A container registry that uses the AT Protocol for manifest storage and S3 for blob storage. atcr.io

Corrupted Docker login credentials? #29

Closed calico.lgbt
Closed opened by calico.lgbt

Hi, I think I've managed to corrupt docker-credential-atcr's saved credentials somehow...

If anybody else has experienced this or knows where the credentials are stored, please instruct me on how to fix this.

I ran docker compose pull whilst logged into an old, now non-existent account, then ran docker logout atcr, then I tried logging in again via both the myimage and the fallback, and the values just do not seem to save. I have also tried updating docker-credential-atcr, to no result.

Using default tag: latest
Error: no accounts configured for atcr.io
Run: docker-credential-atcr login
error getting credentials - err: exit status 1, out: ``

.docker/config.json

{
	"credHelpers": {
		"atcr.io": "atcr"
	}
}

(This will be populated such as the below when successfully running docker login atcr.io)

{
	"auths": {
		"atcr.io": {}
	},
	"credHelpers": {
		"atcr.io": "atcr"
	}
}

What OS are you on? mac? linux? windows?

I'm not sure how you would have ended up in this scenario, I'm not sure what command you used to install the credential helper

I found you in the logs and it's not a client problem. Nothing is wrong with your account or your credentials.

Your PDS at greenhouse.lgbt is intermittently returning 502s. When atcr checks your credential during that window, the check takes 22 seconds and then fails. The credential helper reads that failure as "this login is dead" and deletes the saved credential, so the next command tells you there are no accounts configured. Log in, PDS hiccups, credential erased, repeat. That's why it looks like nothing saves.

The deletion is our bug, not yours. It should not throw away a working login just because your PDS was briefly unreachable, and I'm fixing that.

A few things for now:

  1. docker logout atcr should be docker logout atcr.io. The short name doesn't match anything, so that command was a no-op.
  2. merchantcalico.bsky.social works reliably and shows 200s throughout the logs, because it doesn't depend on greenhouse.lgbt being up. Use that until your PDS is stable.
  3. Avoid docker login atcr.io with a password. It silently stores nothing and still reports success, which is a separate bug I'm also fixing. Use docker-credential-atcr login instead.

Thank you, that's good to hear. I shall try again a little later.

My PDS should be stable, I mainly just had it off for a short duration whilst attempting to pull. Mistake on my end.

Thank you so much! I've managed to log in correctly.

Sign up or Sign in to add to the discussion.