bool run(const string& , BSONObj& cmdObj, string& errmsg, BSONObjBuilder& result, bool){
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 ) ){
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 old = info->findChunk( find );
return _split( result , errmsg , ns , info , old , cmdObj.getObjectField( "middle" ) );
}
ChunkManagerPtr DBConfig::shardCollection( const string& ns , ShardKeyPattern fieldsAndOrder , bool unique ) {
uassert( 8042 , "db doesn't have sharding enabled" , _shardingEnabled );
scoped_lock lk( _lock );
CollectionInfo& ci = _collections[ns];
uassert( 8043 , "collection already sharded" , ! ci.isSharded() );
log() << "enable sharding on: " << ns << " with shard key: " << fieldsAndOrder << endl;
// From this point on, 'ns' is going to be treated as a sharded collection. We assume this is the first
// time it is seen by the sharded system and thus create the first chunk for the collection. All the remaining
// chunks will be created as a by-product of splitting.
ci.shard( ns , fieldsAndOrder , unique );
ChunkManagerPtr cm = ci.getCM();
uassert( 13449 , "collections already sharded" , (cm->numChunks() == 0) );
cm->createFirstChunk( getPrimary() );
_save();
try {
cm->maybeChunkCollection();
}
catch ( UserException& e ) {
// failure to chunk is not critical enough to abort the command (and undo the _save()'d configDB state)
log() << "couldn't chunk recently created collection: " << ns << " " << e << endl;
}
return cm;
}
void _insert( Request& r , DbMessage& d, ChunkManagerPtr manager ){
while ( d.moreJSObjs() ){
BSONObj o = d.nextJsObj();
if ( ! manager->hasShardKey( o ) ){
bool bad = true;
if ( manager->getShardKey().partOfShardKey( "_id" ) ){
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
b.appendElements( o );
o = b.obj();
bad = ! manager->hasShardKey( o );
}
if ( bad ){
log() << "tried to insert object without shard key: " << r.getns() << " " << o << endl;
throw UserException( 8011 , "tried to insert object without shard key" );
}
}
ChunkPtr c = manager->findChunk( o );
log(4) << " server:" << c->getShard().toString() << " " << o << endl;
insert( c->getShard() , r.getns() , o );
r.gotInsert();
c->splitIfShould( o.objsize() );
}
}
void _update( Request& r , DbMessage& d, ChunkManagerPtr manager ) {
int flags = d.pullInt();
BSONObj query = d.nextJsObj();
uassert( 13506 , "$atomic not supported sharded" , query["$atomic"].eoo() );
uassert( 10201 , "invalid update" , d.moreJSObjs() );
BSONObj toupdate = d.nextJsObj();
BSONObj chunkFinder = query;
bool upsert = flags & UpdateOption_Upsert;
bool multi = flags & UpdateOption_Multi;
uassert( 10202 , "can't mix multi and upsert and sharding" , ! ( upsert && multi ) );
if (upsert) {
uassert(8012, "can't upsert something without shard key",
(manager->hasShardKey(toupdate) ||
(toupdate.firstElement().fieldName()[0] == '$' && manager->hasShardKey(query))));
BSONObj key = manager->getShardKey().extractKey(query);
BSONForEach(e, key) {
uassert(13465, "shard key in upsert query must be an exact match", getGtLtOp(e) == BSONObj::Equality);
}
}
/**
* Returns true if request is a query for sharded indexes.
*/
static bool doShardedIndexQuery(OperationContext* txn, Request& r, const QuerySpec& qSpec) {
// Extract the ns field from the query, which may be embedded within the "query" or
// "$query" field.
auto nsField = qSpec.filter()["ns"];
if (nsField.eoo()) {
return false;
}
const NamespaceString indexNSSQuery(nsField.str());
auto status = grid.catalogCache()->getDatabase(txn, indexNSSQuery.db().toString());
if (!status.isOK()) {
return false;
}
shared_ptr<DBConfig> config = status.getValue();
if (!config->isSharded(indexNSSQuery.ns())) {
return false;
}
// if you are querying on system.indexes, we need to make sure we go to a shard
// that actually has chunks. This is not a perfect solution (what if you just
// look at all indexes), but better than doing nothing.
