C++ OplogReader::conn方法代码示例

    bool BackgroundSync::isRollbackRequired(OplogReader& r) {
        string hn = r.conn()->getServerAddress();
        if (!r.more()) {
            // In vanilla Mongo, this happened for one of the
            // following reasons:
            //  - we were ahead of what we are syncing from (don't
            //    think that is possible anymore)
            //  - remote oplog is empty for some weird reason
            // in either case, if it (strangely) happens, we'll just return
            // and our caller will simply try again after a short sleep.
            log() << "replSet error empty query result from " << hn << " oplog, attempting rollback" << rsLog;
             return true;
        }

        BSONObj o = r.nextSafe();
        uint64_t ts = o["ts"]._numberLong();
        uint64_t lastHash = o["h"].numberLong();
        GTID gtid = getGTIDFromBSON("_id", o);

        if( !theReplSet->gtidManager->rollbackNeeded(gtid, ts, lastHash)) {
            log() << "Rollback NOT needed! Our GTID" << gtid << endl;
            return false;
        }

        log() << "Rollback needed! Our GTID" <<
            theReplSet->gtidManager->getLiveState().toString() <<
            " remote GTID: " << gtid.toString() << ". Attempting rollback." << rsLog;

        runRollback(r, ts);
        return true;
    }

    bool isRollbackRequired(OplogReader& r, uint64_t *lastTS) {
        string hn = r.conn()->getServerAddress();
        verify(r.more());
        BSONObj rollbackStatus;
        bool found = getRollbackStatus(rollbackStatus);
        if (found) {
            // we have a rollback in progress,
            // must complete it
            log() << "Rollback needed, found rollbackStatus: " << rollbackStatus << rsLog;
            return true;
        }

        BSONObj o = r.nextSafe();
        uint64_t ts = o["ts"]._numberLong();
        uint64_t lastHash = o["h"].numberLong();
        GTID gtid = getGTIDFromBSON("_id", o);

        if (!theReplSet->gtidManager->rollbackNeeded(gtid, ts, lastHash)) {
            log() << "Rollback NOT needed! " << gtid << endl;
            return false;
        }

        log() << "Rollback needed! Our GTID: " <<
            theReplSet->gtidManager->getLiveState().toString() <<
            ", remote GTID: " << gtid.toString() << ". Attempting rollback." << rsLog;

        *lastTS = ts;
        return true;
    }

    void BackgroundSync::getOplogReader(OplogReader& r) {
        const Member *target = NULL, *stale = NULL;
        BSONObj oldest;

        {
            boost::unique_lock<boost::mutex> lock(_mutex);
            if (_lastOpTimeFetched.isNull()) {
                // then we're initial syncing and we're still waiting for this to be set
                _currentSyncTarget = NULL;
                return;
            }

            // Wait until we've applied the ops we have before we choose a sync target
            while (!_appliedBuffer) {
                _condvar.wait(lock);
            }
        }

        verify(r.conn() == NULL);

        while ((target = theReplSet->getMemberToSyncTo()) != NULL) {
            string current = target->fullName();

            if (!r.connect(current)) {
                LOG(2) << "replSet can't connect to " << current << " to read operations" << rsLog;
                r.resetConnection();
                theReplSet->veto(current);
                continue;
            }

            if (isStale(r, oldest)) {
                r.resetConnection();
                theReplSet->veto(current, 600);
                stale = target;
                continue;
            }

            // if we made it here, the target is up and not stale
            {
                boost::unique_lock<boost::mutex> lock(_mutex);
                _currentSyncTarget = target;
            }

            return;
        }

        // the only viable sync target was stale
        if (stale) {
            theReplSet->goStale(stale, oldest);
            sleepsecs(120);
        }

