void runSyncThread() { Client::initThread("rsSync"); AuthorizationSession::get(cc())->grantInternalAuthorization(); ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator(); // Set initial indexPrefetch setting const std::string& prefetch = replCoord->getSettings().rsIndexPrefetch; if (!prefetch.empty()) { BackgroundSync::IndexPrefetchConfig prefetchConfig = BackgroundSync::PREFETCH_ALL; if (prefetch == "none") prefetchConfig = BackgroundSync::PREFETCH_NONE; else if (prefetch == "_id_only") prefetchConfig = BackgroundSync::PREFETCH_ID_ONLY; else if (prefetch == "all") prefetchConfig = BackgroundSync::PREFETCH_ALL; else { warning() << "unrecognized indexPrefetch setting " << prefetch << ", defaulting " << "to \"all\""; } BackgroundSync::get()->setIndexPrefetchConfig(prefetchConfig); } while (!inShutdown()) { // After a reconfig, we may not be in the replica set anymore, so // check that we are in the set (and not an arbiter) before // trying to sync with other replicas. // TODO(spencer): Use a condition variable to await loading a config if (replCoord->getMemberState().startup()) { warning() << "did not receive a valid config yet"; sleepsecs(1); continue; } const MemberState memberState = replCoord->getMemberState(); // An arbiter can never transition to any other state, and doesn't replicate, ever if (memberState.arbiter()) { break; } // If we are removed then we don't belong to the set anymore if (memberState.removed()) { sleepsecs(5); continue; } try { if (memberState.primary() && !replCoord->isWaitingForApplierToDrain()) { sleepsecs(1); continue; } bool initialSyncRequested = BackgroundSync::get()->getInitialSyncRequestedFlag(); // Check criteria for doing an initial sync: // 1. If the oplog is empty, do an initial sync // 2. If minValid has _initialSyncFlag set, do an initial sync // 3. If initialSyncRequested is true if (getGlobalReplicationCoordinator()->getMyLastOptime().isNull() || getInitialSyncFlag() || initialSyncRequested) { syncDoInitialSync(); continue; // start from top again in case sync failed. } if (!replCoord->setFollowerMode(MemberState::RS_RECOVERING)) { continue; } /* we have some data. continue tailing. */ SyncTail tail(BackgroundSync::get(), multiSyncApply); tail.oplogApplication(); } catch (const DBException& e) { log() << "Received exception while syncing: " << e.toString(); sleepsecs(10); } catch (const std::exception& e) { log() << "Received exception while syncing: " << e.what(); sleepsecs(10); } } }
void runSyncThread() { Client::initThread("rsSync"); AuthorizationSession::get(cc())->grantInternalAuthorization(); ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator(); // Overwrite prefetch index mode in BackgroundSync if ReplSettings has a mode set. ReplSettings replSettings = replCoord->getSettings(); if (replSettings.isPrefetchIndexModeSet()) BackgroundSync::get()->setIndexPrefetchConfig(replSettings.getPrefetchIndexMode()); while (!inShutdown()) { // After a reconfig, we may not be in the replica set anymore, so // check that we are in the set (and not an arbiter) before // trying to sync with other replicas. // TODO(spencer): Use a condition variable to await loading a config if (replCoord->getMemberState().startup()) { warning() << "did not receive a valid config yet"; sleepsecs(1); continue; } const MemberState memberState = replCoord->getMemberState(); // An arbiter can never transition to any other state, and doesn't replicate, ever if (memberState.arbiter()) { break; } // If we are removed then we don't belong to the set anymore if (memberState.removed()) { sleepsecs(5); continue; } try { if (memberState.primary() && !replCoord->isWaitingForApplierToDrain()) { sleepsecs(1); continue; } bool initialSyncRequested = BackgroundSync::get()->getInitialSyncRequestedFlag(); // Check criteria for doing an initial sync: // 1. If the oplog is empty, do an initial sync // 2. If minValid has _initialSyncFlag set, do an initial sync // 3. If initialSyncRequested is true if (getGlobalReplicationCoordinator()->getMyLastOptime().isNull() || getInitialSyncFlag() || initialSyncRequested) { syncDoInitialSync(); continue; // start from top again in case sync failed. } if (!replCoord->setFollowerMode(MemberState::RS_RECOVERING)) { continue; } /* we have some data. continue tailing. */ SyncTail tail(BackgroundSync::get(), multiSyncApply); tail.oplogApplication(); } catch (...) { std::terminate(); } } }
void runSyncThread() { Client::initThread("rsSync"); replLocalAuth(); ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator(); // Set initial indexPrefetch setting std::string& prefetch = replCoord->getSettings().rsIndexPrefetch; if (!prefetch.empty()) { BackgroundSync::IndexPrefetchConfig prefetchConfig = BackgroundSync::PREFETCH_ALL; if (prefetch == "none") prefetchConfig = BackgroundSync::PREFETCH_NONE; else if (prefetch == "_id_only") prefetchConfig = BackgroundSync::PREFETCH_ID_ONLY; else if (prefetch == "all") prefetchConfig = BackgroundSync::PREFETCH_ALL; else { warning() << "unrecognized indexPrefetch setting " << prefetch << ", defaulting " << "to \"all\""; } BackgroundSync::get()->setIndexPrefetchConfig(prefetchConfig); } while (!inShutdown()) { // After a reconfig, we may not be in the replica set anymore, so // check that we are in the set (and not an arbiter) before // trying to sync with other replicas. // TODO(spencer): Use a condition variable to await loading a config if (replCoord->getReplicationMode() != ReplicationCoordinator::modeReplSet) { log() << "replSet warning did not receive a valid config yet, sleeping 5 seconds " << rsLog; sleepsecs(5); continue; } const MemberState memberState = replCoord->getCurrentMemberState(); if (replCoord->getCurrentMemberState().arbiter()) { break; } try { if (memberState.primary()) { sleepsecs(1); continue; } bool initialSyncRequested = BackgroundSync::get()->getInitialSyncRequestedFlag(); // Check criteria for doing an initial sync: // 1. If the oplog is empty, do an initial sync // 2. If minValid has _initialSyncFlag set, do an initial sync // 3. If initialSyncRequested is true if (getGlobalReplicationCoordinator()->getMyLastOptime().isNull() || getInitialSyncFlag() || initialSyncRequested) { syncDoInitialSync(); continue; // start from top again in case sync failed. } replCoord->setFollowerMode(MemberState::RS_RECOVERING); /* we have some data. continue tailing. */ SyncTail tail(BackgroundSync::get(), multiSyncApply); tail.oplogApplication(); } catch(const DBException& e) { log() << "Received exception while syncing: " << e.toString(); sleepsecs(10); } catch(...) { sethbmsg("unexpected exception in syncThread()"); // TODO : SET NOT SECONDARY here? sleepsecs(60); } } cc().shutdown(); }
