Restarting main process? about puma_worker_killer HOT 5 CLOSED

If you kill the main process nothing will restart it. So it wouldn't really help, it would instead kill your app.

The main process doesn't serve requests and shouldn't grow in size. So this shouldn't be needed. If it was it would mean a memory leak in the puma  library.

—
Sent from Mailbox

On Wed, Dec 24, 2014 at 10:09 AM, Leooo [email protected] wrote:

Hello,
I may be missing someting totally obvious here, but:
This gem is great to mitigate memory leaks on applications using Puma. Though it only helps to restart workers, not to restart the main process when it is the one whose RAM is getting glutted.

Is this doable to restart the main process (and maybe all the workers at the same time then, not sure how it works) when this is the one consuming the most memory?

Reply to this email directly or view it on GitHub:
#10

from puma_worker_killer.

OK so I did miss something obvious then.

I agree this would be killing the whole app (which may still be needed sometimes), but I thought the main process was actually serving requests: number of processes running = (nb workers + 1). In my ActionController::Live controller I thought I saw requests were sometimes run by it, not by workers. And we can always run Puma with 0 workers, there is still a process running, and serving requests? But I must be wrong I will check more. Thanks for the quick answer.

from puma_worker_killer.

That +1 process you're seeing in puma the master process. When you have workers >=1 then the master process doesn't serve requests. Rather the master gets the request and then passes it to another worker to serve the request. It needs to be available to route the requests. Though it could be possible that i'm wrong and not understanding something but i'm pretty sure that's how it works cc/ @evanphx

from puma_worker_killer.

In cluster mode, the master process never runs requests.

from puma_worker_killer.

Thanks for the confirmation! My initial doubt came from the fact that I am checking which worker is running in my SSE thread using its index from the following line:

index=ObjectSpace.each_object(Puma::Cluster::Worker).map { |obj| obj }.first.try(:index)

but sometimes the index is empty, so I thought this may have been the master process.

L

from puma_worker_killer.