        {
            boost::unique_lock<boost::mutex> lock(_mutex);
            _currentSyncTarget = NULL;
        }
    }

    bool BackgroundSync::isRollbackRequired(OplogReader& r) {
        string hn = r.conn()->getServerAddress();

        if (!r.more()) {
            try {
                BSONObj theirLastOp = r.getLastOp(rsoplog);
                if (theirLastOp.isEmpty()) {
                    log() << "replSet error empty query result from " << hn << " oplog" << rsLog;
                    sleepsecs(2);
                    return true;
                }
                OpTime theirTS = theirLastOp["ts"]._opTime();
                if (theirTS < _lastOpTimeFetched) {
                    log() << "replSet we are ahead of the sync source, will try to roll back"
                          << rsLog;
                    theReplSet->syncRollback(r);
                    return true;
                }
                /* we're not ahead?  maybe our new query got fresher data.  best to come back and try again */
                log() << "replSet syncTail condition 1" << rsLog;
                sleepsecs(1);
            }
            catch(DBException& e) {
                log() << "replSet error querying " << hn << ' ' << e.toString() << rsLog;
                sleepsecs(2);
            }
            return true;
        }

        BSONObj o = r.nextSafe();
        OpTime ts = o["ts"]._opTime();
        long long h = o["h"].numberLong();
        if( ts != _lastOpTimeFetched || h != _lastH ) {
            log() << "replSet our last op time fetched: " << _lastOpTimeFetched.toStringPretty() << rsLog;
            log() << "replset source's GTE: " << ts.toStringPretty() << rsLog;
            theReplSet->syncRollback(r);
            return true;
        }

        return false;
    }

    bool BackgroundSync::_rollbackIfNeeded(OperationContext* txn, OplogReader& r) {
        string hn = r.conn()->getServerAddress();

        if (!r.more()) {
            try {
                BSONObj theirLastOp = r.getLastOp(rsOplogName.c_str());
                if (theirLastOp.isEmpty()) {
                    error() << "empty query result from " << hn << " oplog";
                    sleepsecs(2);
                    return true;
                }
                OpTime theirOpTime = extractOpTime(theirLastOp);
                if (theirOpTime < _lastOpTimeFetched) {
                    log() << "we are ahead of the sync source, will try to roll back";
                    syncRollback(txn, _replCoord->getMyLastOptime(), &r, _replCoord);
                    return true;
                }
                /* we're not ahead?  maybe our new query got fresher data.  best to come back and try again */
                log() << "syncTail condition 1";
                sleepsecs(1);
            }
            catch(DBException& e) {
                error() << "querying " << hn << ' ' << e.toString();
                sleepsecs(2);
            }
            return true;
        }

        BSONObj o = r.nextSafe();
        OpTime opTime = extractOpTime(o);
        long long hash = o["h"].numberLong();
        if ( opTime != _lastOpTimeFetched || hash != _lastFetchedHash ) {
            log() << "our last op time fetched: " << _lastOpTimeFetched;
            log() << "source's GTE: " << opTime;
            syncRollback(txn, _replCoord->getMyLastOptime(), &r, _replCoord);
            return true;
        }

        return false;
    }

    void BackgroundSync::getOplogReader(OplogReader& r) {
        const Member *target = NULL, *stale = NULL;
        BSONObj oldest;

        verify(r.conn() == NULL);
        while ((target = theReplSet->getMemberToSyncTo()) != NULL) {
            string current = target->fullName();

            if (!r.connect(current)) {
                LOG(2) << "replSet can't connect to " << current << " to read operations" << rsLog;
                r.resetConnection();
                theReplSet->veto(current);
                continue;
            }

            // if we made it here, the target is up and not stale
            {
                boost::unique_lock<boost::mutex> lock(_mutex);
                _currentSyncTarget = target;
            }

            return;
        }

        // the only viable sync target was stale
        if (stale) {
            GTID remoteOldestGTID = getGTIDFromBSON("_id", oldest);
            theReplSet->goStale(stale, remoteOldestGTID);
            // vanilla Mongo used to do a sleep of 120 seconds here
            // We removed it. It seems excessive, and if this machine is doing
            // nothing anyway, sleeping won't help. It might as well
            // return with a null sync target, and produce() will handle
            // that fact and sleep one second
        }

