spakman / jobby Goto Github PK

Despite being engineered to avoid that, I recently noticed some zombied child processes on our production server.

Will investigate more when I have time at work.

When stopping a jobbyd with a USR1 signal, the receiving socket is closed and the queue is honoured.

This allows another jobbyd to start on the same socket (desirable), but the new jobbyd does not know how many jobs are running or are in the queue of the old jobbyd. This means that the --max-children for that socket will potentially be --max-children of the old jobbyd plus the --max-children of the new jobbyd until the old jobs have finished.