DBClientConnection * DBClientReplicaSet::checkMaster() {
ReplicaSetMonitorPtr monitor = _getMonitor();
HostAndPort h = monitor->getMasterOrUassert();
if ( h == _masterHost && _master ) {
// a master is selected. let's just make sure connection didn't die
if ( ! _master->isFailed() )
return _master.get();
monitor->failedHost( _masterHost );
h = monitor->getMasterOrUassert(); // old master failed, try again.
}
_masterHost = h;
ConnectionString connStr(_masterHost);
string errmsg;
DBClientConnection* newConn = NULL;
try {
// Needs to perform a dynamic_cast because we need to set the replSet
// callback. We should eventually not need this after we remove the
// callback.
newConn = dynamic_cast<DBClientConnection*>(
connStr.connect(errmsg, _so_timeout));
}
catch (const AssertionException& ex) {
errmsg = ex.toString();
}
if (newConn == NULL || !errmsg.empty()) {
monitor->failedHost(_masterHost);
uasserted(13639, str::stream() << "can't connect to new replica set master ["
<< _masterHost.toString() << "]"
<< (errmsg.empty()? "" : ", err: ") << errmsg);
}
_master.reset(newConn);
_master->setReplSetClientCallback(this);
_master->setRunCommandHook(_runCommandHook);
_master->setPostRunCommandHook(_postRunCommandHook);
_auth( _master.get() );
return _master.get();
}
void DBClientReplicaSet::isntMaster() {
log() << "got not master for: " << _masterHost << endl;
// Can't use _getMonitor because that will create a new monitor from the cached seed if
// the monitor doesn't exist.
ReplicaSetMonitorPtr monitor = ReplicaSetMonitor::get( _setName );
if ( monitor ) {
monitor->failedHost( _masterHost );
}
_master.reset();
}
void DBClientConnection::handleNotMasterResponse(const BSONObj& replyBody,
StringData errorMsgFieldName) {
const BSONElement errorMsgElem = replyBody[errorMsgFieldName];
const BSONElement codeElem = replyBody["code"];
if (!isNotMasterErrorString(errorMsgElem) &&
!ErrorCodes::isNotMasterError(ErrorCodes::Error(codeElem.numberInt()))) {
return;
}
ReplicaSetMonitorPtr monitor = ReplicaSetMonitor::get(_parentReplSetName);
if (monitor) {
monitor->failedHost(_serverAddress,
{ErrorCodes::NotMaster,
str::stream() << "got not master from: " << _serverAddress
<< " of repl set: "
<< _parentReplSetName});
}
_markFailed(kSetFlag);
}
// Ensure nothing breaks when out-of-band failedHost is called during scan
TEST(ReplicaSetMonitorTests, OutOfBandFailedHost) {
SetStatePtr state = boost::make_shared<SetState>("name", basicSeedsSet);
ReplicaSetMonitorPtr rsm = boost::make_shared<ReplicaSetMonitor>(state);
Refresher refresher = rsm->startOrContinueRefresh();
for (size_t i = 0; i != basicSeeds.size(); ++i) {
NextStep ns = refresher.getNextStep();
}
for (size_t i = 0; i != basicSeeds.size(); ++i) {
bool primary = (i == 0);
refresher.receivedIsMaster(basicSeeds[i], -1, BSON(
"setName" << "name"
<< "ismaster" << primary
<< "secondary" << !primary
<< "hosts" << BSON_ARRAY("a" << "b" << "c")
<< "ok" << true
));
if (i >= 1) {
HostAndPort a("a");
rsm->failedHost(a);
Node* node = state->findNode(a);
ASSERT(node);
ASSERT(!node->isUp);
ASSERT(!node->isMaster);
}
else {
Node* node = state->findNode(HostAndPort("a"));
ASSERT(node);
ASSERT(node->isUp);
ASSERT(node->isMaster);
}
}
}