void run() {
DurTransaction txn;
create();
ASSERT_EQUALS( 2, nExtents() );
DiskLoc l[ 8 ];
for ( int i = 0; i < 8; ++i ) {
StatusWith<DiskLoc> status = collection()->insertDocument( &txn, bigObj(), true );
ASSERT( status.isOK() );
l[ i ] = status.getValue();
ASSERT( !l[ i ].isNull() );
//ASSERT_EQUALS( i < 2 ? i + 1 : 3 + i % 2, nRecords() );
//if ( i > 3 )
// ASSERT( l[ i ] == l[ i - 4 ] );
}
ASSERT( nRecords() == 8 );
// Too big
BSONObjBuilder bob;
bob.appendOID( "_id", NULL, true );
bob.append( "a", string( MinExtentSize + 500, 'a' ) ); // min extent size is now 4096
BSONObj bigger = bob.done();
StatusWith<DiskLoc> status = collection()->insertDocument( &txn, bigger, false );
ASSERT( !status.isOK() );
ASSERT_EQUALS( 0, nRecords() );
}
void getMe(BSONObj& me) {
const string myname = getHostName();
Client::Transaction transaction(0);
// local.me is an identifier for a server for getLastError w:2+
if (!Collection::findOne("local.me", BSONObj(), me) ||
!me.hasField("host") || me["host"].String() != myname) {
// cleaning out local.me requires write
// lock. This is a rare operation, so it should
// be ok
if (!Lock::isWriteLocked("local")) {
throw RetryWithWriteLock();
}
// clean out local.me
deleteObjects("local.me", BSONObj(), false, false);
// repopulate
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
b.append( "host", myname );
me = b.obj();
updateObjects("local.me", me, BSONObj(), true, false);
}
transaction.commit(0);
}
bool VersionManager::initShardVersionCB( DBClientBase * conn_in, BSONObj& result ){
WriteBackListener::init( *conn_in );
DBClientBase* conn = getVersionable( conn_in );
verify( conn ); // errors thrown above
BSONObjBuilder cmdBuilder;
cmdBuilder.append( "setShardVersion" , "" );
cmdBuilder.appendBool( "init", true );
cmdBuilder.append( "configdb" , configServer.modelServer() );
cmdBuilder.appendOID( "serverID" , &serverID );
cmdBuilder.appendBool( "authoritative" , true );
BSONObj cmd = cmdBuilder.obj();
LOG(1) << "initializing shard connection to " << conn->toString() << endl;
LOG(2) << "initial sharding settings : " << cmd << endl;
bool ok = conn->runCommand( "admin" , cmd , result );
// HACK for backwards compatibility with v1.8.x, v2.0.0 and v2.0.1
// Result is false, but will still initialize serverID and configdb
if( ! ok && ! result["errmsg"].eoo() && ( result["errmsg"].String() == "need to specify namespace"/* 2.0.1/2 */ ||
result["errmsg"].String() == "need to speciy namespace" /* 1.8 */ ))
{
ok = true;
}
LOG(3) << "initial sharding result : " << result << endl;
return ok;
}
static BSONObj bigObj() {
BSONObjBuilder b;
b.appendOID("_id", 0, true);
string as( 187, 'a' );
b.append( "a", as );
return b.obj();
}
void SyncSourceFeedback::ensureMe(OperationContext* txn) {
string myname = getHostName();
{
Lock::DBLock dlk(txn->lockState(), "local", MODE_X);
WriteUnitOfWork wunit(txn);
Client::Context ctx(txn, "local");
// local.me is an identifier for a server for getLastError w:2+
if (!Helpers::getSingleton(txn, "local.me", _me) ||
!_me.hasField("host") ||
_me["host"].String() != myname) {
// clean out local.me
Helpers::emptyCollection(txn, "local.me");
// repopulate
BSONObjBuilder b;
b.appendOID("_id", 0, true);
b.append("host", myname);
_me = b.obj();
Helpers::putSingleton(txn, "local.me", _me);
}
wunit.commit();
// _me is used outside of a read lock, so we must copy it out of the mmap
_me = _me.getOwned();
}
}
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() );
}
}
BSONObj StorageEngine::getoid(string oid) {
