Status MetadataLoader::initCollection( const string& ns, const string& shard, CollectionMetadata* metadata ) const { // // Bring collection entry from the config server. // BSONObj collDoc; { try { ScopedDbConnection conn( _configLoc.toString(), 30 ); collDoc = conn->findOne( CollectionType::ConfigNS, QUERY(CollectionType::ns()<<ns)); conn.done(); } catch ( const DBException& e ) { string errMsg = str::stream() << "could not query collection metadata" << causedBy( e ); // We deliberately do not return conn to the pool, since it was involved // with the error here. return Status( ErrorCodes::HostUnreachable, errMsg ); } } string errMsg; if ( collDoc.isEmpty() ) { errMsg = str::stream() << "could not load metadata, collection " << ns << " not found"; warning() << errMsg << endl; return Status( ErrorCodes::NamespaceNotFound, errMsg ); } CollectionType collInfo; if ( !collInfo.parseBSON( collDoc, &errMsg ) || !collInfo.isValid( &errMsg ) ) { errMsg = str::stream() << "could not parse metadata for collection " << ns << causedBy( errMsg ); warning() << errMsg << endl; return Status( ErrorCodes::FailedToParse, errMsg ); } log() << "Collection Info: " << collInfo << endl; if ( collInfo.isDroppedSet() && collInfo.getDropped() ) { errMsg = str::stream() << "could not load metadata, collection " << ns << " was dropped"; warning() << errMsg << endl; return Status( ErrorCodes::NamespaceNotFound, errMsg ); } if ( collInfo.isKeyPatternSet() && !collInfo.getKeyPattern().isEmpty() ) { // Sharded collection, need to load chunks metadata->_keyPattern = collInfo.getKeyPattern(); metadata->_shardVersion = ChunkVersion( 0, 0, collInfo.getEpoch() ); metadata->_collVersion = ChunkVersion( 0, 0, collInfo.getEpoch() ); metadata->_linkedNS = collInfo.getLinked(); return Status::OK(); } else if ( collInfo.isPrimarySet() && collInfo.getPrimary() == shard ) { // A collection with a non-default primary // Empty primary field not allowed if set dassert( collInfo.getPrimary() != "" ); metadata->_keyPattern = BSONObj(); metadata->_shardVersion = ChunkVersion( 1, 0, collInfo.getEpoch() ); metadata->_collVersion = metadata->_shardVersion; return Status::OK(); } else { // A collection with a primary that doesn't match this shard or is empty, the primary // may have changed before we loaded. errMsg = // br str::stream() << "collection " << ns << " does not have a shard key " << "and primary " << ( collInfo.isPrimarySet() ? collInfo.getPrimary() : "" ) << " does not match this shard " << shard; warning() << errMsg << endl; metadata->_collVersion = ChunkVersion( 0, 0, OID() ); return Status( ErrorCodes::RemoteChangeDetected, errMsg ); } }
Status MetadataLoader::initChunks( const string& ns, const string& shard, const CollectionMetadata* oldMetadata, CollectionMetadata* metadata ) const { map<string, ChunkVersion> versionMap; // Preserve the epoch versionMap[shard] = metadata->_shardVersion; OID epoch = metadata->getCollVersion().epoch(); bool fullReload = true; // Check to see if we should use the old version or not. if ( oldMetadata ) { // If our epochs are compatible, it's useful to use the old metadata for diffs if ( oldMetadata->getCollVersion().hasCompatibleEpoch( epoch ) ) { fullReload = false; dassert( oldMetadata->isValid() ); versionMap[shard] = oldMetadata->_shardVersion; metadata->_collVersion = oldMetadata->_collVersion; // TODO: This could be made more efficient if copying not required, but // not as frequently reloaded as in mongos. metadata->_chunksMap = oldMetadata->_chunksMap; LOG( 2 ) << "loading new chunks for collection " << ns << " using old metadata w/ version " << oldMetadata->getShardVersion() << " and " << metadata->_chunksMap.size() << " chunks" << endl; } else { warning() << "reloading collection metadata for " << ns << " with new epoch " << epoch.toString() << ", the current epoch is " << oldMetadata->getCollVersion().epoch().toString() << endl; } } // Exposes the new metadata's range map and version to the "differ," who // would ultimately be