bool WriteBatchExecutor::applyUpdate( const string& ns,
const BatchedUpdateDocument& updateOp,
CurOp* currentOp,
WriteStats* stats,
BatchedErrorDetail* error ) {
OpDebug& opDebug = currentOp->debug();
_opCounters->gotUpdate();
BSONObj queryObj = updateOp.getQuery();
BSONObj updateObj = updateOp.getUpdateExpr();
bool multi = updateOp.isMultiSet() ? updateOp.getMulti() : false;
bool upsert = updateOp.isUpsertSet() ? updateOp.getUpsert() : false;
currentOp->setQuery( queryObj );
opDebug.op = dbUpdate;
opDebug.query = queryObj;
bool resExisting = false;
long long resNum = 0;
OID resUpserted = OID();
try {
const NamespaceString requestNs( ns );
UpdateRequest request( requestNs );
request.setQuery( queryObj );
request.setUpdates( updateObj );
request.setUpsert( upsert );
request.setMulti( multi );
request.setUpdateOpLog();
UpdateResult res = update( request, &opDebug );
resExisting = res.existing;
resNum = res.numMatched;
resUpserted = res.upserted;
stats->numUpdated += !resUpserted.isSet() ? resNum : 0;
stats->numUpserted += resUpserted.isSet() ? 1 : 0;
}
catch ( const UserException& ex ) {
opDebug.exceptionInfo = ex.getInfo();
toBatchedError( ex, error );
return false;
}
_le->recordUpdate( resExisting, resNum, resUpserted );
return true;
}
void ClientCursor::updateSlaveLocation(OperationContext* txn, CurOp& curop) {
if (_slaveReadTill.isNull())
return;
verify(str::startsWith(_ns.c_str(), "local.oplog."));
Client* c = curop.getClient();
verify(c);
OID rid = c->getRemoteID();
if (!rid.isSet())
return;
repl::getGlobalReplicationCoordinator()->setLastOptimeForSlave(rid, _slaveReadTill);
}
void WriteBackManager::queueWriteBack( const string& remote , const BSONObj& o ) {
static mongo::mutex xxx( "WriteBackManager::queueWriteBack tmp" );
static OID lastOID;
scoped_lock lk( xxx );
const BSONElement& e = o["id"];
if ( lastOID.isSet() ) {
if ( e.OID() < lastOID ) {
log() << "this could fail" << endl;
printStackTrace();
}
}
lastOID = e.OID();
getWritebackQueue( remote )->queue.push( o );
}
void CollectionType::parseBSON(BSONObj source) {
clear();
bool ok = true;
ok &= FieldParser::extract(source, ns, "", &_ns);
ok &= FieldParser::extract(source, primary, "", &_primary);
ok &= FieldParser::extract(source, keyPattern, BSONObj(), &_keyPattern);
ok &= FieldParser::extract(source, unique, false, &_unique);
ok &= FieldParser::extract(source, createdAt, 0ULL, &_createdAt);
ok &= FieldParser::extract(source, noBalance, false, &_noBalance);
ok &= FieldParser::extract(source, epoch, OID(), &_epoch);
if (! ok) {
clear();
return;
}
//
// backward compatibility
//
// 'createAt' used to be called 'lastmod' up to 2.2.
Date_t lastmod;
if (! FieldParser::extract(source, DEPRECATED_lastmod, 0ULL, &lastmod)) {
clear();
return;
}
else if (lastmod != 0ULL) {
_createdAt = lastmod;
}
// There was a flag to mark a collection as loggically dropped, up to 2.2.