OID oidobj;
oidobj.init(oid);
BSONObjBuilder ob;
ob.appendOID("_id",&oidobj);
return ob.obj();
}
bool initShardVersion( DBClientBase& conn_in, BSONObj& result ){
WriteBackListener::init( conn_in );
DBClientBase* conn = getVersionable( &conn_in );
assert( conn ); // errors thrown above
BSONObjBuilder cmdBuilder;
cmdBuilder.append( "setShardVersion" , "" );
cmdBuilder.appendBool( "init", true );
cmdBuilder.append( "configdb" , configServer.modelServer() );
cmdBuilder.appendOID( "serverID" , &serverID );
cmdBuilder.appendBool( "authoritative" , true );
BSONObj cmd = cmdBuilder.obj();
LOG(1) << "initializing shard connection to " << conn->toString() << endl;
LOG(2) << "initial sharding settings : " << cmd << endl;
bool ok = conn->runCommand( "admin" , cmd , result );
LOG(3) << "initial sharding result : " << result << endl;
return ok;
}
bool VersionManager::initShardVersionCB( DBClientBase * conn_in, BSONObj& result ){
WriteBackListener::init( *conn_in );
bool ok;
DBClientBase* conn = NULL;
try {
// May throw if replica set primary is down
conn = getVersionable( conn_in );
dassert( conn ); // errors thrown above
BSONObjBuilder cmdBuilder;
cmdBuilder.append( "setShardVersion" , "" );
cmdBuilder.appendBool( "init", true );
cmdBuilder.append( "configdb" , configServer.modelServer() );
cmdBuilder.appendOID( "serverID" , &serverID );
cmdBuilder.appendBool( "authoritative" , true );
BSONObj cmd = cmdBuilder.obj();
LOG(1) << "initializing shard connection to " << conn->toString() << endl;
LOG(2) << "initial sharding settings : " << cmd << endl;
ok = conn->runCommand("admin", cmd, result, 0);
}
catch( const DBException& ex ) {
bool ignoreFailure = ShardConnection::ignoreInitialVersionFailure
&& conn_in->type() == ConnectionString::SET;
if ( !ignoreFailure )
throw;
// Using initShardVersion is not strictly required when talking to replica sets - it is
// preferred to do so because it registers mongos early with the mongod. This info is
// also sent by checkShardVersion before a connection is used for a write or read.
OCCASIONALLY {
warning() << "failed to initialize new replica set connection version, "
<< "will initialize on first use" << endl;
}
return true;
}
// HACK for backwards compatibility with v1.8.x, v2.0.0 and v2.0.1
// Result is false, but will still initialize serverID and configdb
if( ! ok && ! result["errmsg"].eoo() && ( result["errmsg"].String() == "need to specify namespace"/* 2.0.1/2 */ ||
result["errmsg"].String() == "need to speciy namespace" /* 1.8 */ ))
{
ok = true;
}
LOG(3) << "initial sharding result : " << result << endl;
return ok;
}
BSONObj BatchedCommandResponse::toBSON() const {
BSONObjBuilder builder;
if (_isOkSet)
builder.append(ok(), _ok);
if (_isErrCodeSet)
builder.append(errCode(), _errCode);
if (_isErrMessageSet)
builder.append(errMessage(), _errMessage);
if (_isNModifiedSet)
builder.appendNumber(nModified(), _nModified);
if (_isNSet)
builder.appendNumber(n(), _n);
if (_upsertDetails.get()) {
BSONArrayBuilder upsertedBuilder(builder.subarrayStart(upsertDetails()));
for (std::vector<BatchedUpsertDetail*>::const_iterator it = _upsertDetails->begin();
it != _upsertDetails->end();
++it) {
BSONObj upsertedDetailsDocument = (*it)->toBSON();
upsertedBuilder.append(upsertedDetailsDocument);
}
upsertedBuilder.done();
}
if (_isLastOpSet) {
if (_lastOp.getTerm() != repl::OpTime::kUninitializedTerm) {
_lastOp.append(&builder, "opTime");
} else {
builder.append("opTime", _lastOp.getTimestamp());
}
}
if (_isElectionIdSet)