responsible of filling them up. SCMConfigDiffTracker differ( shard ); log() << "Metadata attach" << endl; differ.attach( ns, metadata->_linkedNS, metadata->_chunksMap, metadata->_collVersion, versionMap ); try { ScopedDbConnection conn( _configLoc.toString(), 30 ); auto_ptr<DBClientCursor> cursor = conn->query( ChunkType::ConfigNS, differ.configDiffQuery() ); if ( !cursor.get() ) { // Make our metadata invalid metadata->_collVersion = ChunkVersion( 0, 0, OID() ); metadata->_chunksMap.clear(); conn.done(); return Status( ErrorCodes::HostUnreachable, "problem opening chunk metadata cursor" ); } // // The diff tracker should always find at least one chunk (the highest chunk we saw // last time). If not, something has changed on the config server (potentially between // when we read the collection data and when we read the chunks data). // int diffsApplied = differ.calculateConfigDiff( *cursor ); if ( diffsApplied > 0 ) { // Chunks found, return ok LOG(2) << "loaded " << diffsApplied << " chunks into new metadata for " << ns << " with version " << metadata->_collVersion << endl; metadata->_shardVersion = versionMap[shard]; metadata->fillRanges(); conn.done(); dassert( metadata->isValid() ); return Status::OK(); } else if ( diffsApplied == 0 ) { // No chunks found, the collection is dropping or we're confused // If this is a full reload, assume it is a drop for backwards compatibility // TODO: drop the config.collections entry *before* the chunks and eliminate this // ambiguity string errMsg = str::stream() << "no chunks found when reloading " << ns << ", previous version was " << metadata->_collVersion.toString() << ( fullReload ? ", this is a drop" : "" ); warning() << errMsg << endl; metadata->_collVersion = ChunkVersion( 0, 0, OID() ); metadata->_chunksMap.clear(); conn.done(); return fullReload ? Status( ErrorCodes::NamespaceNotFound, errMsg ) : Status( ErrorCodes::RemoteChangeDetected, errMsg ); } else { // Invalid chunks found, our epoch may have changed because we dropped/recreated // the collection. string errMsg = // br str::stream() << "invalid chunks found when reloading " << ns << ", previous version was " << metadata->_collVersion.toString() << ", this should be rare"; warning() << errMsg << endl; metadata->_collVersion = ChunkVersion( 0, 0, OID() ); metadata->_chunksMap.clear(); conn.done(); return Status( ErrorCodes::RemoteChangeDetected, errMsg ); } } catch ( const DBException& e ) { string errMsg = str::stream() << "problem querying chunks metadata" << causedBy( e ); // We deliberately do not return connPtr to the pool, since it was involved // with the error here. return Status( ErrorCodes::HostUnreachable, errMsg ); } }
/* **************************************************************************** * * mongoUnsubscribeContextAvailability - */ HttpStatusCode mongoUnsubscribeContextAvailability(UnsubscribeContextAvailabilityRequest* requestP, UnsubscribeContextAvailabilityResponse* responseP, const std::string& tenant) { reqSemTake(__FUNCTION__, "ngsi9 unsubscribe request"); LM_T(LmtMongo, ("Unsubscribe Context Availability")); DBClientConnection* connection = getMongoConnection(); /* No matter if success or failure, the subscriptionId in the response is always the one * in the request */ responseP->subscriptionId = requestP->subscriptionId; /* Look for document */ BSONObj sub; try { OID id = OID(requestP->subscriptionId.get()); LM_T(LmtMongo, ("findOne() in '%s' collection _id '%s'}", getSubscribeContextAvailabilityCollectionName(tenant).c_str(), requestP->subscriptionId.get().c_str())); mongoSemTake(__FUNCTION__, "findOne in SubscribeContextAvailabilityCollection"); sub = connection->findOne(getSubscribeContextAvailabilityCollectionName(tenant).c_str(), BSON("_id" << id)); mongoSemGive(__FUNCTION__, "findOne in SubscribeContextAvailabilityCollection"); if (sub.isEmpty()) { responseP->statusCode.fill(SccContextElementNotFound); reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (no subscriptions)"); return SccOk; } } catch( const