bool dropped;
if (! FieldParser::extract(source, DEPRECATED_dropped, false, &dropped) || dropped) {
clear();
return;
}
// 'lastmodEpoch' was a transition format to 'epoch', up to 2.2
OID lastmodEpoch;
if (! FieldParser::extract(source, DEPRECATED_lastmodEpoch, OID(), &lastmodEpoch)) {
clear();
return;
}
else if (lastmodEpoch.isSet()) {
_epoch = lastmodEpoch;
}
}
bool mergeChunks( OperationContext* txn,
const NamespaceString& nss,
const BSONObj& minKey,
const BSONObj& maxKey,
const OID& epoch,
string* errMsg ) {
//
// Get sharding state up-to-date
//
ConnectionString configLoc = ConnectionString::parse( shardingState.getConfigServer(),
*errMsg );
if ( !configLoc.isValid() ){
warning() << *errMsg << endl;
return false;
}
//
// Get the distributed lock
//
ScopedDistributedLock collLock( configLoc, nss.ns() );
collLock.setLockMessage( stream() << "merging chunks in " << nss.ns() << " from "
<< minKey << " to " << maxKey );
Status acquisitionStatus = collLock.tryAcquire();
if (!acquisitionStatus.isOK()) {
*errMsg = stream() << "could not acquire collection lock for " << nss.ns()
<< " to merge chunks in [" << minKey << "," << maxKey << ")"
<< causedBy(acquisitionStatus);
warning() << *errMsg << endl;
return false;
}
//
// We now have the collection lock, refresh metadata to latest version and sanity check
//
ChunkVersion shardVersion;
Status status = shardingState.refreshMetadataNow(txn, nss.ns(), &shardVersion);
if ( !status.isOK() ) {
*errMsg = str::stream() << "could not merge chunks, failed to refresh metadata for "
<< nss.ns() << causedBy( status.reason() );
warning() << *errMsg << endl;
return false;
}
if ( epoch.isSet() && shardVersion.epoch() != epoch ) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " has changed" << " since merge was sent" << "(sent epoch : "
<< epoch.toString()
<< ", current epoch : " << shardVersion.epoch().toString() << ")";
warning() << *errMsg << endl;
return false;
}
CollectionMetadataPtr metadata = shardingState.getCollectionMetadata( nss.ns() );
if ( !metadata || metadata->getKeyPattern().isEmpty() ) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " is not sharded";
warning() << *errMsg << endl;
return false;
}
dassert( metadata->getShardVersion().equals( shardVersion ) );
if ( !metadata->isValidKey( minKey ) || !metadata->isValidKey( maxKey ) ) {
*errMsg = stream() << "could not merge chunks, the range "
<< rangeToString( minKey, maxKey ) << " is not valid"
<< " for collection " << nss.ns() << " with key pattern "
<< metadata->getKeyPattern();
warning() << *errMsg << endl;
return false;
}
//
// Get merged chunk information
//
ChunkVersion mergeVersion = metadata->getCollVersion();
mergeVersion.incMinor();
OwnedPointerVector<ChunkType> chunksToMerge;
ChunkType itChunk;
itChunk.setMin( minKey );
itChunk.setMax( minKey );
itChunk.setNS( nss.ns() );
itChunk.setShard( shardingState.getShardName() );
while ( itChunk.getMax().woCompare( maxKey ) < 0 &&
metadata->getNextChunk( itChunk.getMax(), &itChunk ) ) {
auto_ptr<ChunkType> saved( new ChunkType );
itChunk.cloneTo( saved.get() );
chunksToMerge.mutableVector().push_back( saved.release() );
}
if ( chunksToMerge.empty() ) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " range starting at " << minKey
<< " and ending at " << maxKey
<< " does not belong to shard " << shardingState.getShardName();
warning() << *errMsg << endl;
return false;
}
//
// Validate the range starts and ends at chunks and has no holes, error if not valid
//
BSONObj firstDocMin = ( *chunksToMerge.begin() )->getMin();
BSONObj firstDocMax = ( *chunksToMerge.begin() )->getMax();
// minKey is inclusive
bool minKeyInRange = rangeContains( firstDocMin, firstDocMax, minKey );
if ( !minKeyInRange ) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " range starting at " << minKey
<< " does not belong to shard " << shardingState.getShardName();
warning() << *errMsg << endl;
return false;
}
BSONObj lastDocMin = ( *chunksToMerge.rbegin() )->getMin();
BSONObj lastDocMax = ( *chunksToMerge.rbegin() )->getMax();
// maxKey is exclusive