builder.appendOID(electionId(), const_cast<OID*>(&_electionId));
if (_writeErrorDetails.get()) {
BSONArrayBuilder errDetailsBuilder(builder.subarrayStart(writeErrors()));
for (std::vector<WriteErrorDetail*>::const_iterator it = _writeErrorDetails->begin();
it != _writeErrorDetails->end();
++it) {
BSONObj errDetailsDocument = (*it)->toBSON();
errDetailsBuilder.append(errDetailsDocument);
}
errDetailsBuilder.done();
}
if (_wcErrDetails.get()) {
builder.append(writeConcernError(), _wcErrDetails->toBSON());
}
return builder.obj();
}
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 run(){
int secsToSleep = 0;
while ( 1 ){
try {
ScopedDbConnection conn( _addr );
BSONObj result;
{
BSONObjBuilder cmd;
cmd.appendOID( "writebacklisten" , &serverID );
if ( ! conn->runCommand( "admin" , cmd.obj() , result ) ){
log() << "writebacklisten command failed! " << result << endl;
conn.done();
continue;
}
}
log(1) << "writebacklisten result: " << result << endl;
BSONObj data = result.getObjectField( "data" );
if ( data.getBoolField( "writeBack" ) ){
string ns = data["ns"].valuestrsafe();
int len;
Message m( (void*)data["msg"].binData( len ) , false );
massert( 10427 , "invalid writeback message" , m.header()->valid() );
grid.getDBConfig( ns )->getChunkManager( ns , true );
Request r( m , 0 );
r.process();
}
else {
log() << "unknown writeBack result: " << result << endl;
}
conn.done();
secsToSleep = 0;
}
catch ( std::exception e ){
log() << "WriteBackListener exception : " << e.what() << endl;
}
catch ( ... ){
log() << "WriteBackListener uncaught exception!" << endl;
}
secsToSleep++;
sleepsecs(secsToSleep);
if ( secsToSleep > 10 )
secsToSleep = 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 );
}
}
void insert( const BSONObj &o ) {
if ( o["_id"].eoo() ) {
BSONObjBuilder b;
OID oid;
oid.init();
b.appendOID( "_id", &oid );
b.appendElements( o );
_collection->insertDocument( &_txn, b.obj(), false );
}
else {
_collection->insertDocument( &_txn, o, false );
}
}
void insert(const char* s) {
WriteUnitOfWork wunit(&_txn);
const BSONObj o = fromjson(s);
if (o["_id"].eoo()) {
BSONObjBuilder b;
OID oid;
oid.init();
b.appendOID("_id", &oid);
b.appendElements(o);
_collection->insertDocument(&_txn, b.obj(), false);
} else {
_collection->insertDocument(&_txn, o, false);
}
wunit.commit();
}
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 insert(const char* s) {
WriteUnitOfWork wunit(&_opCtx);
const BSONObj o = fromjson(s);
OpDebug* const nullOpDebug = nullptr;
if (o["_id"].eoo()) {
BSONObjBuilder b;
OID oid;
oid.init();
b.appendOID("_id", &oid);
b.appendElements(o);
_collection->insertDocument(&_opCtx, b.obj(), nullOpDebug, false);
} else {
_collection->insertDocument(&_opCtx, o, nullOpDebug, false);
}
wunit.commit();
}
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();
}
bool setShardVersion( DBClientBase & conn , const string& ns , ShardChunkVersion version , bool authoritative , BSONObj& result ){
BSONObjBuilder cmdBuilder;
cmdBuilder.append( "setShardVersion" , ns.c_str() );
cmdBuilder.append( "configdb" , configServer.modelServer() );
cmdBuilder.appendTimestamp( "version" , version );
cmdBuilder.appendOID( "serverID" , &serverID );
if ( authoritative )
cmdBuilder.appendBool( "authoritative" , 1 );
Shard s = Shard::make( conn.getServerAddress() );
cmdBuilder.append( "shard" , s.getName() );
cmdBuilder.append( "shardHost" , s.getConnString() );
BSONObj cmd = cmdBuilder.obj();
log(1) << " setShardVersion " << s.getName() << " " << conn.getServerAddress() << " " << ns << " " << cmd << " " << &conn << endl;
