Disable auto restart for docker images?

[MisterE2002]

I created a custom Dockerfile based on Alpine.
With the last line

Code:

CMD /path/to/script.sh

My custom image should run, do it's thing, and stop. In the ideal scenario the Container can terminate. But Scale seems to have enabled a "auto restart" feature. (assumes it crashed?). Is this possible on a per image basis?

If not, what would be the best solution to avoid the main proces to vanish (without creating some infinite loop->delay or something). In that case a second run of the same image should tear down the than old idle container and start a new instance.

 

[sretalla]

Scale seems to have enabled a "auto restart" feature.

SCALE uses kubernetes... which effectively assumes that all containers (workloads/pods) are required to be running all the time.

The simplest way to have a container running only once and then stopping is to have that logic in the container. (where it doesn't quit when finished and can self-check the work and decide if it should do the work or not)

 

[MisterE2002]

It is a one-shot process. So, the only workaround seems to let the container running until the next is started and use "Update Strategy = Kill existing pods before creating new ones".

Most solutions seems to create a infinite loop with a "sleep" inside. Not sure if this is the most elegant solution (Will it really give the resources back to the host)

 

[sretalla]

5 or even 300 seconds of sleep inside a loop takes such a small amount of resources you couldn't measure it.

 

[MisterE2002]

This seems to work somewhat.

Code:

logfile="${log_folder}/result_$(date +"%Y_%m_%d_%I_%M_%p").log"
echo "Started: $(date)" >> "$logfile"
# actual work
echo "Finished: $(date)" >> "$logfile"
sleep infinity
echo "Line after infinity (should not happen): $(date)" >> "$logfile"

But i get 2 log files (or it is appending if in the same minute). But only the "Started" line. So it seems to create 2 instances for some reason.

 

[browntiger]

That is NOT what you should be using dockers for. Dockers can be restarted on another IP at any time.
From outside of a docker, you can set replicas: 1 to scheduler a docker creation and to 0 to stop a docker.
It sounds that all you need in reality to scheduler (cron some job).

 

[MisterE2002]

euh...not sure if i understand. But Scale does not has Jail (LXC) support. I do not like to run complete VM's so a custom Container seems a nice alternative. But i have some containers which has to run once and should close afterwards (in the ideal case). This is normal Docker behavior but iX changed this. Great for auto-healing but not for my case.

Changing a running container with some commands i also saw, but that defeats the "as-a-appliance" way and after each start i have script something on the scale OS. If i understand correctly you suggest to add a crontab in Scale with some custom scripting to change this property. Which seems a solution but the "sleep infinity" is the more lazy (and elegant) route imho.

And idling the container seems to work fine *but* it creates 2 instances. One does not complete. Was on my way to file a bug. But maybe i still missing something. Will create the bugreport in the next post

 

[MisterE2002]

If i import a custom made docker image i see two instances running. Tested with Scale RC1 (in virtualbox/linux host)


Create our own custom dockerfile:

Dockerfile:

Code:

FROM ghcr.io/linuxserver/baseimage-alpine:3.17
COPY --chown=568:568 script_bin /script_bin
CMD /script_bin/bin.sh

Including a script to log outside of the container

Code:

log_folder="/script_log"
logfile="${log_folder}/output.log"


echo "Started: $(date), $(hostname), $$" >> "$logfile"

# fake our actual payload
waittime=60
echo "wait for $waittime seconds: $(date), $(hostname), $$" >> "$logfile"
sleep $waittime

echo "Finished: $(date), $(hostname), $$" >> "$logfile"

# make sure to run to infinity (and beyond)
sleep infinity
echo "Line after infinity (should not happen): $(date)" >> "$logfile"

build docker tarfile:
docker build -t myinfinity .
docker save myinfinity:latest > myinfinity.tar




# transfer the tar to scale

on scale:
create a dataset "tank/logs"
docker load < myinfinity.tar

In TrueNAS -> Launch Docker Image
Name = myinfinity
Image repository = myinfinity
Container Environment Variables
PUID = 568
PGID = 568
storage -> host = /mnt/tank/logs container = /script_log



If we start the container we see 2 container instances created and run (/mnt/tank/logs/output.log)
Started: Mon Nov 28 13:18:34 UTC 2022, myinfinity-ix-chart-5f5578fdbb-vwscc, 141
wait for 60 seconds: Mon Nov 28 13:18:34 UTC 2022, myinfinity-ix-chart-5f5578fdbb-vwscc, 141
Started: Mon Nov 28 13:18:36 UTC 2022, myinfinity-ix-chart-66cfb9b6-xj2gb, 141
wait for 60 seconds: Mon Nov 28 13:18:36 UTC 2022, myinfinity-ix-chart-66cfb9b6-xj2gb, 141
Finished: Mon Nov 28 13:19:36 UTC 2022, myinfinity-ix-chart-66cfb9b6-xj2gb, 141



Glancing over the Scale logs

==> k3s_daemon.log <==
Nov 28 05:18:36 truenas k3s[5796]: I1128 05:18:36.209659 5796 event.go:294] "Event occurred" object="ix-myinfinity/myinfinity-ix-chart" fieldPath="" kind="Deployment" apiVersion="apps/v1" type="Normal" reason="ScalingReplicaSet" message="Scaled down replica set myinfinity-ix-chart-5f5578fdbb to 0 from 1"
Nov 28 05:18:36 truenas k3s[5796]: I1128 05:18:36.259661 5796 event.go:294] "Event occurred" object="ix-myinfinity/myinfinity-ix-chart-5f5578fdbb" fieldPath="" kind="ReplicaSet" apiVersion="apps/v1" type="Normal" reason="SuccessfulDelete" message="Deleted pod: myinfinity-ix-chart-5f5578fdbb-vwscc"

 

[MisterE2002]

update: they found the culprit with 2 instances and if i saw correctly they also include a option to not automatically restart containers.