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 );
}
}
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 );
}
}
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);
}
}
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 _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();
}
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 );
}
}
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;
}
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();
}
}
}
}
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.
}
}
void Balancer::_doBalanceRound( DBClientBase& conn, vector<CandidateChunkPtr>* candidateChunks ) {
verify( candidateChunks );
//
// 1. Check whether there is any sharded collection to be balanced by querying
// the ShardsNS::collections collection
//
auto_ptr<DBClientCursor> cursor = conn.query(CollectionType::ConfigNS, BSONObj());
if ( NULL == cursor.get() ) {
warning() << "could not query " << CollectionType::ConfigNS
<< " while trying to balance" << endl;
return;
}
vector< string > collections;
while ( cursor->more() ) {
BSONObj col = cursor->nextSafe();
// sharded collections will have a shard "key".
if ( ! col[CollectionType::keyPattern()].eoo() &&
! col[CollectionType::noBalance()].trueValue() ){
collections.push_back( col[CollectionType::ns()].String() );
}
else if( col[CollectionType::noBalance()].trueValue() ){
LOG(1) << "not balancing collection " << col[CollectionType::ns()].String()
<< ", explicitly disabled" << endl;
}
}
cursor.reset();
if ( collections.empty() ) {
LOG(1) << "no collections to balance" << endl;
return;
}
//
// 2. Get a list of all the shards that are participating in this balance round
// along with any maximum allowed quotas and current utilization. We get the
// latter by issuing db.serverStatus() (mem.mapped) to all shards.
//
// TODO: skip unresponsive shards and mark information as stale.
//
ShardInfoMap shardInfo;
Status loadStatus = DistributionStatus::populateShardInfoMap(&shardInfo);
if (!loadStatus.isOK()) {
warning() << "failed to load shard metadata" << causedBy(loadStatus) << endl;
return;
}
if (shardInfo.size() < 2) {
LOG(1) << "can't balance without more active shards" << endl;
return;
}
OCCASIONALLY warnOnMultiVersion( shardInfo );
//
// 3. For each collection, check if the balancing policy recommends moving anything around.
//
for (vector<string>::const_iterator it = collections.begin(); it != collections.end(); ++it ) {
const string& ns = *it;
OwnedPointerMap<string, OwnedPointerVector<ChunkType> > shardToChunksMap;
cursor = conn.query(ChunkType::ConfigNS,
QUERY(ChunkType::ns(ns)).sort(ChunkType::min()));
set<BSONObj> allChunkMinimums;
while ( cursor->more() ) {
BSONObj chunkDoc = cursor->nextSafe().getOwned();
auto_ptr<ChunkType> chunk(new ChunkType());
string errmsg;
if (!chunk->parseBSON(chunkDoc, &errmsg)) {
error() << "bad chunk format for " << chunkDoc
<< ": " << errmsg << endl;
return;
}
allChunkMinimums.insert(chunk->getMin().getOwned());
OwnedPointerVector<ChunkType>*& chunkList =
shardToChunksMap.mutableMap()[chunk->getShard()];
if (chunkList == NULL) {
chunkList = new OwnedPointerVector<ChunkType>();
}
chunkList->mutableVector().push_back(chunk.release());
}
cursor.reset();
if (shardToChunksMap.map().empty()) {
LOG(1) << "skipping empty collection (" << ns << ")";
continue;
}
for (ShardInfoMap::const_iterator i = shardInfo.begin(); i != shardInfo.end(); ++i) {
// this just makes sure there is an entry in shardToChunksMap for every shard
OwnedPointerVector<ChunkType>*& chunkList =
shardToChunksMap.mutableMap()[i->first];
if (chunkList == NULL) {
chunkList = new OwnedPointerVector<ChunkType>();
}
}
DistributionStatus status(shardInfo, shardToChunksMap.map());
// load tags
Status result = clusterCreateIndex(TagsType::ConfigNS,
BSON(TagsType::ns() << 1 << TagsType::min() << 1),
true, // unique
WriteConcernOptions::AllConfigs,
NULL);
if ( !result.isOK() ) {
warning() << "could not create index tags_1_min_1: " << result.reason() << endl;
continue;
}
cursor = conn.query(TagsType::ConfigNS,
QUERY(TagsType::ns(ns)).sort(TagsType::min()));
vector<TagRange> ranges;
while ( cursor->more() ) {
BSONObj tag = cursor->nextSafe();
TagRange tr(tag[TagsType::min()].Obj().getOwned(),
tag[TagsType::max()].Obj().getOwned(),
tag[TagsType::tag()].String());
ranges.push_back(tr);
uassert(16356,
str::stream() << "tag ranges not valid for: " << ns,
status.addTagRange(tr) );
}
cursor.reset();
DBConfigPtr cfg = grid.getDBConfig( ns );
if ( !cfg ) {
warning() << "could not load db config to balance " << ns << " collection" << endl;
continue;
}
// This line reloads the chunk manager once if this process doesn't know the collection
// is sharded yet.