return conn.runCommand( "admin" , cmd , result );
}
void Model::save( bool check ){
ScopedDbConnection conn( modelServer() );
BSONObjBuilder b;
serialize( b );
if ( _id.isEmpty() ){
OID oid;
oid.init();
b.appendOID( "_id" , &oid );
BSONObj o = b.obj();
conn->insert( getNS() , o );
_id = o["_id"].wrap().getOwned();
log(4) << "inserted new model " << getNS() << " " << o << endl;
}
else {
BSONElement id = _id["_id"];
b.append( id );
BSONObjBuilder qb;
qb.append( id );
BSONObj q = qb.obj();
BSONObj o = b.obj();
log(4) << "updated old model" << getNS() << " " << q << " " << o << endl;
conn->update( getNS() , q , o );
}
string errmsg = "";
if ( check )
errmsg = conn->getLastError();
conn.done();
if ( check && errmsg.size() )
throw UserException( (string)"error on Model::save: " + errmsg );
}
void SyncSourceFeedback::ensureMe() {
string myname = getHostName();
{
Client::WriteContext ctx("local");
// local.me is an identifier for a server for getLastError w:2+
if (!Helpers::getSingleton("local.me", _me) ||
!_me.hasField("host") ||
_me["host"].String() != myname) {
// clean out local.me
Helpers::emptyCollection("local.me");
// repopulate
BSONObjBuilder b;
b.appendOID("_id", 0, true);
b.append("host", myname);
_me = b.obj();
Helpers::putSingleton("local.me", _me);
}
}
}
// Adds a wOpTime and a wElectionId field to a set of gle options
static BSONObj buildGLECmdWithOpTime( const BSONObj& gleOptions,
const OpTime& opTime,
const OID& electionId ) {
BSONObjBuilder builder;
BSONObjIterator it( gleOptions );
for ( int i = 0; it.more(); ++i ) {
BSONElement el = it.next();
// Make sure first element is getLastError : 1
if ( i == 0 ) {
StringData elName( el.fieldName() );
if ( !elName.equalCaseInsensitive( "getLastError" ) ) {
builder.append( "getLastError", 1 );
}
}
builder.append( el );
}
builder.appendTimestamp( "wOpTime", opTime.asDate() );
builder.appendOID( "wElectionId", const_cast<OID*>(&electionId) );
return builder.obj();
}
bool replHandshake(DBClientConnection *conn) {
BSONObj me;
{
dblock l;
// local.me is an identifier for a server for getLastError w:2+
if ( ! Helpers::getSingleton( "local.me" , me ) ) {
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
me = b.obj();
Helpers::putSingleton( "local.me" , me );
}
}
BSONObjBuilder cmd;
cmd.appendAs( me["_id"] , "handshake" );
BSONObj res;
bool ok = conn->runCommand( "admin" , cmd.obj() , res );
// ignoring for now on purpose for older versions
log(ok) << "replHandshake res not: " << ok << " res: " << res << endl;
return true;
}
bool replHandshake(DBClientConnection *conn) {
string myname = getHostName();
BSONObj me;
{
Lock::DBWrite l("local");
// local.me is an identifier for a server for getLastError w:2+
if ( ! Helpers::getSingleton( "local.me" , me ) ||
! me.hasField("host") ||
me["host"].String() != myname ) {
// clean out local.me
Helpers::emptyCollection("local.me");
// repopulate
BSONObjBuilder b;
b.appendOID( "_id" , 0 , true );
b.append( "host", myname );
me = b.obj();
Helpers::putSingleton( "local.me" , me );
}
}
BSONObjBuilder cmd;
cmd.appendAs( me["_id"] , "handshake" );
if (theReplSet) {
cmd.append("member", theReplSet->selfId());
cmd.append("config", theReplSet->myConfig().asBson());
}
BSONObj res;
bool ok = conn->runCommand( "admin" , cmd.obj() , res );
// ignoring for now on purpose for older versions
LOG( ok ? 1 : 0 ) << "replHandshake res not: " << ok << " res: " << res << endl;
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.
