automatic container update fails other containers

[tuxbass]

Have an issue with docker image updates.

There are 2 containers - Authelia & Redis. Former is depending on Redis.

 

If both containers were to be updated automatically by Unraid, then Authelia fails to start up, as Redis - following its update - hasn't started yet.

 

Is there a way to force some delays on upgrade, or other methods to mitigate this nuisance?

 

[wgstarks]

You can set a delay. In the docker tab switch to advance view and you’ll see a delay setting. IIRC changing the display order also changes the startup order.

[tuxbass]

On 4/2/2022 at 3:49 PM, wgstarks said:

You can set a delay. In the docker tab switch to advance view and you’ll see a delay setting. IIRC changing the display order also changes the startup order.

Startup order, yes. Delay setting works perfectly for the initial startup. But when it comes to image upgrades (which will cause container restart), then delays don't mean anything. Example, say we have following containers:

- redis
- mariadb - delay 10s -
- authelia

Now when we have a situation where both redis & authelia containers are updated, then we can have a situation where authelia fails to start as redis is not up. If updates happened in the listed order, this problem would likely not occur.

[JonathanM]

Don't auto upgrade containers that support other containers. There really is no good way to automate dependencies currently. You could write your own scripts to get things done, but a universal fix in Unraid will require major rewrites, and I don't see that being a high priority when there is a simple solution available now, to just disable automatic updates on containers that must stay running to avoid breaking other containers.

 

Your fix to update in order only partially solves the root issue. You could still run into situations where the full startup sequence is needed to get things running again, so the only way I can see doing automatic updates with the current startup order is to bring down all the containers, apply all updates, then run through the startup routine. Otherwise people will still have situations where some containers get stuck in limbo when their parent containers get updated.

 

Unattended updates should not be a thing on systems that are important to keep running, too many things can go wrong and require intervention to fix.