bool maxKeyInRange = lastDocMin.woCompare( maxKey ) < 0 &&
lastDocMax.woCompare( maxKey ) >= 0;
if ( !maxKeyInRange ) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " range ending at " << maxKey
<< " does not belong to shard " << shardingState.getShardName();
warning() << *errMsg << endl;
return false;
}
bool validRangeStartKey = firstDocMin.woCompare( minKey ) == 0;
bool validRangeEndKey = lastDocMax.woCompare( maxKey ) == 0;
if ( !validRangeStartKey || !validRangeEndKey ) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " does not contain a chunk "
<< ( !validRangeStartKey ? "starting at " + minKey.toString() : "" )
<< ( !validRangeStartKey && !validRangeEndKey ? " or " : "" )
<< ( !validRangeEndKey ? "ending at " + maxKey.toString() : "" );
warning() << *errMsg << endl;
return false;
}
if ( chunksToMerge.size() == 1 ) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " already contains chunk for " << rangeToString( minKey, maxKey );
warning() << *errMsg << endl;
return false;
}
bool holeInRange = false;
// Look for hole in range
ChunkType* prevChunk = *chunksToMerge.begin();
ChunkType* nextChunk = NULL;
for ( OwnedPointerVector<ChunkType>::const_iterator it = chunksToMerge.begin();
it != chunksToMerge.end(); ++it ) {
if ( it == chunksToMerge.begin() ) continue;
nextChunk = *it;
if ( prevChunk->getMax().woCompare( nextChunk->getMin() ) != 0 ) {
holeInRange = true;
break;
}
prevChunk = nextChunk;
}
if ( holeInRange ) {
dassert( NULL != nextChunk );
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " has a hole in the range " << rangeToString( minKey, maxKey )
<< " at " << rangeToString( prevChunk->getMax(),
nextChunk->getMin() );
warning() << *errMsg << endl;
return false;
}
//
// Run apply ops command
//
BSONObj applyOpsCmd = buildApplyOpsCmd( chunksToMerge,
shardVersion,
mergeVersion );
bool ok;
BSONObj result;
try {
ScopedDbConnection conn( configLoc, 30.0 );
ok = conn->runCommand( "config", applyOpsCmd, result );
if ( !ok ) *errMsg = result.toString();
conn.done();
}
catch( const DBException& ex ) {
ok = false;
*errMsg = ex.toString();
}
if ( !ok ) {
*errMsg = stream() << "could not merge chunks for " << nss.ns()
<< ", writing to config failed" << causedBy( errMsg );
warning() << *errMsg << endl;
return false;
}
//
// Install merged chunk metadata
//
{
Lock::DBLock writeLk(txn->lockState(), nss.db(), newlm::MODE_X);
shardingState.mergeChunks(txn, nss.ns(), minKey, maxKey, mergeVersion);
}
//
// Log change
//
BSONObj mergeLogEntry = buildMergeLogEntry( chunksToMerge,
shardVersion,
mergeVersion );
configServer.logChange( "merge", nss.ns(), mergeLogEntry );
return true;
}
/**
* Upgrade v3 to v4 described here.
*
* This upgrade takes a config server without collection epochs (potentially) and adds
* epochs to all mongo processes.
*
*/
bool doUpgradeV3ToV4(const ConnectionString& configLoc,
const VersionType& lastVersionInfo,
string* errMsg)
{
string dummy;
if (!errMsg) errMsg = &dummy;
verify(lastVersionInfo.getCurrentVersion() == UpgradeHistory_NoEpochVersion);
if (lastVersionInfo.isUpgradeIdSet() && lastVersionInfo.getUpgradeId().isSet()) {
//
// Another upgrade failed, so cleanup may be necessary
//
BSONObj lastUpgradeState = lastVersionInfo.getUpgradeState();
bool inCriticalSection;
if (!FieldParser::extract(lastUpgradeState,
inCriticalSectionField,
&inCriticalSection,
errMsg))
{
*errMsg = stream() << "problem reading previous upgrade state" << causedBy(errMsg);
return false;
}
if (inCriticalSection) {
// Manual intervention is needed here. Somehow our upgrade didn't get applied
// consistently across config servers.
*errMsg = cannotCleanupMessage;
return false;
}
if (!_cleanupUpgradeState(configLoc, lastVersionInfo.getUpgradeId(), errMsg)) {
// If we can't cleanup the old upgrade state, the user might have done it for us,
// not a fatal problem (we'll just end up with extra collections).
warning() << "could not cleanup previous upgrade state" << causedBy(errMsg) << endl;
*errMsg = "";
}
}
//
// Check the versions of other mongo processes in the cluster before upgrade.