}
}
void v8ToMongoElement( BSONObjBuilder & b , v8::Handle<v8::String> name , const string sname , v8::Handle<v8::Value> value ){
if ( value->IsString() ){
if ( sname == "$where" )
b.appendCode( sname.c_str() , toSTLString( value ).c_str() );
else
b.append( sname.c_str() , toSTLString( value ).c_str() );
return;
}
if ( value->IsFunction() ){
b.appendCode( sname.c_str() , toSTLString( value ).c_str() );
return;
}
if ( value->IsNumber() ){
b.append( sname.c_str() , value->ToNumber()->Value() );
return;
}
if ( value->IsArray() ){
BSONObj sub = v8ToMongo( value->ToObject() );
b.appendArray( sname.c_str() , sub );
return;
}
if ( value->IsDate() ){
b.appendDate( sname.c_str() , (unsigned long long )(v8::Date::Cast( *value )->NumberValue()) );
return;
}
if ( value->IsObject() ){
string s = toSTLString( value );
if ( s.size() && s[0] == '/' ){
s = s.substr( 1 );
string r = s.substr( 0 , s.find( "/" ) );
string o = s.substr( s.find( "/" ) + 1 );
b.appendRegex( sname.c_str() , r.c_str() , o.c_str() );
}
else if ( value->ToObject()->GetPrototype()->IsObject() &&
value->ToObject()->GetPrototype()->ToObject()->HasRealNamedProperty( String::New( "isObjectId" ) ) ){
OID oid;
oid.init( toSTLString( value ) );
b.appendOID( sname.c_str() , &oid );
}
else {
BSONObj sub = v8ToMongo( value->ToObject() );
b.append( sname.c_str() , sub );
}
return;
}
if ( value->IsBoolean() ){
b.appendBool( sname.c_str() , value->ToBoolean()->Value() );
return;
}
else if ( value->IsUndefined() ){
return;
}
else if ( value->IsNull() ){
b.appendNull( sname.c_str() );
return;
}
cout << "don't know how to covert to mongo field [" << name << "]\t" << value << endl;
}
void WriteBackListener::run() {
int secsToSleep = 0;
while ( ! inShutdown() ) {
if ( ! Shard::isAShardNode( _addr ) ) {
log(1) << _addr << " is not a shard node" << endl;
sleepsecs( 60 );
continue;
}
try {
ScopedDbConnection conn( _addr );
BSONObj result;
{
BSONObjBuilder cmd;
cmd.appendOID( "writebacklisten" , &serverID ); // Command will block for data
if ( ! conn->runCommand( "admin" , cmd.obj() , result ) ) {
log() << "writebacklisten command failed! " << result << endl;
conn.done();
continue;
}
}
log(1) << "writebacklisten result: " << result << endl;
BSONObj data = result.getObjectField( "data" );
if ( data.getBoolField( "writeBack" ) ) {
string ns = data["ns"].valuestrsafe();
ConnectionIdent cid( "" , 0 );
OID wid;
if ( data["connectionId"].isNumber() && data["id"].type() == jstOID ) {
string s = "";
if ( data["instanceIdent"].type() == String )
s = data["instanceIdent"].String();
cid = ConnectionIdent( s , data["connectionId"].numberLong() );
wid = data["id"].OID();
}
else {
warning() << "mongos/mongod version mismatch (1.7.5 is the split)" << endl;
}
int len; // not used, but needed for next call
Message m( (void*)data["msg"].binData( len ) , false );
massert( 10427 , "invalid writeback message" , m.header()->valid() );
DBConfigPtr db = grid.getDBConfig( ns );
ShardChunkVersion needVersion( data["version"] );
LOG(1) << "connectionId: " << cid << " writebackId: " << wid << " needVersion : " << needVersion.toString()
<< " mine : " << db->getChunkManager( ns )->getVersion().toString() << endl;// TODO change to log(3)
if ( logLevel ) log(1) << debugString( m ) << endl;
if ( needVersion.isSet() && needVersion <= db->getChunkManager( ns )->getVersion() ) {
// this means when the write went originally, the version was old
// if we're here, it means we've already updated the config, so don't need to do again
//db->getChunkManager( ns , true ); // SERVER-1349
}
else {
// we received a writeback object that was sent to a previous version of a shard
