021-06-03 11:21:44 +08:00
|
|
|
state ClientDBInfo clientInfo = db->serverInfo->get().client;
|
2021-02-13 10:55:01 +08:00
|
|
|
|
|
|
|
if (globalConfigVersion.present()) {
|
2021-03-16 09:03:54 +08:00
|
|
|
// Since the history keys end with versionstamps, they
|
|
|
|
// should be sorted correctly (versionstamps are stored in
|
|
|
|
// big-endian order).
|
2021-05-04 04:14:16 +08:00
|
|
|
RangeResult globalConfigHistory =
|
2021-03-16 09:03:54 +08:00
|
|
|
wait(tr.getRange(globalConfigHistoryKeys, CLIENT_KNOBS->TOO_MANY));
|
|
|
|
// If the global configuration version key has been set,
|
|
|
|
// the history should contain at least one item.
|
|
|
|
ASSERT(globalConfigHistory.size() > 0);
|
|
|
|
clientInfo.history.clear();
|
|
|
|
|
|
|
|
for (const auto& kv : globalConfigHistory) {
|
|
|
|
ObjectReader reader(kv.value.begin(), IncludeVersion());
|
|
|
|
if (reader.protocolVersion() != g_network->protocolVersion()) {
|
|
|
|
// If the protocol version has changed, the
|
|
|
|
// GlobalConfig actor should refresh its view by
|
|
|
|
// reading the entire global configuration key
|
2021-03-20 04:28:03 +08:00
|
|
|
// range. Setting the version to the max int64_t
|
|
|
|
// will always cause the global configuration
|
|
|
|
// updater to refresh its view of the configuration
|
|
|
|
// keyspace.
|
2021-03-16 09:03:54 +08:00
|
|
|
clientInfo.history.clear();
|
2021-03-20 04:28:03 +08:00
|
|
|
clientInfo.history.emplace_back(std::numeric_limits<Version>::max());
|
2021-03-16 09:03:54 +08:00
|
|
|
break;
|
2021-02-13 10:55:01 +08:00
|
|
|
}
|
2021-03-20 04:28:03 +08:00
|
|
|
|
|
|
|
VersionHistory vh;
|
2021-03-16 09:03:54 +08:00
|
|
|
reader.deserialize(vh);
|
2021-02-13 10:55:01 +08:00
|
|
|
|
2021-03-16 09:03:54 +08:00
|
|
|
// Read commit version out of versionstamp at end of key.
|
|
|
|
BinaryReader versionReader =
|
|
|
|
BinaryReader(kv.key.removePrefix(globalConfigHistoryPrefix), Unversioned());
|
|
|
|
Version historyCommitVersion;
|
|
|
|
versionReader >> historyCommitVersion;
|
|
|
|
historyCommitVersion = bigEndian64(historyCommitVersion);
|
|
|
|
vh.version = historyCommitVersion;
|
|
|
|
|
2021-03-18 11:41:46 +08:00
|
|
|
clientInfo.history.push_back(std::move(vh));
|
2021-02-13 10:55:01 +08:00
|
|
|
}
|
|
|
|
|
2021-03-24 07:22:39 +08:00
|
|
|
if (clientInfo.history.size() > 0) {
|
|
|
|
// The first item in the historical list of mutations
|
|
|
|
// is only used to:
|
|
|
|
// a) Recognize that some historical changes may have
|
|
|
|
// been missed, and the entire global
|
|
|
|
// configuration keyspace needs to be read, or..
|
|
|
|
// b) Check which historical updates have already
|
|
|
|
// been applied. If this is the case, the first
|
|
|
|
// history item must have a version greater than
|
|
|
|
// or equal to whatever version the global
|
|
|
|
// configuration was last updated at, and
|
|
|
|
// therefore won't need to be applied again.
|
|
|
|
clientInfo.history[0].mutations = Standalone<VectorRef<MutationRef>>();
|
|
|
|
}
|
|
|
|
|
2019-05-11 05:01:52 +08:00
|
|
|
clientInfo.id = deterministicRandom()->randomUniqueID();
|
2021-06-03 11:21:44 +08:00
|
|
|
// Update ServerDBInfo so fdbserver processes receive updated history.
|
|
|
|
ServerDBInfo serverInfo = db->serverInfo->get();
|
|
|
|
serverInfo.id = deterministicRandom()->randomUniqueID();
|
|
|
|
serverInfo.infoGeneration = ++db->dbInfoCount;
|
|
|
|
serverInfo.client = clientInfo;
|
|
|
|
db->serverInfo->set(serverInfo);
|
|
|
|
|
|
|
|
// Update ClientDBInfo so client processes receive updated history.
|
2017-05-26 04:48:44 +08:00
|
|
|
db->clientInfo->set(clientInfo);
|
|
|
|
}
|
2019-02-01 10:20:14 +08:00
|
|
|
|
2021-02-13 10:55:01 +08:00
|
|
|
state Future<Void> globalConfigFuture = tr.watch(globalConfigVersionKey);
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(tr.commit());
|
2021-03-24 07:22:39 +08:00
|
|
|
wait(globalConfigFuture);
|
|
|
|
break;
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(tr.onError(e));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> updatedChangingDatacenters(ClusterControllerData* self) {
|
|
|
|
// do not change the cluster controller until all the processes have had a chance to register
|
|
|
|
wait(delay(SERVER_KNOBS->WAIT_FOR_GOOD_RECRUITMENT_DELAY));
|
2018-02-10 08:48:55 +08:00
|
|
|
loop {
|
2018-06-29 16:11:59 +08:00
|
|
|
state Future<Void> onChange = self->desiredDcIds.onChange();
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!self->desiredDcIds.get().present()) {
|
|
|
|
self->changingDcIds.set(std::make_pair(false, self->desiredDcIds.get()));
|
2018-06-29 16:11:59 +08:00
|
|
|
} else {
|
2018-06-29 15:10:29 +08:00
|
|
|
auto& worker = self->id_worker[self->clusterControllerProcessId];
|
2021-03-11 02:06:03 +08:00
|
|
|
uint8_t newFitness = ClusterControllerPriorityInfo::calculateDCFitness(
|
|
|
|
worker.details.interf.locality.dcId(), self->desiredDcIds.get().get());
|
|
|
|
self->changingDcIds.set(
|
|
|
|
std::make_pair(worker.priorityInfo.dcFitness > newFitness, self->desiredDcIds.get()));
|
|
|
|
|
|
|
|
TraceEvent("UpdateChangingDatacenter", self->id)
|
|
|
|
.detail("OldFitness", worker.priorityInfo.dcFitness)
|
|
|
|
.detail("NewFitness", newFitness);
|
|
|
|
if (worker.priorityInfo.dcFitness > newFitness) {
|
2018-06-29 15:10:29 +08:00
|
|
|
worker.priorityInfo.dcFitness = newFitness;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!worker.reply.isSet()) {
|
|
|
|
worker.reply.send(RegisterWorkerReply(worker.details.processClass, worker.priorityInfo));
|
2018-06-29 15:10:29 +08:00
|
|
|
}
|
2018-06-29 16:11:59 +08:00
|
|
|
} else {
|
|
|
|
state int currentFit = ProcessClass::BestFit;
|
2021-03-11 02:06:03 +08:00
|
|
|
while (currentFit <= ProcessClass::NeverAssign) {
|
2018-06-29 16:11:59 +08:00
|
|
|
bool updated = false;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto& it : self->id_worker) {
|
|
|
|
if ((!it.second.priorityInfo.isExcluded &&
|
|
|
|
it.second.priorityInfo.processClassFitness == currentFit) ||
|
|
|
|
currentFit == ProcessClass::NeverAssign) {
|
|
|
|
uint8_t fitness = ClusterControllerPriorityInfo::calculateDCFitness(
|
|
|
|
it.second.details.interf.locality.dcId(), self->changingDcIds.get().second.get());
|
|
|
|
if (it.first != self->clusterControllerProcessId &&
|
|
|
|
it.second.priorityInfo.dcFitness != fitness) {
|
2018-06-29 16:11:59 +08:00
|
|
|
updated = true;
|
|
|
|
it.second.priorityInfo.dcFitness = fitness;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!it.second.reply.isSet()) {
|
|
|
|
it.second.reply.send(
|
|
|
|
RegisterWorkerReply(it.second.details.processClass, it.second.priorityInfo));
|
2018-06-29 16:11:59 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
if (updated && currentFit < ProcessClass::NeverAssign) {
|
|
|
|
wait(delay(SERVER_KNOBS->CC_CLASS_DELAY));
|
2018-06-29 16:11:59 +08:00
|
|
|
}
|
|
|
|
currentFit++;
|
|
|
|
}
|
2018-02-10 08:48:55 +08:00
|
|
|
}
|
|
|
|
}
|
2018-06-29 15:10:29 +08:00
|
|
|
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(onChange);
|
2018-02-10 08:48:55 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> updatedChangedDatacenters(ClusterControllerData* self) {
|
2018-02-10 08:48:55 +08:00
|
|
|
state Future<Void> changeDelay = delay(SERVER_KNOBS->CC_CHANGE_DELAY);
|
2018-06-29 16:11:59 +08:00
|
|
|
state Future<Void> onChange = self->changingDcIds.onChange();
|
2018-02-10 08:48:55 +08:00
|
|
|
loop {
|
|
|
|
choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(onChange)) {
|
2018-06-29 16:11:59 +08:00
|
|
|
changeDelay = delay(SERVER_KNOBS->CC_CHANGE_DELAY);
|
|
|
|
onChange = self->changingDcIds.onChange();
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(changeDelay)) {
|
2018-02-10 08:48:55 +08:00
|
|
|
changeDelay = Never();
|
2018-06-29 16:11:59 +08:00
|
|
|
onChange = self->changingDcIds.onChange();
|
|
|
|
|
2018-06-29 15:10:29 +08:00
|
|
|
self->changedDcIds.set(self->changingDcIds.get());
|
2021-03-11 02:06:03 +08:00
|
|
|
if (self->changedDcIds.get().second.present()) {
|
2018-07-03 01:06:54 +08:00
|
|
|
TraceEvent("UpdateChangedDatacenter", self->id).detail("CCFirst", self->changedDcIds.get().first);
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!self->changedDcIds.get().first) {
|
2018-06-29 15:10:29 +08:00
|
|
|
auto& worker = self->id_worker[self->clusterControllerProcessId];
|
2021-03-11 02:06:03 +08:00
|
|
|
uint8_t newFitness = ClusterControllerPriorityInfo::calculateDCFitness(
|
|
|
|
worker.details.interf.locality.dcId(), self->changedDcIds.get().second.get());
|
|
|
|
if (worker.priorityInfo.dcFitness != newFitness) {
|
2018-06-29 15:10:29 +08:00
|
|
|
worker.priorityInfo.dcFitness = newFitness;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!worker.reply.isSet()) {
|
|
|
|
worker.reply.send(
|
|
|
|
RegisterWorkerReply(worker.details.processClass, worker.priorityInfo));
|
2018-06-29 15:10:29 +08:00
|
|
|
}
|
2018-02-10 08:48:55 +08:00
|
|
|
}
|
2018-06-29 16:11:59 +08:00
|
|
|
} else {
|
|
|
|
state int currentFit = ProcessClass::BestFit;
|
2021-03-11 02:06:03 +08:00
|
|
|
while (currentFit <= ProcessClass::NeverAssign) {
|
2018-06-29 16:11:59 +08:00
|
|
|
bool updated = false;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto& it : self->id_worker) {
|
|
|
|
if ((!it.second.priorityInfo.isExcluded &&
|
|
|
|
it.second.priorityInfo.processClassFitness == currentFit) ||
|
|
|
|
currentFit == ProcessClass::NeverAssign) {
|
|
|
|
uint8_t fitness = ClusterControllerPriorityInfo::calculateDCFitness(
|
|
|
|
it.second.details.interf.locality.dcId(),
|
|
|
|
self->changedDcIds.get().second.get());
|
|
|
|
if (it.first != self->clusterControllerProcessId &&
|
|
|
|
it.second.priorityInfo.dcFitness != fitness) {
|
2018-06-29 16:11:59 +08:00
|
|
|
updated = true;
|
|
|
|
it.second.priorityInfo.dcFitness = fitness;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!it.second.reply.isSet()) {
|
|
|
|
it.second.reply.send(RegisterWorkerReply(it.second.details.processClass,
|
|
|
|
it.second.priorityInfo));
|
2018-06-29 16:11:59 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
if (updated && currentFit < ProcessClass::NeverAssign) {
|
|
|
|
wait(delay(SERVER_KNOBS->CC_CLASS_DELAY));
|
2018-06-29 16:11:59 +08:00
|
|
|
}
|
|
|
|
currentFit++;
|
|
|
|
}
|
2018-02-10 08:48:55 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> updateDatacenterVersionDifference(ClusterControllerData* self) {
|
2018-06-22 07:34:36 +08:00
|
|
|
state double lastLogTime = 0;
|
2018-06-14 09:14:14 +08:00
|
|
|
loop {
|
|
|
|
self->versionDifferenceUpdated = false;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (self->db.serverInfo->get().recoveryState >= RecoveryState::ACCEPTING_COMMITS &&
|
|
|
|
self->db.config.usableRegions == 1) {
|
|
|
|
bool oldDifferenceTooLarge = !self->versionDifferenceUpdated ||
|
|
|
|
self->datacenterVersionDifference >= SERVER_KNOBS->MAX_VERSION_DIFFERENCE;
|
2018-06-14 09:14:14 +08:00
|
|
|
self->versionDifferenceUpdated = true;
|
|
|
|
self->datacenterVersionDifference = 0;
|
2018-07-07 05:44:11 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (oldDifferenceTooLarge) {
|
2018-07-07 05:44:11 +08:00
|
|
|
checkOutstandingRequests(self);
|
|
|
|
}
|
|
|
|
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(self->db.serverInfo->onChange());
|
2018-06-14 09:14:14 +08:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
|
|
|
state Optional<TLogInterface> primaryLog;
|
|
|
|
state Optional<TLogInterface> remoteLog;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (self->db.serverInfo->get().recoveryState >= RecoveryState::ALL_LOGS_RECRUITED) {
|
|
|
|