// We can't upgrade if there are active pre-v2.2 processes in the cluster
//
Status mongoVersionStatus = checkClusterMongoVersions(configLoc,
string(minMongoProcessVersion));
if (!mongoVersionStatus.isOK()) {
*errMsg = stream() << "cannot upgrade with pre-v" << minMongoProcessVersion
<< " mongo processes active in the cluster"
<< causedBy(mongoVersionStatus);
return false;
}
VersionType newVersionInfo;
lastVersionInfo.cloneTo(&newVersionInfo);
// Set our upgrade id and state
OID upgradeId = OID::gen();
newVersionInfo.setUpgradeId(upgradeId);
newVersionInfo.setUpgradeState(BSONObj());
// Write our upgrade id and state
{
scoped_ptr<ScopedDbConnection> connPtr;
try {
connPtr.reset(ScopedDbConnection::getInternalScopedDbConnection(configLoc, 30));
ScopedDbConnection& conn = *connPtr;
verify(newVersionInfo.isValid(NULL));
conn->update(VersionType::ConfigNS,
BSON("_id" << 1 << VersionType::version_DEPRECATED(3)),
newVersionInfo.toBSON());
_checkGLE(conn);
}
catch (const DBException& e) {
*errMsg = stream() << "could not initialize version info for upgrade"
<< causedBy(e);
return false;
}
connPtr->done();
}
//
// First lock all collection namespaces that exist
//
OwnedPointerMap<string, CollectionType> ownedCollections;
const map<string, CollectionType*>& collections = ownedCollections.map();
Status findCollectionsStatus = findAllCollectionsV3(configLoc, &ownedCollections);
if (!findCollectionsStatus.isOK()) {
*errMsg = stream() << "could not read collections from config server"
<< causedBy(findCollectionsStatus);
return false;
}
//
// Acquire locks for all sharded collections
// Something that didn't involve getting thousands of locks would be better.
//
OwnedPointerVector<ScopedDistributedLock> collectionLocks;
log() << "acquiring locks for " << collections.size() << " sharded collections..." << endl;
// WARNING - this string is used programmatically when forcing locks, be careful when
// changing!
// TODO: Add programmatic "why" field to lock collection
string lockMessage = str::stream() << "ensuring epochs for config upgrade"
<< " (" << upgradeId.toString() << ")";
if (!_acquireAllCollectionLocks(configLoc,
collections,
lockMessage,
20 * 60 * 1000,
&collectionLocks,
errMsg))
{
*errMsg = stream() << "could not acquire all namespace locks for upgrade"
<< " (" << upgradeId.toString() << ")"
<< causedBy(errMsg);
return false;
}
// We are now preventing all splits and migrates for all sharded collections
// Get working and backup suffixes
string workingSuffix = genWorkingSuffix(upgradeId);
string backupSuffix = genBackupSuffix(upgradeId);
log() << "copying collection and chunk metadata to working and backup collections..."
<< endl;
// Get a backup and working copy of the config.collections and config.chunks collections
Status copyStatus = copyFrozenCollection(configLoc,
CollectionType::ConfigNS,
CollectionType::ConfigNS + workingSuffix);
if (!copyStatus.isOK()) {
*errMsg = stream() << "could not copy " << CollectionType::ConfigNS << " to "
<< (CollectionType::ConfigNS + workingSuffix)
<< causedBy(copyStatus);
return false;
}
copyStatus = copyFrozenCollection(configLoc,
CollectionType::ConfigNS,
CollectionType::ConfigNS + backupSuffix);
if (!copyStatus.isOK()) {
*errMsg = stream() << "could not copy " << CollectionType::ConfigNS << " to "
<< (CollectionType::ConfigNS + backupSuffix) << causedBy(copyStatus);
return false;
}
copyStatus = copyFrozenCollection(configLoc,
ChunkType::ConfigNS,
ChunkType::ConfigNS + workingSuffix);
if (!copyStatus.isOK()) {
*errMsg = stream() << "could not copy " << ChunkType::ConfigNS << " to "
<< (ChunkType::ConfigNS + workingSuffix) << causedBy(copyStatus);
return false;
}
copyStatus = copyFrozenCollection(configLoc,
