bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
if ( ! okForConfigChanges( errmsg ) )
return false;
ShardConnection::sync();
Timer t;
string ns = cmdObj.firstElement().valuestrsafe();
if ( ns.size() == 0 ) {
errmsg = "no ns";
return false;
}
DBConfigPtr config = grid.getDBConfig( ns );
if ( ! config->isSharded( ns ) ) {
config->reload();
if ( ! config->isSharded( ns ) ) {
errmsg = "ns not sharded. have to shard before we can move a chunk";
return false;
}
}
BSONObj find = cmdObj.getObjectField( "find" );
if ( find.isEmpty() ) {
errmsg = "need to specify find. see help";
return false;
}
string toString = cmdObj["to"].valuestrsafe();
if ( ! toString.size() ) {
errmsg = "you have to specify where you want to move the chunk";
return false;
}
Shard to = Shard::make( toString );
// so far, chunk size serves test purposes; it may or may not become a supported parameter
long long maxChunkSizeBytes = cmdObj["maxChunkSizeBytes"].numberLong();
if ( maxChunkSizeBytes == 0 ) {
maxChunkSizeBytes = Chunk::MaxChunkSize;
}
tlog() << "CMD: movechunk: " << cmdObj << endl;
ChunkManagerPtr info = config->getChunkManager( ns );
ChunkPtr c = info->findChunk( find );
const Shard& from = c->getShard();
if ( from == to ) {
errmsg = "that chunk is already on that shard";
return false;
}
BSONObj res;
if ( ! c->moveAndCommit( to , maxChunkSizeBytes , res ) ) {
errmsg = "move failed";
result.append( "cause" , res );
return false;
}
// preemptively reload the config to get new version info
config->getChunkManager( ns , true );
result.append( "millis" , t.millis() );
return true;
}
bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
if ( ! okForConfigChanges( errmsg ) )
return false;
ShardConnection::sync();
string ns = cmdObj.firstElement().valuestrsafe();
if ( ns.size() == 0 ) {
errmsg = "no ns";
return false;
}
DBConfigPtr config = grid.getDBConfig( ns );
if ( ! config->isSharded( ns ) ) {
config->reload();
if ( ! config->isSharded( ns ) ) {
errmsg = "ns not sharded. have to shard before can split";
return false;
}
}
BSONObj find = cmdObj.getObjectField( "find" );
if ( find.isEmpty() ) {
find = cmdObj.getObjectField( "middle" );
if ( find.isEmpty() ) {
errmsg = "need to specify find or middle";
return false;
}
}
ChunkManagerPtr info = config->getChunkManager( ns );
ChunkPtr chunk = info->findChunk( find );
BSONObj middle = cmdObj.getObjectField( "middle" );
assert( chunk.get() );
log() << "splitting: " << ns << " shard: " << chunk << endl;
BSONObj res;
bool worked;
if ( middle.isEmpty() ) {
BSONObj ret = chunk->singleSplit( true /* force a split even if not enough data */ , res );
worked = !ret.isEmpty();
}
else {
// sanity check if the key provided is a valid split point
if ( ( middle == chunk->getMin() ) || ( middle == chunk->getMax() ) ) {
errmsg = "cannot split on initial or final chunk's key";
return false;
}
if (!fieldsMatch(middle, info->getShardKey().key())){
errmsg = "middle has different fields (or different order) than shard key";
return false;
}
vector<BSONObj> splitPoints;
splitPoints.push_back( middle );
worked = chunk->multiSplit( splitPoints , res );
}
if ( !worked ) {
errmsg = "split failed";
result.append( "cause" , res );
return false;
}
config->getChunkManager( ns , true );
return true;
}
/**
* @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...
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, 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, 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;
}
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;
}
int Balancer::_moveChunks(const vector<CandidateChunkPtr>* candidateChunks,
const WriteConcernOptions* writeConcern,
bool waitForDelete) {
int movedCount = 0;
for (vector<CandidateChunkPtr>::const_iterator it = candidateChunks->begin();
it != candidateChunks->end();
++it) {
// If the balancer was disabled since we started this round, don't start new
// chunks moves.
SettingsType balancerConfig;
std::string errMsg;
if (!grid.getBalancerSettings(&balancerConfig, &errMsg)) {
warning() << errMsg;
// No point in continuing the round if the config servers are unreachable.
return movedCount;
}
if ((balancerConfig.isKeySet() && // balancer config doc exists
!grid.shouldBalance(balancerConfig)) ||
MONGO_FAIL_POINT(skipBalanceRound)) {
LOG(1) << "Stopping balancing round early as balancing was disabled";
return movedCount;
}
// Changes to metadata, borked metadata, and connectivity problems between shards should
// cause us to abort this chunk move, but shouldn't cause us to abort the entire round
// of chunks.
// TODO(spencer): We probably *should* abort the whole round on issues communicating
// with the config servers, but its impossible to distinguish those types of failures
// at the moment.
// TODO: Handle all these things more cleanly, since they're expected problems
const CandidateChunk& chunkInfo = *it->get();
try {
DBConfigPtr cfg = grid.getDBConfig(chunkInfo.ns);
verify(cfg);
// NOTE: We purposely do not reload metadata here, since _doBalanceRound already
// tried to do so once.