AssertionException &e ) { /* This happens when OID format is wrong */ // FIXME: this checking should be done at parsing stage, without progressing to // mongoBackend. By the moment we can live this here, but we should remove in the future // (odl issues #95) mongoSemGive(__FUNCTION__, "findOne in SubscribeContextAvailabilityCollection (mongo assertion exception)"); reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (mongo assertion exception)"); responseP->statusCode.fill(SccContextElementNotFound); return SccOk; } catch( const DBException &e ) { mongoSemGive(__FUNCTION__, "findOne in SubscribeContextAvailabilityCollection (mongo db exception)"); reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (mongo db exception)"); responseP->statusCode.fill(SccReceiverInternalError, std::string("collection: ") + getSubscribeContextAvailabilityCollectionName(tenant).c_str() + " - findOne() _id: " + requestP->subscriptionId.get() + " - exception: " + e.what()); return SccOk; } catch(...) { mongoSemGive(__FUNCTION__, "findOne in SubscribeContextAvailabilityCollection (mongo generic exception)"); reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (mongo generic exception)"); responseP->statusCode.fill(SccReceiverInternalError, std::string("collection: ") + getSubscribeContextAvailabilityCollectionName(tenant).c_str() + " - findOne() _id: " + requestP->subscriptionId.get() + " - exception: " + "generic"); return SccOk; } /* Remove document in MongoDB */ // FIXME: I would prefer to do the find and remove in a single operation. Is the some similar // to findAndModify for this? try { LM_T(LmtMongo, ("remove() in '%s' collection _id '%s'}", getSubscribeContextAvailabilityCollectionName(tenant).c_str(), requestP->subscriptionId.get().c_str())); mongoSemTake(__FUNCTION__, "remove in SubscribeContextAvailabilityCollection"); connection->remove(getSubscribeContextAvailabilityCollectionName(tenant).c_str(), BSON("_id" << OID(requestP->subscriptionId.get()))); mongoSemGive(__FUNCTION__, "remove in SubscribeContextAvailabilityCollection"); } catch( const DBException &e ) { mongoSemGive(__FUNCTION__, "remove in SubscribeContextAvailabilityCollection (mongo db exception)"); reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (mongo db exception)"); responseP->statusCode.fill(SccReceiverInternalError, std::string("collection: ") + getSubscribeContextAvailabilityCollectionName(tenant).c_str() + " - remove() _id: " + requestP->subscriptionId.get().c_str() + " - exception: " + e.what()); return SccOk; } catch(...) { mongoSemGive(__FUNCTION__, "remove in SubscribeContextAvailabilityCollection (mongo generic exception)"); reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request (mongo generic exception)"); responseP->statusCode.fill(SccReceiverInternalError, std::string("collection: ") + getSubscribeContextAvailabilityCollectionName(tenant).c_str() + " - remove() _id: " + requestP->subscriptionId.get().c_str() + " - exception: " + "generic"); return SccOk; } responseP->statusCode.fill(SccOk); reqSemGive(__FUNCTION__, "ngsi9 unsubscribe request"); return SccOk; }
void BatchWriteExec::executeBatch( const BatchedCommandRequest& clientRequest, BatchedCommandResponse* clientResponse ) { LOG( 4 ) << "starting execution of write batch of size " << static_cast<int>( clientRequest.sizeWriteOps() ) << " for " << clientRequest.getNS() << endl; BatchWriteOp batchOp; batchOp.initClientRequest( &clientRequest ); // Current batch status bool refreshedTargeter = false; int rounds = 0; int numCompletedOps = 0; int numRoundsWithoutProgress = 0; while ( !batchOp.isFinished() ) { // // Get child batches to send using the targeter // // Targeting errors can be caused by remote metadata changing (the collection could have // been dropped and recreated, for example with a new shard key). If a remote metadata // change occurs *before* a client sends us a batch, we need to make sure that we don't // error out just because we're staler than the client - otherwise