Handle podman containers as a runit service?

[bryanasdev000]

Hi folks, hope everyone is well.

There are any examples of runit services invoking podman containers?

I tried a few variations but it only start works ok, stop and restart does not work as expected (it does not kill/stop/restart the container).

I am probably missing something basic.

So far I have a service that calls a shell script that does the podman invocation.

The script:

$ cat jellyfin.sh

#!/usr/bin/env bash

set -euxo pipefail

podman run \
 --rm \
 --name jellyfin \
 --publish 127.0.0.1:8096:8096/tcp \
 --publish 0.0.0.0:1900:1900/udp \
 --publish 0.0.0.0:7359:7359/udp \
 --device /dev/dri/:/dev/dri/ \
 --user 1000:1000 \
 --userns keep-id \
 --volume jellyfin-cache:/cache:Z \
 --volume jellyfin-config:/config:Z \
 --mount type=bind,source=/home/jellyfin,destination=/media,ro=true
,relabel=private \
 docker.io/jellyfin/jellyfin:10.10.1

The service:

# cat /etc/sv/container-jellyfin/run

#!/bin/sh
exec 2>&1
exec chpst -u user:user /home/user/jellyfin.sh 2>&1

I have seen a few other services using the finish script, not sure if it's the way to handle this.

Starting the service:

# sv start container-jellyfin
ok: run: container-jellyfin: (pid 978) 1s, normally down

$ podman ps
CONTAINER ID  IMAGE                                COMMAND     CREATED         STATUS                    PORTS                                                                               NAMES
6c6913bdec2f  docker.io/jellyfin/jellyfin:10.10.1              28 seconds ago  Up 27 seconds (starting)  127.0.0.1:8096->8096/tcp, 0.0.0.0:1900->1900/udp, 0.0.0.0:7359->7359/udp, 8096/tcp  jellyfin

# sv stop container-jellyfin
ok: down: container-jellyfin: 0s

$ podman ps
CONTAINER ID  IMAGE                                COMMAND     CREATED         STATUS                    PORTS                                                                               NAMES
6c6913bdec2f  docker.io/jellyfin/jellyfin:10.10.1              57 seconds ago  Up 56 seconds (starting)  127.0.0.1:8096->8096/tcp, 0.0.0.0:1900->1900/udp, 0.0.0.0:7359->7359/udp, 8096/tcp  jellyfin

# pstree
runit─┬─anacron
      ├─conmon───jellyfin───17*[{jellyfin}]
      ├─passt.avx2
      ├─podman─┬─exe───9*[{exe}]
      │        └─7*[{podman}]
      ├─podman pause
      └─runsvdir─┬─6*[runsv───agetty]
                 ├─runsv─┬─udevd───udevd
                 │       └─vlogger
                 ├─runsv─┬─dhcpcd
                 │       └─vlogger
                 ├─runsv─┬─sshd─┬─sshd-session───sshd-session───bash───su───bash───pstree
                 │       │      └─sshd-session───sshd-session───bash
                 │       └─vlogger
                 ├─runsv─┬─socklog
                 │       └─svlogd
                 ├─runsv───nanoklogd
                 ├─runsv─┬─cronie-crond
                 │       └─vlogger
                 ├─runsv─┬─chronyd
                 │       └─vlogger
                 ├─2*[runsv─┬─pause]
                 │          └─vlogger]
                 ├─runsv─┬─caddy───9*[{caddy}]
                 │       └─vlogger
                 ├─runsv───vlogger
                 ├─runsv─┬─prometheus───9*[{prometheus}]
                 │       └─vlogger
                 ├─runsv─┬─alertmanager───8*[{alertmanager}]
                 │       └─vlogger
                 ├─runsv─┬─node_exporter───5*[{node_exporter}]
                 │       └─vlogger
                 └─runsv─┬─blackbox_export───9*[{blackbox_export}]
                         └─vlogger

Thanks in advance.

EDIT: Format and more info.

[Johnnynator]

Some remarks
a) 2>&1 in a line with exec <program> does essentially nothing, the exec 2>&1 before does the redirection already
b) You really want to use exec podman and not have bash lingering in the middle. (Or just skip that jellyfin.sh script entirely and just run exec chpt -u user:user podman args... directly in run.

Your current setup will have bash receive SIGTERM, bash will exit, podman continues to run and runit only ever knew of bash and not podman.

[bryanasdev000]

Welp, it seens to work like a charm, still testing but thanks!

The service is now as:

# cat /etc/sv/container-jellyfin/run

#!/bin/sh
exec 2>&1
exec chpst -u user:user podman run \
 --rm \
 --name jellyfin \
 --publish 127.0.0.1:8096:8096/tcp \
 --publish 0.0.0.0:1900:1900/udp \
 --publish 0.0.0.0:7359:7359/udp \
 --device /dev/dri/:/dev/dri/ \
 --user 1000:1000 \
 --userns keep-id \
 --volume jellyfin-cache:/cache:Z \
 --volume jellyfin-config:/config:Z \
 --mount type=bind,source=/home/jellyfin,destination=/media,ro=true,relabel=private \
 docker.io/jellyfin/jellyfin:10.10.1

It seems to be working as expected, still need to mess with a service account and podman rootless now :P

Thank you!

Will test a little bit more and will mark as resolved.

[bryanasdev000]

With exec 2>&1 only at the top and no more on the service call, I got no logs, so instead, I changed the service to this:

#!/bin/sh

exec chpst -u user:user podman run \
 --rm \
 --name jellyfin \
 --publish 127.0.0.1:8096:8096/tcp \
 --publish 0.0.0.0:1900:1900/udp \
 --publish 0.0.0.0:7359:7359/udp \
 --device /dev/dri/:/dev/dri/ \
 --user 1000:1000 \
 --userns keep-id \
 --volume jellyfin-cache:/cache:Z \
 --volume jellyfin-config:/config:Z \
 --mount type=bind,source=/home/jellyfin,destination=/media,ro=true,relabel=private \
 docker.io/jellyfin/jellyfin:10.10.1 2>&1

Also, I got a log/run file that contains:

#!/bin/sh
exec vlogger -t container-jellyfin -p daemon

Service and log working flawless.

Thanks @Johnnynator

[Anachron]

May I suggest creating podman containers using podman-compose and then starting/stopping them in service-files?
It will be as easy as cd /path/to/compose.yml && podman compose start and cd /path/to/compose.yml && podman compose stop to start/stop all. You can then use podman compose wait to wait for all containers to stop inside the run file.

Another way would be to make use of pods, and run podman pod start <PODNAME>, but there is no podman pod wait, so you would have to podman compose wait in your run file still.