// the actual shard may not have the object the writeback operation is for
// we need to reload the chunk manager and get the new shard versions
db->getChunkManager( ns , true );
}
// do request and then call getLastError
// we have to call getLastError so we can return the right fields to the user if they decide to call getLastError
BSONObj gle;
try {
Request r( m , 0 );
r.init();
ClientInfo * ci = r.getClientInfo();
if (!noauth) {
ci->getAuthenticationInfo()->authorize("admin", internalSecurity.user);
}
ci->noAutoSplit();
r.process();
ci->newRequest(); // this so we flip prev and cur shards
BSONObjBuilder b;
if ( ! ci->getLastError( BSON( "getLastError" << 1 ) , b , true ) ) {
b.appendBool( "commandFailed" , true );
}
gle = b.obj();
ci->clearSinceLastGetError();
}
catch ( DBException& e ) {
error() << "error processing writeback: " << e << endl;
BSONObjBuilder b;
b.append( "err" , e.toString() );
e.getInfo().append( b );
gle = b.obj();
}
{
scoped_lock lk( _seenWritebacksLock );
WBStatus& s = _seenWritebacks[cid];
s.id = wid;
s.gle = gle;
}
}
else if ( result["noop"].trueValue() ) {
// no-op
}
else {
log() << "unknown writeBack result: " << result << endl;
}
conn.done();
secsToSleep = 0;
continue;
}
catch ( std::exception& e ) {
if ( inShutdown() ) {
// we're shutting down, so just clean up
return;
}
log() << "WriteBackListener exception : " << e.what() << endl;
// It's possible this shard was removed
Shard::reloadShardInfo();
}
catch ( ... ) {
log() << "WriteBackListener uncaught exception!" << endl;
}
secsToSleep++;
sleepsecs(secsToSleep);
if ( secsToSleep > 10 )
secsToSleep = 0;
}
log() << "WriteBackListener exiting : address no longer in cluster " << _addr;
}
void WriteBackListener::run() {
int secsToSleep = 0;
scoped_ptr<ChunkVersion> lastNeededVersion;
int lastNeededCount = 0;
bool needsToReloadShardInfo = false;
while ( ! inShutdown() ) {
if ( ! Shard::isAShardNode( _addr ) ) {
LOG(1) << _addr << " is not a shard node" << endl;
sleepsecs( 60 );
continue;
}
try {
if (needsToReloadShardInfo) {
// It's possible this shard was removed
Shard::reloadShardInfo();
needsToReloadShardInfo = false;
}
scoped_ptr<ScopedDbConnection> conn(
ScopedDbConnection::getInternalScopedDbConnection( _addr ) );
BSONObj result;
{
BSONObjBuilder cmd;
cmd.appendOID( "writebacklisten" , &serverID ); // Command will block for data
if ( ! conn->get()->runCommand( "admin" , cmd.obj() , result ) ) {
result = result.getOwned();
log() << "writebacklisten command failed! " << result << endl;
conn->done();
continue;
}
}
conn->done();
LOG(1) << "writebacklisten result: " << result << endl;
BSONObj data = result.getObjectField( "data" );
if ( data.getBoolField( "writeBack" ) ) {
string ns = data["ns"].valuestrsafe();
ConnectionIdent cid( "" , 0 );
OID wid;
if ( data["connectionId"].isNumber() && data["id"].type() == jstOID ) {
string s = "";
if ( data["instanceIdent"].type() == String )
s = data["instanceIdent"].String();
cid = ConnectionIdent( s , data["connectionId"].numberLong() );
wid = data["id"].OID();
}
else {
warning() << "mongos/mongod version mismatch (1.7.5 is the split)" << endl;
}
int len; // not used, but needed for next call
Message msg( (void*)data["msg"].binData( len ) , false );
massert( 10427 , "invalid writeback message" , msg.header()->valid() );
DBConfigPtr db = grid.getDBConfig( ns );
ChunkVersion needVersion = ChunkVersion::fromBSON( data, "version" );
//
// TODO: Refactor the sharded strategy to correctly handle all sharding state changes itself,
// we can't rely on WBL to do this for us b/c anything could reset our state in-between.
// We should always reload here for efficiency when possible, but staleness is also caught in the
// loop below.