mongos will be have // unpredictable behavior. // // (If a metadata change happens *during* or *after* a client sends us a batch, however, // we make no guarantees about delivery.) // // For this reason, we don't record targeting errors until we've refreshed our targeting // metadata at least once *after* receiving the client batch - at that point, we know: // // 1) our new metadata is the same as the metadata when the client sent a batch, and so // targeting errors are real. // OR // 2) our new metadata is a newer version than when the client sent a batch, and so // the metadata must have changed after the client batch was sent. We don't need to // deliver in this case, since for all the client knows we may have gotten the batch // exactly when the metadata changed. // OwnedPointerVector<TargetedWriteBatch> childBatchesOwned; vector<TargetedWriteBatch*>& childBatches = childBatchesOwned.mutableVector(); // If we've already had a targeting error, we've refreshed the metadata once and can // record target errors definitively. bool recordTargetErrors = refreshedTargeter; Status targetStatus = batchOp.targetBatch( *_targeter, recordTargetErrors, &childBatches ); if ( !targetStatus.isOK() ) { // Don't do anything until a targeter refresh _targeter->noteCouldNotTarget(); refreshedTargeter = true; ++_stats->numTargetErrors; dassert( childBatches.size() == 0u ); } // // Send all child batches // size_t numSent = 0; size_t numToSend = childBatches.size(); bool remoteMetadataChanging = false; while ( numSent != numToSend ) { // Collect batches out on the network, mapped by endpoint OwnedHostBatchMap ownedPendingBatches; OwnedHostBatchMap::MapType& pendingBatches = ownedPendingBatches.mutableMap(); // // Send side // // Get as many batches as we can at once for ( vector<TargetedWriteBatch*>::iterator it = childBatches.begin(); it != childBatches.end(); ++it ) { // // Collect the info needed to dispatch our targeted batch // TargetedWriteBatch* nextBatch = *it; // If the batch is NULL, we sent it previously, so skip if ( nextBatch == NULL ) continue; // Figure out what host we need to dispatch our targeted batch ConnectionString shardHost; Status resolveStatus = _resolver->chooseWriteHost( nextBatch->getEndpoint() .shardName, &shardHost ); if ( !resolveStatus.isOK() ) { ++_stats->numResolveErrors; // Record a resolve failure // TODO: It may be necessary to refresh the cache if stale, or maybe just // cancel and retarget the batch WriteErrorDetail error; buildErrorFrom( resolveStatus, &error ); LOG( 4 ) << "unable to send write batch to " << shardHost.toString() << causedBy( resolveStatus.toString() ) << endl; batchOp.noteBatchError( *nextBatch, error ); // We're done with this batch // Clean up when we can't resolve a host delete *it; *it = NULL; --numToSend; continue; } // If we already have a batch for this host, wait until the next time OwnedHostBatchMap::MapType::iterator pendingIt = pendingBatches.find( shardHost ); if ( pendingIt != pendingBatches.end() ) continue; // // We now have all the info needed to dispatch the batch // BatchedCommandRequest request( clientRequest.getBatchType() ); batchOp.buildBatchRequest( *nextBatch, &request ); // Internally we use full namespaces for request/response, but we send the // command to a database with the collection name in the request. NamespaceString nss( request.getNS() ); request.setNS( nss.coll() ); LOG( 4 ) << "sending write batch to " << shardHost.toString() << ": " << request.toString() << endl; _dispatcher->addCommand( shardHost, nss.db(), request ); // Indicate we're done by setting the batch to NULL // We'll only get duplicate hostEndpoints if we have broadcast and non-broadcast // endpoints for the same host, so this should be pretty efficient without // moving stuff around. *it = NULL; // Recv-side is responsible for cleaning up the nextBatch when used pendingBatches.insert( make_pair( shardHost, nextBatch ) ); } // Send them all out _dispatcher->sendAll(); numSent += pendingBatches.size(); // // Recv