ChunkType::ConfigNS,
ChunkType::ConfigNS + backupSuffix);
if (!copyStatus.isOK()) {
*errMsg = stream() << "could not copy " << ChunkType::ConfigNS << " to "
<< (ChunkType::ConfigNS + backupSuffix) << causedBy(copyStatus);
return false;
}
//
// Go through sharded collections one-by-one and add epochs where missing
//
for (map<string, CollectionType*>::const_iterator it = collections.begin();
it != collections.end(); ++it)
{
// Create a copy so that we can change the epoch later
CollectionType collection;
it->second->cloneTo(&collection);
log() << "checking epochs for " << collection.getNS() << " collection..." << endl;
OID epoch = collection.getEpoch();
//
// Go through chunks to find epoch if we haven't found it or to verify epoch is the same
//
OwnedPointerVector<ChunkType> ownedChunks;
const vector<ChunkType*>& chunks = ownedChunks.vector();
Status findChunksStatus = findAllChunks(configLoc, collection.getNS(), &ownedChunks);
if (!findChunksStatus.isOK()) {
*errMsg = stream() << "could not read chunks from config server"
<< causedBy(findChunksStatus);
return false;
}
for (vector<ChunkType*>::const_iterator chunkIt = chunks.begin();
chunkIt != chunks.end(); ++chunkIt)
{
const ChunkType& chunk = *(*chunkIt);
// If our chunk epoch is set and doesn't match
if (epoch.isSet() && chunk.getVersion().epoch().isSet()
&& chunk.getVersion().epoch() != epoch)
{
*errMsg = stream() << "chunk epoch for " << chunk.toString() << " in "
<< collection.getNS() << " does not match found epoch "
<< epoch;
return false;
}
else if (!epoch.isSet() && chunk.getVersion().epoch().isSet()) {
epoch = chunk.getVersion().epoch();
}
}
//
// Write collection epoch if needed
//
if (!collection.getEpoch().isSet()) {
OID newEpoch = OID::gen();
log() << "writing new epoch " << newEpoch << " for " << collection.getNS()
<< " collection..." << endl;
scoped_ptr<ScopedDbConnection> connPtr;
try {
connPtr.reset(ScopedDbConnection::getInternalScopedDbConnection(configLoc, 30));
ScopedDbConnection& conn = *connPtr;
conn->update(CollectionType::ConfigNS + workingSuffix,
BSON(CollectionType::ns(collection.getNS())),
BSON("$set" << BSON(CollectionType::DEPRECATED_lastmodEpoch(newEpoch))));
_checkGLE(conn);
}
catch (const DBException& e) {
*errMsg = stream() << "could not write a new epoch for " << collection.getNS()
<< causedBy(e);
return false;
}
connPtr->done();
collection.setEpoch(newEpoch);
}
epoch = collection.getEpoch();
verify(epoch.isSet());
//
// Now write verified epoch to all chunks
//
log() << "writing epoch " << epoch << " for " << chunks.size() << " chunks in "
<< collection.getNS() << " collection..." << endl;
{
scoped_ptr<ScopedDbConnection> connPtr;
try {
connPtr.reset(ScopedDbConnection::getInternalScopedDbConnection(configLoc, 30));
ScopedDbConnection& conn = *connPtr;
// Multi-update of all chunks
conn->update(ChunkType::ConfigNS + workingSuffix,
BSON(ChunkType::ns(collection.getNS())),
BSON("$set" << BSON(ChunkType::DEPRECATED_epoch(epoch))),
false,
true); // multi
_checkGLE(conn);
}
catch (const DBException& e) {
*errMsg = stream() << "could not write a new epoch " << epoch.toString()
<< " for chunks in " << collection.getNS() << causedBy(e);
return false;
}
connPtr->done();
}
}
//
// Paranoid verify the collection writes
//
{
scoped_ptr<ScopedDbConnection> connPtr;
try {
connPtr.reset(ScopedDbConnection::getInternalScopedDbConnection(configLoc, 30));
ScopedDbConnection& conn = *connPtr;
// Find collections with no epochs
BSONObj emptyDoc =
conn->findOne(CollectionType::ConfigNS + workingSuffix,
BSON("$unset" << BSON(CollectionType::DEPRECATED_lastmodEpoch() << 1)));
if (!emptyDoc.isEmpty()) {
*errMsg = stream() << "collection " << emptyDoc
