> On Tue, 14 May 2013 17:40:55 +0200 > Marcin Deranek <[email protected]> wrote: > >> I was trying to use ftok to have consistent semaphores over >> uwsgi restarts and I noticed that ftok does not play nicely with Zerg >> dance (using uWSGI 1.4.8). When started I get a bunch of semaphores >> created; when zerg dance gets completed all semaphores are gone and >> trying to serve any request generates: >> >> semop(): Invalid argument [core/lock.c line 482] >> >> although content is served as usual. > > This is the case for 1.9.10 too. I believe what happens here is that > new instance does not create semaphores as they already exist (created > by old instance) and old instance destroys them on exit. This can be > nicely seen on each odd dance. On even dance semaphores are > properly created as they did not exist in the previous one. > > Marcin >
You basically have a special condition of multiple instances using the same --ftok. Am i right ? As ipcsem are the only kind of locking to allows multiple-instances locking i think it is whorty to make a special case for it. -- Roberto De Ioris http://unbit.it _______________________________________________ uWSGI mailing list [email protected] http://lists.unbit.it/cgi-bin/mailman/listinfo/uwsgi
