void _delete( Request& r , DbMessage& d, ChunkManager* manager ){
int flags = d.pullInt();
bool justOne = flags & 1;
uassert( 10203 , "bad delete message" , d.moreJSObjs() );
BSONObj pattern = d.nextJsObj();
vector<Chunk*> chunks;
manager->getChunksForQuery( chunks , pattern );
cout << "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<string> seen;
for ( vector<Chunk*>::iterator i=chunks.begin(); i!=chunks.end(); i++){
Chunk * c = *i;
if ( seen.count( c->getShard() ) )
continue;
seen.insert( c->getShard() );
doWrite( dbDelete , r , c->getShard() );
}
}
void _insert( Request& r , DbMessage& d, ChunkManager* 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" );
}
}
Chunk& c = manager->findChunk( o );
log(4) << " server:" << c.getShard() << " " << o << endl;
insert( c.getShard() , r.getns() , o );
c.splitIfShould( o.objsize() );
}
}
void _delete( Request& r , DbMessage& d, ShardManager* manager ){
int flags = d.pullInt();
bool justOne = flags & 1;
uassert( "bad delete message" , d.moreJSObjs() );
BSONObj pattern = d.nextJsObj();
if ( manager->hasShardKey( pattern ) ){
Shard& s = manager->findShard( pattern );
doWrite( dbDelete , r , s.getServer() );
return;
}
if ( ! justOne && ! pattern.hasField( "_id" ) )
throw UserException( "can only delete with a non-shard key pattern if can delete as many as we find" );
vector<Shard*> shards;
manager->getShardsForQuery( shards , pattern );
set<string> seen;
for ( vector<Shard*>::iterator i=shards.begin(); i!=shards.end(); i++){
Shard * s = *i;
if ( seen.count( s->getServer() ) )
continue;
seen.insert( s->getServer() );
doWrite( dbDelete , r , s->getServer() );
}
}
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);
}
}
/**
* Semantics for insert are ContinueOnError - to match mongod semantics :
* 1) Error is thrown immediately for corrupt objects
* 2) Error is thrown only for UserExceptions during the insert process, if last obj had error that's thrown
*/
void _insert( Request& r , DbMessage& d, ChunkManagerPtr manager ){
vector<BSONObj> insertsRemaining;
while ( d.moreJSObjs() ){
insertsRemaining.push_back( d.nextJsObj() );
}
map<ChunkPtr, vector<BSONObj> > insertsForChunks; // Map for bulk inserts to diff chunks
_insert( r, d, manager, insertsRemaining, insertsForChunks );
}
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 _insert( Request& r , DbMessage& d, ShardManager* manager ){
while ( d.moreJSObjs() ){
BSONObj o = d.nextJsObj();
if ( ! manager->hasShardKey( o ) ){
log() << "tried to insert object without shard key: " << r.getns() << " " << o << endl;
throw UserException( "tried to insert object without shard key" );
}
Shard& s = manager->findShard( o );
log(4) << " server:" << s.getServer() << " " << o << endl;
insert( s.getServer() , r.getns() , o );
s.splitIfShould( o.objsize() );
}
}
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 _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 _update( Request& r , DbMessage& d, ShardManager* manager ){
int flags = d.pullInt();
BSONObj query = d.nextJsObj();
uassert( "invalid update" , d.moreJSObjs() );
BSONObj toupdate = d.nextJsObj();
bool upsert = flags & 1;
if ( upsert && ! manager->hasShardKey( toupdate ) )
throw UserException( "can't upsert something without shard key" );
if ( ! manager->hasShardKey( query ) )
throw UserException( "can't do update with query that doesn't have the shard key" );
if ( manager->hasShardKey( toupdate ) && manager->getShardKey().compare( query , toupdate ) )
throw UserException( "change would move shards!" );
Shard& s = manager->findShard( toupdate );
doWrite( dbUpdate , r , s.getServer() );
s.splitIfShould( d.msg().data->dataLen() );
}
void _update( Request& r , DbMessage& d, ChunkManager* 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] == '$' ){
// TODO: check for $set, etc.. on shard key
}
else if ( manager->hasShardKey( toupdate ) && manager->getShardKey().compare( query , toupdate ) ){
throw UserException( 8014 , "change would move shards!" );
}
}
if ( multi ){