ChunkManagerPtr cm = cfg->getChunkManagerIfExists( ns, true );
if ( !cm ) {
warning() << "could not load chunks to balance " << ns << " collection" << endl;
continue;
}
// loop through tags to make sure no chunk spans tags; splits on tag min. for all chunks
bool didAnySplits = false;
for ( unsigned i = 0; i < ranges.size(); i++ ) {
BSONObj min = ranges[i].min;
min = cm->getShardKey().extendRangeBound( min, false );
if ( allChunkMinimums.count( min ) > 0 )
continue;
didAnySplits = true;
log() << "ns: " << ns << " need to split on "
<< min << " because there is a range there" << endl;
ChunkPtr c = cm->findIntersectingChunk( min );
vector<BSONObj> splitPoints;
splitPoints.push_back( min );
BSONObj res;
if ( !c->multiSplit( splitPoints, res ) ) {
error() << "split failed: " << res << endl;
}
else {
LOG(1) << "split worked: " << res << endl;
}
break;
}
if ( didAnySplits ) {
// state change, just wait till next round
continue;
}
CandidateChunk* p = _policy->balance( ns, status, _balancedLastTime );
if ( p ) candidateChunks->push_back( CandidateChunkPtr( p ) );
}
}
void Balancer::_doBalanceRound( DBClientBase& conn, vector<CandidateChunkPtr>* candidateChunks ) {
verify( candidateChunks );
//
// 1. Check whether there is any sharded collection to be balanced by querying
// the ShardsNS::collections collection
//
auto_ptr<DBClientCursor> cursor = conn.query(CollectionType::ConfigNS, BSONObj());
vector< string > collections;
while ( cursor->more() ) {
BSONObj col = cursor->nextSafe();
// sharded collections will have a shard "key".
if ( ! col[CollectionType::keyPattern()].eoo() &&
! col[CollectionType::noBalance()].trueValue() ){
collections.push_back( col[CollectionType::ns()].String() );
}
else if( col[CollectionType::noBalance()].trueValue() ){
LOG(1) << "not balancing collection " << col[CollectionType::ns()].String()
<< ", explicitly disabled" << endl;
}
}
cursor.reset();
if ( collections.empty() ) {
LOG(1) << "no collections to balance" << endl;
return;
}
//
// 2. Get a list of all the shards that are participating in this balance round
// along with any maximum allowed quotas and current utilization. We get the
// latter by issuing db.serverStatus() (mem.mapped) to all shards.
//
// TODO: skip unresponsive shards and mark information as stale.
//
vector<Shard> allShards;
Shard::getAllShards( allShards );
if ( allShards.size() < 2) {
LOG(1) << "can't balance without more active shards" << endl;
return;
}
ShardInfoMap shardInfo;
for ( vector<Shard>::const_iterator it = allShards.begin(); it != allShards.end(); ++it ) {
const Shard& s = *it;
ShardStatus status = s.getStatus();
shardInfo[ s.getName() ] = ShardInfo( s.getMaxSize(),
status.mapped(),
s.isDraining(),
status.hasOpsQueued(),
s.tags()
);
}
//
// 3. For each collection, check if the balancing policy recommends moving anything around.
//
for (vector<string>::const_iterator it = collections.begin(); it != collections.end(); ++it ) {
const string& ns = *it;
map< string,vector<BSONObj> > shardToChunksMap;
cursor = conn.query(ChunkType::ConfigNS,
QUERY(ChunkType::ns(ns)).sort(ChunkType::min()));
set<BSONObj> allChunkMinimums;
while ( cursor->more() ) {
BSONObj chunk = cursor->nextSafe().getOwned();
vector<BSONObj>& chunks = shardToChunksMap[chunk[ChunkType::shard()].String()];
allChunkMinimums.insert( chunk[ChunkType::min()].Obj() );
chunks.push_back( chunk );
}
cursor.reset();
if (shardToChunksMap.empty()) {
LOG(1) << "skipping empty collection (" << ns << ")";
continue;
}
for ( vector<Shard>::iterator i=allShards.begin(); i!=allShards.end(); ++i ) {
// this just makes sure there is an entry in shardToChunksMap for every shard
Shard s = *i;
shardToChunksMap[s.getName()].size();
}
DistributionStatus status( shardInfo, shardToChunksMap );
// load tags
conn.ensureIndex(TagsType::ConfigNS,
BSON(TagsType::ns() << 1 << TagsType::min() << 1),
true);
cursor = conn.query(TagsType::ConfigNS,
QUERY(TagsType::ns(ns)).sort(TagsType::min()));
vector<TagRange> ranges;
while ( cursor->more() ) {
BSONObj tag = cursor->nextSafe();
TagRange tr(tag[TagsType::min()].Obj().getOwned(),
tag[TagsType::max()].Obj().getOwned(),
tag[TagsType::tag()].String());
ranges.push_back(tr);
uassert(16356,
str::stream() << "tag ranges not valid for: " << ns,
status.addTagRange(tr) );
}
cursor.reset();
DBConfigPtr cfg = grid.getDBConfig( ns );
verify( cfg );
ChunkManagerPtr cm = cfg->getChunkManager( ns );
verify( cm );