        {
            boost::unique_lock<boost::mutex> lock(_mutex);
            _currentSyncTarget = NULL;
        }
    }

    /**
     * Do the initial sync for this member.  There are several steps to this process:
     *
     *     0. Add _initialSyncFlag to minValid to tell us to restart initial sync if we
     *        crash in the middle of this procedure
     *     1. Record start time.
     *     2. Clone.
     *     3. Set minValid1 to sync target's latest op time.
     *     4. Apply ops from start to minValid1, fetching missing docs as needed.
     *     5. Set minValid2 to sync target's latest op time.
     *     6. Apply ops from minValid1 to minValid2.
     *     7. Build indexes.
     *     8. Set minValid3 to sync target's latest op time.
     *     9. Apply ops from minValid2 to minValid3.
          10. Clean up minValid and remove _initialSyncFlag field
     *
     * At that point, initial sync is finished.  Note that the oplog from the sync target is applied
     * three times: step 4, 6, and 8.  4 may involve refetching, 6 should not.  By the end of 6,
     * this member should have consistent data.  8 is "cosmetic," it is only to get this member
     * closer to the latest op time before it can transition to secondary state.
     */
    void ReplSetImpl::_syncDoInitialSync() {
        replset::InitialSync init(replset::BackgroundSync::get());
        replset::SyncTail tail(replset::BackgroundSync::get());
        sethbmsg("initial sync pending",0);

        // if this is the first node, it may have already become primary
        if ( box.getState().primary() ) {
            sethbmsg("I'm already primary, no need for initial sync",0);
            return;
        }

        const Member *source = getMemberToSyncTo();
        if (!source) {
            sethbmsg("initial sync need a member to be primary or secondary to do our initial sync", 0);
            sleepsecs(15);
            return;
        }

        string sourceHostname = source->h().toString();
        init.setHostname(sourceHostname);
        OplogReader r;
        if( !r.connect(sourceHostname) ) {
            sethbmsg( str::stream() << "initial sync couldn't connect to " << source->h().toString() , 0);
            sleepsecs(15);
            return;
        }

        BSONObj lastOp = r.getLastOp(rsoplog);
        if( lastOp.isEmpty() ) {
            sethbmsg("initial sync couldn't read remote oplog", 0);
            sleepsecs(15);
            return;
        }

        // written by applyToHead calls
        BSONObj minValid;

        if (replSettings.fastsync) {
            log() << "fastsync: skipping database clone" << rsLog;

            // prime oplog
            init.oplogApplication(lastOp, lastOp);
            return;
        }
        else {
            // Add field to minvalid document to tell us to restart initial sync if we crash
            theReplSet->setInitialSyncFlag();

            sethbmsg("initial sync drop all databases", 0);
            dropAllDatabasesExceptLocal();

            sethbmsg("initial sync clone all databases", 0);

            list<string> dbs = r.conn()->getDatabaseNames();

            Cloner cloner;
            if (!_syncDoInitialSync_clone(cloner, sourceHostname.c_str(), dbs, true)) {
                veto(source->fullName(), 600);
                sleepsecs(300);
                return;
            }

            sethbmsg("initial sync data copy, starting syncup",0);

            log() << "oplog sync 1 of 3" << endl;
            if ( ! _syncDoInitialSync_applyToHead( init, &r , source , lastOp , minValid ) ) {
                return;
            }

            lastOp = minValid;

            // Now we sync to the latest op on the sync target _again_, as we may have recloned ops
            // that were "from the future" compared with minValid. During this second application,
            // nothing should need to be recloned.
            log() << "oplog sync 2 of 3" << endl;
            if (!_syncDoInitialSync_applyToHead(tail, &r , source , lastOp , minValid)) {
                return;
            }
            // data should now be consistent
//.........这里部分代码省略.........