for (auto& logSet : self->db.serverInfo->get().logSystemConfig.tLogs) {
|
|
|
|
if (logSet.isLocal && logSet.locality != tagLocalitySatellite) {
|
|
|
|
for (auto& tLog : logSet.tLogs) {
|
|
|
|
if (tLog.present()) {
|
2018-06-14 09:14:14 +08:00
|
|
|
primaryLog = tLog.interf();
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!logSet.isLocal) {
|
|
|
|
for (auto& tLog : logSet.tLogs) {
|
|
|
|
if (tLog.present()) {
|
2018-06-14 09:14:14 +08:00
|
|
|
remoteLog = tLog.interf();
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!primaryLog.present() || !remoteLog.present()) {
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(self->db.serverInfo->onChange());
|
2018-06-14 09:14:14 +08:00
|
|
|
continue;
|
|
|
|
}
|
2018-06-22 07:31:52 +08:00
|
|
|
|
2018-06-14 09:14:14 +08:00
|
|
|
state Future<Void> onChange = self->db.serverInfo->onChange();
|
|
|
|
loop {
|
2021-03-11 02:06:03 +08:00
|
|
|
state Future<TLogQueuingMetricsReply> primaryMetrics =
|
|
|
|
brokenPromiseToNever(primaryLog.get().getQueuingMetrics.getReply(TLogQueuingMetricsRequest()));
|
|
|
|
state Future<TLogQueuingMetricsReply> remoteMetrics =
|
|
|
|
brokenPromiseToNever(remoteLog.get().getQueuingMetrics.getReply(TLogQueuingMetricsRequest()));
|
2018-06-22 07:31:52 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
wait((success(primaryMetrics) && success(remoteMetrics)) || onChange);
|
|
|
|
if (onChange.isReady()) {
|
2018-06-14 09:14:14 +08:00
|
|
|
break;
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
|
|
|
|
if (primaryMetrics.get().v > 0 && remoteMetrics.get().v > 0) {
|
|
|
|
bool oldDifferenceTooLarge = !self->versionDifferenceUpdated ||
|
|
|
|
self->datacenterVersionDifference >= SERVER_KNOBS->MAX_VERSION_DIFFERENCE;
|
2020-05-11 08:49:09 +08:00
|
|
|
self->versionDifferenceUpdated = true;
|
|
|
|
self->datacenterVersionDifference = primaryMetrics.get().v - remoteMetrics.get().v;
|
2018-06-14 09:14:14 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (oldDifferenceTooLarge && self->datacenterVersionDifference < SERVER_KNOBS->MAX_VERSION_DIFFERENCE) {
|
2020-05-11 08:49:09 +08:00
|
|
|
checkOutstandingRequests(self);
|
|
|
|
}
|
2018-07-07 05:44:11 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if (now() - lastLogTime > SERVER_KNOBS->CLUSTER_CONTROLLER_LOGGING_DELAY) {
|
2020-05-11 08:49:09 +08:00
|
|
|
lastLogTime = now();
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEvent("DatacenterVersionDifference", self->id)
|
|
|
|
.detail("Difference", self->datacenterVersionDifference);
|
2020-05-11 08:49:09 +08:00
|
|
|
}
|
2018-06-22 07:31:52 +08:00
|
|
|
}
|
2018-06-14 09:14:14 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(delay(SERVER_KNOBS->VERSION_LAG_METRIC_INTERVAL) || onChange);
|
|
|
|
if (onChange.isReady()) {
|
2018-06-14 09:14:14 +08:00
|
|
|
break;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
// A background actor that periodically checks remote DC health, and `checkOutstandingRequests` if remote DC
|
|
|
|
// recovers.
|
2021-09-10 11:29:28 +08:00
|
|
|
ACTOR Future<Void> updateRemoteDCHealth(ClusterControllerData* self) {
|
2021-10-19 03:04:22 +08:00
|
|
|
// The purpose of the initial delay is to wait for the cluster to achieve a steady state before checking remote
|
|
|
|
// DC health, since remote DC healthy may trigger a failover, and we don't want that to happen too frequently.
|
2021-09-10 11:29:28 +08:00
|
|
|
wait(delay(SERVER_KNOBS->INITIAL_UPDATE_CROSS_DC_INFO_DELAY));
|
2021-10-08 10:46:37 +08:00
|
|
|
|
|
|
|
self->remoteDCMonitorStarted = true;
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
// When the remote DC health just start, we may just recover from a health degradation. Check if we can failback
|
|
|
|
// if we are currently in the remote DC in the database configuration.
|
2021-10-08 10:46:37 +08:00
|
|
|
if (!self->remoteTransactionSystemDegraded) {
|
|
|
|
checkOutstandingRequests(self);
|
|
|
|
}
|
|
|
|
|
2021-09-10 11:29:28 +08:00
|
|
|
loop {
|
|
|
|
bool oldRemoteTransactionSystemDegraded = self->remoteTransactionSystemDegraded;
|
|
|
|
self->remoteTransactionSystemDegraded = self->remoteTransactionSystemContainsDegradedServers();
|
|
|
|
|
|
|
|
if (oldRemoteTransactionSystemDegraded && !self->remoteTransactionSystemDegraded) {
|
|
|
|
checkOutstandingRequests(self);
|
|
|
|
}
|
|
|
|
wait(delay(SERVER_KNOBS->CHECK_REMOTE_HEALTH_INTERVAL));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2019-02-19 06:54:28 +08:00
|
|
|
ACTOR Future<Void> doEmptyCommit(Database cx) {
|
|
|
|
state Transaction tr(cx);
|
|
|
|
loop {
|
|
|
|
try {
|
|
|
|
tr.setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
|
|
|
tr.setOption(FDBTransactionOptions::LOCK_AWARE);
|
|
|
|
tr.makeSelfConflicting();
|
|
|
|
wait(tr.commit());
|
|
|
|
return Void();
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
wait(tr.onError(e));
|
2019-02-19 06:54:28 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> handleForcedRecoveries(ClusterControllerData* self, ClusterControllerFullInterface interf) {
|
2019-02-19 06:54:28 +08:00
|
|
|
loop {
|
2021-03-11 02:06:03 +08:00
|
|
|
state ForceRecoveryRequest req = waitNext(interf.clientInterface.forceRecovery.getFuture());
|
|
|
|
TraceEvent("ForcedRecoveryStart", self->id)
|
|
|
|
.detail("ClusterControllerDcId", self->clusterControllerDcId)
|
|
|
|
.detail("DcId", req.dcId.printable());
|
2019-02-19 06:54:28 +08:00
|
|
|
state Future<Void> fCommit = doEmptyCommit(self->cx);
|
2019-02-20 08:05:20 +08:00
|
|
|
wait(fCommit || delay(SERVER_KNOBS->FORCE_RECOVERY_CHECK_DELAY));
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!fCommit.isReady() || fCommit.isError()) {
|
2019-03-23 08:08:58 +08:00
|
|
|
if (self->clusterControllerDcId != req.dcId) {
|
2021-09-17 08:42:34 +08:00
|
|
|
std::vector<Optional<Key>> dcPriority;
|
2019-02-19 06:54:28 +08:00
|
|
|
dcPriority.push_back(req.dcId);
|
|
|
|
dcPriority.push_back(self->clusterControllerDcId);
|
|
|
|
self->desiredDcIds.set(dcPriority);
|
|
|
|
} else {
|
|
|
|
self->db.forceRecovery = true;
|
|
|
|
self->db.forceMasterFailure.trigger();
|
|
|
|
}
|
|
|
|
wait(fCommit);
|
|
|
|
}
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("ForcedRecoveryFinish", self->id).log();
|
2019-02-19 06:54:28 +08:00
|
|
|
self->db.forceRecovery = false;
|
|
|
|
req.reply.send(Void());
|
2019-02-19 09:09:55 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
ACTOR Future<Void> startDataDistributor(ClusterControllerData* self) {
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(delay(0.0)); // If master fails at the same time, give it a chance to clear master PID.
|
2018-12-14 05:31:37 +08:00
|
|
|
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("CCStartDataDistributor", self->id).log();
|
2019-02-14 03:54:35 +08:00
|
|
|
loop {
|
2019-01-31 01:05:12 +08:00
|
|
|
try {
|
2021-09-13 21:58:38 +08:00
|
|
|
state bool noDistributor = !self->db.serverInfo->get().distributor.present();
|
2021-03-11 02:06:03 +08:00
|
|
|
while (!self->masterProcessId.present() ||
|
|
|
|
self->masterProcessId != self->db.serverInfo->get().master.locality.processId() ||
|
|
|
|
self->db.serverInfo->get().recoveryState < RecoveryState::ACCEPTING_COMMITS) {
|
2019-03-21 07:03:36 +08:00
|
|
|
wait(self->db.serverInfo->onChange() || delay(SERVER_KNOBS->WAIT_FOR_GOOD_RECRUITMENT_DELAY));
|
2019-01-31 01:05:12 +08:00
|
|
|
}
|
2021-09-13 21:58:38 +08:00
|
|
|
if (noDistributor && self->db.serverInfo->get().distributor.present()) {
|
|
|
|
// Existing distributor registers while waiting, so skip.
|
|
|
|
return Void();
|
2019-01-31 01:05:12 +08:00
|
|
|
}
|
2019-02-01 02:51:25 +08:00
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
std::map<Optional<Standalone<StringRef>>, int> idUsed = self->getUsedIds();
|
|
|
|
WorkerFitnessInfo ddWorker = self->getWorkerForRoleInDatacenter(self->clusterControllerDcId,
|
|
|
|
ProcessClass::DataDistributor,
|
|
|
|
ProcessClass::NeverAssign,
|
|
|
|
self->db.config,
|
|
|
|
idUsed);
|
|
|
|
InitializeDataDistributorRequest req(deterministicRandom()->randomUniqueID());
|
|
|
|
state WorkerDetails worker = ddWorker.worker;
|
2019-03-20 02:29:19 +08:00
|
|
|
if (self->onMasterIsBetter(worker, ProcessClass::DataDistributor)) {
|
|
|
|
worker = self->id_worker[self->masterProcessId.get()].details;
|
|
|
|
}
|
2020-01-08 11:53:09 +08:00
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
self->recruitingDistributorID = req.reqId;
|
|
|
|
TraceEvent("CCRecruitDataDistributor", self->id)
|
|
|
|
.detail("Addr", worker.interf.address())
|
|
|
|
.detail("DDID", req.reqId);
|
2019-01-31 01:05:12 +08:00
|
|
|
|
2021-09-13 21:58:38 +08:00
|
|
|
ErrorOr<DataDistributorInterface> ddInterf = wait(worker.interf.dataDistributor.getReplyUnlessFailedFor(
|
2021-03-11 02:06:03 +08:00
|
|
|
req, SERVER_KNOBS->WAIT_FOR_DISTRIBUTOR_JOIN_DELAY, 0));
|
2021-09-13 21:58:38 +08:00
|
|
|
|
|
|
|
if (ddInterf.present()) {
|
|
|
|
self->recruitDistributor.set(false);
|
|
|
|
self->recruitingDistributorID = ddInterf.get().id();
|
|
|
|
const auto& distributor = self->db.serverInfo->get().distributor;
|
|
|
|
TraceEvent("CCDataDistributorRecruited", self->id)
|
|
|
|
.detail("Addr", worker.interf.address())
|
|
|
|
.detail("DDID", ddInterf.get().id());
|
|
|
|
if (distributor.present() && distributor.get().id() != ddInterf.get().id() &&
|
|
|
|
self->id_worker.count(distributor.get().locality.processId())) {
|
|
|
|
|
|
|
|
TraceEvent("CCHaltDataDistributorAfterRecruit", self->id)
|
|
|
|
.detail("DDID", distributor.get().id())
|
|
|
|
.detail("DcID", printable(self->clusterControllerDcId));
|
|
|
|
|
|
|
|
DataDistributorSingleton(distributor).halt(self, distributor.get().locality.processId());
|
|
|
|
}
|
|
|
|
if (!distributor.present() || distributor.get().id() != ddInterf.get().id()) {
|
|
|
|
self->db.setDistributor(ddInterf.get());
|
|
|
|
}
|
|
|
|
checkOutstandingRequests(self);
|
|
|
|
return Void();
|
2019-01-31 01:05:12 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2019-07-05 23:12:25 +08:00
|
|
|
TraceEvent("CCDataDistributorRecruitError", self->id).error(e);
|
2021-03-11 02:06:03 +08:00
|
|
|
if (e.code() != error_code_no_more_servers) {
|
2019-01-31 01:05:12 +08:00
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(lowPriorityDelay(SERVER_KNOBS->ATTEMPT_RECRUITMENT_DELAY));
|
2018-12-14 05:31:37 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> monitorDataDistributor(ClusterControllerData* self) {
|
|
|
|
while (self->db.serverInfo->get().recoveryState < RecoveryState::ACCEPTING_COMMITS) {
|
2019-03-25 07:48:24 +08:00
|
|
|
wait(self->db.serverInfo->onChange());
|
2018-12-14 05:31:37 +08:00
|
|
|
}
|
|
|
|
|
2019-02-01 02:51:25 +08:00
|
|
|
loop {
|
2021-09-13 21:58:38 +08:00
|
|
|
if (self->db.serverInfo->get().distributor.present() && !self->recruitDistributor.get()) {
|
|
|
|
choose {
|
|
|
|
when(wait(waitFailureClient(self->db.serverInfo->get().distributor.get().waitFailure,
|
2021-09-16 01:41:59 +08:00
|
|
|
SERVER_KNOBS->DD_FAILURE_TIME))) {
|
2021-09-13 21:58:38 +08:00
|
|
|
TraceEvent("CCDataDistributorDied", self->id)
|
|
|
|
.detail("DDID", self->db.serverInfo->get().distributor.get().id());
|
|
|
|
self->db.clearInterf(ProcessClass::DataDistributorClass);
|
|
|
|
}
|
|
|
|
when(wait(self->recruitDistributor.onChange())) {}
|
|
|
|
}
|
2019-02-13 07:50:44 +08:00
|
|
|
} else {
|
2021-09-13 21:58:38 +08:00
|
|
|
wait(startDataDistributor(self));
|
2019-02-15 08:24:46 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> startRatekeeper(ClusterControllerData* self) {
|
|
|
|
wait(delay(0.0)); // If master fails at the same time, give it a chance to clear master PID.
