Make rsyslog container aware remove pidfile on start

There was a commit to rsyslogd that made the rsyslog daemon container-aware by checking if it was running as PID 1. If so, it will avoid writing a pidfile to disk. This is very handy when using it in a container environment where you might have a lot of container restarts or recreations, and you don’t want the daemon to check if a pidfile exists, because when it does, rsyslogd will refuse to start.

I am using Debian Stretch in my Docker image, but a version of rsyslog including this commit does not exists in the Debian apt repositories.

A work-around is to delete the pidfile manually whenever a container starts up by using a custom ENTRYPOINT, which is what I ended up with. In which case, here’s a file to include in your custom rsyslog image bin/

# Clear the pid file from old runs. It doesn't restart otherwise.
# NOTE: This won't be necessary after upgrading to version 8.34.0 of rsyslogd
rm /var/run/

# Run command after configuration
exec $@

Then reference this file in the Dockerfile:

# Use stretch-backports to get rsyslog version with mongodb uristr option
FROM debian:stretch-backports

RUN apt update && apt -t stretch-backports install -y \
    rsyslog \
    rsyslog-relp \

# This must be updated if more protocols are added
EXPOSE 514/tcp

# rsyslogd will die on QUIT, TERM or INT (man rsyslogd)

COPY rsyslog.conf /etc/rsyslog.conf

COPY bin/ /usr/bin

# This will enable debug output toggling with SIGUSR1 (requires -d)

# -n : Avoid auto-backgrounding
# -d : Enable debug mode
CMD ["rsyslogd", "-n", "-d"]

An alternative to this is to build your own image of rsyslog from source, but I didn’t want to do that.