<< " is still missing epoch after config upgrade";
connPtr->done();
return false;
}
// Find collections with empty epochs
emptyDoc = conn->findOne(CollectionType::ConfigNS + workingSuffix,
BSON(CollectionType::DEPRECATED_lastmodEpoch(OID())));
if (!emptyDoc.isEmpty()) {
*errMsg = stream() << "collection " << emptyDoc
<< " still has empty epoch after config upgrade";
connPtr->done();
return false;
}
// Find chunks with no epochs
emptyDoc =
conn->findOne(ChunkType::ConfigNS + workingSuffix,
BSON("$unset" << BSON(ChunkType::DEPRECATED_epoch() << 1)));
if (!emptyDoc.isEmpty()) {
*errMsg = stream() << "chunk " << emptyDoc
<< " is still missing epoch after config upgrade";
connPtr->done();
return false;
}
// Find chunks with empty epochs
emptyDoc = conn->findOne(ChunkType::ConfigNS + workingSuffix,
BSON(ChunkType::DEPRECATED_epoch(OID())));
if (!emptyDoc.isEmpty()) {
*errMsg = stream() << "chunk " << emptyDoc
<< " still has empty epoch after config upgrade";
connPtr->done();
return false;
}
}
catch (const DBException& e) {
*errMsg = stream() << "could not verify epoch writes" << causedBy(e);
return false;
}
connPtr->done();
}
//
// Double check that our collections haven't changed
//
Status idCheckStatus = checkIdsTheSame(configLoc,
CollectionType::ConfigNS,
CollectionType::ConfigNS + workingSuffix);
if (!idCheckStatus.isOK()) {
*errMsg = stream() << CollectionType::ConfigNS
<< " was modified while working on upgrade"
<< causedBy(idCheckStatus);
return false;
}
idCheckStatus = checkIdsTheSame(configLoc,
ChunkType::ConfigNS,
ChunkType::ConfigNS + workingSuffix);
if (!idCheckStatus.isOK()) {
*errMsg = stream() << ChunkType::ConfigNS << " was modified while working on upgrade"
<< causedBy(idCheckStatus);
return false;
}
//
// ENTER CRITICAL SECTION
//
newVersionInfo.setUpgradeState(BSON(inCriticalSectionField(true)));
{
scoped_ptr<ScopedDbConnection> connPtr;
try {
connPtr.reset(ScopedDbConnection::getInternalScopedDbConnection(configLoc, 30));
ScopedDbConnection& conn = *connPtr;
verify(newVersionInfo.isValid(NULL));
conn->update(VersionType::ConfigNS,
BSON("_id" << 1 << VersionType::version_DEPRECATED(3)),
newVersionInfo.toBSON());
_checkGLE(conn);
}
catch (const DBException& e) {
// No cleanup message here since we're not sure if we wrote or not, and
// not dangerous either way except to prevent further updates (at which point
// the message is printed)
*errMsg = stream()
<< "could not update version info to enter critical update section"
<< causedBy(e);
return false;
}
// AT THIS POINT ANY FAILURE REQUIRES MANUAL INTERVENTION!
connPtr->done();
}
log() << "entered critical section for config upgrade" << endl;
Status overwriteStatus = overwriteCollection(configLoc,
CollectionType::ConfigNS + workingSuffix,
CollectionType::ConfigNS);
if (!overwriteStatus.isOK()) {
error() << cleanupMessage << endl;
*errMsg = stream() << "could not overwrite collection " << CollectionType::ConfigNS
<< " with working collection "
<< (CollectionType::ConfigNS + workingSuffix)
<< causedBy(overwriteStatus);
return false;
}
overwriteStatus = overwriteCollection(configLoc,
ChunkType::ConfigNS + workingSuffix,
ChunkType::ConfigNS);
if (!overwriteStatus.isOK()) {
error() << cleanupMessage << endl;
*errMsg = stream() << "could not overwrite collection " << ChunkType::ConfigNS
<< " with working collection "
<< (ChunkType::ConfigNS + workingSuffix)
<< causedBy(overwriteStatus);
return false;
}
//
// Finally update the version to latest and add clusterId to version
//
OID newClusterId = OID::gen();
// Note: hardcoded versions, since this is a very particular upgrade
// Note: DO NOT CLEAR the config version unless bumping the minCompatibleVersion,
// we want to save the excludes that were set.
