Worker is leaving zombie haproxy processes

[David Shrewsbury]

Restarting HAProxy is leaving zombie haproxy processes.

Should follow the suggestions laid out here:

http://stackoverflow.com/questions/2760652/how-to-kill-or-avoid-zombie-processes-with-subprocess-module

[David Shrewsbury (dshrews)]

Problem resolved during development w/o the source being identified. Not repeatable.

[Andrew Hutchings (linuxjedi)]

http://comments.gmane.org/gmane.comp.web.haproxy/11045

[David Shrewsbury (dshrews)]

See:

http://comments.gmane.org/gmane.comp.web.haproxy/11045

[David Shrewsbury (dshrews)]

Theory, based on the bug link above: requests coming in too fast might trigger this.

I think we could add some code to the worker to basically verify haproxy has restarted fully (PID file present?) before returning the Gearman response. Perhaps add a timeout, too.

[David Shrewsbury (dshrews)]

Gah, that won't work. I already do this in the worker.