ShardPtr shard;
ChunkManagerPtr cm;
config->getChunkManagerOrPrimary(indexNSSQuery.ns(), cm, shard);
if (cm) {
set<ShardId> shardIds;
cm->getAllShardIds(&shardIds);
verify(shardIds.size() > 0);
shard = grid.shardRegistry()->getShard(*shardIds.begin());
}
ShardConnection dbcon(shard->getConnString(), r.getns());
DBClientBase& c = dbcon.conn();
string actualServer;
Message response;
bool ok = c.call(r.m(), response, true, &actualServer);
uassert(10200, "mongos: error calling db", ok);
{
QueryResult::View qr = response.singleData().view2ptr();
if (qr.getResultFlags() & ResultFlag_ShardConfigStale) {
dbcon.done();
// Version is zero b/c this is deprecated codepath
throw RecvStaleConfigException(r.getns(),
"Strategy::doQuery",
ChunkVersion(0, 0, OID()),
ChunkVersion(0, 0, OID()));
}
}
r.reply(response, actualServer.size() ? actualServer : c.getServerAddress());
dbcon.done();
return true;
}
int Balancer::_moveChunks( const vector<CandidateChunkPtr>* candidateChunks , bool secondaryThrottle ) {
int movedCount = 0;
for ( vector<CandidateChunkPtr>::const_iterator it = candidateChunks->begin(); it != candidateChunks->end(); ++it ) {
const CandidateChunk& chunkInfo = *it->get();
DBConfigPtr cfg = grid.getDBConfig( chunkInfo.ns );
verify( cfg );
ChunkManagerPtr cm = cfg->getChunkManager( chunkInfo.ns );
verify( cm );
ChunkPtr c = cm->findChunk( 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->findChunk( 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 , 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->findChunk( 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++;
}
}
}
return movedCount;
}
bool run(const string& , BSONObj& cmdObj, string& errmsg, BSONObjBuilder& result, bool) {
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 ) ) {
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;
ChunkPtr p;
if ( middle.isEmpty() ) {
p = chunk->singleSplit( true /* force a split even if not enough data */ , res );
}
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;
}
vector<BSONObj> splitPoints;
splitPoints.push_back( middle );
p = chunk->multiSplit( splitPoints , res );
}
if ( p.get() == NULL ) {
errmsg = "split failed";
result.append( "cause" , res );
return false;
}
return true;
}
bool run(const string& , BSONObj& cmdObj, string& errmsg, BSONObjBuilder& result, bool){
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 ) ){
errmsg = "ns not sharded. have to shard before 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;
}
result.append( "millis" , t.millis() );
return true;
}
int Balancer::_moveChunks( const vector<CandidateChunkPtr>* candidateChunks ) {
int movedCount = 0;
for ( vector<CandidateChunkPtr>::const_iterator it = candidateChunks->begin(); it != candidateChunks->end(); ++it ) {
const CandidateChunk& chunkInfo = *it->get();
DBConfigPtr cfg = grid.getDBConfig( chunkInfo.ns );
assert( cfg );
ChunkManagerPtr cm = cfg->getChunkManager( chunkInfo.ns );
assert( cm );
const BSONObj& chunkToMove = chunkInfo.chunk;
ChunkPtr c = cm->findChunk( chunkToMove["min"].Obj() );
if ( c->getMin().woCompare( chunkToMove["min"].Obj() ) || c->getMax().woCompare( chunkToMove["max"].Obj() ) ) {
// likely a split happened somewhere
cm = cfg->getChunkManager( chunkInfo.ns , true /* reload */);
assert( cm );
c = cm->findChunk( chunkToMove["min"].Obj() );
if ( c->getMin().woCompare( chunkToMove["min"].Obj() ) || c->getMax().woCompare( chunkToMove["max"].Obj() ) ) {
log() << "chunk mismatch after reload, ignoring will retry issue cm: "
<< c->getMin() << " min: " << chunkToMove["min"].Obj() << endl;
continue;
}
}
BSONObj res;
if ( c->moveAndCommit( Shard::make( chunkInfo.to ) , Chunk::MaxChunkSize , res ) ) {
movedCount++;
continue;
}
// the move requires acquiring the collection metadata's lock, which can fail
log() << "balacer move failed: " << res << " from: " << chunkInfo.from << " to: " << chunkInfo.to
<< " chunk: " << chunkToMove << endl;
if ( res["chunkTooBig"].trueValue() ) {
// reload just to be safe
cm = cfg->getChunkManager( chunkInfo.ns );
assert( cm );
c = cm->findChunk( chunkToMove["min"].Obj() );
log() << "forcing a split because migrate failed for size reasons" << endl;
res = BSONObj();
c->singleSplit( true , res );
log() << "forced split results: " << res << endl;
// TODO: if the split fails, mark as jumbo SERVER-2571
}
}
return movedCount;
}
void checkShardVersion( DBClientBase& conn , const string& ns , bool authoritative ){
// TODO: cache, optimize, etc...
