It seems that under certain circumstances, if a worker process exits abnormally, ProcessPool will spawn two new processes to replace it. This ignores the maximum worker limit the ProcessPool is configured with, so eventually enough processes will be spawned so as to consume all available system resources.
I don't completely understand why this happens, but I know that the "second" replacement process is spawned by the startAWorker call attached to the stopAWorker Deferred in _cb_doWork. This may be because the Shutdown call issued by stopAWorker fails with ProcessTerminated. This results in the Deferred succeeding instead of failing, allowing the startAWorker callback to run. Additionally, when the process exits, it gets restarted by the logic in _addProcess. Just a guess, though.
(In general I think all this code should (and could be) be a lot simpler.)
heyakyra
commented
5 years ago
It seems that under certain circumstances, if a worker process exits abnormally, ProcessPool will spawn two new processes to replace it. This ignores the maximum worker limit the ProcessPool is configured with, so eventually enough processes will be spawned so as to consume all available system resources.
I don't completely understand why this happens, but I know that the "second" replacement process is spawned by the
startAWorkercall attached to thestopAWorkerDeferred in_cb_doWork. This may be because theShutdowncall issued bystopAWorkerfails withProcessTerminated. This results in the Deferred succeeding instead of failing, allowing thestartAWorkercallback to run. Additionally, when the process exits, it gets restarted by the logic in_addProcess. Just a guess, though.(In general I think all this code should (and could be) be a lot simpler.)