|
2019-03-21 07:03:36 +08:00
|
|
|
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("CCStartRatekeeper", self->id).log();
|
2019-02-15 08:24:46 +08:00
|
|
|
loop {
|
|
|
|
try {
|
2020-04-12 10:30:05 +08:00
|
|
|
state bool no_ratekeeper = !self->db.serverInfo->get().ratekeeper.present();
|
2021-03-11 02:06:03 +08:00
|
|
|
while (!self->masterProcessId.present() ||
|
|
|
|
self->masterProcessId != self->db.serverInfo->get().master.locality.processId() ||
|
|
|
|
self->db.serverInfo->get().recoveryState < RecoveryState::ACCEPTING_COMMITS) {
|
2019-03-21 07:03:36 +08:00
|
|
|
wait(self->db.serverInfo->onChange() || delay(SERVER_KNOBS->WAIT_FOR_GOOD_RECRUITMENT_DELAY));
|
2019-02-15 08:24:46 +08:00
|
|
|
}
|
2020-04-12 10:30:05 +08:00
|
|
|
if (no_ratekeeper && self->db.serverInfo->get().ratekeeper.present()) {
|
2019-03-22 13:20:00 +08:00
|
|
|
// Existing ratekeeper registers while waiting, so skip.
|
|
|
|
return Void();
|
2019-02-15 08:24:46 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
std::map<Optional<Standalone<StringRef>>, int> id_used = self->getUsedIds();
|
2021-03-11 02:06:03 +08:00
|
|
|
WorkerFitnessInfo rkWorker = self->getWorkerForRoleInDatacenter(self->clusterControllerDcId,
|
|
|
|
ProcessClass::Ratekeeper,
|
|
|
|
ProcessClass::NeverAssign,
|
|
|
|
self->db.config,
|
|
|
|
id_used);
|
2019-05-11 05:01:52 +08:00
|
|
|
InitializeRatekeeperRequest req(deterministicRandom()->randomUniqueID());
|
2019-03-20 02:29:19 +08:00
|
|
|
state WorkerDetails worker = rkWorker.worker;
|
2019-03-27 23:24:25 +08:00
|
|
|
if (self->onMasterIsBetter(worker, ProcessClass::Ratekeeper)) {
|
2019-03-20 02:29:19 +08:00
|
|
|
worker = self->id_worker[self->masterProcessId.get()].details;
|
|
|
|
}
|
2019-03-23 09:22:45 +08:00
|
|
|
|
2019-03-20 02:29:19 +08:00
|
|
|
self->recruitingRatekeeperID = req.reqId;
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEvent("CCRecruitRatekeeper", self->id)
|
|
|
|
.detail("Addr", worker.interf.address())
|
|
|
|
.detail("RKID", req.reqId);
|
2019-02-15 08:24:46 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ErrorOr<RatekeeperInterface> interf = wait(
|
|
|
|
worker.interf.ratekeeper.getReplyUnlessFailedFor(req, SERVER_KNOBS->WAIT_FOR_RATEKEEPER_JOIN_DELAY, 0));
|
2019-02-15 08:24:46 +08:00
|
|
|
if (interf.present()) {
|
2019-03-23 09:22:45 +08:00
|
|
|
self->recruitRatekeeper.set(false);
|
2019-03-25 02:04:39 +08:00
|
|
|
self->recruitingRatekeeperID = interf.get().id();
|
2020-04-12 10:30:05 +08:00
|
|
|
const auto& ratekeeper = self->db.serverInfo->get().ratekeeper;
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEvent("CCRatekeeperRecruited", self->id)
|
|
|
|
.detail("Addr", worker.interf.address())
|
|
|
|
.detail("RKID", interf.get().id());
|
|
|
|
if (ratekeeper.present() && ratekeeper.get().id() != interf.get().id() &&
|
|
|
|
self->id_worker.count(ratekeeper.get().locality.processId())) {
|
|
|
|
TraceEvent("CCHaltRatekeeperAfterRecruit", self->id)
|
|
|
|
.detail("RKID", ratekeeper.get().id())
|
|
|
|
.detail("DcID", printable(self->clusterControllerDcId));
|
2021-09-13 21:58:38 +08:00
|
|
|
RatekeeperSingleton(ratekeeper).halt(self, ratekeeper.get().locality.processId());
|
2021-03-11 02:06:03 +08:00
|
|
|
}
|
|
|
|
if (!ratekeeper.present() || ratekeeper.get().id() != interf.get().id()) {
|
2019-03-23 08:56:16 +08:00
|
|
|
self->db.setRatekeeper(interf.get());
|
|
|
|
}
|
2019-03-23 09:22:45 +08:00
|
|
|
checkOutstandingRequests(self);
|
2019-03-21 04:54:15 +08:00
|
|
|
return Void();
|
2019-02-15 08:24:46 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2019-07-05 23:12:25 +08:00
|
|
|
TraceEvent("CCRatekeeperRecruitError", self->id).error(e);
|
2021-03-11 02:06:03 +08:00
|
|
|
if (e.code() != error_code_no_more_servers) {
|
2019-02-15 08:24:46 +08:00
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(lowPriorityDelay(SERVER_KNOBS->ATTEMPT_RECRUITMENT_DELAY));
|
2019-02-15 08:24:46 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> monitorRatekeeper(ClusterControllerData* self) {
|
|
|
|
while (self->db.serverInfo->get().recoveryState < RecoveryState::ACCEPTING_COMMITS) {
|
2019-03-25 07:48:24 +08:00
|
|
|
wait(self->db.serverInfo->onChange());
|
2019-02-15 08:24:46 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
loop {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (self->db.serverInfo->get().ratekeeper.present() && !self->recruitRatekeeper.get()) {
|
2019-03-23 09:22:45 +08:00
|
|
|
choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(waitFailureClient(self->db.serverInfo->get().ratekeeper.get().waitFailure,
|
|
|
|
SERVER_KNOBS->RATEKEEPER_FAILURE_TIME))) {
|
2019-07-05 23:12:25 +08:00
|
|
|
TraceEvent("CCRatekeeperDied", self->id)
|
2021-03-11 02:06:03 +08:00
|
|
|
.detail("RKID", self->db.serverInfo->get().ratekeeper.get().id());
|
2019-03-27 23:24:25 +08:00
|
|
|
self->db.clearInterf(ProcessClass::RatekeeperClass);
|
2019-03-21 01:00:31 +08:00
|
|
|
}
|
2019-03-23 09:22:45 +08:00
|
|
|
when(wait(self->recruitRatekeeper.onChange())) {}
|
2019-03-20 06:21:46 +08:00
|
|
|
}
|
2019-02-15 08:24:46 +08:00
|
|
|
} else {
|
2021-03-11 02:06:03 +08:00
|
|
|
wait(startRatekeeper(self));
|
2018-12-14 05:31:37 +08:00
|
|
|
}
|
2019-02-19 06:54:28 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-15 23:35:58 +08:00
|
|
|
// Acquires the BM lock by getting the next epoch no.
|
|
|
|
ACTOR Future<int64_t> getNextBMEpoch(ClusterControllerData* self) {
|
|
|
|
state Reference<ReadYourWritesTransaction> tr = makeReference<ReadYourWritesTransaction>(self->cx);
|
|
|
|
|
|
|
|
loop {
|
|
|
|
tr->setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
tr->setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
|
|
|
try {
|
|
|
|
Optional<Value> oldEpoch = wait(tr->get(blobManagerEpochKey));
|
|
|
|
state int64_t newEpoch = oldEpoch.present() ? decodeBlobManagerEpochValue(oldEpoch.get()) + 1 : 1;
|
|
|
|
tr->set(blobManagerEpochKey, blobManagerEpochValueFor(newEpoch));
|
|
|
|
|
|
|
|
wait(tr->commit());
|
|
|
|
return newEpoch;
|
|
|
|
} catch (Error& e) {
|
|
|
|
wait(tr->onError(e));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
ACTOR Future<Void> startEncryptKeyProxy(ClusterControllerData* self) {
|
|
|
|
wait(delay(0.0)); // If master fails at the same time, give it a chance to clear master PID.
|
|
|
|
|
|
|
|
TraceEvent("CCEKP_Start", self->id).log();
|
|
|
|
loop {
|
|
|
|
try {
|
|
|
|
// EncryptKeyServer interface is critical in recovering tlog encrypted transactions,
|
|
|
|
// hence, the process only waits for the master recruitment and not the full cluster recovery.
|
|
|
|
state bool noEncryptKeyServer = !self->db.serverInfo->get().encryptKeyProxy.present();
|
|
|
|
while (!self->masterProcessId.present() ||
|
|
|
|
self->masterProcessId != self->db.serverInfo->get().master.locality.processId() ||
|
|
|
|
self->db.serverInfo->get().recoveryState < RecoveryState::LOCKING_CSTATE) {
|
|
|
|
wait(self->db.serverInfo->onChange() || delay(SERVER_KNOBS->WAIT_FOR_GOOD_RECRUITMENT_DELAY));
|
|
|
|
}
|
|
|
|
if (noEncryptKeyServer && self->db.serverInfo->get().encryptKeyProxy.present()) {
|
|
|
|
// Existing encryptKeyServer registers while waiting, so skip.