side // while ( _dispatcher->numPending() > 0 ) { // Get the response ConnectionString shardHost; BatchedCommandResponse response; Status dispatchStatus = _dispatcher->recvAny( &shardHost, &response ); // Get the TargetedWriteBatch to find where to put the response dassert( pendingBatches.find( shardHost ) != pendingBatches.end() ); TargetedWriteBatch* batch = pendingBatches.find( shardHost )->second; if ( dispatchStatus.isOK() ) { TrackedErrors trackedErrors; trackedErrors.startTracking( ErrorCodes::StaleShardVersion ); LOG( 4 ) << "write results received from " << shardHost.toString() << ": " << response.toString() << endl; // Dispatch was ok, note response batchOp.noteBatchResponse( *batch, response, &trackedErrors ); // Note if anything was stale const vector<ShardError*>& staleErrors = trackedErrors.getErrors( ErrorCodes::StaleShardVersion ); if ( staleErrors.size() > 0 ) { noteStaleResponses( staleErrors, _targeter ); ++_stats->numStaleBatches; } // Remember if the shard is actively changing metadata right now if ( isShardMetadataChanging( staleErrors ) ) { remoteMetadataChanging = true; } // Remember that we successfully wrote to this shard // NOTE: This will record lastOps for shards where we actually didn't update // or delete any documents, which preserves old behavior but is conservative _stats->noteWriteAt( shardHost, response.isLastOpSet() ? response.getLastOp() : OpTime(), response.isElectionIdSet() ? response.getElectionId() : OID()); } else { // Error occurred dispatching, note it stringstream msg; msg << "write results unavailable from " << shardHost.toString() << causedBy( dispatchStatus.toString() ); WriteErrorDetail error; buildErrorFrom( Status( ErrorCodes::RemoteResultsUnavailable, msg.str() ), &error ); LOG( 4 ) << "unable to receive write results from " << shardHost.toString() << causedBy( dispatchStatus.toString() ) << endl; batchOp.noteBatchError( *batch, error ); } } } ++rounds; ++_stats->numRounds; // If we're done, get out if ( batchOp.isFinished() ) break; // MORE WORK TO DO // // Refresh the targeter if we need to (no-op if nothing stale) // bool targeterChanged = false; Status refreshStatus = _targeter->refreshIfNeeded( &targeterChanged ); if ( !refreshStatus.isOK() ) { // It's okay if we can't refresh, we'll just record errors for the ops if // needed. warning() << "could not refresh targeter" << causedBy( refreshStatus.reason() ) << endl; } // // Ensure progress is being made toward completing the batch op // int currCompletedOps = batchOp.numWriteOpsIn( WriteOpState_Completed ); if ( currCompletedOps == numCompletedOps && !targeterChanged && !remoteMetadataChanging ) { ++numRoundsWithoutProgress; } else { numRoundsWithoutProgress = 0; } numCompletedOps = currCompletedOps; if ( numRoundsWithoutProgress > kMaxRoundsWithoutProgress ) { stringstream msg; msg << "no progress was made executing batch write op in " << clientRequest.getNS() << " after " << kMaxRoundsWithoutProgress << " rounds (" << numCompletedOps << " ops completed in " << rounds << " rounds total)"; WriteErrorDetail error; buildErrorFrom( Status( ErrorCodes::NoProgressMade, msg.str() ), &error ); batchOp.abortBatch( error ); break; } } batchOp.buildClientResponse( clientResponse ); LOG( 4 ) << "finished execution of write batch" << ( clientResponse->isErrDetailsSet() ? " with write errors" : "") << ( clientResponse->isErrDetailsSet() && clientResponse->isWriteConcernErrorSet() ? " and" : "" ) << ( clientResponse->isWriteConcernErrorSet() ? " with write concern error" : "" ) << " for " << clientRequest.getNS() << endl; }
