ngx.pipe.spawn shell mode child process not killed after openresty exits

[fffonion]

When using ngx.pipe.spawn in shell mode (args is a string), sh -c in invoked internally.
While OpenResty exists, SIGKILL is sent to sh process, but it's not able to propogate to child process.
The child process is then left alive.

POC:

init_work_by_lua_block {
  require("ngx.pipe").spawn("/tmp/a.py")
}

$ cat /tmp/a.py:

#!/usr/bin/env python3
import time
time.sleep(23333)

Then signal stop the OpenResty/Nginx process.

[spacewander]

It is expected as there is no way to store the children's pid created by sh -c inside OpenResty.

[fffonion]

Thank you for the reply. Does it make sense to force a `sh -c "exec ...."` then? In that case sh replace itself with the spawned child process so it can be killed properly. 罗泽轩 <notifications@github.com>于2021年2月20日 周六上午9:03写道：

It is expected as there is no way to store the children's pid created by sh -c inside OpenResty. — You are receiving this because you authored the thread. Reply to this email directly, view it on GitHub <#1855 (comment)>, or unsubscribe <https://github.com/notifications/unsubscribe-auth/AAQJLFKLTIFUIXLWAUPSZXDS74C45ANCNFSM4X4ZAFBQ> .

-- Regards fffonion (B607 2745 84E8 D5E5 <https://yooooo.us/gpg.txt>)

[spacewander]

sh -c "exec ...." might be a break change, and it doesn't work with cases like sh -c "www || ls".

[fffonion]

sh -c "exec ...." might be a break change, and it doesn't work with cases like sh -c "www || ls".

That's goind point. Maybe we just notice the behaviour on lua-resty-core? If you feel okay I can propose a PR.

[spacewander]

PR is welcome!