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Recruit EncryptKeyProxy in the same datacenter as the ClusterController.
|
|
|
|
// This should always be possible, given EncryptKeyProxy is stateless, we can recruit EncryptKeyProxy
|
|
|
|
// on the same process as the CluserController.
|
|
|
|
state std::map<Optional<Standalone<StringRef>>, int> id_used;
|
|
|
|
self->updateKnownIds(&id_used);
|
|
|
|
state WorkerFitnessInfo ekpWorker = self->getWorkerForRoleInDatacenter(self->clusterControllerDcId,
|
|
|
|
ProcessClass::EncryptKeyProxy,
|
|
|
|
ProcessClass::NeverAssign,
|
|
|
|
self->db.config,
|
|
|
|
id_used);
|
|
|
|
|
|
|
|
InitializeEncryptKeyProxyRequest req(deterministicRandom()->randomUniqueID());
|
|
|
|
state WorkerDetails worker = ekpWorker.worker;
|
|
|
|
if (self->onMasterIsBetter(worker, ProcessClass::EncryptKeyProxy)) {
|
|
|
|
worker = self->id_worker[self->masterProcessId.get()].details;
|
|
|
|
}
|
|
|
|
|
|
|
|
self->recruitingEncryptKeyProxyID = req.reqId;
|
|
|
|
TraceEvent("CCEKP_Recruit", self->id).detail("Addr", worker.interf.address()).detail("Id", req.reqId);
|
|
|
|
|
|
|
|
ErrorOr<EncryptKeyProxyInterface> interf = wait(worker.interf.encryptKeyProxy.getReplyUnlessFailedFor(
|
|
|
|
req, SERVER_KNOBS->WAIT_FOR_ENCRYPT_KEY_PROXY_JOIN_DELAY, 0));
|
|
|
|
if (interf.present()) {
|
|
|
|
self->recruitEncryptKeyProxy.set(false);
|
|
|
|
self->recruitingEncryptKeyProxyID = interf.get().id();
|
|
|
|
const auto& encryptKeyProxy = self->db.serverInfo->get().encryptKeyProxy;
|
|
|
|
TraceEvent("CCEKP_Recruited", self->id)
|
|
|
|
.detail("Addr", worker.interf.address())
|
|
|
|
.detail("Id", interf.get().id())
|
|
|
|
.detail("ProcessId", interf.get().locality.processId());
|
|
|
|
if (encryptKeyProxy.present() && encryptKeyProxy.get().id() != interf.get().id() &&
|
|
|
|
self->id_worker.count(encryptKeyProxy.get().locality.processId())) {
|
|
|
|
TraceEvent("CCEKP_HaltAfterRecruit", self->id)
|
|
|
|
.detail("Id", encryptKeyProxy.get().id())
|
|
|
|
.detail("DcId", printable(self->clusterControllerDcId));
|
|
|
|
EncryptKeyProxySingleton(encryptKeyProxy).halt(self, encryptKeyProxy.get().locality.processId());
|
|
|
|
}
|
|
|
|
if (!encryptKeyProxy.present() || encryptKeyProxy.get().id() != interf.get().id()) {
|
|
|
|
self->db.setEncryptKeyProxy(interf.get());
|
|
|
|
TraceEvent("CCEKP_UpdateInf", self->id)
|
|
|
|
.detail("Id", self->db.serverInfo->get().encryptKeyProxy.get().id());
|
|
|
|
}
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
} catch (Error& e) {
|
|
|
|
TraceEvent("CCEKP_RecruitError", self->id).error(e);
|
|
|
|
if (e.code() != error_code_no_more_servers) {
|
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
wait(lowPriorityDelay(SERVER_KNOBS->ATTEMPT_RECRUITMENT_DELAY));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
ACTOR Future<Void> monitorEncryptKeyProxy(ClusterControllerData* self) {
|
|
|
|
loop {
|
|
|
|
if (self->db.serverInfo->get().encryptKeyProxy.present() && !self->recruitEncryptKeyProxy.get()) {
|
|
|
|
choose {
|
|
|
|
when(wait(waitFailureClient(self->db.serverInfo->get().encryptKeyProxy.get().waitFailure,
|
|
|
|
SERVER_KNOBS->ENCRYPT_KEY_PROXY_FAILURE_TIME))) {
|
|
|
|
TraceEvent("CCEKP_Died", self->id);
|
|
|
|
self->db.clearInterf(ProcessClass::EncryptKeyProxyClass);
|
|
|
|
}
|
|
|
|
when(wait(self->recruitEncryptKeyProxy.onChange())) {}
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
wait(startEncryptKeyProxy(self));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-15 23:35:58 +08:00
|
|
|
ACTOR Future<Void> startBlobManager(ClusterControllerData* self) {
|
|
|
|
wait(delay(0.0)); // If master fails at the same time, give it a chance to clear master PID.
|
|
|
|
|
|
|
|
TraceEvent("CCStartBlobManager", self->id).log();
|
|
|
|
loop {
|
|
|
|
try {
|
|
|
|
state bool noBlobManager = !self->db.serverInfo->get().blobManager.present();
|
|
|
|
while (!self->masterProcessId.present() ||
|
|
|
|
self->masterProcessId != self->db.serverInfo->get().master.locality.processId() ||
|
|
|
|
self->db.serverInfo->get().recoveryState < RecoveryState::ACCEPTING_COMMITS) {
|
|
|
|
wait(self->db.serverInfo->onChange() || delay(SERVER_KNOBS->WAIT_FOR_GOOD_RECRUITMENT_DELAY));
|
|
|
|
}
|
|
|
|
if (noBlobManager && self->db.serverInfo->get().blobManager.present()) {
|
|
|
|
// Existing blob manager registers while waiting, so skip.
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
|
|
|
state std::map<Optional<Standalone<StringRef>>, int> id_used = self->getUsedIds();
|
|
|
|
state WorkerFitnessInfo bmWorker = self->getWorkerForRoleInDatacenter(self->clusterControllerDcId,
|
|
|
|
ProcessClass::BlobManager,
|
|
|
|
ProcessClass::NeverAssign,
|
|
|
|
self->db.config,
|
|
|
|
id_used);
|
|
|
|
|
|
|
|
int64_t nextEpoch = wait(getNextBMEpoch(self));
|
2021-12-05 09:08:23 +08:00
|
|
|
if (!self->masterProcessId.present() ||
|
|
|
|
self->masterProcessId != self->db.serverInfo->get().master.locality.processId() ||
|
|
|
|
self->db.serverInfo->get().recoveryState < RecoveryState::ACCEPTING_COMMITS) {
|
|
|
|
continue;
|
|
|
|
}
|
2021-09-15 23:35:58 +08:00
|
|
|
InitializeBlobManagerRequest req(deterministicRandom()->randomUniqueID(), nextEpoch);
|
|
|
|
state WorkerDetails worker = bmWorker.worker;
|
|
|
|
if (self->onMasterIsBetter(worker, ProcessClass::BlobManager)) {
|
|
|
|
worker = self->id_worker[self->masterProcessId.get()].details;
|
|
|
|
}
|
|
|
|
|
|
|
|
self->recruitingBlobManagerID = req.reqId;
|
|
|
|
TraceEvent("CCRecruitBlobManager", self->id)
|
|
|
|
.detail("Addr", worker.interf.address())
|
|
|
|
.detail("BMID", req.reqId);
|
|
|
|
|
|
|
|
ErrorOr<BlobManagerInterface> interf = wait(worker.interf.blobManager.getReplyUnlessFailedFor(
|
|
|
|
req, SERVER_KNOBS->WAIT_FOR_BLOB_MANAGER_JOIN_DELAY, 0));
|
|
|
|
if (interf.present()) {
|
|
|
|
self->recruitBlobManager.set(false);
|
|
|
|
self->recruitingBlobManagerID = interf.get().id();
|
|
|
|
const auto& blobManager = self->db.serverInfo->get().blobManager;
|
|
|
|
TraceEvent("CCBlobManagerRecruited", self->id)
|
|
|
|
.detail("Addr", worker.interf.address())
|
|
|
|
.detail("BMID", interf.get().id());
|
|
|
|
if (blobManager.present() && blobManager.get().id() != interf.get().id() &&
|
|
|
|
self->id_worker.count(blobManager.get().locality.processId())) {
|
|
|
|
TraceEvent("CCHaltBlobManagerAfterRecruit", self->id)
|
|
|
|
.detail("BMID", blobManager.get().id())
|
|
|
|
.detail("DcID", printable(self->clusterControllerDcId));
|
|
|
|
BlobManagerSingleton(blobManager).halt(self, blobManager.get().locality.processId());
|
|
|
|
}
|
|
|
|
if (!blobManager.present() || blobManager.get().id() != interf.get().id()) {
|
|
|
|
self->db.setBlobManager(interf.get());
|
|
|
|
}
|
|
|
|
checkOutstandingRequests(self);
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
} catch (Error& e) {
|
|
|
|
TraceEvent("CCBlobManagerRecruitError", self->id).error(e);
|
|
|
|
if (e.code() != error_code_no_more_servers) {
|
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
wait(lowPriorityDelay(SERVER_KNOBS->ATTEMPT_RECRUITMENT_DELAY));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-12-08 13:43:58 +08:00
|
|
|
ACTOR Future<Void> watchBlobGranulesConfigKey(ClusterControllerData* self) {
|
|
|
|
state Reference<ReadYourWritesTransaction> tr = makeReference<ReadYourWritesTransaction>(self->cx);
|
2021-12-11 03:05:59 +08:00
|
|
|
state Key blobGranuleConfigKey = configKeysPrefix.withSuffix("blob_granules_enabled"_sr);
|
2021-12-08 13:43:58 +08:00
|
|
|
|
|
|
|
loop {
|
|
|
|
try {
|
2021-12-10 06:06:18 +08:00
|
|
|
tr->reset();
|
2021-12-08 13:43:58 +08:00
|
|
|
tr->setOption(FDBTransactionOptions::ACCESS_SYSTEM_KEYS);
|
|
|
|
tr->setOption(FDBTransactionOptions::PRIORITY_SYSTEM_IMMEDIATE);
|
2021-12-10 06:06:18 +08:00
|
|
|
|
2022-01-06 02:48:53 +08:00
|
|
|
Optional<Value> blobConfig = wait(tr->get(blobGranuleConfigKey));
|
|
|
|
if (blobConfig.present()) {
|
|
|
|
self->db.blobGranulesEnabled.set(blobConfig.get() == LiteralStringRef("1"));
|
|
|
|
}
|
2021-12-10 05:22:11 +08:00
|
|
|
|
2022-01-06 02:48:53 +08:00
|
|
|
state Future<Void> watch = tr->watch(blobGranuleConfigKey);
|
|
|
|
wait(tr->commit());
|
|
|
|
wait(watch);
|
2021-12-08 13:43:58 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
wait(tr->onError(e));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-09-15 23:35:58 +08:00
|
|
|
ACTOR Future<Void> monitorBlobManager(ClusterControllerData* self) {
|
|
|
|
while (self->db.serverInfo->get().recoveryState < RecoveryState::ACCEPTING_COMMITS) {
|
|
|
|
wait(self->db.serverInfo->onChange());
|
|
|
|
}
|
|
|
|
|
|
|
|
loop {
|
|
|
|
if (self->db.serverInfo->get().blobManager.present() && !self->recruitBlobManager.get()) {
|
2021-12-11 03:05:59 +08:00
|
|
|
state Future<Void> wfClient = waitFailureClient(self->db.serverInfo->get().blobManager.get().waitFailure,
|
|
|
|
SERVER_KNOBS->BLOB_MANAGER_FAILURE_TIME);
|
|
|
|
loop {
|
|
|
|
choose {
|
|
|
|
when(wait(wfClient)) {
|
|
|
|
TraceEvent("CCBlobManagerDied", self->id)
|
|
|
|
.detail("BMID", self->db.serverInfo->get().blobManager.get().id());
|
|
|
|
self->db.clearInterf(ProcessClass::BlobManagerClass);
|
|
|
|
break;
|
|
|
|
}
|
|
|
|
when(wait(self->recruitBlobManager.onChange())) { break; }
|
|
|
|
when(wait(self->db.blobGranulesEnabled.onChange())) {
|
|
|
|
// if there is a blob manager present but blob granules are now disabled, stop the BM
|
|
|
|
if (!self->db.blobGranulesEnabled.get()) {
|
|
|
|
const auto& blobManager = self->db.serverInfo->get().blobManager;
|
|
|
|
BlobManagerSingleton(blobManager)
|
|
|
|
.haltBlobGranules(self, blobManager.get().locality.processId());
|
|
|
|
break;
|
|
|
|
}
|
2021-12-08 13:43:58 +08:00
|
|
|
}
|
|
|
|
}
|
2021-09-15 23:35:58 +08:00
|
|
|
}
|
2021-12-10 05:22:11 +08:00
|
|
|
} else if (self->db.blobGranulesEnabled.get()) {
|
2021-12-09 03:12:44 +08:00
|
|
|
// if there is no blob manager present but blob granules are now enabled, recruit a BM
|
|
|
|
wait(startBlobManager(self));
|
2021-09-15 23:35:58 +08:00
|
|
|
} else {
|
2021-12-09 03:12:44 +08:00
|
|
|
// if there is no blob manager present and blob granules are disabled, wait for a config change
|
2021-12-10 05:22:11 +08:00
|
|
|
wait(self->db.blobGranulesEnabled.onChange());
|
2021-09-15 23:35:58 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> dbInfoUpdater(ClusterControllerData* self) {
|
2020-04-06 14:09:36 +08:00
|
|
|
state Future<Void> dbInfoChange = self->db.serverInfo->onChange();
|
|
|
|
state Future<Void> updateDBInfo = self->updateDBInfo.onTrigger();
|
|
|
|
loop {
|
|
|
|
choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(updateDBInfo)) { wait(delay(SERVER_KNOBS->DBINFO_BATCH_DELAY) || dbInfoChange); }
|
2020-04-06 14:09:36 +08:00
|
|
|
when(wait(dbInfoChange)) {}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
|
2020-04-18 06:05:01 +08:00
|
|
|
UpdateServerDBInfoRequest req;
|
2021-03-11 02:06:03 +08:00
|
|
|
if (dbInfoChange.isReady()) {
|
|
|
|
for (auto& it : self->id_worker) {
|
2020-04-18 06:05:01 +08:00
|
|
|
req.broadcastInfo.push_back(it.second.details.interf.updateServerDBInfo.getEndpoint());
|
2020-04-06 14:09:36 +08:00
|
|
|
}
|
|
|
|
} else {
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto it : self->removedDBInfoEndpoints) {
|
2020-04-18 07:45:22 +08:00
|
|
|
self->updateDBInfoEndpoints.erase(it);
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
req.broadcastInfo =
|
|
|
|
std::vector<Endpoint>(self->updateDBInfoEndpoints.begin(), self->updateDBInfoEndpoints.end());
|
2020-04-06 14:09:36 +08:00
|
|
|
}
|
|
|
|
|
2020-04-18 06:05:01 +08:00
|
|
|
self->updateDBInfoEndpoints.clear();
|
2020-04-06 14:09:36 +08:00
|
|
|
self->removedDBInfoEndpoints.clear();
|
2021-03-11 02:06:03 +08:00
|
|
|
|
2020-04-06 14:09:36 +08:00
|
|
|
dbInfoChange = self->db.serverInfo->onChange();
|
|
|
|
updateDBInfo = self->updateDBInfo.onTrigger();
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
req.serializedDbInfo =
|
|
|
|
BinaryWriter::toValue(self->db.serverInfo->get(), AssumeVersion(g_network->protocolVersion()));
|
2020-04-06 14:09:36 +08:00
|
|
|
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("DBInfoStartBroadcast", self->id).log();
|
2020-04-06 14:09:36 +08:00
|
|
|
choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(std::vector<Endpoint> notUpdated =
|
|
|
|
wait(broadcastDBInfoRequest(req, SERVER_KNOBS->DBINFO_SEND_AMOUNT, Optional<Endpoint>(), false))) {
|
2020-09-01 01:37:00 +08:00
|
|
|
TraceEvent("DBInfoFinishBroadcast", self->id).detail("NotUpdated", notUpdated.size());
|
2021-03-11 02:06:03 +08:00
|
|
|
if (notUpdated.size()) {
|
2020-04-18 07:45:22 +08:00
|
|
|
self->updateDBInfoEndpoints.insert(notUpdated.begin(), notUpdated.end());
|
2020-04-06 14:09:36 +08:00
|
|
|
self->updateDBInfo.trigger();
|
|
|
|
}
|
|
|
|
}
|
2020-04-11 08:02:11 +08:00
|
|
|
when(wait(dbInfoChange)) {}
|
2020-04-06 14:09:36 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-06-29 12:45:02 +08:00
|
|
|
// The actor that periodically monitors the health of tracked workers.