    /**
     * Do the initial sync for this member.
     */
    void ReplSetImpl::_syncDoInitialSync() {
        replset::InitialSync init(replset::BackgroundSync::get());
        sethbmsg("initial sync pending",0);

        // if this is the first node, it may have already become primary
        if ( box.getState().primary() ) {
            sethbmsg("I'm already primary, no need for initial sync",0);
            return;
        }

        const Member *source = getMemberToSyncTo();
        if (!source) {
            sethbmsg("initial sync need a member to be primary or secondary to do our initial sync", 0);
            sleepsecs(15);
            return;
        }

        string sourceHostname = source->h().toString();
        init.setHostname(sourceHostname);
        OplogReader r;
        if( !r.connect(sourceHostname) ) {
            sethbmsg( str::stream() << "initial sync couldn't connect to " << source->h().toString() , 0);
            sleepsecs(15);
            return;
        }

        BSONObj lastOp = r.getLastOp(rsoplog);
        if( lastOp.isEmpty() ) {
            sethbmsg("initial sync couldn't read remote oplog", 0);
            sleepsecs(15);
            return;
        }

        if (replSettings.fastsync) {
            log() << "fastsync: skipping database clone" << rsLog;

            // prime oplog
            init.oplogApplication(lastOp, lastOp);
            return;
        }
        else {
            sethbmsg("initial sync drop all databases", 0);
            dropAllDatabasesExceptLocal();

            sethbmsg("initial sync clone all databases", 0);

            list<string> dbs = r.conn()->getDatabaseNames();

            if ( ! _syncDoInitialSync_clone( sourceHostname.c_str(), dbs, true ) ) {
                veto(source->fullName(), 600);
                sleepsecs(300);
                return;
            }

            sethbmsg("initial sync data copy, starting syncup",0);
            
            BSONObj minValid;
            if ( ! _syncDoInitialSync_applyToHead( init, &r , source , lastOp , minValid ) ) {
                return;
            }

            lastOp = minValid;

            // reset state, as that "didn't count"
            emptyOplog(); 
            lastOpTimeWritten = OpTime();
            lastH = 0;

            sethbmsg("initial sync building indexes",0);
            if ( ! _syncDoInitialSync_clone( sourceHostname.c_str(), dbs, false ) ) {
                veto(source->fullName(), 600);
                sleepsecs(300);
                return;
            }
        }

        sethbmsg("initial sync query minValid",0);

        BSONObj minValid;
        if ( ! _syncDoInitialSync_applyToHead( init, &r, source, lastOp, minValid ) ) {
            return;
        }
        
        // ---------


        sethbmsg("initial sync finishing up",0);

        verify( !box.getState().primary() ); // wouldn't make sense if we were.

        {
            Client::WriteContext cx( "local." );
            cx.ctx().db()->flushFiles(true);
            try {
                log() << "replSet set minValid=" << minValid["ts"]._opTime().toString() << rsLog;
            }
            catch(...) { }
//.........这里部分代码省略.........

    /**
     * Do the initial sync for this member.  There are several steps to this process:
     *
     *     1. Record start time.
     *     2. Clone.
     *     3. Set minValid1 to sync target's latest op time.
     *     4. Apply ops from start to minValid1, fetching missing docs as needed.
     *     5. Set minValid2 to sync target's latest op time.
     *     6. Apply ops from minValid1 to minValid2.
     *     7. Build indexes.
     *     8. Set minValid3 to sync target's latest op time.
     *     9. Apply ops from minValid2 to minValid3.
     *
     * At that point, initial sync is finished.  Note that the oplog from the sync target is applied
     * three times: step 4, 6, and 8.  4 may involve refetching, 6 should not.  By the end of 6,
     * this member should have consistent data.  8 is "cosmetic," it is only to get this member
     * closer to the latest op time before it can transition to secondary state.
     */
    void ReplSetImpl::_syncDoInitialSync() {
        replset::InitialSync init(replset::BackgroundSync::get());
        replset::SyncTail tail(replset::BackgroundSync::get());
        sethbmsg("initial sync pending",0);