WriteBackListener::init( conn );
DBConfigPtr conf = grid.getDBConfig( ns );
if ( ! conf )
return;
ShardChunkVersion version = 0;
unsigned long long officialSequenceNumber = 0;
ChunkManagerPtr manager;
const bool isSharded = conf->isSharded( ns );
if ( isSharded ){
manager = conf->getChunkManager( ns , authoritative );
officialSequenceNumber = manager->getSequenceNumber();
}
unsigned long long & sequenceNumber = checkShardVersionLastSequence[ make_pair(&conn,ns) ];
if ( sequenceNumber == officialSequenceNumber )
return;
if ( isSharded ){
version = manager->getVersion( Shard::make( conn.getServerAddress() ) );
}
log(2) << " have to set shard version for conn: " << &conn << " ns:" << ns
<< " my last seq: " << sequenceNumber << " current: " << officialSequenceNumber
<< " version: " << version << " manager: " << manager.get()
<< endl;
BSONObj result;
if ( setShardVersion( conn , ns , version , authoritative , result ) ){
// success!
log(1) << " setShardVersion success!" << endl;
sequenceNumber = officialSequenceNumber;
dassert( sequenceNumber == checkShardVersionLastSequence[ make_pair(&conn,ns) ] );
return;
}
log(1) << " setShardVersion failed!\n" << result << endl;
if ( result.getBoolField( "need_authoritative" ) )
massert( 10428 , "need_authoritative set but in authoritative mode already" , ! authoritative );
if ( ! authoritative ){
checkShardVersion( conn , ns , 1 );
return;
}
log() << " setShardVersion failed: " << result << endl;
massert( 10429 , (string)"setShardVersion failed! " + result.jsonString() , 0 );
}
void _insert( Request& r , DbMessage& d, ChunkManagerPtr manager ) {
while ( d.moreJSObjs() ) {
BSONObj o = d.nextJsObj();
if ( ! manager->hasShardKey( o ) ) {
bool bad = true;
if ( manager->getShardKey().partOfShardKey( "_id" ) ) {
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
b.appendElements( o );
o = b.obj();
bad = ! manager->hasShardKey( o );
}
if ( bad ) {
log() << "tried to insert object without shard key: " << r.getns() << " " << o << endl;
throw UserException( 8011 , "tried to insert object without shard key" );
}
}
// Many operations benefit from having the shard key early in the object
o = manager->getShardKey().moveToFront(o);
const int maxTries = 10;
bool gotThrough = false;
for ( int i=0; i<maxTries; i++ ) {
try {
ChunkPtr c = manager->findChunk( o );
log(4) << " server:" << c->getShard().toString() << " " << o << endl;
insert( c->getShard() , r.getns() , o );
r.gotInsert();
if ( r.getClientInfo()->autoSplitOk() )
c->splitIfShould( o.objsize() );
gotThrough = true;
break;
}
catch ( StaleConfigException& e ) {
log( i < ( maxTries / 2 ) ) << "retrying insert because of StaleConfigException: " << e << " object: " << o << endl;
r.reset();
manager = r.getChunkManager();
uassert(14804, "collection no longer sharded", manager);
}
sleepmillis( i * 200 );
}
assert( inShutdown() || gotThrough );
}
}
/**
* Returns true if request is a query for sharded indexes.
*/
static bool doShardedIndexQuery( Request& r, const QuerySpec& qSpec ) {
// Extract the ns field from the query, which may be embedded within the "query" or
// "$query" field.
string indexNSQuery(qSpec.filter()["ns"].str());
DBConfigPtr config = grid.getDBConfig( r.getns() );
if ( !config->isSharded( indexNSQuery )) {
return false;
}
// if you are querying on system.indexes, we need to make sure we go to a shard
// that actually has chunks. This is not a perfect solution (what if you just
// look at all indexes), but better than doing nothing.
ShardPtr shard;
ChunkManagerPtr cm;
config->getChunkManagerOrPrimary( indexNSQuery, cm, shard );
if ( cm ) {
set<Shard> shards;
cm->getAllShards( shards );
verify( shards.size() > 0 );
shard.reset( new Shard( *shards.begin() ) );
}
ShardConnection dbcon( *shard , r.getns() );
DBClientBase &c = dbcon.conn();
string actualServer;
Message response;
bool ok = c.call( r.m(), response, true , &actualServer );
uassert( 10200 , "mongos: error calling db", ok );
{
QueryResult *qr = (QueryResult *) response.singleData();
if ( qr->resultFlags() & ResultFlag_ShardConfigStale ) {
dbcon.done();
// Version is zero b/c this is deprecated codepath
throw RecvStaleConfigException( r.getns(),
"Strategy::doQuery",
ChunkVersion( 0, 0, OID() ),
ChunkVersion( 0, 0, OID() ));
}
}
r.reply( response , actualServer.size() ? actualServer : c.getServerAddress() );
dbcon.done();
return true;
}
ChunkManagerPtr DBConfig::getChunkManager( const string& ns , bool reload ){
scoped_lock lk( _lock );
ChunkManagerPtr m = _shards[ns];
if ( m && ! reload )
return m;
uassert( 10181 , (string)"not sharded:" + ns , _isSharded( ns ) );
if ( m && reload )
log() << "reloading shard info for: " << ns << endl;
m.reset( new ChunkManager( this , ns , _sharded[ ns ].key , _sharded[ns].unique ) );
_shards[ns] = m;
return m;
}
// TODO: Same limitations as other mongos metadata commands, sometimes we'll be stale here
// and fail. Need to better integrate targeting with commands.
