Label to cleanup images by container

[DeltaTangoLima]

Feature Request Checklist

• I checked existing issues and discussions for duplicates.
• This is a feature request, not a support question.

Problem Description

I have three use cases I want to use Watchtower for:

✅ Auto pull and notify only

• I use this for mission critical containers, so I can do controlled updates (eg. Pihole, Home Assistant, etc)
• This is the smallest group of the three

✅ Auto pull, update containers, and notify

• I use this for all other containers, as I don't mind taking the risk of auto-updating, but want the old image there in case I need to quickly rollback (eg. Immich, Forgejo, etc)
• This would be a moderately sized group, if I had the below option

❌ Auto pull, update containers, cleanup old images, and notify

• I want to use this for containers I don't really care about even having to rollback to the older version, I'll just forward fix the problem when I can, as they're not terribly important (eg. Calibre, Piped, etc)
• Of my 40+ docker hosts, this group would represent the vast majority of containers, meaning less administrative overhead

Desired Solution

I'm asking for consideration of a container-level cleanup option, using a boolean label such as com.centurylinklabs.watchtower.cleanup

Current Alternatives or Workarounds

Scripted cleanup of specific containers' older images, but that isn't terribly compatible with my how I manage my docker fleet (40+ hosts) using Portainer.

Ideally, I'd comb my Discord messages each weekend (my Watchtower updates are cron'd for the early hours of Saturday morning), test the auto-updated containers, and only have to cleanup the two smaller groups of images as I go.

Additional Context

Understand this may not be possible without a significant rewrite of part of Watchtower. I haven't looked at the code myself, but I'm assuming the reason the cleanup option is host-wide is because Watchtower uses a docker prune (or similar) to achieve it.

If so, then what I'm asking for would require individual assessment at a container level to determine which older images should and shouldn't be removed.

[nicholas-fedor]

This is possible to implement. It just requires accounting for multi-container, same-image deployments and other configurations.

For the time being, the workaround is simply using multiple instances of Watchtower with different scopes. This allows for more fine-grained handling on a per-group basis, such as scheduling updates so you don't have infrastructure-level services being updated at the same time as your other services and applications.

For more info: https://watchtower.nickfedor.com/latest/advanced-features/running-multiple-instances/

Given what you described, the latter is probably better overall, but adding per-container handling is not unreasonable.

[DeltaTangoLima]

This is possible to implement. It just requires accounting for multi-container, same-image deployments and other configurations.

Ah, that use case hadn't occurred to me. I like the workaround you've suggested - that'll help me reduce some of the admin overhead on my digger docker hosts. I'll give it a try. Thanks.