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(){
OID lastID;
lastID.clear();
int secsToSleep = 0;
while ( ! inShutdown() && Shard::isMember( _addr ) ){
if ( lastID.isSet() ){
scoped_lock lk( _seenWritebacksLock );
_seenWritebacks.insert( lastID );
lastID.clear();
}
try {
ScopedDbConnection conn( _addr );
BSONObj result;
{
BSONObjBuilder cmd;
cmd.appendOID( "writebacklisten" , &serverID ); // Command will block for data
if ( ! conn->runCommand( "admin" , cmd.obj() , result ) ){
log() << "writebacklisten command failed! " << result << endl;
conn.done();
continue;
}
}
log(1) << "writebacklisten result: " << result << endl;
BSONObj data = result.getObjectField( "data" );
if ( data.getBoolField( "writeBack" ) ){
string ns = data["ns"].valuestrsafe();
{
BSONElement e = data["id"];
if ( e.type() == jstOID )
lastID = e.OID();
}
int len;
Message m( (void*)data["msg"].binData( len ) , false );
massert( 10427 , "invalid writeback message" , m.header()->valid() );
DBConfigPtr db = grid.getDBConfig( ns );
ShardChunkVersion needVersion( data["version"] );
log(1) << "writeback id: " << lastID << " needVersion : " << needVersion.toString()
<< " mine : " << db->getChunkManager( ns )->getVersion().toString() << endl;// TODO change to log(3)
if ( logLevel ) log(1) << debugString( m ) << endl;
if ( needVersion.isSet() && needVersion <= db->getChunkManager( ns )->getVersion() ){
// this means when the write went originally, the version was old
// if we're here, it means we've already updated the config, so don't need to do again
//db->getChunkManager( ns , true ); // SERVER-1349
}
else {
// we received a writeback object that was sent to a previous version of a shard
// the actual shard may not have the object the writeback operation is for
// we need to reload the chunk manager and get the new shard versions
db->getChunkManager( ns , true );
}
Request r( m , 0 );
r.init();
r.process();
}
else if ( result["noop"].trueValue() ){
// no-op
}
else {
log() << "unknown writeBack result: " << result << endl;
}
conn.done();
secsToSleep = 0;
continue;
}
catch ( std::exception e ){
if ( inShutdown() ){
// we're shutting down, so just clean up
return;
}
log() << "WriteBackListener exception : " << e.what() << endl;
// It's possible this shard was removed
Shard::reloadShardInfo();
}
catch ( ... ){
log() << "WriteBackListener uncaught exception!" << endl;
}
secsToSleep++;
sleepsecs(secsToSleep);
if ( secsToSleep > 10 )
secsToSleep = 0;
}
log() << "WriteBackListener exiting : address no longer in cluster " << _addr;
}