ShardPtr guessMergeShard( const NamespaceString& nss, const BSONObj& minKey ) {
DBConfigPtr config = grid.getDBConfig( nss.ns() );
if ( !config->isSharded( nss ) ) {
config->reload();
if ( !config->isSharded( nss ) ) {
return ShardPtr();
}
}
ChunkManagerPtr manager = config->getChunkManager( nss );
if ( !manager ) return ShardPtr();
ChunkPtr chunk = manager->findChunkForDoc( minKey );
if ( !chunk ) return ShardPtr();
return ShardPtr( new Shard( chunk->getShard() ) );
}
void _delete( Request& r , DbMessage& d, ChunkManagerPtr manager ){
int flags = d.pullInt();
bool justOne = flags & 1;
uassert( 10203 , "bad delete message" , d.moreJSObjs() );
BSONObj pattern = d.nextJsObj();
vector<shared_ptr<ChunkRange> > chunks;
manager->getChunksForQuery( chunks , pattern );
log(2) << "delete : " << pattern << " \t " << chunks.size() << " justOne: " << justOne << endl;
if ( chunks.size() == 1 ){
doWrite( dbDelete , r , chunks[0]->getShard() );
return;
}
if ( justOne && ! pattern.hasField( "_id" ) )
throw UserException( 8015 , "can only delete with a non-shard key pattern if can delete as many as we find" );
set<Shard> seen;
for ( vector<shared_ptr<ChunkRange> >::iterator i=chunks.begin(); i!=chunks.end(); i++){
shared_ptr<ChunkRange> c = *i;
if ( seen.count( c->getShard() ) )
continue;
seen.insert( c->getShard() );
doWrite( dbDelete , r , c->getShard() );
}
}
void _insert( Request& r , DbMessage& d, ChunkManagerPtr manager ){
while ( d.moreJSObjs() ){
BSONObj o = d.nextJsObj();
if ( ! manager->hasShardKey( o ) ){
bool bad = true;
if ( manager->getShardKey().partOfShardKey( "_id" ) ){
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
b.appendElements( o );
o = b.obj();
bad = ! manager->hasShardKey( o );
}
if ( bad ){
log() << "tried to insert object without shard key: " << r.getns() << " " << o << endl;
throw UserException( 8011 , "tried to insert object without shard key" );
}
}
bool gotThrough = false;
for ( int i=0; i<10; i++ ){
try {
ChunkPtr c = manager->findChunk( o );
log(4) << " server:" << c->getShard().toString() << " " << o << endl;
insert( c->getShard() , r.getns() , o );
r.gotInsert();
c->splitIfShould( o.objsize() );
gotThrough = true;
break;
}
catch ( StaleConfigException& ){
log(1) << "retrying insert because of StaleConfigException: " << o << endl;
r.reset();
manager = r.getChunkManager();
}
sleepmillis( i * 200 );
}
assert( gotThrough );
}
}
void handleIndexWrite( int op , Request& r ) {
DbMessage& d = r.d();
if ( op == dbInsert ) {
while( d.moreJSObjs() ) {
BSONObj o = d.nextJsObj();
const char * ns = o["ns"].valuestr();
if ( r.getConfig()->isSharded( ns ) ) {
BSONObj newIndexKey = o["key"].embeddedObjectUserCheck();
uassert( 10205 , (string)"can't use unique indexes with sharding ns:" + ns +
" key: " + o["key"].embeddedObjectUserCheck().toString() ,
IndexDetails::isIdIndexPattern( newIndexKey ) ||
! o["unique"].trueValue() ||
r.getConfig()->getChunkManager( ns )->getShardKey().isPrefixOf( newIndexKey ) );
ChunkManagerPtr cm = r.getConfig()->getChunkManager( ns );
assert( cm );
set<Shard> shards;
cm->getAllShards(shards);
for (set<Shard>::const_iterator it=shards.begin(), end=shards.end(); it != end; ++it)
doWrite( op , r , *it );
}
else {
doWrite( op , r , r.primaryShard() );
}
r.gotInsert();
}
}
else if ( op == dbUpdate ) {
throw UserException( 8050 , "can't update system.indexes" );
}
else if ( op == dbDelete ) {
// TODO
throw UserException( 8051 , "can't delete indexes on sharded collection yet" );
}
else {
log() << "handleIndexWrite invalid write op: " << op << endl;
throw UserException( 8052 , "handleIndexWrite invalid write op" );
}
}
ChunkManagerPtr DBConfig::shardCollection( const string& ns , ShardKeyPattern fieldsAndOrder , bool unique ){
if ( ! _shardingEnabled )
throw UserException( 8042 , "db doesn't have sharding enabled" );
scoped_lock lk( _lock );
ChunkManagerPtr info = _shards[ns];
if ( info )
return info;
if ( _isSharded( ns ) )
throw UserException( 8043 , "already sharded" );
log() << "enable sharding on: " << ns << " with shard key: " << fieldsAndOrder << endl;
_sharded[ns] = CollectionInfo( fieldsAndOrder , unique );
info.reset( new ChunkManager( this , ns , fieldsAndOrder , unique ) );
_shards[ns] = info;
return info;
}
/**
* Splits the chunks touched based from the targeter stats if needed.