vector<Chunk*> chunks;
manager->getChunksForQuery( chunks , chunkFinder );
set<string> seen;
for ( vector<Chunk*>::iterator i=chunks.begin(); i!=chunks.end(); i++){
Chunk * c = *i;
if ( seen.count( c->getShard() ) )
continue;
doWrite( dbUpdate , r , c->getShard() );
seen.insert( c->getShard() );
}
}
else {
Chunk& c = manager->findChunk( chunkFinder );
doWrite( dbUpdate , r , c.getShard() );
c.splitIfShould( d.msg().data->dataLen() );
}
}
void _update( Request& r , DbMessage& d, ChunkManagerPtr manager ){
int flags = d.pullInt();
BSONObj query = d.nextJsObj();
uassert( 10201 , "invalid update" , d.moreJSObjs() );
BSONObj toupdate = d.nextJsObj();
BSONObj chunkFinder = query;
bool upsert = flags & UpdateOption_Upsert;
bool multi = flags & UpdateOption_Multi;
uassert( 10202 , "can't mix multi and upsert and sharding" , ! ( upsert && multi ) );
if ( upsert && !(manager->hasShardKey(toupdate) ||
(toupdate.firstElement().fieldName()[0] == '$' && 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 _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
}
}
}
void _insert( Request& r , DbMessage& d, ChunkManagerPtr manager, vector<BSONObj>& insertsRemaining, map<ChunkPtr, vector<BSONObj> > insertsForChunks, int retries = 0 ) {
uassert( 16055, str::stream() << "too many retries during bulk insert, " << insertsRemaining.size() << " inserts remaining", retries < 30 );
uassert( 16056, str::stream() << "shutting down server during bulk insert, " << insertsRemaining.size() << " inserts remaining", ! inShutdown() );
const int flags = d.reservedField() | InsertOption_ContinueOnError; // ContinueOnError is always on when using sharding.
_groupInserts( manager, insertsRemaining, insertsForChunks );
while( ! insertsForChunks.empty() ){
ChunkPtr c = insertsForChunks.begin()->first;
vector<BSONObj>& objs = insertsForChunks.begin()->second;
const Shard& shard = c->getShard();
const string& ns = r.getns();
ShardConnection dbcon( shard, ns, manager );
try {
LOG(4) << " server:" << c->getShard().toString() << " bulk insert " << objs.size() << " documents" << endl;
// Taken from single-shard bulk insert, should not need multiple methods in future
// insert( c->getShard() , r.getns() , objs , flags);
// It's okay if the version is set here, an exception will be thrown if the version is incompatible
dbcon.setVersion();
dbcon->insert( ns , objs , flags);
// TODO: Option for safe inserts here - can then use this for all inserts
dbcon.done();
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();
}
// TODO: The only reason we're grouping by chunks here is for auto-split, more efficient
// to track this separately and bulk insert to shards
if ( r.getClientInfo()->autoSplitOk() )
c->splitIfShould( bytesWritten );
}
catch ( StaleConfigException& e ) {
// Cleanup the connection
dbcon.done();
// Assume the inserts did *not* succeed, so we don't want to erase them
int logLevel = retries < 2;
LOG( logLevel ) << "retrying bulk insert of " << objs.size() << " documents to chunk " << c << " because of StaleConfigException: " << e << endl;
if( retries > 2 ){
versionManager.forceRemoteCheckShardVersionCB( e.getns() );
}
// TODO: Replace with actual chunk handling code, simplify request
r.reset();
manager = r.getChunkManager();
if( ! manager ) {
// TODO : We can probably handle this better?
uasserted( 14804, "collection no longer sharded" );
}
// End TODO
// We may need to regroup at least some of our inserts since our chunk manager may have changed
_insert( r, d, manager, insertsRemaining, insertsForChunks, retries + 1 );
return;
}
catch( UserException& ){
// Unexpected exception, so don't clean up the conn
dbcon.kill();
// These inserts won't be retried, as something weird happened here
insertsForChunks.erase( insertsForChunks.begin() );
// Throw if this is the last chunk bulk-inserted to
if( insertsForChunks.empty() ){
throw;
}
}
insertsForChunks.erase( insertsForChunks.begin() );
}
}
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() );
}
}