void appendReplicationInfo(OperationContext* txn, BSONObjBuilder& result, int level) { ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator(); if (replCoord->getSettings().usingReplSets()) { IsMasterResponse isMasterResponse; replCoord->fillIsMasterForReplSet(&isMasterResponse); result.appendElements(isMasterResponse.toBSON()); if (level) { replCoord->appendSlaveInfoData(&result); } return; } // TODO(dannenberg) replAllDead is bad and should be removed when master slave is removed if (replAllDead) { result.append("ismaster", 0); string s = string("dead: ") + replAllDead; result.append("info", s); } else { result.appendBool("ismaster", getGlobalReplicationCoordinator()->isMasterForReportingPurposes()); } if (level) { BSONObjBuilder sources(result.subarrayStart("sources")); int n = 0; list<BSONObj> src; { const char* localSources = "local.sources"; AutoGetCollectionForRead ctx(txn, localSources); unique_ptr<PlanExecutor> exec( InternalPlanner::collectionScan(txn, localSources, ctx.getCollection())); BSONObj obj; PlanExecutor::ExecState state; while (PlanExecutor::ADVANCED == (state = exec->getNext(&obj, NULL))) { src.push_back(obj); } } for (list<BSONObj>::const_iterator i = src.begin(); i != src.end(); i++) { BSONObj s = *i; BSONObjBuilder bb; bb.append(s["host"]); string sourcename = s["source"].valuestr(); if (sourcename != "main") bb.append(s["source"]); { BSONElement e = s["syncedTo"]; BSONObjBuilder t(bb.subobjStart("syncedTo")); t.appendDate("time", e.timestampTime()); t.append("inc", e.timestampInc()); t.done(); } if (level > 1) { wassert(!txn->lockState()->isLocked()); // note: there is no so-style timeout on this connection; perhaps we should have // one. ScopedDbConnection conn(s["host"].valuestr()); DBClientConnection* cliConn = dynamic_cast<DBClientConnection*>(&conn.conn()); if (cliConn && replAuthenticate(cliConn)) { BSONObj first = conn->findOne((string) "local.oplog.$" + sourcename, Query().sort(BSON("$natural" << 1))); BSONObj last = conn->findOne((string) "local.oplog.$" + sourcename, Query().sort(BSON("$natural" << -1))); bb.appendDate("masterFirst", first["ts"].timestampTime()); bb.appendDate("masterLast", last["ts"].timestampTime()); const auto lag = (last["ts"].timestampTime() - s["syncedTo"].timestampTime()); bb.append("lagSeconds", durationCount<Milliseconds>(lag) / 1000.0); } conn.done(); } sources.append(BSONObjBuilder::numStr(n++), bb.obj()); } sources.done(); replCoord->appendSlaveInfoData(&result); } }
void appendReplicationInfo(OperationContext* opCtx, BSONObjBuilder& result, int level) { ReplicationCoordinator* replCoord = ReplicationCoordinator::get(opCtx); if (replCoord->getSettings().usingReplSets()) { IsMasterResponse isMasterResponse; replCoord->fillIsMasterForReplSet(&isMasterResponse); result.appendElements(isMasterResponse.toBSON()); if (level) { replCoord->appendSlaveInfoData(&result); } return; } result.appendBool("ismaster", ReplicationCoordinator::get(opCtx)->isMasterForReportingPurposes()); if (level) { BSONObjBuilder sources(result.subarrayStart("sources")); int n = 0; list<BSONObj> src; { const NamespaceString localSources{"local.sources"}; AutoGetCollectionForReadCommand ctx(opCtx, localSources); auto exec = InternalPlanner::collectionScan( opCtx, localSources.ns(), ctx.getCollection(), PlanExecutor::NO_YIELD); BSONObj obj; PlanExecutor::ExecState state; while (PlanExecutor::ADVANCED == (state = exec->getNext(&obj, NULL))) { src.push_back(obj.getOwned()); } // Non-yielding collection scans from InternalPlanner will never error. invariant(PlanExecutor::IS_EOF == state); } for (list<BSONObj>::const_iterator i = src.begin(); i != src.end(); i++) { BSONObj s = *i; BSONObjBuilder bb; bb.append(s["host"]); string sourcename = s["source"].valuestr(); if (sourcename != "main") bb.append(s["source"]); { BSONElement e = s["syncedTo"]; BSONObjBuilder t(bb.subobjStart("syncedTo")); t.appendDate("time", e.timestampTime()); t.append("inc", e.timestampInc()); t.done(); } if (level > 1) { invariant(!opCtx->lockState()->isLocked()); // note: there is no so-style timeout on this connection; perhaps we should have // one. ScopedDbConnection conn(s["host"].valuestr()); DBClientConnection* cliConn = dynamic_cast<DBClientConnection*>(&conn.conn()); if (cliConn && replAuthenticate(cliConn)) { BSONObj first = conn->findOne((string) "local.oplog.