|
|
|
|
ACTOR Future<Void> workerHealthMonitor(ClusterControllerData* self) {
|
|
|
|
loop {
|
|
|
|
try {
|
|
|
|
while (!self->goodRecruitmentTime.isReady()) {
|
2021-07-26 13:37:08 +08:00
|
|
|
wait(lowPriorityDelay(SERVER_KNOBS->CC_WORKER_HEALTH_CHECKING_INTERVAL));
|
2021-06-29 12:45:02 +08:00
|
|
|
}
|
|
|
|
|
|
|
|
self->degradedServers = self->getServersWithDegradedLink();
|
|
|
|
|
|
|
|
// Compare `self->degradedServers` with `self->excludedDegradedServers` and remove those that have
|
|
|
|
// recovered.
|
|
|
|
for (auto it = self->excludedDegradedServers.begin(); it != self->excludedDegradedServers.end();) {
|
|
|
|
if (self->degradedServers.find(*it) == self->degradedServers.end()) {
|
|
|
|
self->excludedDegradedServers.erase(it++);
|
|
|
|
} else {
|
|
|
|
++it;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
if (!self->degradedServers.empty()) {
|
|
|
|
std::string degradedServerString;
|
|
|
|
for (const auto& server : self->degradedServers) {
|
|
|
|
degradedServerString += server.toString() + " ";
|
|
|
|
}
|
|
|
|
TraceEvent("ClusterControllerHealthMonitor").detail("DegradedServers", degradedServerString);
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
// Check if the cluster controller should trigger a recovery to exclude any degraded servers from
|
|
|
|
// the transaction system.
|
2021-06-29 12:45:02 +08:00
|
|
|
if (self->shouldTriggerRecoveryDueToDegradedServers()) {
|
|
|
|
if (SERVER_KNOBS->CC_HEALTH_TRIGGER_RECOVERY) {
|
|
|
|
if (self->recentRecoveryCountDueToHealth() < SERVER_KNOBS->CC_MAX_HEALTH_RECOVERY_COUNT) {
|
|
|
|
self->recentHealthTriggeredRecoveryTime.push(now());
|
|
|
|
self->excludedDegradedServers = self->degradedServers;
|
|
|
|
TraceEvent("DegradedServerDetectedAndTriggerRecovery")
|
|
|
|
.detail("RecentRecoveryCountDueToHealth", self->recentRecoveryCountDueToHealth());
|
|
|
|
self->db.forceMasterFailure.trigger();
|
|
|
|
}
|
|
|
|
} else {
|
|
|
|
self->excludedDegradedServers.clear();
|
2021-07-27 10:55:10 +08:00
|
|
|
TraceEvent("DegradedServerDetectedAndSuggestRecovery").log();
|
2021-06-29 12:45:02 +08:00
|
|
|
}
|
2021-09-10 11:29:28 +08:00
|
|
|
} else if (self->shouldTriggerFailoverDueToDegradedServers()) {
|
2021-10-08 10:46:37 +08:00
|
|
|
double ccUpTime = now() - machineStartTime();
|
|
|
|
if (SERVER_KNOBS->CC_HEALTH_TRIGGER_FAILOVER &&
|
|
|
|
ccUpTime > SERVER_KNOBS->INITIAL_UPDATE_CROSS_DC_INFO_DELAY) {
|
2021-09-10 11:29:28 +08:00
|
|
|
TraceEvent("DegradedServerDetectedAndTriggerFailover").log();
|
2021-10-07 07:16:24 +08:00
|
|
|
std::vector<Optional<Key>> dcPriority;
|
2021-09-10 11:29:28 +08:00
|
|
|
auto remoteDcId = self->db.config.regions[0].dcId == self->clusterControllerDcId.get()
|
|
|
|
? self->db.config.regions[1].dcId
|
|
|
|
: self->db.config.regions[0].dcId;
|
2021-09-11 05:51:55 +08:00
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
// Switch the current primary DC and remote DC in desiredDcIds, so that the remote DC
|
|
|
|
// becomes the new primary, and the primary DC becomes the new remote.
|
2021-09-10 11:29:28 +08:00
|
|
|
dcPriority.push_back(remoteDcId);
|
|
|
|
dcPriority.push_back(self->clusterControllerDcId);
|
|
|
|
self->desiredDcIds.set(dcPriority);
|
|
|
|
} else {
|
2021-10-08 10:46:37 +08:00
|
|
|
TraceEvent("DegradedServerDetectedAndSuggestFailover").detail("CCUpTime", ccUpTime);
|
2021-09-10 11:29:28 +08:00
|
|
|
}
|
2021-06-29 12:45:02 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
|
|
|
wait(delay(SERVER_KNOBS->CC_WORKER_HEALTH_CHECKING_INTERVAL));
|
|
|
|
} catch (Error& e) {
|
|
|
|
TraceEvent(SevWarnAlways, "ClusterControllerHealthMonitorError").error(e);
|
2021-08-20 07:31:31 +08:00
|
|
|
if (e.code() == error_code_actor_cancelled) {
|
|
|
|
throw;
|
|
|
|
}
|
2021-06-29 12:45:02 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> clusterControllerCore(ClusterControllerFullInterface interf,
|
|
|
|
Future<Void> leaderFail,
|
|
|
|
ServerCoordinators coordinators,
|
2021-05-19 01:47:16 +08:00
|
|
|
LocalityData locality,
|
2021-08-07 14:18:10 +08:00
|
|
|
ConfigDBType configDBType) {
|
2021-04-27 06:54:08 +08:00
|
|
|
state ClusterControllerData self(interf, locality, coordinators);
|
2021-08-07 14:18:10 +08:00
|
|
|
state ConfigBroadcaster configBroadcaster(coordinators, configDBType);
|
2021-03-11 02:06:03 +08:00
|
|
|
state Future<Void> coordinationPingDelay = delay(SERVER_KNOBS->WORKER_COORDINATION_PING_DELAY);
|
2017-05-26 04:48:44 +08:00
|
|
|
state uint64_t step = 0;
|
2021-03-11 02:06:03 +08:00
|
|
|
state Future<ErrorOr<Void>> error = errorOr(actorCollection(self.addActor.getFuture()));
|
|
|
|
|
2022-01-14 03:11:01 +08:00
|
|
|
// EncryptKeyProxy is necessary for TLog recovery, recruit it as the first process
|
|
|
|
if (SERVER_KNOBS->ENABLE_ENCRYPT_KEY_PROXY) {
|
|
|
|
self.addActor.send(monitorEncryptKeyProxy(&self));
|
|
|
|
}
|
2022-01-07 04:15:51 +08:00
|
|
|
self.addActor.send(clusterWatchDatabase(&self, &self.db, coordinators, leaderFail)); // Start the master database
|
2021-03-11 02:06:03 +08:00
|
|
|
self.addActor.send(self.updateWorkerList.init(self.db.db));
|
2021-06-11 11:57:50 +08:00
|
|
|
self.addActor.send(statusServer(interf.clientInterface.databaseStatus.getFuture(),
|
|
|
|
&self,
|
|
|
|
coordinators,
|
2021-08-07 14:18:10 +08:00
|
|
|
(configDBType == ConfigDBType::DISABLED) ? nullptr : &configBroadcaster));
|
2021-03-11 02:06:03 +08:00
|
|
|
self.addActor.send(timeKeeper(&self));
|
|
|
|
self.addActor.send(monitorProcessClasses(&self));
|
|
|
|
self.addActor.send(monitorServerInfoConfig(&self.db));
|
2021-03-16 09:03:54 +08:00
|
|
|
self.addActor.send(monitorGlobalConfig(&self.db));
|
2021-03-11 02:06:03 +08:00
|
|
|
self.addActor.send(updatedChangingDatacenters(&self));
|
|
|
|
self.addActor.send(updatedChangedDatacenters(&self));
|
|
|
|
self.addActor.send(updateDatacenterVersionDifference(&self));
|
|
|
|
self.addActor.send(handleForcedRecoveries(&self, interf));
|
|
|
|
self.addActor.send(monitorDataDistributor(&self));
|
|
|
|
self.addActor.send(monitorRatekeeper(&self));
|
2021-12-08 13:43:58 +08:00
|
|
|
self.addActor.send(monitorBlobManager(&self));
|
2021-12-10 05:22:11 +08:00
|
|
|
self.addActor.send(watchBlobGranulesConfigKey(&self));
|
2021-03-11 02:06:03 +08:00
|
|
|
self.addActor.send(dbInfoUpdater(&self));
|
|
|
|
self.addActor.send(traceCounters("ClusterControllerMetrics",
|
|
|
|
self.id,
|
|
|
|
SERVER_KNOBS->STORAGE_LOGGING_DELAY,
|
|
|
|
&self.clusterControllerMetrics,
|
|
|
|
self.id.toString() + "/ClusterControllerMetrics"));
|
|
|
|
self.addActor.send(traceRole(Role::CLUSTER_CONTROLLER, interf.id()));
|
|
|
|
// printf("%s: I am the cluster controller\n", g_network->getLocalAddress().toString().c_str());
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-06-29 12:45:02 +08:00
|
|
|
if (SERVER_KNOBS->CC_ENABLE_WORKER_HEALTH_MONITOR) {
|
|
|
|
self.addActor.send(workerHealthMonitor(&self));
|
2021-10-07 07:16:24 +08:00
|
|
|
self.addActor.send(updateRemoteDCHealth(&self));
|
2021-06-29 12:45:02 +08:00
|
|
|
}
|
|
|
|
|
2017-05-26 04:48:44 +08:00
|
|
|
loop choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(ErrorOr<Void> err = wait(error)) {
|
2022-01-07 04:15:51 +08:00
|
|
|
if (err.isError() && err.getError().code() != error_code_restart_cluster_controller) {
|
2018-09-06 06:06:14 +08:00
|
|
|
endRole(Role::CLUSTER_CONTROLLER, interf.id(), "Stop Received Error", false, err.getError());
|
2021-03-11 02:06:03 +08:00
|
|
|
} else {
|
2018-09-06 06:06:14 +08:00
|
|
|
endRole(Role::CLUSTER_CONTROLLER, interf.id(), "Stop Received Signal", true);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
// We shut down normally even if there was a serious error (so this fdbserver may be re-elected cluster
|
|
|
|
// controller)
|
2017-05-26 04:48:44 +08:00
|
|
|
return Void();
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(OpenDatabaseRequest req = waitNext(interf.clientInterface.openDatabase.getFuture())) {
|
2019-10-04 06:29:11 +08:00
|
|
|
++self.openDatabaseRequests;
|
2019-07-26 08:15:31 +08:00
|
|
|
self.addActor.send(clusterOpenDatabase(&self.db, req));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(RecruitStorageRequest req = waitNext(interf.recruitStorage.getFuture())) {
|
|
|
|
clusterRecruitStorage(&self, req);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-09-21 03:42:20 +08:00
|
|
|
when(RecruitBlobWorkerRequest req = waitNext(interf.recruitBlobWorker.getFuture())) {
|
|
|
|
clusterRecruitBlobWorker(&self, req);
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(RegisterWorkerRequest req = waitNext(interf.registerWorker.getFuture())) {
|
2019-10-04 06:29:11 +08:00
|
|
|
++self.registerWorkerRequests;
|
2022-02-02 14:27:12 +08:00
|
|
|
registerWorker(
|
|
|
|
req, &self, coordinators, (configDBType == ConfigDBType::DISABLED) ? nullptr : &configBroadcaster);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(GetWorkersRequest req = waitNext(interf.getWorkers.getFuture())) {
|
2019-10-04 06:29:11 +08:00
|
|
|
++self.getWorkersRequests;
|