bool run( const string& dbname, BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool fromRepl ) { // // Correct behavior here is very finicky. // // 1. The first step is to append the error that occurred on the previous operation. // This adds an "err" field to the command, which is *not* the command failing. // // 2. Next we parse and validate write concern options. If these options are invalid // the command fails no matter what, even if we actually had an error earlier. The // reason for checking here is to match legacy behavior on these kind of failures - // we'll still get an "err" field for the write error. // // 3. If we had an error on the previous operation, we then return immediately. // // 4. Finally, we actually enforce the write concern. All errors *except* timeout are // reported with ok : 0.0, to match legacy behavior. // // There is a special case when "wOpTime" and "wElectionId" are explicitly provided by // the client (mongos) - in this case we *only* enforce the write concern if it is // valid. // // We always need to either report "err" (if ok : 1) or "errmsg" (if ok : 0), even if // err is null. // LastError *le = lastError.disableForCommand(); // Always append lastOp and connectionId Client& c = cc(); c.appendLastOp( result ); // for sharding; also useful in general for debugging result.appendNumber( "connectionId" , c.getConnectionId() ); OpTime lastOpTime; BSONField<OpTime> wOpTimeField("wOpTime"); FieldParser::FieldState extracted = FieldParser::extract(cmdObj, wOpTimeField, &lastOpTime, &errmsg); if (!extracted) { result.append("badGLE", cmdObj); appendCommandStatus(result, false, errmsg); return false; } bool lastOpTimePresent = extracted != FieldParser::FIELD_NONE; if (!lastOpTimePresent) { // Use the client opTime if no wOpTime is specified lastOpTime = cc().getLastOp(); } OID electionId; BSONField<OID> wElectionIdField("wElectionId"); extracted = FieldParser::extract(cmdObj, wElectionIdField, &electionId, &errmsg); if (!extracted) { result.append("badGLE", cmdObj); appendCommandStatus(result, false, errmsg); return false; } bool electionIdPresent = extracted != FieldParser::FIELD_NONE; bool errorOccurred = false; // Errors aren't reported when wOpTime is used if ( !lastOpTimePresent ) { if ( le->nPrev != 1 ) { errorOccurred = LastError::noError.appendSelf( result, false ); le->appendSelfStatus( result ); } else { errorOccurred = le->appendSelf( result, false ); } } BSONObj writeConcernDoc = cmdObj; // Use the default options if we have no gle options aside from wOpTime/wElectionId const int nFields = cmdObj.nFields(); bool useDefaultGLEOptions = (nFields == 1) || (nFields == 2 && lastOpTimePresent) || (nFields == 3 && lastOpTimePresent && electionIdPresent); if ( useDefaultGLEOptions && getLastErrorDefault ) { writeConcernDoc = *getLastErrorDefault; } // // Validate write concern no matter what, this matches 2.4 behavior // WriteConcernOptions writeConcern; Status status = writeConcern.parse( writeConcernDoc ); if ( status.isOK() ) { // Ensure options are valid for this host status = validateWriteConcern( writeConcern ); } if ( !status.isOK() ) { result.append( "badGLE", writeConcernDoc ); return appendCommandStatus( result, status ); } // Don't wait for replication if there was an error reported - this matches 2.4 behavior if ( errorOccurred ) { dassert( !lastOpTimePresent ); return true; } // No error occurred, so we won't duplicate these fields with write concern errors dassert( result.asTempObj()["err"].eoo() ); dassert( result.asTempObj()["code"].eoo() ); // If we got an electionId, make sure it matches if (electionIdPresent) { if (!theReplSet) { // Ignore electionIds of 0 from mongos. if (electionId != OID()) { errmsg = "wElectionId passed but no replication active"; result.append("code", ErrorCodes::BadValue); return false; } } else { if (electionId != theReplSet->getElectionId()) { LOG(3) << "oid passed in is " << electionId << ", but our id is " << theReplSet->getElectionId(); errmsg = "election occurred after write"; result.append("code", ErrorCodes::WriteConcernFailed); return false; } } } cc().curop()->setMessage( "waiting for write concern" ); WriteConcernResult wcResult; status = waitForWriteConcern( writeConcern, lastOpTime, &wcResult ); wcResult.appendTo( &result ); // For backward compatibility with 2.4, wtimeout returns ok : 1.0 if ( wcResult.wTimedOut ) { dassert( !wcResult.err.empty() ); // so we always report err dassert( !status.isOK() ); result.append( "errmsg", "timed out waiting for slaves" ); result.append( "code", status.code() ); return true; } return appendCommandStatus( result, status ); }
