podman with healthcheck fills journal with error

[HerrFrutti]

problem

My containers are healthy but my journal gets filled with error messages. The error seems to be related with healthchecks.

workaround

For now I disabled all healthchecks and the error disappeared.

logs

example line from my journal:

conmon[11328]: conmon 62de48d99d5858ab629f <error>: Unable to send container stderr message to parent Broken pipe

additional info

conmon version 2.1.7
podman version 4.6.1
podman-compose version 1.0.6
OS: Fedora release 37 (Thirty Seven)

I'm using podman-compose to deploy my containers.

[mufeedali]

Any known workarounds for this other than just disabling the health checks?

[hdonnay]

This is still a problem on Fedora 39

[mufeedali]

I have a couple of health checks and memory usage for journald is way beyond normal due to this. Until this bug gets fixed, I might have to disable my health checks, which is kind of sad :(

[benblasco]

@mufeedali I am experiencing this issue too. How have you been disabling the health checks?

[mufeedali]

@benblasco by simply removing them from my podman-compose files

[shyzus]

@benblasco If you are dealing with images like the gitlab image which provides health checks by default. You can use this option to disable it --health-cmd=none

[stryan]

I've been noticing this problem on my OpenSUSE Leap 15.5 box. Containers are run through quadlet with podman 4.7.1.

[symysak]

I am also facing this problem with Ubuntu 22.04 LTS 64-bit. I am not using podman-compose.

raspberrypi@raspberrypi:~$ podman -v podman version 4.6.2

[benblasco]

@benblasco If you are dealing with images like the gitlab image which provides health checks by default. You can use this option to disable it --health-cmd=none

Thanks for sharing this info. I will have to figure out how to do this with quadlet!

[benblasco]

Going through the documentation it appears as though the healthcmd directive can only be used for container units, and not pod units. Is there a way to override the healthcmd for a pod unit? Am I just reading the documentation incorrectly?

[rhatdan]

I don't think healthcmds effect Pods, they are a container concept. Pod's only have an infra container, which does not use healthcmds. Individual containers within the pod can have healthcmds.

[hdonnay]

I ended up adding the following to my .container file:

[Service]
# The following is for a conmon bug where it emits warnings for expected behavior.
LogFilterPatterns=~Broken pipe

See systemd.exec(1) for more information. YMMV if your container expects to have Broken pipe in its logging output during normal operation.

[daniarla]

I'm having the same issue on debian unstable. Hoping for a fix soon!