2021-09-17 08:42:34 +08:00
|
|
|
std::vector<WorkerDetails> workers;
|
2017-10-25 03:58:54 +08:00
|
|
|
|
2021-08-15 07:51:39 +08:00
|
|
|
for (auto const& [id, worker] : self.id_worker) {
|
2021-03-11 02:06:03 +08:00
|
|
|
if ((req.flags & GetWorkersRequest::NON_EXCLUDED_PROCESSES_ONLY) &&
|
2021-08-15 07:51:39 +08:00
|
|
|
self.db.config.isExcludedServer(worker.details.interf.addresses())) {
|
2017-10-25 03:58:54 +08:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
if ((req.flags & GetWorkersRequest::TESTER_CLASS_ONLY) &&
|
2021-08-15 07:51:39 +08:00
|
|
|
worker.details.processClass.classType() != ProcessClass::TesterClass) {
|
2017-10-25 03:58:54 +08:00
|
|
|
continue;
|
|
|
|
}
|
|
|
|
|
2021-08-15 07:51:39 +08:00
|
|
|
workers.push_back(worker.details);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2017-10-25 03:58:54 +08:00
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
req.reply.send(workers);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(GetClientWorkersRequest req = waitNext(interf.clientInterface.getClientWorkers.getFuture())) {
|
2019-10-04 06:29:11 +08:00
|
|
|
++self.getClientWorkersRequests;
|
2021-09-17 08:42:34 +08:00
|
|
|
std::vector<ClientWorkerInterface> workers;
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto& it : self.id_worker) {
|
2019-03-09 00:25:07 +08:00
|
|
|
if (it.second.details.processClass.classType() != ProcessClass::TesterClass) {
|
|
|
|
workers.push_back(it.second.details.interf.clientInterface);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
req.reply.send(workers);
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(coordinationPingDelay)) {
|
2017-05-26 04:48:44 +08:00
|
|
|
CoordinationPingMessage message(self.id, step++);
|
2021-03-11 02:06:03 +08:00
|
|
|
for (auto& it : self.id_worker)
|
2019-03-09 00:25:07 +08:00
|
|
|
it.second.details.interf.coordinationPing.send(message);
|
2021-03-11 02:06:03 +08:00
|
|
|
coordinationPingDelay = delay(SERVER_KNOBS->WORKER_COORDINATION_PING_DELAY);
|
2017-05-26 04:48:44 +08:00
|
|
|
TraceEvent("CoordinationPingSent", self.id).detail("TimeStep", message.timeStep);
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(RegisterMasterRequest req = waitNext(interf.registerMaster.getFuture())) {
|
2019-10-04 06:29:11 +08:00
|
|
|
++self.registerMasterRequests;
|
2021-03-11 02:06:03 +08:00
|
|
|
clusterRegisterMaster(&self, req);
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-06-24 05:59:00 +08:00
|
|
|
when(UpdateWorkerHealthRequest req = waitNext(interf.updateWorkerHealth.getFuture())) {
|
2021-06-29 12:45:02 +08:00
|
|
|
if (SERVER_KNOBS->CC_ENABLE_WORKER_HEALTH_MONITOR) {
|
2021-06-24 05:59:00 +08:00
|
|
|
self.updateWorkerHealth(req);
|
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(GetServerDBInfoRequest req = waitNext(interf.getServerDBInfo.getFuture())) {
|
|
|
|
self.addActor.send(clusterGetServerInfo(&self.db, req.knownServerInfoID, req.reply));
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
when(ReplyPromise<Void> ping = waitNext(interf.clientInterface.ping.getFuture())) { ping.send(Void()); }
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> replaceInterface(ClusterControllerFullInterface interf) {
|
2019-05-30 07:57:13 +08:00
|
|
|
loop {
|
2021-03-11 02:06:03 +08:00
|
|
|
if (interf.hasMessage()) {
|
2019-05-30 07:57:13 +08:00
|
|
|
wait(delay(SERVER_KNOBS->REPLACE_INTERFACE_DELAY));
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
wait(delay(SERVER_KNOBS->REPLACE_INTERFACE_CHECK_DELAY));
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-03-11 02:06:03 +08:00
|
|
|
ACTOR Future<Void> clusterController(ServerCoordinators coordinators,
|
|
|
|
Reference<AsyncVar<Optional<ClusterControllerFullInterface>>> currentCC,
|
|
|
|
bool hasConnected,
|
|
|
|
Reference<AsyncVar<ClusterControllerPriorityInfo>> asyncPriorityInfo,
|
2021-05-19 01:47:16 +08:00
|
|
|
LocalityData locality,
|
2021-08-07 14:18:10 +08:00
|
|
|
ConfigDBType configDBType) {
|
2017-05-26 04:48:44 +08:00
|
|
|
loop {
|
|
|
|
state ClusterControllerFullInterface cci;
|
|
|
|
state bool inRole = false;
|
|
|
|
cci.initEndpoints();
|
|
|
|
try {
|
2021-03-11 02:06:03 +08:00
|
|
|
// Register as a possible leader; wait to be elected
|
|
|
|
state Future<Void> leaderFail =
|
|
|
|
tryBecomeLeader(coordinators, cci, currentCC, hasConnected, asyncPriorityInfo);
|
|
|
|
state Future<Void> shouldReplace = replaceInterface(cci);
|
2017-05-26 04:48:44 +08:00
|
|
|
|
|
|
|
while (!currentCC->get().present() || currentCC->get().get() != cci) {
|
|
|
|
choose {
|
2021-03-11 02:06:03 +08:00
|
|
|
when(wait(currentCC->onChange())) {}
|
|
|
|
when(wait(leaderFail)) {
|
|
|
|
ASSERT(false);
|
|
|
|
throw internal_error();
|
|
|
|
}
|
|
|
|
when(wait(shouldReplace)) { break; }
|
2017-05-26 04:48:44 +08:00
|
|
|
}
|
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
if (!shouldReplace.isReady()) {
|
2019-05-30 07:57:13 +08:00
|
|
|
shouldReplace = Future<Void>();
|
|
|
|
hasConnected = true;
|
|
|
|
startRole(Role::CLUSTER_CONTROLLER, cci.id(), UID());
|
|
|
|
inRole = true;
|
2017-05-26 04:48:44 +08:00
|
|
|
|
2021-08-07 14:18:10 +08:00
|
|
|
wait(clusterControllerCore(cci, leaderFail, coordinators, locality, configDBType));
|
2019-05-30 07:57:13 +08:00
|
|
|
}
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
2017-05-26 04:48:44 +08:00
|
|
|
if (inRole)
|
2021-03-11 02:06:03 +08:00
|
|
|
endRole(Role::CLUSTER_CONTROLLER,
|
|
|
|
cci.id(),
|
|
|
|
"Error",
|
|
|
|
e.code() == error_code_actor_cancelled || e.code() == error_code_coordinators_changed,
|
|
|
|
e);
|
2017-05-26 04:48:44 +08:00
|
|
|
else
|
2021-03-11 02:06:03 +08:00
|
|
|
TraceEvent(e.code() == error_code_coordinators_changed ? SevInfo : SevError,
|
|
|
|
"ClusterControllerCandidateError",
|
|
|
|
cci.id())
|
|
|
|
.error(e);
|
2017-05-26 04:48:44 +08:00
|
|
|
throw;
|
|
|
|
}
|
|
|
|
}
|
|
|
|
}
|
|
|
|
|
2021-10-11 11:44:56 +08:00
|
|
|
ACTOR Future<Void> clusterController(Reference<IClusterConnectionRecord> connRecord,
|
2021-03-11 02:06:03 +08:00
|
|
|
Reference<AsyncVar<Optional<ClusterControllerFullInterface>>> currentCC,
|
|
|
|
Reference<AsyncVar<ClusterControllerPriorityInfo>> asyncPriorityInfo,
|
|
|
|
Future<Void> recoveredDiskFiles,
|
2021-05-19 01:47:16 +08:00
|
|
|
LocalityData locality,
|
2021-08-07 14:18:10 +08:00
|
|
|
ConfigDBType configDBType) {
|
2018-08-11 04:57:10 +08:00
|
|
|
wait(recoveredDiskFiles);
|
2017-05-26 04:48:44 +08:00
|
|
|
state bool hasConnected = false;
|
|
|
|
loop {
|
|
|
|
try {
|
2022-02-25 08:34:30 +08:00
|
|
|
wait(connRecord->resolveHostnames());
|
2021-10-11 11:44:56 +08:00
|
|
|
ServerCoordinators coordinators(connRecord);
|
2021-08-07 14:18:10 +08:00
|
|
|
wait(clusterController(coordinators, currentCC, hasConnected, asyncPriorityInfo, locality, configDBType));
|
2021-03-11 02:06:03 +08:00
|
|
|
} catch (Error& e) {
|
|
|
|
if (e.code() != error_code_coordinators_changed)
|
2017-05-26 04:48:44 +08:00
|
|
|
throw; // Expected to terminate fdbserver
|
|
|
|
}
|
|
|
|
|
|
|
|
hasConnected = true;
|
|
|
|
}
|
2018-05-09 08:17:17 +08:00
|
|
|
}
|
2021-06-24 05:59:00 +08:00
|
|
|
|
|
|
|
namespace {
|
|
|
|
|
|
|
|
// Tests `ClusterControllerData::updateWorkerHealth()` can update `ClusterControllerData::workerHealth` based on
|
|
|
|
// `UpdateWorkerHealth` request correctly.
|
|
|
|
TEST_CASE("/fdbserver/clustercontroller/updateWorkerHealth") {
|
2021-08-28 08:07:47 +08:00
|
|
|
// Create a testing ClusterControllerData. Most of the internal states do not matter in this test.
|
|
|
|
state ClusterControllerData data(ClusterControllerFullInterface(),
|
|
|
|
LocalityData(),
|
2021-10-11 11:44:56 +08:00
|
|
|
ServerCoordinators(Reference<IClusterConnectionRecord>(
|
|
|
|
new ClusterConnectionMemoryRecord(ClusterConnectionString()))));
|
2021-08-28 08:07:47 +08:00
|
|
|
state NetworkAddress workerAddress(IPAddress(0x01010101), 1);
|
|
|
|
state NetworkAddress badPeer1(IPAddress(0x02020202), 1);
|
|
|
|
state NetworkAddress badPeer2(IPAddress(0x03030303), 1);
|
|
|
|
state NetworkAddress badPeer3(IPAddress(0x04040404), 1);
|
|
|
|
|
|
|
|
// Create a `UpdateWorkerHealthRequest` with two bad peers, and they should appear in the `workerAddress`'s
|
|
|
|
// degradedPeers.
|
|
|
|
{
|
|
|
|
UpdateWorkerHealthRequest req;
|
|
|
|
req.address = workerAddress;
|
|
|
|
req.degradedPeers.push_back(badPeer1);
|
|
|
|
req.degradedPeers.push_back(badPeer2);
|
|
|
|
data.updateWorkerHealth(req);
|
|
|
|
ASSERT(data.workerHealth.find(workerAddress) != data.workerHealth.end());
|
|
|
|
auto& health = data.workerHealth[workerAddress];
|
|
|
|
ASSERT_EQ(health.degradedPeers.size(), 2);
|
|
|
|
ASSERT(health.degradedPeers.find(badPeer1) != health.degradedPeers.end());
|
|
|
|
ASSERT_EQ(health.degradedPeers[badPeer1].startTime, health.degradedPeers[badPeer1].lastRefreshTime);
|
|
|
|
ASSERT(health.degradedPeers.find(badPeer2) != health.degradedPeers.end());
|
|
|
|
}
|
|
|
|
|
|
|
|
// Create a `UpdateWorkerHealthRequest` with two bad peers, one from the previous test and a new one.
