There's a reliance on an atomic bool and cmpxchg to
prevent the producer of notification from queueing
himself if he's already got a notification scheduled.
There's an ordering issue though where the producers code
can execute, see the flag is true and not bother queueing
a notification, yet the consumer side is about to clear the
flag and finish. The notification thus never gets queued
and the producer side thinks he will get a notification.
In my terminology:
producer is ConnMap::notifyPausedConnection
consumer is ConnMap::notifyAllPausedConnections
Change-Id: Id324b6369c5ee3a6b6758a7a93e017a4ff7c4a78
Reviewed-on: http://review.couchbase.org/63934
Well-Formed: buildbot <build@couchbase.com>
Reviewed-by: Dave Rigby <daver@couchbase.com>
Tested-by: buildbot <build@couchbase.com>
connection_t &conn = queue.front();
Notifiable *tp = dynamic_cast<Notifiable*>(conn.get());
if (tp && tp->isPaused() && conn->isReserved()) {
+ tp->setNotificationScheduled(false);
engine.notifyIOComplete(conn->getCookie(), ENGINE_SUCCESS);
tp->setNotifySent(true);
}
- tp->setNotificationScheduled(false);
queue.pop();
}
}