Checkmk in Docker with MAIL_RELAY_HOST won't start

Troubleshooting
1

Hi,

I’m running checkmk/check-mk-raw:latest (6 days ago).

I have a docker-compose.yml I’ve been using for years now on multiple instances - it has 2 ‘services’ - checkmk and ‘mailer’ (SMTP relay, GitHub - namshi/docker-smtp: SMTP docker container). The latest server I’m trying to set this up is giving me issues:

checkmk  | ### PREPARE POSTFIX (Hostname: checkmk, Relay host: mailer)
checkmk  | ### STARTING MAIL SERVICES
checkmk  | syslogd: timed out waiting for child
checkmk exited with code 1
checkmk  | ### STARTING MAIL SERVICES
checkmk  | syslogd: timed out waiting for child

htop on host is showing 100% CPU for syslogd.

If I edit docker-compose.yml and comment out

   environment:
      MAIL_RELAY_HOST: mailer

it starts normally (but obviously isn’t using the mailer container as a SMTP relay), but as soon as this variable is present the error returns.

If I start mailer by itself, it works (mailer container does not give any errors by itself). However, checkmk is stuck on trying to start mail service.

Any ideas are welcome.
Thanks!

2

New development: this only seems to happen on Linode, now Akamai…and strangely matches the timeframe since Akamai took over.
Tried the same thing on Hetzner and it worked without issues.

What I don’t understand is what can cause this? checkmk Docker image with env var set…and it ends up in a loop…

3

I had this as I noted in an earlier post below. I am self-hosted on Debian 11 / docker-ce and like you the container will not start with that parameter set. What I do is comment it out and I have a custom mail.cf bound in which works fine. The issue I was asking help for last week was that although the container starts, the postfix service doesn’t so I have to manually hop in to the running container and issue a “service start postfix” after which notifications work as they should.

I had assumed from the lack of response that it was something daft I’d done but I guess I’m “pleased” to hear it isn’t just me although clearly not a universal issue.

4

I’ve created a new VPS on Linode, and it works with the default checkmk Docker image - syslogd doesn’t hang. My favorite type of issues :slight_smile:

5

Hi, I have the same problem with a archlinux and docker host (version 23.0.1).

For the moment I commented the “syslogd” call in the docker-entrypoint.sh (before the postfix start service) and the container started up and I received the mail notifications without any problems.

Is a workaround, but for me is enought for now. I don’t know the collateral effects of do that :stuck_out_tongue:

I hope this help you.

1 Like
6

Just happened on Hetzner, so I’d like to revert my statement that it only happens on Linode/Akami.

7

The issue can be easily reproduced:

$ docker run --rm -it ubuntu bash
apt-get update && apt-get install -y inetutils-syslogd
# ...
$ time syslogd
syslogd: timed out waiting for child

real    0m30.001s
user    0m0.001s
sys     0m0.000s

there seems to be a problem with this variant of syslogd.

(starting syslogd with -d for debugging output seems to fix the issue)

there is also an additional error message after some time:

syslogd: /dev/xconsole: No such file or directory

but if you replace the installed syslogd with the one that busybox provides, it works.
so my fix is:

#!/bin/bash
set -e
apt-get update
apt-get remove -y inetutils-syslogd
apt-get install -y --no-install-recommends busybox-syslogd

mounted via dockers -v option to /docker-entrypoint.d/pre-entrypoint/fixsyslogd. don’t forget to set the executable bit, otherwise the hook will not be executed.