//
ChunkManagerPtr manager;
ShardPtr primary;
db->getChunkManagerOrPrimary( ns, manager, primary );
ChunkVersion currVersion;
if( manager ) currVersion = manager->getVersion();
LOG(1) << "connectionId: " << cid << " writebackId: " << wid << " needVersion : " << needVersion.toString()
<< " mine : " << currVersion.toString() << endl;
LOG(1) << msg.toString() << endl;
//
// We should reload only if we need to update our version to be compatible *and* we
// haven't already done so. This avoids lots of reloading when we remove/add a sharded collection
//
bool alreadyReloaded = lastNeededVersion &&
lastNeededVersion->isEquivalentTo( needVersion );
if( alreadyReloaded ){
LOG(1) << "wbl already reloaded config information for version "
<< needVersion << ", at version " << currVersion << endl;
}
else if( lastNeededVersion ) {
log() << "new version change detected to " << needVersion.toString()
<< ", " << lastNeededCount << " writebacks processed at "
<< lastNeededVersion->toString() << endl;
lastNeededCount = 0;
}
//
// Set our lastNeededVersion for next time
//
lastNeededVersion.reset( new ChunkVersion( needVersion ) );
lastNeededCount++;
//
// Determine if we should reload, if so, reload
//
bool shouldReload = ! needVersion.isWriteCompatibleWith( currVersion ) &&
! alreadyReloaded;
if( shouldReload && currVersion.isSet()
&& needVersion.isSet()
&& currVersion.hasCompatibleEpoch( needVersion ) )
{
//
// If we disagree about versions only, reload the chunk manager
//
db->getChunkManagerIfExists( ns, true );
}
else if( shouldReload ){
//
// If we disagree about anything else, reload the full db
//
warning() << "reloading config data for " << db->getName() << ", "
<< "wanted version " << needVersion.toString()
<< " but currently have version " << currVersion.toString() << endl;
db->reload();
}
// do request and then call getLastError
// we have to call getLastError so we can return the right fields to the user if they decide to call getLastError
BSONObj gle;
int attempts = 0;
while ( true ) {
attempts++;
try {
Request r( msg , 0 );
r.init();
r.d().reservedField() |= Reserved_FromWriteback;
ClientInfo * ci = r.getClientInfo();
if (!noauth) {
ci->getAuthorizationManager()->grantInternalAuthorization(
"_writebackListener");
}
ci->noAutoSplit();
r.process( attempts );
ci->newRequest(); // this so we flip prev and cur shards
BSONObjBuilder b;
string errmsg;
if ( ! ci->getLastError( "admin",
BSON( "getLastError" << 1 ),
b,
errmsg,
true ) )
{
b.appendBool( "commandFailed" , true );
if( ! b.hasField( "errmsg" ) ){
b.append( "errmsg", errmsg );
gle = b.obj();
}
else if( errmsg.size() > 0 ){
// Rebuild GLE object with errmsg
// TODO: Make this less clumsy by improving GLE interface
gle = b.obj();
if( gle["errmsg"].type() == String ){
BSONObj gleNoErrmsg =
gle.filterFieldsUndotted( BSON( "errmsg" << 1 ),
false );
BSONObjBuilder bb;
bb.appendElements( gleNoErrmsg );
bb.append( "errmsg", gle["errmsg"].String() +
" ::and:: " +
errmsg );
gle = bb.obj().getOwned();
}
}
}
else{
gle = b.obj();
}
if ( gle["code"].numberInt() == 9517 ) {
log() << "new version change detected, "
<< lastNeededCount << " writebacks processed previously" << endl;
lastNeededVersion.reset();
lastNeededCount = 1;
log() << "writeback failed because of stale config, retrying attempts: " << attempts << endl;
LOG(1) << "writeback error : " << gle << endl;
//
// Bringing this in line with the similar retry logic elsewhere
//
// TODO: Reloading the chunk manager may not help if we dropped a
// collection, but we don't actually have that info in the writeback
// error
//
if( attempts <= 2 ){
db->getChunkManagerIfExists( ns, true );
}
else{
versionManager.forceRemoteCheckShardVersionCB( ns );
sleepsecs( attempts - 1 );
}
uassert( 15884, str::stream()
<< "Could not reload chunk manager after "
<< attempts << " attempts.", attempts <= 4 );
continue;
}
ci->clearSinceLastGetError();
}
catch ( DBException& e ) {
error() << "error processing writeback: " << e << endl;
BSONObjBuilder b;
e.getInfo().append( b, "err", "code" );
gle = b.obj();
}
break;
}
{
scoped_lock lk( _seenWritebacksLock );
WBStatus& s = _seenWritebacks[cid];
s.id = wid;
s.gle = gle;
}
}
else if ( result["noop"].trueValue() ) {
// no-op
}
else {
log() << "unknown writeBack result: " << result << endl;
}
secsToSleep = 0;
continue;
}
catch ( std::exception& e ) {
// Attention! Do not call any method that would throw an exception
// (or assert) in this block.