$" + sourcename, Query().sort(BSON("$natural" << 1))); BSONObj last = conn->findOne((string) "local.oplog.$" + sourcename, Query().sort(BSON("$natural" << -1))); bb.appendDate("masterFirst", first["ts"].timestampTime()); bb.appendDate("masterLast", last["ts"].timestampTime()); const auto lag = (last["ts"].timestampTime() - s["syncedTo"].timestampTime()); bb.append("lagSeconds", durationCount<Milliseconds>(lag) / 1000.0); } conn.done(); } sources.append(BSONObjBuilder::numStr(n++), bb.obj()); } sources.done(); replCoord->appendSlaveInfoData(&result); } }
void appendReplicationInfo(OperationContext* txn, BSONObjBuilder& result, int level) { ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator(); if (replCoord->getSettings().usingReplSets()) { if (replCoord->getReplicationMode() != ReplicationCoordinator::modeReplSet || replCoord->getCurrentMemberState().shunned()) { result.append("ismaster", false); result.append("secondary", false); result.append("info", ReplSet::startupStatusMsg.get()); result.append( "isreplicaset" , true ); } else { theReplSet->fillIsMaster(result); } return; } if ( replAllDead ) { result.append("ismaster", 0); string s = string("dead: ") + replAllDead; result.append("info", s); } else { result.appendBool("ismaster", getGlobalReplicationCoordinator()->isMasterForReportingPurposes()); } if (level && replCoord->getSettings().usingReplSets()) { result.append( "info" , "is replica set" ); } else if ( level ) { BSONObjBuilder sources( result.subarrayStart( "sources" ) ); int n = 0; list<BSONObj> src; { const char* localSources = "local.sources"; Client::ReadContext ctx(txn, localSources); auto_ptr<PlanExecutor> exec( InternalPlanner::collectionScan(txn, localSources, ctx.ctx().db()->getCollection(txn, localSources))); BSONObj obj; Runner::RunnerState state; while (Runner::RUNNER_ADVANCED == (state = exec->getNext(&obj, NULL))) { src.push_back(obj); } } for( list<BSONObj>::const_iterator i = src.begin(); i != src.end(); i++ ) { BSONObj s = *i; BSONObjBuilder bb; bb.append( s["host"] ); string sourcename = s["source"].valuestr(); if ( sourcename != "main" ) bb.append( s["source"] ); { BSONElement e = s["syncedTo"]; BSONObjBuilder t( bb.subobjStart( "syncedTo" ) ); t.appendDate( "time" , e.timestampTime() ); t.append( "inc" , e.timestampInc() ); t.done(); } if ( level > 1 ) { wassert(txn->lockState()->threadState() == 0); // note: there is no so-style timeout on this connection; perhaps we should have one. ScopedDbConnection conn(s["host"].valuestr()); DBClientConnection *cliConn = dynamic_cast< DBClientConnection* >( &conn.conn() ); if ( cliConn && replAuthenticate(cliConn) ) { BSONObj first = conn->findOne( (string)"local.oplog.$" + sourcename, Query().sort( BSON( "$natural" << 1 ) ) ); BSONObj last = conn->findOne( (string)"local.oplog.$" + sourcename, Query().sort( BSON( "$natural" << -1 ) ) ); bb.appendDate( "masterFirst" , first["ts"].timestampTime() ); bb.appendDate( "masterLast" , last["ts"].timestampTime() ); double lag = (double) (last["ts"].timestampTime() - s["syncedTo"].timestampTime()); bb.append( "lagSeconds" , lag / 1000 ); } conn.done(); } sources.append( BSONObjBuilder::numStr( n++ ) , bb.obj() ); } sources.done(); } }