/** * @return true if had to do something */ bool checkShardVersion( DBClientBase * conn_in , const string& ns , ChunkManagerPtr refManager, bool authoritative , int tryNumber ) { // TODO: cache, optimize, etc... WriteBackListener::init( *conn_in ); DBConfigPtr conf = grid.getDBConfig( ns ); if ( ! conf ) return false; DBClientBase* conn = getVersionable( conn_in ); verify(conn); // errors thrown above unsigned long long officialSequenceNumber = 0; ChunkManagerPtr manager; const bool isSharded = conf->isSharded( ns ); if ( isSharded ) { manager = conf->getChunkManagerIfExists( ns , authoritative ); // It's possible the chunk manager was reset since we checked whether sharded was true, // so must check this here. if( manager ) officialSequenceNumber = manager->getSequenceNumber(); } // Check this manager against the reference manager if( isSharded && manager ){ Shard shard = Shard::make( conn->getServerAddress() ); if( refManager && ! refManager->compatibleWith( manager, shard ) ){ throw SendStaleConfigException( ns, str::stream() << "manager (" << manager->getVersion( shard ).toString() << " : " << manager->getSequenceNumber() << ") " << "not compatible with reference manager (" << refManager->getVersion( shard ).toString() << " : " << refManager->getSequenceNumber() << ") " << "on shard " << shard.getName() << " (" << shard.getAddress().toString() << ")", refManager->getVersion( shard ), manager->getVersion( shard ) ); } } else if( refManager ){ Shard shard = Shard::make( conn->getServerAddress() ); string msg( str::stream() << "not sharded (" << ( (manager.get() == 0) ? string( "<none>" ) : str::stream() << manager->getSequenceNumber() ) << ") but has reference manager (" << refManager->getSequenceNumber() << ") " << "on conn " << conn->getServerAddress() << " (" << conn_in->getServerAddress() << ")" ); throw SendStaleConfigException( ns, msg, refManager->getVersion( shard ), ChunkVersion( 0, OID() )); } // has the ChunkManager been reloaded since the last time we updated the connection-level version? // (ie., last time we issued the setShardVersions below) unsigned long long sequenceNumber = connectionShardStatus.getSequence(conn,ns); if ( sequenceNumber == officialSequenceNumber ) { return false; } ChunkVersion version = ChunkVersion( 0, OID() ); if ( isSharded && manager ) { version = manager->getVersion( Shard::make( conn->getServerAddress() ) ); } if( ! version.isSet() ){ LOG(0) << "resetting shard version of " << ns << " on " << conn->getServerAddress() << ", " << ( ! isSharded ? "no longer sharded" : ( ! manager ? "no chunk manager found" : "version is zero" ) ) << endl; } LOG(2).stream() << " have to set shard version for conn: " << conn->getServerAddress() << " ns:" << ns << " my last seq: " << sequenceNumber << " current: " << officialSequenceNumber << " version: " << version << " manager: " << manager.get() << endl; const string versionableServerAddress(conn->getServerAddress()); BSONObj result; if ( setShardVersion( *conn , ns , version , manager , authoritative , result ) ) { // success! LOG(1) << " setShardVersion success: " << result << endl; connectionShardStatus.setSequence( conn , ns , officialSequenceNumber ); return true; } LOG(1) << " setShardVersion failed!\n" << result << endl; if ( result["need_authoritative"].trueValue() ) massert( 10428 , "need_authoritative set but in authoritative mode already" , ! authoritative ); if ( ! authoritative ) { // use the original connection and get a fresh versionable connection // since conn can be invalidated (or worse, freed) after the failure checkShardVersion(conn_in, ns, refManager, 1, tryNumber + 1); return true; } if ( result["reloadConfig"].trueValue() ) { if( result["version"].timestampTime() == 0 ){ warning() << "reloading full configuration for " << conf->getName() << ", connection state indicates significant version changes" << endl; // reload db conf->reload(); } else { // reload config conf->getChunkManager( ns , true ); } } const int maxNumTries = 7; if ( tryNumber < maxNumTries ) { LOG( tryNumber < ( maxNumTries / 2 ) ? 