Logrotate postrotate and docker exec strange behaviour

Question

I'm trying to configure logrotate with docker container. I'm running httpd as background process in docker container, and after logrotate I need to reload it to use new log files. I don't want to restart the container because of possible downtime. Sending SIGHUP with docker kill --signal=HUP <container> is not working, as my entrypoint is bash script which does not handle signals. I tried to do it like this in logrotate config:

...
sharedscripts
postrotate
    service httpd reload > /dev/null 2>/dev/null || true
    docker exec some-container kill -HUP $(ps -e | awk '{print $1}')>>/tmp/exec-out.log 2>>/tmp/exec-out.log || true
endscript

but I got

kill: sending signal to 30 failed: No such process
kill: sending signal to 31 failed: No such process
kill: sending signal to 32 failed: No such process
kill: sending signal to 33 failed: No such process
kill: sending signal to 34 failed: No such process
kill: sending signal to 35 failed: No such process
kill: sending signal to 36 failed: No such process
kill: sending signal to 37 failed: No such process
kill: sending signal to 38 failed: No such process

I'm quite new to docker and linux and I don't really understand why docker gets process ids that do not exist.

EDIT: I also would not like to change the bash script to trap SIGHUP if possible, but instead solve the problem in logrotate config.

Answer 1

I believe the $(ps -e | awk '{print $1}')>>/tmp/exec-out.log 2>>/tmp/exec-out.log || true is being run in the context of the host not the docker and so gets the wrong pid as you can see.

If you ran the docker container with pid=host the pids will work.

Alternatively you can get the pid like so:

docker inspect --format {{.State.Pid}} <container>

But you don't actually need the pid, you can send a signal yourself using docker kill like so:

docker kill --signal=HUP some-container