[service plugin] Correct way to handle process termination

[asanikovich]

We run some processes by RR service plugin.
One of them has a long run logic and very important, which should not be broken (by termination)

In case of supervisor (we used before instead of RR service plugin), we can block SIGTERM signal (in place where important logic start) and increase timeout to wait before SIGKILL.
How to solve it with RR?

[rustatian]

Hey @asanikovich 👋🏻
You may send any signal to the PHP process, stop it, kill, no limitations on that. If you don't have this option, process will be finished. Otherwise it will be restarted.

[asanikovich]

I have option remain_after_exit enabled.
My case about graceful termination on Kubernetes new deploy.
In case of supervisor, supervisor send signals to child processes.

[rustatian]

RR sends SIGINT system call to the underlying PHP process (if you stop the whole RR or just terminate the service via RPC by its name). Then RR waits 5 seconds for the process to gracefully exit, and if the process still exists - sends SIGKILL.

[asanikovich]

1. In case of k8s send signal to pod, RR sends SIGINT to services process?
2. Can I modify default 5 seconds?

[rustatian]

1. As far as I remember, k8s sends a SIGTERM signal. When receiving this signal, RR will initiate a graceful stop, here is the RR timeout (not service): https://github.com/roadrunner-server/roadrunner/blob/master/.rr.yaml#L2169. And yeah, RR will then send SIGINT to the underlying services.
2. Unfortunately no, this timeout is fairly new, so still experimental. You may create a feature request for you previous request with the service_name_in_logs via RPC and this one in one FR.