if ( inShutdown() ) {
// we're shutting down, so just clean up
return;
}
log() << "WriteBackListener exception : " << e.what() << endl;
needsToReloadShardInfo = true;
}
catch ( ... ) {
log() << "WriteBackListener uncaught exception!" << endl;
}
secsToSleep++;
sleepsecs(secsToSleep);
if ( secsToSleep > 10 )
secsToSleep = 0;
}
log() << "WriteBackListener exiting : address no longer in cluster " << _addr;
}
bool VersionManager::initShardVersionCB( DBClientBase * conn_in, BSONObj& result ){
WriteBackListener::init( *conn_in );
bool ok;
DBClientBase* conn = NULL;
try {
// May throw if replica set primary is down
conn = getVersionable( conn_in );
dassert( conn ); // errors thrown above
BSONObjBuilder cmdBuilder;
cmdBuilder.append( "setShardVersion" , "" );
cmdBuilder.appendBool( "init", true );
cmdBuilder.append( "configdb" , configServer.modelServer() );
cmdBuilder.appendOID( "serverID" , &serverID );
cmdBuilder.appendBool( "authoritative" , true );
BSONObj cmd = cmdBuilder.obj();
LOG(1) << "initializing shard connection to " << conn->toString() << endl;
LOG(2) << "initial sharding settings : " << cmd << endl;
ok = conn->runCommand("admin", cmd, result, 0);
}
catch( const DBException& ) {
if ( conn_in->type() != ConnectionString::SET ) {
throw;
}
// NOTE: Only old-style cluster operations will talk via DBClientReplicaSets - using
// checkShardVersion is required (which includes initShardVersion information) if these
// connections are used.
OCCASIONALLY {
warning() << "failed to initialize new replica set connection version, "
<< "will initialize on first use" << endl;
}
return true;
}
// HACK for backwards compatibility with v1.8.x, v2.0.0 and v2.0.1
// Result is false, but will still initialize serverID and configdb
if( ! ok && ! result["errmsg"].eoo() && ( result["errmsg"].String() == "need to specify namespace"/* 2.0.1/2 */ ||
result["errmsg"].String() == "need to speciy namespace" /* 1.8 */ ))
{
ok = true;
}
// Record the connection wire version if sent in the response, initShardVersion is a
// handshake for mongos->mongod connections.
if ( !result["minWireVersion"].eoo() ) {
int minWireVersion = result["minWireVersion"].numberInt();
int maxWireVersion = result["maxWireVersion"].numberInt();
conn->setWireVersions( minWireVersion, maxWireVersion );
}
LOG(3) << "initial sharding result : " << result << endl;
return ok;
}
void pass(int p) {
DurTransaction txn;
create();
ASSERT_EQUALS( 2, nExtents() );
BSONObj b = bigObj();
int N = MinExtentSize / b.objsize() * nExtents() + 5;
int T = N - 4;
DiskLoc truncAt;
//DiskLoc l[ 8 ];
for ( int i = 0; i < N; ++i ) {
BSONObj bb = bigObj();
StatusWith<DiskLoc> status = collection()->insertDocument( &txn, bb, true );
ASSERT( status.isOK() );
DiskLoc a = status.getValue();
if( T == i )
truncAt = a;
ASSERT( !a.isNull() );
/*ASSERT_EQUALS( i < 2 ? i + 1 : 3 + i % 2, nRecords() );
if ( i > 3 )
ASSERT( l[ i ] == l[ i - 4 ] );*/
}
ASSERT( nRecords() < N );
DiskLoc last, first;
{
auto_ptr<Runner> runner(InternalPlanner::collectionScan(ns(),
collection(),
InternalPlanner::BACKWARD));
runner->getNext(NULL, &last);
ASSERT( !last.isNull() );
}
{
auto_ptr<Runner> runner(InternalPlanner::collectionScan(ns(),
collection(),
InternalPlanner::FORWARD));
runner->getNext(NULL, &first);
ASSERT( !first.isNull() );
ASSERT( first != last ) ;
}
collection()->temp_cappedTruncateAfter(&txn, truncAt, false);
ASSERT_EQUALS( collection()->numRecords() , 28u );
{
DiskLoc loc;
auto_ptr<Runner> runner(InternalPlanner::collectionScan(ns(),
collection(),
InternalPlanner::FORWARD));
runner->getNext(NULL, &loc);
ASSERT( first == loc);
}
{
auto_ptr<Runner> runner(InternalPlanner::collectionScan(ns(),
collection(),
InternalPlanner::BACKWARD));
DiskLoc loc;
runner->getNext(NULL, &loc);
ASSERT( last != loc );
ASSERT( !last.isNull() );
}
// Too big
BSONObjBuilder bob;
bob.appendOID("_id", 0, true);
bob.append( "a", string( MinExtentSize + 300, 'a' ) );
BSONObj bigger = bob.done();
StatusWith<DiskLoc> status = collection()->insertDocument( &txn, bigger, true );
ASSERT( !status.isOK() );
ASSERT_EQUALS( 0, nRecords() );
}