newVersionInfo.setMinCompatibleVersion(UpgradeHistory_NoEpochVersion);
newVersionInfo.setCurrentVersion(UpgradeHistory_MandatoryEpochVersion);
newVersionInfo.setClusterId(newClusterId);
// Leave critical section
newVersionInfo.unsetUpgradeId();
newVersionInfo.unsetUpgradeState();
log() << "writing new version info and clusterId " << newClusterId << "..." << endl;
{
scoped_ptr<ScopedDbConnection> connPtr;
try {
connPtr.reset(ScopedDbConnection::getInternalScopedDbConnection(configLoc, 30));
ScopedDbConnection& conn = *connPtr;
verify(newVersionInfo.isValid(NULL));
conn->update(VersionType::ConfigNS,
BSON("_id" << 1 << VersionType::version_DEPRECATED(UpgradeHistory_NoEpochVersion)),
newVersionInfo.toBSON());
_checkGLE(conn);
}
catch (const DBException& e) {
error() << cleanupMessage << endl;
*errMsg = stream() << "could not write new version info "
<< "and exit critical upgrade section" << causedBy(e);
return false;
}
connPtr->done();
}
//
// END CRITICAL SECTION
//
return true;
}
void LocksType::setLockID(const OID& lockID) {
invariant(lockID.isSet());
_lockID = lockID;
}
bool mergeChunks(OperationContext* txn,
const NamespaceString& nss,
const BSONObj& minKey,
const BSONObj& maxKey,
const OID& epoch,
string* errMsg) {
// Get the distributed lock
string whyMessage = stream() << "merging chunks in " << nss.ns() << " from " << minKey << " to "
<< maxKey;
auto scopedDistLock = grid.catalogManager(txn)->distLock(
txn, nss.ns(), whyMessage, DistLockManager::kSingleLockAttemptTimeout);
if (!scopedDistLock.isOK()) {
*errMsg = stream() << "could not acquire collection lock for " << nss.ns()
<< " to merge chunks in [" << minKey << "," << maxKey << ")"
<< causedBy(scopedDistLock.getStatus());
warning() << *errMsg;
return false;
}
ShardingState* shardingState = ShardingState::get(txn);
//
// We now have the collection lock, refresh metadata to latest version and sanity check
//
ChunkVersion shardVersion;
Status status = shardingState->refreshMetadataNow(txn, nss.ns(), &shardVersion);
if (!status.isOK()) {
*errMsg = str::stream() << "could not merge chunks, failed to refresh metadata for "
<< nss.ns() << causedBy(status.reason());
warning() << *errMsg;
return false;
}
if (epoch.isSet() && shardVersion.epoch() != epoch) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns() << " has changed"
<< " since merge was sent"
<< "(sent epoch : " << epoch.toString()
<< ", current epoch : " << shardVersion.epoch().toString() << ")";
warning() << *errMsg;
return false;
}
shared_ptr<CollectionMetadata> metadata = shardingState->getCollectionMetadata(nss.ns());
if (!metadata || metadata->getKeyPattern().isEmpty()) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " is not sharded";
warning() << *errMsg;
return false;
}
dassert(metadata->getShardVersion().equals(shardVersion));
if (!metadata->isValidKey(minKey) || !metadata->isValidKey(maxKey)) {
*errMsg = stream() << "could not merge chunks, the range " << rangeToString(minKey, maxKey)
<< " is not valid"
<< " for collection " << nss.ns() << " with key pattern "
<< metadata->getKeyPattern();
warning() << *errMsg;
return false;
}
//
// Get merged chunk information
//
ChunkVersion mergeVersion = metadata->getCollVersion();
mergeVersion.incMinor();
std::vector<ChunkType> chunksToMerge;
ChunkType itChunk;
itChunk.setMin(minKey);
itChunk.setMax(minKey);
itChunk.setNS(nss.ns());
itChunk.setShard(shardingState->getShardName());
while (itChunk.getMax().woCompare(maxKey) < 0 &&
metadata->getNextChunk(itChunk.getMax(), &itChunk)) {
chunksToMerge.push_back(itChunk);
}
if (chunksToMerge.empty()) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " range starting at " << minKey << " and ending at " << maxKey
<< " does not belong to shard " << shardingState->getShardName();
warning() << *errMsg;
return false;
}
//
// Validate the range starts and ends at chunks and has no holes, error if not valid
//
BSONObj firstDocMin = chunksToMerge.front().getMin();
BSONObj firstDocMax = chunksToMerge.front().getMax();
// minKey is inclusive
bool minKeyInRange = rangeContains(firstDocMin, firstDocMax, minKey);
if (!minKeyInRange) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " range starting at " << minKey << " does not belong to shard "
<< shardingState->getShardName();
warning() << *errMsg;
return false;
}
BSONObj lastDocMin = chunksToMerge.back().getMin();
BSONObj lastDocMax = chunksToMerge.back().getMax();
// maxKey is exclusive
bool maxKeyInRange = lastDocMin.woCompare(maxKey) < 0 && lastDocMax.woCompare(maxKey) >= 0;