*/
static void splitIfNeeded( const string& ns, const TargeterStats& stats ) {
if ( !Chunk::ShouldAutoSplit ) {
return;
}
DBConfigPtr config;
try {
config = grid.getDBConfig( ns );
}
catch ( const DBException& ex ) {
warning() << "failed to get database config for " << ns
<< " while checking for auto-split: " << causedBy( ex ) << endl;
return;
}
ChunkManagerPtr chunkManager;
ShardPtr dummyShard;
config->getChunkManagerOrPrimary( ns, chunkManager, dummyShard );
if ( !chunkManager ) {
return;
}
for ( map<BSONObj, int>::const_iterator it = stats.chunkSizeDelta.begin();
it != stats.chunkSizeDelta.end(); ++it ) {
ChunkPtr chunk;
try {
chunk = chunkManager->findIntersectingChunk( it->first );
}
catch ( const AssertionException& ex ) {
warning() << "could not find chunk while checking for auto-split: "
<< causedBy( ex ) << endl;
return;
}
chunk->splitIfShould( it->second );
}
}
int Balancer::_moveChunks( const vector<CandidateChunkPtr>* candidateChunks ) {
int movedCount = 0;
for ( vector<CandidateChunkPtr>::const_iterator it = candidateChunks->begin(); it != candidateChunks->end(); ++it ){
const CandidateChunk& chunkInfo = *it->get();
DBConfigPtr cfg = grid.getDBConfig( chunkInfo.ns );
assert( cfg );
ChunkManagerPtr cm = cfg->getChunkManager( chunkInfo.ns );
assert( cm );
const BSONObj& chunkToMove = chunkInfo.chunk;
ChunkPtr c = cm->findChunk( chunkToMove["min"].Obj() );
if ( c->getMin().woCompare( chunkToMove["min"].Obj() ) ){
// likely a split happened somewhere
cm = cfg->getChunkManager( chunkInfo.ns , true );
assert( cm );
c = cm->findChunk( chunkToMove["min"].Obj() );
if ( c->getMin().woCompare( chunkToMove["min"].Obj() ) ){
log() << "balancer: chunk mismatch after reload, ignoring will retry issue cm: "
<< c->getMin() << " min: " << chunkToMove["min"].Obj() << endl;
continue;
}
}
string errmsg;
if ( c->moveAndCommit( Shard::make( chunkInfo.to ) , errmsg ) ){
movedCount++;
continue;
}
log() << "balancer: MOVE FAILED **** " << errmsg << "\n"
<< " from: " << chunkInfo.from << " to: " << " chunk: " << chunkToMove << endl;
}
return movedCount;
}
void insertSharded( DBConfigPtr conf, const char* ns, BSONObj& o, int flags ) {
ChunkManagerPtr manager = conf->getChunkManager(ns);
if ( ! manager->hasShardKey( o ) ) {
bool bad = true;
if ( manager->getShardKey().partOfShardKey( "_id" ) ) {
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
b.appendElements( o );
o = b.obj();
bad = ! manager->hasShardKey( o );
}
if ( bad ) {
log() << "tried to insert object without shard key: " << ns << " " << o << endl;
uasserted( 14842 , "tried to insert object without shard key" );
}
}
// Many operations benefit from having the shard key early in the object
o = manager->getShardKey().moveToFront(o);
const int maxTries = 30;
for ( int i=0; i<maxTries; i++ ) {
try {
ChunkPtr c = manager->findChunk( o );
log(4) << " server:" << c->getShard().toString() << " " << o << endl;
insert( c->getShard() , ns , o , flags);
// r.gotInsert();
// if ( r.getClientInfo()->autoSplitOk() )
c->splitIfShould( o.objsize() );
break;
}
catch ( StaleConfigException& e ) {
int logLevel = i < ( maxTries / 2 );
LOG( logLevel ) << "retrying insert because of StaleConfigException: " << e << " object: " << o << endl;
// r.reset();
unsigned long long old = manager->getSequenceNumber();
manager = conf->getChunkManager(ns);
LOG( logLevel ) << " sequenece number - old: " << old << " new: " << manager->getSequenceNumber() << endl;
if (!manager) {
uasserted(14843, "collection no longer sharded");
}
}
sleepmillis( i * 20 );
}
}
bool run(const string& , BSONObj& cmdObj, int, string& errmsg, BSONObjBuilder& result, bool) {
string ns = cmdObj.firstElement().valuestrsafe();
if ( ns.size() == 0 ) {
errmsg = "need to specify fully namespace";
return false;
}
DBConfigPtr config = grid.getDBConfig( ns );
if ( ! config->isSharded( ns ) ) {
errmsg = "ns not sharded.";
return false;
}
ChunkManagerPtr cm = config->getChunkManagerIfExists( ns );
if ( ! cm ) {
errmsg = "no chunk manager?";
return false;
}
cm->_printChunks();
result.appendTimestamp( "version" , cm->getVersion().toLong() );
return 1;
}
void _delete( Request& r , DbMessage& d, ChunkManagerPtr manager ){
int flags = d.pullInt();