        // if this is the first node, it may have already become primary
        if ( box.getState().primary() ) {
            sethbmsg("I'm already primary, no need for initial sync",0);
            return;
        }

        const Member *source = getMemberToSyncTo();
        if (!source) {
            sethbmsg("initial sync need a member to be primary or secondary to do our initial sync", 0);
            sleepsecs(15);
            return;
        }

        string sourceHostname = source->h().toString();
        init.setHostname(sourceHostname);
        OplogReader r;
        if( !r.connect(sourceHostname) ) {
            sethbmsg( str::stream() << "initial sync couldn't connect to " << source->h().toString() , 0);
            sleepsecs(15);
            return;
        }

        BSONObj lastOp = r.getLastOp(rsoplog);
        if( lastOp.isEmpty() ) {
            sethbmsg("initial sync couldn't read remote oplog", 0);
            sleepsecs(15);
            return;
        }

        // written by applyToHead calls
        BSONObj minValid;

        if (replSettings.fastsync) {
            log() << "fastsync: skipping database clone" << rsLog;

            // prime oplog
            init.oplogApplication(lastOp, lastOp);
            return;
        }
        else {
            sethbmsg("initial sync drop all databases", 0);
            dropAllDatabasesExceptLocal();

            sethbmsg("initial sync clone all databases", 0);

            list<string> dbs = r.conn()->getDatabaseNames();

            if ( ! _syncDoInitialSync_clone( sourceHostname.c_str(), dbs, true ) ) {
                veto(source->fullName(), 600);
                sleepsecs(300);
                return;
            }

            sethbmsg("initial sync data copy, starting syncup",0);

            log() << "oplog sync 1 of 3" << endl;
            if ( ! _syncDoInitialSync_applyToHead( init, &r , source , lastOp , minValid ) ) {
                return;
            }

            lastOp = minValid;

            // Now we sync to the latest op on the sync target _again_, as we may have recloned ops
            // that were "from the future" compared with minValid. During this second application,
            // nothing should need to be recloned.
            log() << "oplog sync 2 of 3" << endl;
            if (!_syncDoInitialSync_applyToHead(tail, &r , source , lastOp , minValid)) {
                return;
            }
            // data should now be consistent

            lastOp = minValid;

            sethbmsg("initial sync building indexes",0);
            if ( ! _syncDoInitialSync_clone( sourceHostname.c_str(), dbs, false ) ) {
                veto(source->fullName(), 600);
                sleepsecs(300);
//.........这里部分代码省略.........

    /**
     * Do the initial sync for this member.
     */
    void ReplSetImpl::_syncDoInitialSync() {
        sethbmsg("initial sync pending",0);

        // if this is the first node, it may have already become primary
        if ( box.getState().primary() ) {
            sethbmsg("I'm already primary, no need for initial sync",0);
            return;
        }
        
        const Member *source = getMemberToSyncTo();
        if (!source) {
            sethbmsg("initial sync need a member to be primary or secondary to do our initial sync", 0);
            sleepsecs(15);
            return;
        }

        string sourceHostname = source->h().toString();
        OplogReader r;
        if( !r.connect(sourceHostname) ) {
            sethbmsg( str::stream() << "initial sync couldn't connect to " << source->h().toString() , 0);
            sleepsecs(15);
            return;
        }

        BSONObj lastOp = r.getLastOp(rsoplog);
        if( lastOp.isEmpty() ) {
            sethbmsg("initial sync couldn't read remote oplog", 0);
            sleepsecs(15);
            return;
        }
        OpTime startingTS = lastOp["ts"]._opTime();

        if (replSettings.fastsync) {
            log() << "fastsync: skipping database clone" << rsLog;
        }
        else {
            sethbmsg("initial sync drop all databases", 0);
            dropAllDatabasesExceptLocal();

            sethbmsg("initial sync clone all databases", 0);