if (!maxKeyInRange) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " range ending at " << maxKey << " does not belong to shard "
<< shardingState->getShardName();
warning() << *errMsg;
return false;
}
bool validRangeStartKey = firstDocMin.woCompare(minKey) == 0;
bool validRangeEndKey = lastDocMax.woCompare(maxKey) == 0;
if (!validRangeStartKey || !validRangeEndKey) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " does not contain a chunk "
<< (!validRangeStartKey ? "starting at " + minKey.toString() : "")
<< (!validRangeStartKey && !validRangeEndKey ? " or " : "")
<< (!validRangeEndKey ? "ending at " + maxKey.toString() : "");
warning() << *errMsg;
return false;
}
if (chunksToMerge.size() == 1) {
*errMsg = stream() << "could not merge chunks, collection " << nss.ns()
<< " already contains chunk for " << rangeToString(minKey, maxKey);
warning() << *errMsg;
return false;
}
// Look for hole in range
for (size_t i = 1; i < chunksToMerge.size(); ++i) {
if (chunksToMerge[i - 1].getMax().woCompare(chunksToMerge[i].getMin()) != 0) {
*errMsg =
stream() << "could not merge chunks, collection " << nss.ns()
<< " has a hole in the range " << rangeToString(minKey, maxKey) << " at "
<< rangeToString(chunksToMerge[i - 1].getMax(), chunksToMerge[i].getMin());
warning() << *errMsg;
return false;
}
}
//
// Run apply ops command
//
Status applyOpsStatus = runApplyOpsCmd(txn, chunksToMerge, shardVersion, mergeVersion);
if (!applyOpsStatus.isOK()) {
warning() << applyOpsStatus;
return false;
}
//
// Install merged chunk metadata
//
{
ScopedTransaction transaction(txn, MODE_IX);
Lock::DBLock writeLk(txn->lockState(), nss.db(), MODE_IX);
Lock::CollectionLock collLock(txn->lockState(), nss.ns(), MODE_X);
shardingState->mergeChunks(txn, nss.ns(), minKey, maxKey, mergeVersion);
}
//
// Log change
//
BSONObj mergeLogEntry = buildMergeLogEntry(chunksToMerge, shardVersion, mergeVersion);
grid.catalogManager(txn)->logChange(txn, "merge", nss.ns(), mergeLogEntry);
return true;
}
void ShardCollectionType::setEpoch(const OID& epoch) {
invariant(epoch.isSet());
_epoch = epoch;
}
void WriteBackListener::run(){
OID lastID;
lastID.clear();
int secsToSleep = 0;
while ( ! inShutdown() && Shard::isMember( _addr ) ){
if ( lastID.isSet() ){
scoped_lock lk( _seenWritebacksLock );
_seenWritebacks.insert( lastID );
lastID.clear();
}
try {
ScopedDbConnection conn( _addr );
BSONObj result;
{
BSONObjBuilder cmd;
cmd.appendOID( "writebacklisten" , &serverID ); // Command will block for data
if ( ! conn->runCommand( "admin" , cmd.obj() , result ) ){
log() << "writebacklisten command failed! " << result << endl;
conn.done();
continue;
}
}
log(1) << "writebacklisten result: " << result << endl;
BSONObj data = result.getObjectField( "data" );
if ( data.getBoolField( "writeBack" ) ){
string ns = data["ns"].valuestrsafe();
{
BSONElement e = data["id"];
if ( e.type() == jstOID )
lastID = e.OID();
}
int len;
Message m( (void*)data["msg"].binData( len ) , false );
massert( 10427 , "invalid writeback message" , m.header()->valid() );
DBConfigPtr db = grid.getDBConfig( ns );
ShardChunkVersion needVersion( data["version"] );
log(1) << "writeback id: " << lastID << " needVersion : " << needVersion.toString()
<< " mine : " << db->getChunkManager( ns )->getVersion().toString() << endl;// TODO change to log(3)
if ( logLevel ) log(1) << debugString( m ) << endl;
if ( needVersion.isSet() && needVersion <= db->getChunkManager( ns )->getVersion() ){
// this means when the write went originally, the version was old
// if we're here, it means we've already updated the config, so don't need to do again
//db->getChunkManager( ns , true ); // SERVER-1349
}
else {
// we received a writeback object that was sent to a previous version of a shard
// the actual shard may not have the object the writeback operation is for
// we need to reload the chunk manager and get the new shard versions
db->getChunkManager( ns , true );
}
Request r( m , 0 );
r.init();
r.process();
}
else if ( result["noop"].trueValue() ){
// no-op
}
else {
log() << "unknown writeBack result: " << result << endl;
}
conn.done();
secsToSleep = 0;
continue;
}
catch ( std::exception e ){
if ( inShutdown() ){
// we're shutting down, so just clean up
return;
}
log() << "WriteBackListener exception : " << e.what() << endl;
// It's possible this shard was removed
Shard::reloadShardInfo();
}
catch ( ... ){
log() << "WriteBackListener uncaught exception!" << endl;
}
secsToSleep++;
sleepsecs(secsToSleep);
if ( secsToSleep > 10 )
secsToSleep = 0;
}
log() << "WriteBackListener exiting : address no longer in cluster " << _addr;
}