bool justOne = flags & 1;
uassert( 10203 , "bad delete message" , d.moreJSObjs() );
BSONObj pattern = d.nextJsObj();
set<Shard> shards;
int left = 5;
while ( true ){
try {
manager->getShardsForQuery( shards , pattern );
log(2) << "delete : " << pattern << " \t " << shards.size() << " justOne: " << justOne << endl;
if ( shards.size() == 1 ){
doWrite( dbDelete , r , *shards.begin() );
return;
}
break;
}
catch ( StaleConfigException& e ){
if ( left <= 0 )
throw e;
left--;
log() << "delete failed b/c of StaleConfigException, retrying "
<< " left:" << left << " ns: " << r.getns() << " patt: " << pattern << endl;
r.reset( false );
shards.clear();
manager = r.getChunkManager();
}
}
if ( justOne && ! pattern.hasField( "_id" ) )
throw UserException( 8015 , "can only delete with a non-shard key pattern if can delete as many as we find" );
for ( set<Shard>::iterator i=shards.begin(); i!=shards.end(); i++){
int * x = (int*)(r.d().afterNS());
x[0] |= RemoveOption_Broadcast;
doWrite( dbDelete , r , *i , false );
}
}
void _groupInserts( ChunkManagerPtr manager, vector<BSONObj>& inserts, map<ChunkPtr,vector<BSONObj> >& insertsForChunks ){
// Redo all inserts for chunks which have changed
map<ChunkPtr,vector<BSONObj> >::iterator i = insertsForChunks.begin();
while( ! insertsForChunks.empty() && i != insertsForChunks.end() ){
if( ! manager->compatibleWith( i->first ) ){
inserts.insert( inserts.end(), i->second.begin(), i->second.end() );
insertsForChunks.erase( i++ );
}
else ++i;
}
// Figure out inserts we haven't chunked yet
for( vector<BSONObj>::iterator i = inserts.begin(); i != inserts.end(); ++i ){
BSONObj o = *i;
if ( ! manager->hasShardKey( o ) ) {
bool bad = true;
// Add autogenerated _id to item and see if we now have a shard key
if ( manager->getShardKey().partOfShardKey( "_id" ) ) {
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
b.appendElements( o );
o = b.obj();
bad = ! manager->hasShardKey( o );
}
if ( bad ) {
// TODO:
log() << "tried to insert object with no valid shard key for " << manager->getShardKey() << " : " << o << endl;
uassert( 8011, str::stream() << "tried to insert object with no valid shard key for " << manager->getShardKey().toString() << " : " << o.toString(), false );
}
}
// Many operations benefit from having the shard key early in the object
o = manager->getShardKey().moveToFront(o);
insertsForChunks[manager->findChunk(o)].push_back(o);
}
inserts.clear();
}
/**
* @return true if had to do something
*/
bool checkShardVersion( DBClientBase& conn , const string& ns , bool authoritative , int tryNumber ) {
// TODO: cache, optimize, etc...
WriteBackListener::init( conn );
DBConfigPtr conf = grid.getDBConfig( ns );
if ( ! conf )
return false;
unsigned long long officialSequenceNumber = 0;
ChunkManagerPtr manager;
const bool isSharded = conf->isSharded( ns );
if ( isSharded ) {
manager = conf->getChunkManager( ns , authoritative );
officialSequenceNumber = manager->getSequenceNumber();
}
// 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;
}
ShardChunkVersion version = 0;
if ( isSharded ) {
version = manager->getVersion( Shard::make( conn.getServerAddress() ) );
}
log(2) << " have to set shard version for conn: " << &conn << " ns:" << ns
<< " my last seq: " << sequenceNumber << " current: " << officialSequenceNumber
<< " version: " << version << " manager: " << manager.get()
<< endl;
BSONObj result;
if ( setShardVersion( conn , ns , version , 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.getBoolField( "need_authoritative" ) )
massert( 10428 , "need_authoritative set but in authoritative mode already" , ! authoritative );
if ( ! authoritative ) {
checkShardVersion( conn , ns , 1 , tryNumber + 1 );
return true;
}
if ( tryNumber < 4 ) {
log(1) << "going to retry checkShardVersion" << endl;
sleepmillis( 10 );
checkShardVersion( conn , ns , 1 , tryNumber + 1 );
return true;
}
log() << " setShardVersion failed: " << result << endl;
massert( 10429 , (string)"setShardVersion failed! " + result.jsonString() , 0 );
return true;
}
void _insert( Request& r , DbMessage& d, ChunkManagerPtr manager ) {
const int flags = d.reservedField() | InsertOption_ContinueOnError; // ContinueOnError is always on when using sharding.