            list<string> dbs = r.conn()->getDatabaseNames();
            for( list<string>::iterator i = dbs.begin(); i != dbs.end(); i++ ) {
                string db = *i;
                if( db != "local" ) {
                    sethbmsg( str::stream() << "initial sync cloning db: " << db , 0);
                    bool ok;
                    {
                        writelock lk(db);
                        Client::Context ctx(db);
                        ok = clone(sourceHostname.c_str(), db);
                    }
                    if( !ok ) {
                        sethbmsg( str::stream() << "initial sync error clone of " << db << " failed sleeping 5 minutes" ,0);
                        sleepsecs(300);
                        return;
                    }
                }
            }
        }

        sethbmsg("initial sync query minValid",0);

        isyncassert( "initial sync source must remain readable throughout our initial sync", source->state().readable() );

        /* our cloned copy will be strange until we apply oplog events that occurred
           through the process.  we note that time point here. */
        BSONObj minValid = r.getLastOp(rsoplog);
        isyncassert( "getLastOp is empty ", !minValid.isEmpty() );
        OpTime mvoptime = minValid["ts"]._opTime();
        assert( !mvoptime.isNull() );

        /* apply relevant portion of the oplog
        */
        {
            isyncassert( str::stream() << "initial sync source must remain readable throughout our initial sync [2] state now: " << source->state().toString() , source->state().readable() );
            if( ! initialSyncOplogApplication(source, /*applyGTE*/startingTS, /*minValid*/mvoptime) ) { // note we assume here that this call does not throw
                log() << "replSet initial sync failed during applyoplog" << rsLog;
                emptyOplog(); // otherwise we'll be up!
                
                lastOpTimeWritten = OpTime();
                lastH = 0;
                
                log() << "replSet cleaning up [1]" << rsLog;
                {
                    writelock lk("local.");
                    Client::Context cx( "local." );
                    cx.db()->flushFiles(true);
                }
                log() << "replSet cleaning up [2]" << rsLog;
                sleepsecs(5);
                return;
            }
        }

        sethbmsg("initial sync finishing up",0);

//.........这里部分代码省略.........

    void BackgroundSync::getOplogReader(OplogReader& r) {
        const Member *target = NULL, *stale = NULL;
        BSONObj oldest;

        {
            boost::unique_lock<boost::mutex> lock(_mutex);
            if (_lastOpTimeFetched.isNull()) {
                // then we're initial syncing and we're still waiting for this to be set
                _currentSyncTarget = NULL;
                return;
            }

            // Wait until we've applied the ops we have before we choose a sync target
            while (!_appliedBuffer) {
                _condvar.wait(lock);
            }
        }

        while (MONGO_FAIL_POINT(rsBgSyncProduce)) {
            sleepmillis(0);
        }

        verify(r.conn() == NULL);

        while ((target = theReplSet->getMemberToSyncTo()) != NULL) {
            string current = target->fullName();

            if (!r.connect(current)) {
                LOG(2) << "replSet can't connect to " << current << " to read operations" << rsLog;
                r.resetConnection();
                theReplSet->veto(current);
                continue;
            }

            if (isStale(r, oldest)) {
                r.resetConnection();
                theReplSet->veto(current, 600);
                stale = target;
                continue;
            }

            // if we made it here, the target is up and not stale
            {
                boost::unique_lock<boost::mutex> lock(_mutex);
                _currentSyncTarget = target;
            }
            {
                // prevent writers from blocking readers during fsync
                SimpleMutex::scoped_lock fsynclk(filesLockedFsync);
                // we don't need the local write lock yet, but it's needed by ensureMe()
                // so we take it preemptively to avoid deadlocking.
                Lock::DBWrite lk("local");

                theReplSet->syncSourceFeedback.connect(target);
            }

            return;
        }

        // the only viable sync target was stale
        if (stale) {
            theReplSet->goStale(stale, oldest);
            sleepsecs(120);
        }

        {
            boost::unique_lock<boost::mutex> lock(_mutex);
            _currentSyncTarget = NULL;
        }
    }