|
|
|
|
// The one from the previous test should have lastRefreshTime updated.
|
|
|
|
// The other one from the previous test not included in this test should be removed.
|
|
|
|
{
|
|
|
|
// Make the time to move so that now() guarantees to return a larger value than before.
|
|
|
|
wait(delay(0.001));
|
|
|
|
UpdateWorkerHealthRequest req;
|
|
|
|
req.address = workerAddress;
|
|
|
|
req.degradedPeers.push_back(badPeer1);
|
|
|
|
req.degradedPeers.push_back(badPeer3);
|
|
|
|
data.updateWorkerHealth(req);
|
|
|
|
ASSERT(data.workerHealth.find(workerAddress) != data.workerHealth.end());
|
|
|
|
auto& health = data.workerHealth[workerAddress];
|
|
|
|
ASSERT_EQ(health.degradedPeers.size(), 2);
|
|
|
|
ASSERT(health.degradedPeers.find(badPeer1) != health.degradedPeers.end());
|
|
|
|
ASSERT_LT(health.degradedPeers[badPeer1].startTime, health.degradedPeers[badPeer1].lastRefreshTime);
|
|
|
|
ASSERT(health.degradedPeers.find(badPeer2) == health.degradedPeers.end());
|
|
|
|
ASSERT(health.degradedPeers.find(badPeer3) != health.degradedPeers.end());
|
|
|
|
}
|
|
|
|
|
|
|
|
// Create a `UpdateWorkerHealthRequest` with empty `degradedPeers`, which should remove the worker from
|
|
|
|
// `workerHealth`.
|
|
|
|
{
|
|
|
|
UpdateWorkerHealthRequest req;
|
|
|
|
req.address = workerAddress;
|
|
|
|
data.updateWorkerHealth(req);
|
|
|
|
ASSERT(data.workerHealth.find(workerAddress) == data.workerHealth.end());
|
|
|
|
}
|
|
|
|
|
|
|
|
return Void();
|
2021-06-24 05:59:00 +08:00
|
|
|
}
|
|
|
|
|
2021-06-29 12:45:02 +08:00
|
|
|
TEST_CASE("/fdbserver/clustercontroller/updateRecoveredWorkers") {
|
|
|
|
// Create a testing ClusterControllerData. Most of the internal states do not matter in this test.
|
|
|
|
ClusterControllerData data(ClusterControllerFullInterface(),
|
|
|
|
LocalityData(),
|
2021-10-11 11:44:56 +08:00
|
|
|
ServerCoordinators(Reference<IClusterConnectionRecord>(
|
|
|
|
new ClusterConnectionMemoryRecord(ClusterConnectionString()))));
|
2021-06-29 12:45:02 +08:00
|
|
|
NetworkAddress worker1(IPAddress(0x01010101), 1);
|
|
|
|
NetworkAddress worker2(IPAddress(0x11111111), 1);
|
|
|
|
NetworkAddress badPeer1(IPAddress(0x02020202), 1);
|
|
|
|
NetworkAddress badPeer2(IPAddress(0x03030303), 1);
|
|
|
|
|
|
|
|
// Create following test scenario:
|
|
|
|
// worker1 -> badPeer1 active
|
|
|
|
// worker1 -> badPeer2 recovered
|
|
|
|
// worker2 -> badPeer2 recovered
|
|
|
|
data.workerHealth[worker1].degradedPeers[badPeer1] = {
|
|
|
|
now() - SERVER_KNOBS->CC_DEGRADED_LINK_EXPIRATION_INTERVAL - 1, now()
|
|
|
|
};
|
|
|
|
data.workerHealth[worker1].degradedPeers[badPeer2] = {
|
|
|
|
now() - SERVER_KNOBS->CC_DEGRADED_LINK_EXPIRATION_INTERVAL - 1,
|
|
|
|
now() - SERVER_KNOBS->CC_DEGRADED_LINK_EXPIRATION_INTERVAL - 1
|
|
|
|
};
|
|
|
|
data.workerHealth[worker2].degradedPeers[badPeer2] = {
|
|
|
|
now() - SERVER_KNOBS->CC_DEGRADED_LINK_EXPIRATION_INTERVAL - 1,
|
|
|
|
now() - SERVER_KNOBS->CC_DEGRADED_LINK_EXPIRATION_INTERVAL - 1
|
|
|
|
};
|
|
|
|
data.updateRecoveredWorkers();
|
|
|
|
|
|
|
|
ASSERT_EQ(data.workerHealth.size(), 1);
|
|
|
|
ASSERT(data.workerHealth.find(worker1) != data.workerHealth.end());
|
|
|
|
ASSERT(data.workerHealth[worker1].degradedPeers.find(badPeer1) != data.workerHealth[worker1].degradedPeers.end());
|
|
|
|
ASSERT(data.workerHealth[worker1].degradedPeers.find(badPeer2) == data.workerHealth[worker1].degradedPeers.end());
|
|
|
|
ASSERT(data.workerHealth.find(worker2) == data.workerHealth.end());
|
|
|
|
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_CASE("/fdbserver/clustercontroller/getServersWithDegradedLink") {
|
|
|
|
// Create a testing ClusterControllerData. Most of the internal states do not matter in this test.
|
|
|
|
ClusterControllerData data(ClusterControllerFullInterface(),
|
|
|
|
LocalityData(),
|
2021-10-11 11:44:56 +08:00
|
|
|
ServerCoordinators(Reference<IClusterConnectionRecord>(
|
|
|
|
new ClusterConnectionMemoryRecord(ClusterConnectionString()))));
|
2021-06-29 12:45:02 +08:00
|
|
|
NetworkAddress worker(IPAddress(0x01010101), 1);
|
|
|
|
NetworkAddress badPeer1(IPAddress(0x02020202), 1);
|
|
|
|
NetworkAddress badPeer2(IPAddress(0x03030303), 1);
|
|
|
|
NetworkAddress badPeer3(IPAddress(0x04040404), 1);
|
|
|
|
NetworkAddress badPeer4(IPAddress(0x05050505), 1);
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
// Test that a reported degraded link should stay for sometime before being considered as a degraded link by
|
|
|
|
// cluster controller.
|
2021-06-29 12:45:02 +08:00
|
|
|
{
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer1] = { now(), now() };
|
|
|
|
ASSERT(data.getServersWithDegradedLink().empty());
|
|
|
|
data.workerHealth.clear();
|
|
|
|
}
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
// Test that when there is only one reported degraded link, getServersWithDegradedLink can return correct
|
|
|
|
// degraded server.
|
2021-06-29 12:45:02 +08:00
|
|
|
{
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer1] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
auto degradedServers = data.getServersWithDegradedLink();
|
|
|
|
ASSERT(degradedServers.size() == 1);
|
|
|
|
ASSERT(degradedServers.find(badPeer1) != degradedServers.end());
|
|
|
|
data.workerHealth.clear();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Test that if both A complains B and B compalins A, only one of the server will be chosen as degraded server.
|
|
|
|
{
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer1] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer1].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
auto degradedServers = data.getServersWithDegradedLink();
|
|
|
|
ASSERT(degradedServers.size() == 1);
|
|
|
|
ASSERT(degradedServers.find(worker) != degradedServers.end() ||
|
|
|
|
degradedServers.find(badPeer1) != degradedServers.end());
|
|
|
|
data.workerHealth.clear();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Test that if B complains A and C complains A, A is selected as degraded server instead of B or C.
|
|
|
|
{
|
|
|
|
ASSERT(SERVER_KNOBS->CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE < 4);
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer1] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer1].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer2] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer2].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
auto degradedServers = data.getServersWithDegradedLink();
|
|
|
|
ASSERT(degradedServers.size() == 1);
|
|
|
|
ASSERT(degradedServers.find(worker) != degradedServers.end());
|
|
|
|
data.workerHealth.clear();
|
|
|
|
}
|
|
|
|
|
|
|
|
// Test that if the number of complainers exceeds the threshold, no degraded server is returned.
|
|
|
|
{
|
|
|
|
ASSERT(SERVER_KNOBS->CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE < 4);
|
|
|
|
data.workerHealth[badPeer1].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer2].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer3].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer4].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
ASSERT(data.getServersWithDegradedLink().empty());
|
|
|
|
data.workerHealth.clear();
|
|
|
|
}
|
|
|
|
|
2021-10-19 03:04:22 +08:00
|
|
|
// Test that if the degradation is reported both ways between A and other 4 servers, no degraded server is
|
|
|
|
// returned.
|
2021-06-29 12:45:02 +08:00
|
|
|
{
|
|
|
|
ASSERT(SERVER_KNOBS->CC_DEGRADED_PEER_DEGREE_TO_EXCLUDE < 4);
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer1] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer1].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer2] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer2].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer3] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer3].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[worker].degradedPeers[badPeer4] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
data.workerHealth[badPeer4].degradedPeers[worker] = { now() - SERVER_KNOBS->CC_MIN_DEGRADATION_INTERVAL - 1,
|
|
|
|
now() };
|
|
|
|
ASSERT(data.getServersWithDegradedLink().empty());
|
|
|
|
data.workerHealth.clear();
|
|
|
|
}
|
|
|
|
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_CASE("/fdbserver/clustercontroller/recentRecoveryCountDueToHealth") {
|
|
|
|
// Create a testing ClusterControllerData. Most of the internal states do not matter in this test.
|
|
|
|
ClusterControllerData data(ClusterControllerFullInterface(),
|
|
|
|
LocalityData(),
|
2021-10-11 11:44:56 +08:00
|
|
|
ServerCoordinators(Reference<IClusterConnectionRecord>(
|
|
|
|
new ClusterConnectionMemoryRecord(ClusterConnectionString()))));
|
2021-06-29 12:45:02 +08:00
|
|
|
|
|
|
|
ASSERT_EQ(data.recentRecoveryCountDueToHealth(), 0);
|
|
|
|
|
|
|
|
data.recentHealthTriggeredRecoveryTime.push(now() - SERVER_KNOBS->CC_TRACKING_HEALTH_RECOVERY_INTERVAL - 1);
|
|
|
|
ASSERT_EQ(data.recentRecoveryCountDueToHealth(), 0);
|
|
|
|
|
|
|
|
data.recentHealthTriggeredRecoveryTime.push(now() - SERVER_KNOBS->CC_TRACKING_HEALTH_RECOVERY_INTERVAL + 1);
|
|
|
|
ASSERT_EQ(data.recentRecoveryCountDueToHealth(), 1);
|
|
|
|
|
|
|
|
data.recentHealthTriggeredRecoveryTime.push(now());
|
|
|
|
ASSERT_EQ(data.recentRecoveryCountDueToHealth(), 2);
|
|
|
|
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
|
|
|
TEST_CASE("/fdbserver/clustercontroller/shouldTriggerRecoveryDueToDegradedServers") {
|
|
|
|
// Create a testing ClusterControllerData. Most of the internal states do not matter in this test.
|
|
|
|
ClusterControllerData data(ClusterControllerFullInterface(),
|
|
|
|
LocalityData(),
|
2021-10-11 11:44:56 +08:00
|
|
|
ServerCoordinators(Reference<IClusterConnectionRecord>(
|
|
|
|
new ClusterConnectionMemoryRecord(ClusterConnectionString()))));
|
2021-06-29 12:45:02 +08:00
|
|
|
NetworkAddress master(IPAddress(0x01010101), 1);
|
|
|
|
NetworkAddress tlog(IPAddress(0x02020202), 1);
|
|
|
|
NetworkAddress satelliteTlog(IPAddress(0x03030303), 1);
|
|
|
|
NetworkAddress remoteTlog(IPAddress(0x04040404), 1);
|
|
|
|
NetworkAddress logRouter(IPAddress(0x05050505), 1);
|
|
|
|
NetworkAddress backup(IPAddress(0x06060606), 1);
|
|
|
|
NetworkAddress proxy(IPAddress(0x07070707), 1);
|
|
|
|
NetworkAddress resolver(IPAddress(0x08080808), 1);
|
2022-01-07 04:15:51 +08:00
|
|
|
NetworkAddress clusterController(IPAddress(0x09090909), 1);
|
2021-09-11 05:51:55 +08:00
|
|
|
UID testUID(1, 2);
|
2021-06-29 12:45:02 +08:00
|
|
|
|
|
|
|
// Create a ServerDBInfo using above addresses.