map<ChunkPtr, vector<BSONObj> > insertsForChunk; // Group bulk insert for appropriate shards
try {
while ( d.moreJSObjs() ) {
BSONObj o = d.nextJsObj();
if ( ! manager->hasShardKey( o ) ) {
bool bad = true;
if ( manager->getShardKey().partOfShardKey( "_id" ) ) {
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
b.appendElements( o );
o = b.obj();
bad = ! manager->hasShardKey( o );
}
if ( bad ) {
log() << "tried to insert object with no valid shard key: " << r.getns() << " " << o << endl;
uasserted( 8011 , "tried to insert object with no valid shard key" );
}
}
// Many operations benefit from having the shard key early in the object
o = manager->getShardKey().moveToFront(o);
insertsForChunk[manager->findChunk(o)].push_back(o);
}
for (map<ChunkPtr, vector<BSONObj> >::iterator it = insertsForChunk.begin(); it != insertsForChunk.end(); ++it) {
ChunkPtr c = it->first;
vector<BSONObj> objs = it->second;
const int maxTries = 30;
bool gotThrough = false;
for ( int i=0; i<maxTries; i++ ) {
try {
LOG(4) << " server:" << c->getShard().toString() << " bulk insert " << objs.size() << " documents" << endl;
insert( c->getShard() , r.getns() , objs , flags);
int bytesWritten = 0;
for (vector<BSONObj>::iterator vecIt = objs.begin(); vecIt != objs.end(); ++vecIt) {
r.gotInsert(); // Record the correct number of individual inserts
bytesWritten += (*vecIt).objsize();
}
if ( r.getClientInfo()->autoSplitOk() )
c->splitIfShould( bytesWritten );
gotThrough = true;
break;
}
catch ( StaleConfigException& e ) {
int logLevel = i < ( maxTries / 2 );
LOG( logLevel ) << "retrying bulk insert of " << objs.size() << " documents because of StaleConfigException: " << e << endl;
r.reset();
manager = r.getChunkManager();
if( ! manager ) {
uasserted(14804, "collection no longer sharded");
}
unsigned long long old = manager->getSequenceNumber();
LOG( logLevel ) << " sequence number - old: " << old << " new: " << manager->getSequenceNumber() << endl;
}
sleepmillis( i * 20 );
}
assert( inShutdown() || gotThrough ); // not caught below
}
} catch (const UserException&){
if (!d.moreJSObjs()){
throw;
}
// Ignore and keep going. ContinueOnError is implied with sharding.
}
}
/**
* @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 ), ShardChunkVersion( 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;
}
ShardChunkVersion version = ShardChunkVersion( 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) << " 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 , 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;
}
virtual void queryOp( Request& r ){
QueryMessage q( r.d() );
log(3) << "shard query: " << q.ns << " " << q.query << endl;
if ( q.ntoreturn == 1 && strstr(q.ns, ".$cmd") )
throw UserException( 8010 , "something is wrong, shouldn't see a command here" );
ChunkManagerPtr info = r.getChunkManager();
assert( info );
Query query( q.query );
set<Shard> shards;
info->getShardsForQuery( shards , query.getFilter() );
set<ServerAndQuery> servers;
for ( set<Shard>::iterator i = shards.begin(); i != shards.end(); i++ ){
servers.insert( ServerAndQuery( i->getConnString() , BSONObj() ) );
}
if ( logLevel > 4 ){
StringBuilder ss;
ss << " shard query servers: " << servers.size() << '\n';
for ( set<ServerAndQuery>::iterator i = servers.begin(); i!=servers.end(); i++ ){
const ServerAndQuery& s = *i;
ss << " " << s.toString() << '\n';
}
log() << ss.str();
}
ClusteredCursor * cursor = 0;
BSONObj sort = query.getSort();
if ( sort.isEmpty() ){
cursor = new SerialServerClusteredCursor( servers , q );
}
else {
cursor = new ParallelSortClusteredCursor( servers , q , sort );
}
assert( cursor );
try {
cursor->init();
log(5) << " cursor type: " << cursor->type() << endl;
shardedCursorTypes.hit( cursor->type() );
if ( query.isExplain() ){
BSONObj explain = cursor->explain();
replyToQuery( 0 , r.p() , r.m() , explain );
delete( cursor );
return;
}
} catch(...) {
delete cursor;
throw;
}
ShardedClientCursorPtr cc (new ShardedClientCursor( q , cursor ));
if ( ! cc->sendNextBatch( r ) ){
return;
}
log(6) << "storing cursor : " << cc->getId() << endl;
cursorCache.store( cc );
}
void _update( Request& r , DbMessage& d, ChunkManagerPtr manager ){
