Hi,
I've got the integration working how I imagined. It is using s6-svscan to
manage the containers, hopefully not abusing something. General idea is to
have supervision control over containers, ie to allow start up in a given
runlevel, have the ability to manage dependency start between given
containers and start/stop using openrc (referenced
https://wiki.gentoo.org/wiki/S6).
Assuming containers are already running, ie given instance of whoami
container:
*docker run -p 80:80 -d -it --name whoami traefik/whoami*
The "run" script, /var/svc.d/whoami/run:
#!/bin/execlineb -P
exec docker start -a whoami
The "finish" script, /var/svc.d/whoami/finish
*#!/bin/execlineb -Ps6-permafailon 60 1 2 exit*
The init.d, conf.d. Cat /etc/conf.d/container.whoami:
*INSTANCE=whoami*
Cat /etc/init.d/container:
*#!/sbin/openrc-rundescription="A supervised test service with a
logger"supervisor=s6s6_service_path=/run/openrc/s6-scan/${INSTANCE}depend()
{ need s6-svscan}stop_pre() { docker stop ${INSTANCE}}*
Finally, [openrc-run, ln -s /etc/init.d/container
/etc/init.d/container.whoami] /etc/initd.d/container.whoami start, stop
work as expected (docker ps |grep whoami does not return anything, after
running "/etc/init.d/container.whoami stop"):
List containers
root@ # *docker ps |grep whoami*
68bd2ed585ed traefik/whoami
"/whoami" 35 minutes ago Up 34 minutes
0.0.0.0:80->80/tcp whoami
root@ # *./container.whoami stop*
container.whoami |whoami
container.whoami | * Stopping container.whoami ...
[ ok ]
root@ # *docker ps |grep whoami*
root@ #* ./container.whoami start*
container.whoami | * Starting container.whoami ...
[ ok ]
root@ # docker ps |grep whoami
68bd2ed585ed traefik/whoami
"/whoami" 35 minutes ago Up 3 seconds
0.0.0.0:80->80/tcp whoami
root@ # *ps xf -o pid,ppid,pgrp,euser,args*
PID PPID PGRP EUSER COMMAND
21056 1 21056 root /bin/s6-svscan /run/openrc/s6-scan
21058 21056 21056 root \_ s6-supervise whoami/log
21059 21056 21056 root \_ s6-supervise whoami
27584 21059 27584 root | \_ docker start -a whoami
Similar to above, using s6-svstat:
root@ # *s6-svstat /run/openrc/s6-scan/whoami*
up (pid 27584) 752 seconds
root@ # *./container.whoami stop*
container.whoami |whoami
container.whoami | * Stopping container.whoami ...
[ ok ]
root@ #* s6-svstat /run/openrc/s6-scan/whoami*
down (exitcode 2) 1 seconds, normally up, ready 1 seconds
root@ #* ./container.whoami start*
container.whoami | * Starting container.whoami ...
[ ok ]
root@h003 /e/init.d # s6-svstat /run/openrc/s6-scan/whoami
up (pid 6722) 3 seconds
The goal of all this has been to incorporate containers into
/etc/runlevels. I am not sure if there is a better way, have I missed
something by not using s6-overlay or the like? I am not familiar enough to
know.
An outstanding issue, in the s6 run script, i would like to parametrize the
instance name, I don't know how to do it, as it is currently hard coded:
The "run" script, /var/svc.d/whoami/run:
#!/bin/execlineb -P
exec docker start -a *whoami*
But once that is done, then all running containers could be incorporated by
updating the conf.d for INSTANCE name, templating a /var/svc.d/<INSTANCE>
folder and linking to /etc/init.d/container, ie:
/etc/init.d/container.whoami
/etc/init.d/container.cadvisor
/etc/init.d/container.traefik
[put under control of s6]
/etc/runlevels/20/s6-svscan
...
This is kind of asymmetric, ie container start is in s6, whereas stop
[docker] is in openrc, but I am not seeing a different way, the goal is to
have robust services running. When system boots, s6-scan will start all the
containers automatically, but then further operations, ie for things like
manual failover etc, is possible using standard platform openrc -
effectively docker - start/stop commands.
kind regards
On Thu, May 13, 2021 at 5:17 AM Damo <dhatchett2@gmail.com> wrote:
> Hi,
>
> I've been running docker containers for a while, where I pass
> "--restart=always" into the run command, so the containers restart
> automatically after reboot. I want to have more control over the startup
> order of the containers, ie integrate into openrc start/stop and put into
> different runlevels.
>
> I've had mixed success so far. I would be interested if someone else has
> working solution. My runlevels look something like this:
>
> rl100
> container.registry
> rl90
> container.auth
> container.router
> boot
> ...
>
> FYI, i've found systemd is doing it nicely, where systemctl start/stop
> <CONTAINER> works as I would expect. I see a hardcoded dependency into the
> container PID in the unit file (podman in this case):
>
> [root@]# cat /usr/lib/systemd/system/container-libvirt-exporter.service
>
> #
>
> [Unit]
> Description=Podman container-libvirtd-exporter.service
> Documentation=man:podman-generate-systemd(1)
>
> [Service]
> Restart=always
> ExecStart=/usr/bin/podman start libvirtd-exporter
> ExecStop=/usr/bin/podman stop -t 10 libvirtd-exporter
> KillMode=none
> Type=forking
>
> PIDFile=/var/run/containers/storage/overlay-containers/9037e389e61ed01eb5dfce16fa750b6f0f01827a67640e4748e6527bbfcb6276/userdata/conmon.pid
>
> [Install]
> WantedBy=multi-user.target
>
>
> Kind regards,
> Damo
>
>