1 : 0 ) << "going to retry checkShardVersion host: " << versionableServerAddress << " " << result << endl; sleepmillis( 10 * tryNumber ); // use the original connection and get a fresh versionable connection // since conn can be invalidated (or worse, freed) after the failure checkShardVersion(conn_in, ns, refManager, true, tryNumber + 1); return true; } string errmsg = str::stream() << "setShardVersion failed host: " << versionableServerAddress << " " << result; log() << " " << errmsg << endl; massert( 10429 , errmsg , 0 ); return true; }
/* **************************************************************************** * * mongoUnsubscribeContext - */ HttpStatusCode mongoUnsubscribeContext(UnsubscribeContextRequest* requestP, UnsubscribeContextResponse* responseP, const std::string& tenant) { bool reqSemTaken; std::string err; reqSemTake(__FUNCTION__, "ngsi10 unsubscribe request", SemWriteOp, &reqSemTaken); LM_T(LmtMongo, ("Unsubscribe Context")); /* No matter if success or failure, the subscriptionId in the response is always the one * in the request */ responseP->subscriptionId = requestP->subscriptionId; if (responseP->subscriptionId.get() == "") { responseP->statusCode.fill(SccContextElementNotFound); LM_W(("Bad Input (no subscriptionId)")); return SccOk; } /* Look for document */ BSONObj sub; OID id; try { id = OID(requestP->subscriptionId.get()); } catch (const AssertionException &e) { reqSemGive(__FUNCTION__, "ngsi10 unsubscribe request (mongo assertion exception)", reqSemTaken); // // This happens when OID format is wrong // FIXME: this checking should be done at parsing stage, without progressing to // mongoBackend. For the moment we can live this here, but we should remove in the future // (old issue #95) // responseP->statusCode.fill(SccContextElementNotFound); LM_W(("Bad Input (invalid OID format)")); return SccOk; } if (!collectionFindOne(getSubscribeContextCollectionName(tenant), BSON("_id" << id), &sub, &err)) { reqSemGive(__FUNCTION__, "ngsi10 unsubscribe request (mongo db exception)", reqSemTaken); responseP->statusCode.fill(SccReceiverInternalError, err); return SccOk; } if (sub.isEmpty()) { reqSemGive(__FUNCTION__, "ngsi10 unsubscribe request (no subscriptions found)", reqSemTaken); responseP->statusCode.fill(SccContextElementNotFound, std::string("subscriptionId: /") + requestP->subscriptionId.get() + "/"); return SccOk; } /* Remove document in MongoDB */ // FIXME: I would prefer to do the find and remove in a single operation. Is there something similar // to findAndModify for this? if (!collectionRemove(getSubscribeContextCollectionName(tenant), BSON("_id" << OID(requestP->subscriptionId.get())), &err)) { reqSemGive(__FUNCTION__, "ngsi10 unsubscribe request (mongo db exception)", reqSemTaken); responseP->statusCode.fill(SccReceiverInternalError, err); return SccOk; } /* Destroy any previous ONTIMEINTERVAL thread */ getNotifier()->destroyOntimeIntervalThreads(requestP->subscriptionId.get()); // FIXME P7: mongoSubCache stuff could be avoided if subscription is not patterned // // Removing subscription from mongo subscription cache // LM_T(LmtMongoSubCache, ("removing subscription '%s' (tenant '%s') from mongo subscription cache", requestP->subscriptionId.get().c_str(), tenant.c_str())); cacheSemTake(__FUNCTION__, "Removing subscription from cache"); CachedSubscription* cSubP = mongoSubCacheItemLookup(tenant.c_str(), requestP->subscriptionId.get().c_str()); if (cSubP != NULL) // Will only enter here if wildcard subscription { mongoSubCacheItemRemove(cSubP); } cacheSemGive(__FUNCTION__, "Removing subscription from cache"); reqSemGive(__FUNCTION__, "ngsi10 unsubscribe request", reqSemTaken); responseP->statusCode.fill(SccOk); return SccOk; }