int flags = d.pullInt();
BSONObj query = d.nextJsObj();
uassert( 10201 , "invalid update" , d.moreJSObjs() );
BSONObj toupdate = d.nextJsObj();
BSONObj chunkFinder = query;
bool upsert = flags & UpdateOption_Upsert;
bool multi = flags & UpdateOption_Multi;
uassert( 10202 , "can't mix multi and upsert and sharding" , ! ( upsert && multi ) );
if ( upsert && !(manager->hasShardKey(toupdate) ||
(toupdate.firstElement().fieldName()[0] == '$' && manager->hasShardKey(query))))
{
throw UserException( 8012 , "can't upsert something without shard key" );
}
bool save = false;
if ( ! manager->hasShardKey( query ) ){
if ( multi ){
}
else if ( strcmp( query.firstElement().fieldName() , "_id" ) || query.nFields() != 1 ){
throw UserException( 8013 , "can't do non-multi update with query that doesn't have the shard key" );
}
else {
save = true;
chunkFinder = toupdate;
}
}
if ( ! save ){
if ( toupdate.firstElement().fieldName()[0] == '$' ){
BSONObjIterator ops(toupdate);
while(ops.more()){
BSONElement op(ops.next());
if (op.type() != Object)
continue;
BSONObjIterator fields(op.embeddedObject());
while(fields.more()){
const string field = fields.next().fieldName();
uassert(13123, "Can't modify shard key's value", ! manager->getShardKey().partOfShardKey(field));
}
}
} else if ( manager->hasShardKey( toupdate ) ){
uassert( 8014, "change would move shards!", manager->getShardKey().compare( query , toupdate ) == 0 );
} else {
uasserted(12376, "shard key must be in update object");
}
}
if ( multi ){
set<Shard> shards;
manager->getShardsForQuery( shards , chunkFinder );
int * x = (int*)(r.d().afterNS());
x[0] |= UpdateOption_Broadcast;
for ( set<Shard>::iterator i=shards.begin(); i!=shards.end(); i++){
doWrite( dbUpdate , r , *i , false );
}
}
else {
int left = 5;
while ( true ){
try {
ChunkPtr c = manager->findChunk( chunkFinder );
doWrite( dbUpdate , r , c->getShard() );
c->splitIfShould( d.msg().header()->dataLen() );
break;
}
catch ( StaleConfigException& e ){
if ( left <= 0 )
throw e;
left--;
log() << "update failed b/c of StaleConfigException, retrying "
<< " left:" << left << " ns: " << r.getns() << " query: " << query << endl;
r.reset( false );
manager = r.getChunkManager();
}
}
}
}
/**
* @return true if had to do something
*/
bool checkShardVersion( DBClientBase& conn_in , const string& ns , bool authoritative , int tryNumber ) {
// TODO: cache, optimize, etc...
WriteBackListener::init( conn_in );
DBConfigPtr conf = grid.getDBConfig( ns );
if ( ! conf )
return false;
DBClientBase* conn = 0;
switch ( conn_in.type() ) {
case ConnectionString::INVALID:
assert(0);
break;
case ConnectionString::MASTER:
// great
conn = &conn_in;
break;
case ConnectionString::PAIR:
assert( ! "pair not support for sharding" );
break;
case ConnectionString::SYNC:
// TODO: we should check later that we aren't actually sharded on this
conn = &conn_in;
break;
case ConnectionString::SET:
DBClientReplicaSet* set = (DBClientReplicaSet*)&conn_in;
conn = &(set->masterConn());
break;
}
assert(conn);
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();
}
// 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;
}
ShardChunkVersion version = 0;
if ( isSharded && manager ) {
version = manager->getVersion( Shard::make( conn->getServerAddress() ) );
}
LOG(2) << " have to set shard version for conn: " << conn << " ns:" << ns
<< " my last seq: " << sequenceNumber << " current: " << officialSequenceNumber
<< " version: " << version << " manager: " << manager.get()
<< endl;
BSONObj result;
if ( setShardVersion( *conn , ns , version , 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 ) {
checkShardVersion( *conn , ns , 1 , tryNumber + 1 );
return true;
}
if ( result["reloadConfig"].trueValue() ) {
if( result["version"].timestampTime() == 0 ){
// 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: " << conn->getServerAddress() << " " << result << endl;
sleepmillis( 10 * tryNumber );
checkShardVersion( *conn , ns , true , tryNumber + 1 );
return true;
}
string errmsg = str::stream() << "setShardVersion failed host: " << conn->getServerAddress() << " " << result;
log() << " " << errmsg << endl;
massert( 10429 , errmsg , 0 );
return true;
}