|
|
|
|
ServerDBInfo testDbInfo;
|
2022-01-07 04:15:51 +08:00
|
|
|
testDbInfo.clusterInterface.changeCoordinators =
|
|
|
|
RequestStream<struct ChangeCoordinatorsRequest>(Endpoint({ clusterController }, UID(1, 2)));
|
|
|
|
|
|
|
|
MasterInterface mInterface;
|
|
|
|
mInterface.getCommitVersion = RequestStream<struct GetCommitVersionRequest>(Endpoint({ master }, UID(1, 2)));
|
|
|
|
testDbInfo.master = mInterface;
|
2021-06-29 12:45:02 +08:00
|
|
|
|
|
|
|
TLogInterface localTLogInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
localTLogInterf.peekMessages = RequestStream<struct TLogPeekRequest>(Endpoint({ tlog }, testUID));
|
2021-06-29 12:45:02 +08:00
|
|
|
TLogInterface localLogRouterInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
localLogRouterInterf.peekMessages = RequestStream<struct TLogPeekRequest>(Endpoint({ logRouter }, testUID));
|
2021-06-29 12:45:02 +08:00
|
|
|
BackupInterface backupInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
backupInterf.waitFailure = RequestStream<ReplyPromise<Void>>(Endpoint({ backup }, testUID));
|
2021-06-29 12:45:02 +08:00
|
|
|
TLogSet localTLogSet;
|
|
|
|
localTLogSet.isLocal = true;
|
|
|
|
localTLogSet.tLogs.push_back(OptionalInterface(localTLogInterf));
|
|
|
|
localTLogSet.logRouters.push_back(OptionalInterface(localLogRouterInterf));
|
|
|
|
localTLogSet.backupWorkers.push_back(OptionalInterface(backupInterf));
|
|
|
|
testDbInfo.logSystemConfig.tLogs.push_back(localTLogSet);
|
|
|
|
|
|
|
|
TLogInterface sateTLogInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
sateTLogInterf.peekMessages = RequestStream<struct TLogPeekRequest>(Endpoint({ satelliteTlog }, testUID));
|
2021-06-29 12:45:02 +08:00
|
|
|
TLogSet sateTLogSet;
|
|
|
|
sateTLogSet.isLocal = true;
|
|
|
|
sateTLogSet.locality = tagLocalitySatellite;
|
|
|
|
sateTLogSet.tLogs.push_back(OptionalInterface(sateTLogInterf));
|
|
|
|
testDbInfo.logSystemConfig.tLogs.push_back(sateTLogSet);
|
|
|
|
|
|
|
|
TLogInterface remoteTLogInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
remoteTLogInterf.peekMessages = RequestStream<struct TLogPeekRequest>(Endpoint({ remoteTlog }, testUID));
|
2021-06-29 12:45:02 +08:00
|
|
|
TLogSet remoteTLogSet;
|
|
|
|
remoteTLogSet.isLocal = false;
|
|
|
|
remoteTLogSet.tLogs.push_back(OptionalInterface(remoteTLogInterf));
|
|
|
|
testDbInfo.logSystemConfig.tLogs.push_back(remoteTLogSet);
|
|
|
|
|
|
|
|
GrvProxyInterface proxyInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
proxyInterf.getConsistentReadVersion = RequestStream<struct GetReadVersionRequest>(Endpoint({ proxy }, testUID));
|
2021-06-29 12:45:02 +08:00
|
|
|
testDbInfo.client.grvProxies.push_back(proxyInterf);
|
|
|
|
|
|
|
|
ResolverInterface resolverInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
resolverInterf.resolve = RequestStream<struct ResolveTransactionBatchRequest>(Endpoint({ resolver }, testUID));
|
2021-06-29 12:45:02 +08:00
|
|
|
testDbInfo.resolvers.push_back(resolverInterf);
|
|
|
|
|
|
|
|
testDbInfo.recoveryState = RecoveryState::ACCEPTING_COMMITS;
|
|
|
|
|
|
|
|
// No recovery when no degraded servers.
|
|
|
|
data.db.serverInfo->set(testDbInfo);
|
|
|
|
ASSERT(!data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
|
|
|
|
// Trigger recovery when master is degraded.
|
|
|
|
data.degradedServers.insert(master);
|
|
|
|
ASSERT(data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
// Trigger recovery when primary TLog is degraded.
|
|
|
|
data.degradedServers.insert(tlog);
|
|
|
|
ASSERT(data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
// No recovery when satellite Tlog is degraded.
|
|
|
|
data.degradedServers.insert(satelliteTlog);
|
|
|
|
ASSERT(!data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
// No recovery when remote tlog is degraded.
|
|
|
|
data.degradedServers.insert(remoteTlog);
|
|
|
|
ASSERT(!data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
// No recovery when log router is degraded.
|
|
|
|
data.degradedServers.insert(logRouter);
|
|
|
|
ASSERT(!data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
// No recovery when backup worker is degraded.
|
|
|
|
data.degradedServers.insert(backup);
|
|
|
|
ASSERT(!data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
// Trigger recovery when proxy is degraded.
|
|
|
|
data.degradedServers.insert(proxy);
|
|
|
|
ASSERT(data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
// Trigger recovery when resolver is degraded.
|
|
|
|
data.degradedServers.insert(resolver);
|
|
|
|
ASSERT(data.shouldTriggerRecoveryDueToDegradedServers());
|
|
|
|
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2021-09-10 11:29:28 +08:00
|
|
|
TEST_CASE("/fdbserver/clustercontroller/shouldTriggerFailoverDueToDegradedServers") {
|
|
|
|
// Create a testing ClusterControllerData. Most of the internal states do not matter in this test.
|
|
|
|
ClusterControllerData data(ClusterControllerFullInterface(),
|
|
|
|
LocalityData(),
|
2021-10-11 11:44:56 +08:00
|
|
|
ServerCoordinators(Reference<IClusterConnectionRecord>(
|
|
|
|
new ClusterConnectionMemoryRecord(ClusterConnectionString()))));
|
2021-09-10 11:29:28 +08:00
|
|
|
NetworkAddress master(IPAddress(0x01010101), 1);
|
|
|
|
NetworkAddress tlog(IPAddress(0x02020202), 1);
|
|
|
|
NetworkAddress satelliteTlog(IPAddress(0x03030303), 1);
|
|
|
|
NetworkAddress remoteTlog(IPAddress(0x04040404), 1);
|
|
|
|
NetworkAddress logRouter(IPAddress(0x05050505), 1);
|
|
|
|
NetworkAddress backup(IPAddress(0x06060606), 1);
|
|
|
|
NetworkAddress proxy(IPAddress(0x07070707), 1);
|
|
|
|
NetworkAddress proxy2(IPAddress(0x08080808), 1);
|
|
|
|
NetworkAddress resolver(IPAddress(0x09090909), 1);
|
2022-01-07 04:15:51 +08:00
|
|
|
NetworkAddress clusterController(IPAddress(0x10101010), 1);
|
2021-09-11 05:51:55 +08:00
|
|
|
UID testUID(1, 2);
|
2021-09-10 11:29:28 +08:00
|
|
|
|
|
|
|
data.db.config.usableRegions = 2;
|
|
|
|
|
|
|
|
// Create a ServerDBInfo using above addresses.
|
|
|
|
ServerDBInfo testDbInfo;
|
2022-01-07 04:15:51 +08:00
|
|
|
testDbInfo.clusterInterface.changeCoordinators =
|
|
|
|
RequestStream<struct ChangeCoordinatorsRequest>(Endpoint({ clusterController }, UID(1, 2)));
|
2021-09-10 11:29:28 +08:00
|
|
|
|
|
|
|
TLogInterface localTLogInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
localTLogInterf.peekMessages = RequestStream<struct TLogPeekRequest>(Endpoint({ tlog }, testUID));
|
2021-09-10 11:29:28 +08:00
|
|
|
TLogInterface localLogRouterInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
localLogRouterInterf.peekMessages = RequestStream<struct TLogPeekRequest>(Endpoint({ logRouter }, testUID));
|
2021-09-10 11:29:28 +08:00
|
|
|
BackupInterface backupInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
backupInterf.waitFailure = RequestStream<ReplyPromise<Void>>(Endpoint({ backup }, testUID));
|
2021-09-10 11:29:28 +08:00
|
|
|
TLogSet localTLogSet;
|
|
|
|
localTLogSet.isLocal = true;
|
|
|
|
localTLogSet.tLogs.push_back(OptionalInterface(localTLogInterf));
|
|
|
|
localTLogSet.logRouters.push_back(OptionalInterface(localLogRouterInterf));
|
|
|
|
localTLogSet.backupWorkers.push_back(OptionalInterface(backupInterf));
|
|
|
|
testDbInfo.logSystemConfig.tLogs.push_back(localTLogSet);
|
|
|
|
|
|
|
|
TLogInterface sateTLogInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
sateTLogInterf.peekMessages = RequestStream<struct TLogPeekRequest>(Endpoint({ satelliteTlog }, testUID));
|
2021-09-10 11:29:28 +08:00
|
|
|
TLogSet sateTLogSet;
|
|
|
|
sateTLogSet.isLocal = true;
|
|
|
|
sateTLogSet.locality = tagLocalitySatellite;
|
|
|
|
sateTLogSet.tLogs.push_back(OptionalInterface(sateTLogInterf));
|
|
|
|
testDbInfo.logSystemConfig.tLogs.push_back(sateTLogSet);
|
|
|
|
|
|
|
|
TLogInterface remoteTLogInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
remoteTLogInterf.peekMessages = RequestStream<struct TLogPeekRequest>(Endpoint({ remoteTlog }, testUID));
|
2021-09-10 11:29:28 +08:00
|
|
|
TLogSet remoteTLogSet;
|
|
|
|
remoteTLogSet.isLocal = false;
|
|
|
|
remoteTLogSet.tLogs.push_back(OptionalInterface(remoteTLogInterf));
|
|
|
|
testDbInfo.logSystemConfig.tLogs.push_back(remoteTLogSet);
|
|
|
|
|
|
|
|
GrvProxyInterface grvProxyInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
grvProxyInterf.getConsistentReadVersion = RequestStream<struct GetReadVersionRequest>(Endpoint({ proxy }, testUID));
|
2021-09-10 11:29:28 +08:00
|
|
|
testDbInfo.client.grvProxies.push_back(grvProxyInterf);
|
|
|
|
|
|
|
|
CommitProxyInterface commitProxyInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
commitProxyInterf.commit = RequestStream<struct CommitTransactionRequest>(Endpoint({ proxy2 }, testUID));
|
2021-09-10 11:29:28 +08:00
|
|
|
testDbInfo.client.commitProxies.push_back(commitProxyInterf);
|
|
|
|
|
|
|
|
ResolverInterface resolverInterf;
|
2021-09-11 05:51:55 +08:00
|
|
|
resolverInterf.resolve = RequestStream<struct ResolveTransactionBatchRequest>(Endpoint({ resolver }, testUID));
|
2021-09-10 11:29:28 +08:00
|
|
|
testDbInfo.resolvers.push_back(resolverInterf);
|
|
|
|
|
|
|
|
testDbInfo.recoveryState = RecoveryState::ACCEPTING_COMMITS;
|
|
|
|
|
|
|
|
// No failover when no degraded servers.
|
|
|
|
data.db.serverInfo->set(testDbInfo);
|
|
|
|
ASSERT(!data.shouldTriggerFailoverDueToDegradedServers());
|
|
|
|
|
|
|
|
// No failover when small number of degraded servers
|
|
|
|
data.degradedServers.insert(master);
|
|
|
|
ASSERT(!data.shouldTriggerFailoverDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
2021-09-11 05:51:55 +08:00
|
|
|
// Trigger failover when enough servers in the txn system are degraded.
|
2021-09-10 11:29:28 +08:00
|
|
|
data.degradedServers.insert(master);
|
|
|
|
data.degradedServers.insert(tlog);
|
|
|
|
data.degradedServers.insert(proxy);
|
|
|
|
data.degradedServers.insert(proxy2);
|
|
|
|
data.degradedServers.insert(resolver);
|
|
|
|
ASSERT(data.shouldTriggerFailoverDueToDegradedServers());
|
|
|
|
|
|
|
|
// No failover when usable region is 1.
|
|
|
|
data.db.config.usableRegions = 1;
|
|
|
|
ASSERT(!data.shouldTriggerFailoverDueToDegradedServers());
|
|
|
|
data.db.config.usableRegions = 2;
|
|
|
|
|
|
|
|
// No failover when remote is also degraded.
|
|
|
|
data.degradedServers.insert(remoteTlog);
|
|
|
|
ASSERT(!data.shouldTriggerFailoverDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
// No failover when some are not from transaction system
|
|
|
|
data.degradedServers.insert(NetworkAddress(IPAddress(0x13131313), 1));
|
|
|
|
data.degradedServers.insert(NetworkAddress(IPAddress(0x13131313), 2));
|
|
|
|
data.degradedServers.insert(NetworkAddress(IPAddress(0x13131313), 3));
|
|
|
|
data.degradedServers.insert(NetworkAddress(IPAddress(0x13131313), 4));
|
|
|
|
data.degradedServers.insert(NetworkAddress(IPAddress(0x13131313), 5));
|
|
|
|
ASSERT(!data.shouldTriggerFailoverDueToDegradedServers());
|
|
|
|
data.degradedServers.clear();
|
|
|
|
|
|
|
|
return Void();
|
|
|
|
}
|
|
|
|
|
2022-02-18 06:50:35 +08:00
|
|
|
} // namespace
|