Broken gitea credential cause garm no never start again

[ChristopherHX]

I played with invalid stuff, then I wanted to restart garm...

Now to fix this, I would have to start garm to use garm-cli.

Type 'dlv help' for list of commands.
time=2025-05-29T16:21:25.498+02:00 level=INFO msg="starting consumer" consumer_id=cache watcher=database
time=2025-05-29T16:21:25.498+02:00 level=INFO msg="Loading provider" provider=kubernetes_external
time=2025-05-29T16:21:41.228+02:00 level=INFO msg="failed to start entity controller: waiting for error group: loading repositories: waiting for error group: starting worker: creating github client: fetching github client: invalid gitea URLs"
Process 75461 has exited with status 1
Detaching

However this is now a state that is hard to solve without deleting the database / change the code to keep running (I will try the second way to unblock my experiment).

[gabriel-samfira]

Yep. I am aware. It's on the todo list to be fixed before a release is cut from main. The plan is to start the workers and add a health check and a retry loop that attempts to start the needed go routines, with a backoff. The events will be added to the entity.

It's more work than I have time for right now, as it requires some thought on how to do it right (retry and give users transparency into what's happening).

[gabriel-samfira]

Did you manually update the database by any chance? The error you got, indicates that either the credentials were saved in the db without the API base URL or that something called the client constructor with an entity that for some reason didn't have credentials preloaded. For create and update credentials, there is theoretically validation that prevents adding credentials without valid URLs.

This needs to be addressed anyway, I am just wondering if there is more to this issue than just making sure GARM continues to function while ignoring the erroneous entity.

[ChristopherHX]

I have experimented with simplification e.g. omiting apiURL.

Then the modified garm server broke the database. E.g. I expected my code to fill the missing url, but it didn't. To restart debugging with code changes ended in this fatal error

[gabriel-samfira]

Gotcha. The validation was supposed to catch any invalid credentials during create/update, which is why any error when creating the client was made to stop everything because the db somehow entered in an erroneous state.

I will still refactor the workers to allow for manual recovery of an entity (keep the worker suspended until the error is corrected by the user) and add an event signaling an error.

[gabriel-samfira]

I have experimented with simplification e.g. omiting apiURL.

This can be done in the client. Make the api url optional, and set the api base url to the same value as baseURL before sending to the api.

I will create a PR soon.