// loop through tags to make sure no chunk spans tags; splits on tag min. for all chunks
bool didAnySplits = false;
for ( unsigned i = 0; i < ranges.size(); i++ ) {
BSONObj min = ranges[i].min;
min = cm->getShardKey().extendRangeBound( min, false );
if ( allChunkMinimums.count( min ) > 0 )
continue;
didAnySplits = true;
log() << "ns: " << ns << " need to split on "
<< min << " because there is a range there" << endl;
ChunkPtr c = cm->findIntersectingChunk( min );
vector<BSONObj> splitPoints;
splitPoints.push_back( min );
BSONObj res;
if ( !c->multiSplit( splitPoints, res ) ) {
error() << "split failed: " << res << endl;
}
else {
LOG(1) << "split worked: " << res << endl;
}
break;
}
if ( didAnySplits ) {
// state change, just wait till next round
continue;
}
CandidateChunk* p = _policy->balance( ns, status, _balancedLastTime );
if ( p ) candidateChunks->push_back( CandidateChunkPtr( p ) );
}
}
void _insert( Request& r , DbMessage& d, ChunkManagerPtr manager ) {
const int flags = d.reservedField();
bool keepGoing = flags & InsertOption_KeepGoing; // modified before assertion if should abort
while ( d.moreJSObjs() ) {
try {
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;
uasserted( 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 = 30;
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 , flags);
r.gotInsert();
if ( r.getClientInfo()->autoSplitOk() )
c->splitIfShould( o.objsize() );
gotThrough = true;
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 = r.getChunkManager();
LOG( logLevel ) << " sequence number - old: " << old << " new: " << manager->getSequenceNumber() << endl;
if (!manager) {
keepGoing = false;
uasserted(14804, "collection no longer sharded");
}
}
sleepmillis( i * 20 );
}
assert( inShutdown() || gotThrough ); // not caught below
} catch (const UserException&){
if (!keepGoing || !d.moreJSObjs()){
throw;
}
// otherwise ignore and keep going
}
}
}
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 );
vector<shared_ptr<ChunkRange> > shards;
info->getChunksForQuery( shards , query.getFilter() );
set<ServerAndQuery> servers;
for ( vector<shared_ptr<ChunkRange> >::iterator i = shards.begin(); i != shards.end(); i++ ){
shared_ptr<ChunkRange> c = *i;
//servers.insert( ServerAndQuery( c->getShard().getConnString() , BSONObj() ) ); // ERH ERH ERH
servers.insert( ServerAndQuery( c->getShard().getConnString() , c->getFilter() ) );
}
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() ){
// 1. no sort, can just hit them in serial
cursor = new SerialServerClusteredCursor( servers , q );
}
else {
int shardKeyOrder = info->getShardKey().canOrder( sort );
if ( shardKeyOrder ){
// 2. sort on shard key, can do in serial intelligently
set<ServerAndQuery> buckets;
for ( vector<shared_ptr<ChunkRange> >::iterator i = shards.begin(); i != shards.end(); i++ ){
shared_ptr<ChunkRange> s = *i;
buckets.insert( ServerAndQuery( s->getShard().getConnString() , s->getFilter() , s->getMin() ) );
}
cursor = new SerialServerClusteredCursor( buckets , q , shardKeyOrder );
}
else {
// 3. sort on non-sharded key, pull back a portion from each server and iterate slowly
cursor = new ParallelSortClusteredCursor( servers , q , sort );
}
}
assert( cursor );
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;
}
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;
if ( multi )
uassert( 10202 , "can't mix multi and upsert and sharding" , ! upsert );
if ( upsert && !(manager->hasShardKey(toupdate) ||
(toupdate.firstElement().fieldName()[0] == '$' && manager->hasShardKey(query))))
{
throw UserException( 8012 , "can't upsert something without shard key" );
}
bool save = false;
if ( ! manager->hasShardKey( query ) ){
if ( multi ){
}
else if ( query.nFields() != 1 || strcmp( query.firstElement().fieldName() , "_id" ) ){
throw UserException( 8013 , "can't do update with query that doesn't have the shard key" );
}
else {
save = true;
chunkFinder = toupdate;
}
}
if ( ! save ){
if ( toupdate.firstElement().fieldName()[0] == '$' ){
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 ){
vector<shared_ptr<ChunkRange> > chunks;
manager->getChunksForQuery( chunks , chunkFinder );
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;
doWrite( dbUpdate , r , c->getShard() );
seen.insert( c->getShard() );
}
}
else {
ChunkPtr c = manager->findChunk( chunkFinder );
doWrite( dbUpdate , r , c->getShard() );
c->splitIfShould( d.msg().header()->dataLen() );
}
}
