Status SyncSourceFeedback::updateUpstream(OperationContext* txn) {
ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
if (replCoord->getMemberState().primary()) {
// primary has no one to update to
return Status::OK();
}
BSONObjBuilder cmd;
{
boost::unique_lock<boost::mutex> lock(_mtx);
if (_handshakeNeeded) {
// Don't send updates if there are nodes that haven't yet been handshaked
return Status(ErrorCodes::NodeNotFound,
"Need to send handshake before updating position upstream");
}
// the command could not be created, likely because the node was removed from the set
if (!replCoord->prepareReplSetUpdatePositionCommand(&cmd)) {
return Status::OK();
}
}
BSONObj res;
LOG(2) << "Sending slave oplog progress to upstream updater: " << cmd.done();
try {
_connection->runCommand("admin", cmd.obj(), res);
}
catch (const DBException& e) {
log() << "SyncSourceFeedback error sending update: " << e.what() << endl;
// blacklist sync target for .5 seconds and find a new one
replCoord->blacklistSyncSource(_syncTarget,
Date_t(curTimeMillis64() + 500));
BackgroundSync::get()->clearSyncTarget();
_resetConnection();
return e.toStatus();
}
Status status = Command::getStatusFromCommandResult(res);
if (!status.isOK()) {
log() << "SyncSourceFeedback error sending update, response: " << res.toString() <<endl;
// blacklist sync target for .5 seconds and find a new one
replCoord->blacklistSyncSource(_syncTarget,
Date_t(curTimeMillis64() + 500));
BackgroundSync::get()->clearSyncTarget();
_resetConnection();
}
return status;
}
Status SyncSourceFeedback::updateUpstream(OperationContext* txn) {
ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
if (replCoord->getMemberState().primary()) {
// primary has no one to update to
return Status::OK();
}
BSONObjBuilder cmd;
{
stdx::unique_lock<stdx::mutex> lock(_mtx);
// the command could not be created, likely because the node was removed from the set
if (!replCoord->prepareReplSetUpdatePositionCommand(&cmd)) {
return Status::OK();
}
}
BSONObj res;
LOG(2) << "Sending slave oplog progress to upstream updater: " << cmd.done();
try {
_connection->runCommand("admin", cmd.obj(), res);
}
catch (const DBException& e) {
log() << "SyncSourceFeedback error sending update: " << e.what() << endl;
// blacklist sync target for .5 seconds and find a new one
replCoord->blacklistSyncSource(_syncTarget, Date_t::now() + Milliseconds(500));
BackgroundSync::get()->clearSyncTarget();
_resetConnection();
return e.toStatus();
}
Status status = Command::getStatusFromCommandResult(res);
if (!status.isOK()) {
log() << "SyncSourceFeedback error sending update, response: " << res.toString() <<endl;
// blacklist sync target for .5 seconds and find a new one, unless we were rejected due
// to the syncsource having a newer config
if (status != ErrorCodes::InvalidReplicaSetConfig || res["cfgver"].eoo() ||
res["cfgver"].numberLong() < replCoord->getConfig().getConfigVersion()) {
replCoord->blacklistSyncSource(_syncTarget, Date_t::now() + Milliseconds(500));
BackgroundSync::get()->clearSyncTarget();
_resetConnection();
}
}
return status;
}
bool SyncSourceFeedback::replHandshake(OperationContext* txn) {
ReplicationCoordinator* replCoord = getGlobalReplicationCoordinator();
if (replCoord->getCurrentMemberState().primary()) {
// primary has no one to handshake to
return true;
}
// construct a vector of handshake obj for us as well as all chained members
std::vector<BSONObj> handshakeObjs;
replCoord->prepareReplSetUpdatePositionCommandHandshakes(txn, &handshakeObjs);
LOG(1) << "handshaking upstream updater";
for (std::vector<BSONObj>::iterator it = handshakeObjs.begin();
it != handshakeObjs.end();
++it) {
BSONObj res;
try {
LOG(2) << "Sending to " << _connection.get()->toString() << " the replication "
"handshake: " << *it;
if (!_connection->runCommand("admin", *it, res)) {
std::string errMsg = res["errmsg"].valuestrsafe();
massert(17447, "upstream updater is not supported by the member from which we"
" are syncing, please update all nodes to 2.6 or later.",
errMsg.find("no such cmd") == std::string::npos);
log() << "replSet error while handshaking the upstream updater: "
<< errMsg;
// sleep half a second if we are not in our sync source's config
// TODO(dannenberg) after 2.8, remove the string comparison
if (res["code"].numberInt() == ErrorCodes::NodeNotFound ||
errMsg.find("could not be found in replica set config while attempting "
"to associate it with") != std::string::npos) {
// black list sync target for 10 seconds and find a new one
replCoord->blacklistSyncSource(_syncTarget,
Date_t(curTimeMillis64() + 10*1000));
BackgroundSync::get()->clearSyncTarget();
}
_resetConnection();
return false;
}
}
catch (const DBException& e) {
log() << "SyncSourceFeedback error sending handshake: " << e.what() << endl;
_resetConnection();
return false;
}
}
return true;
}