ChunkManagerPtr cm = cfg->getChunkManager(chunkInfo.ns);
verify(cm);
ChunkPtr c = cm->findIntersectingChunk(chunkInfo.chunk.min);
if (c->getMin().woCompare(chunkInfo.chunk.min) ||
c->getMax().woCompare(chunkInfo.chunk.max)) {
// likely a split happened somewhere
cm = cfg->getChunkManager(chunkInfo.ns, true /* reload */);
verify(cm);
c = cm->findIntersectingChunk(chunkInfo.chunk.min);
if (c->getMin().woCompare(chunkInfo.chunk.min) ||
c->getMax().woCompare(chunkInfo.chunk.max)) {
log() << "chunk mismatch after reload, ignoring will retry issue "
<< chunkInfo.chunk.toString() << endl;
continue;
}
}
BSONObj res;
if (c->moveAndCommit(Shard::make(chunkInfo.to),
Chunk::MaxChunkSize,
writeConcern,
waitForDelete,
0, /* maxTimeMS */
res)) {
movedCount++;
continue;
}
// the move requires acquiring the collection metadata's lock, which can fail
log() << "balancer move failed: " << res << " from: " << chunkInfo.from
<< " to: " << chunkInfo.to << " chunk: " << chunkInfo.chunk << endl;
if (res["chunkTooBig"].trueValue()) {
// reload just to be safe
cm = cfg->getChunkManager(chunkInfo.ns);
verify(cm);
c = cm->findIntersectingChunk(chunkInfo.chunk.min);
log() << "performing a split because migrate failed for size reasons";
Status status = c->split(Chunk::normal, NULL, NULL);
log() << "split results: " << status << endl;
if (!status.isOK()) {
log() << "marking chunk as jumbo: " << c->toString() << endl;
c->markAsJumbo();
// we increment moveCount so we do another round right away
movedCount++;
}
}
} catch (const DBException& ex) {
warning() << "could not move chunk " << chunkInfo.chunk.toString()
<< ", continuing balancing round" << causedBy(ex) << endl;
}
}
return movedCount;
}
int Balancer::_moveChunks(const vector<CandidateChunkPtr>* candidateChunks,
bool secondaryThrottle,
bool waitForDelete)
{
int movedCount = 0;
for ( vector<CandidateChunkPtr>::const_iterator it = candidateChunks->begin(); it != candidateChunks->end(); ++it ) {
const CandidateChunk& chunkInfo = *it->get();
// Changes to metadata, borked metadata, and connectivity problems should cause us to
// abort this chunk move, but shouldn't cause us to abort the entire round of chunks.
// TODO: Handle all these things more cleanly, since they're expected problems
try {
DBConfigPtr cfg = grid.getDBConfig( chunkInfo.ns );
verify( cfg );
// NOTE: We purposely do not reload metadata here, since _doBalanceRound already
// tried to do so once.
ChunkManagerPtr cm = cfg->getChunkManager( chunkInfo.ns );
verify( cm );
ChunkPtr c = cm->findIntersectingChunk( chunkInfo.chunk.min );
if ( c->getMin().woCompare( chunkInfo.chunk.min ) || c->getMax().woCompare( chunkInfo.chunk.max ) ) {
// likely a split happened somewhere
cm = cfg->getChunkManager( chunkInfo.ns , true /* reload */);
verify( cm );
c = cm->findIntersectingChunk( chunkInfo.chunk.min );
if ( c->getMin().woCompare( chunkInfo.chunk.min ) || c->getMax().woCompare( chunkInfo.chunk.max ) ) {
log() << "chunk mismatch after reload, ignoring will retry issue " << chunkInfo.chunk.toString() << endl;
continue;
}
}
BSONObj res;
if (c->moveAndCommit(Shard::make(chunkInfo.to),
Chunk::MaxChunkSize,
secondaryThrottle,
waitForDelete,
0, /* maxTimeMS */
res)) {
movedCount++;
continue;
}
// the move requires acquiring the collection metadata's lock, which can fail
log() << "balancer move failed: " << res << " from: " << chunkInfo.from << " to: " << chunkInfo.to
<< " chunk: " << chunkInfo.chunk << endl;
if ( res["chunkTooBig"].trueValue() ) {
// reload just to be safe
cm = cfg->getChunkManager( chunkInfo.ns );
verify( cm );
c = cm->findIntersectingChunk( chunkInfo.chunk.min );
log() << "forcing a split because migrate failed for size reasons" << endl;
res = BSONObj();
c->singleSplit( true , res );
log() << "forced split results: " << res << endl;
if ( ! res["ok"].trueValue() ) {
log() << "marking chunk as jumbo: " << c->toString() << endl;
c->markAsJumbo();
// we increment moveCount so we do another round right away
movedCount++;
}
}
}
catch( const DBException& ex ) {
warning() << "could not move chunk " << chunkInfo.chunk.toString()
<< ", continuing balancing round" << causedBy( ex ) << endl;
}
}
return movedCount;
}