Effectively stopping a process in good time.

[cmaughan]

I'm still a little confused about stopping a long running process.
Suppose I have an internal mechanism to stop it, or it will simply finish its work 'soon'.

proc.stop(... wait : 10s, terminate :5s, kill 5s)

Question 1: If I call this function, and after 2 seconds the process naturally exits, my understanding here is that I will still wait for 10 seconds. So how to 'exit early' if the process has cleanly closed, otherwise wait for 10s? i.e. I don't want to wait if I don't have to.

Question 2: if the 10 seconds expire, then the process is sent a SIGTERM after 5 more seconds, and then a SIGKILL after 5 more seconds?

[DaanDeMeyer]

The wait will stop early if the process exits early.

The stop actions work as follows: The action specifies what to do, the timeout specifies how long to wait before going to the next action (or return if it's the last action). So your example will first wait max 10 seconds for the process to exit on its own, if it hasn't exited after 10 seconds, reproc will send SIGTERM and wait max 5 more seconds for the process to exit. If it hasn't exited after 5 seconds, reproc will send SIGKILL and wait max 5 more seconds for the process to exit. If it exits within 5 seconds, reproc returns the exit status, otherwise, reproc will return an error.

[DaanDeMeyer]

Can you try again with the lasest commit on the main branch? It's a long shot but I made some changes that might fix the issue. At the very least, calling reproc_wait with REPROC_INFINITE as the timeout should now return immediately when the child process exits.

[cmaughan]

Sorry for the delay; I plan on digging into this tomorrow; I'll try your update.

[DaanDeMeyer]

I had to remove it unfortunately as it broke other use cases. I would suggest looking into whether the child processes spawned by the server are doing anything fishy. If they inherit all pipes from the server process and outlive it, then there's a good chance that's the cause of the issue.

(Unfortunately, you're in a bit of uncharted territory here, I haven't yet tested reproc with processes that spawn other processes themselves)

[cmaughan]

I noticed on windows at least that you can specify a pipe is not inheritable when created. Would that help?

[DaanDeMeyer]

Yes and reproc uses that internally, but some pipes have to be inherited by the child process, including one pipe which we use to figure out if the process has exited. Unfortunately, there's no way to say that a pipe should only be inherited by a direct child process and not by any processes spawned by that child process.