Cryptic error message when git is not available

[gecube]

Hi!

I got the next error message:

{"level":"error","ts":"2023-09-19T22:10:07.384Z","msg":"failed to checkout and determine revision: unable to list remote for 'ssh://git@****/zodia/infra/kubernetes': unknown error: remote: ","controller":"gitrepository","controllerGroup":"source.toolkit.fluxcd.io","controllerKind":"GitRepository","GitRepository":{"name":"flux-system","namespace":"flux-system"},"namespace":"flux-system","name":"flux-system","reconcileID":"ae69a081-4408-49c3-abe5-f37135fc1573","error":"failed to checkout and determine revision: unable to list remote for 'ssh://git@****/zodia/infra/kubernetes': unknown error: remote: "}

I am sure that in that very moment GitLab was not available, so it means network error. Unfortunately, it is completely opaque what happened from the error message.

My expectations: the error message will clearly state that git repo source is not available (due to network error).

[gecube]

The same error message in slack:

[makkes]

The underlying error is generated by go-git here.

[gecube]

@makkes agree, thanks for prompt answer. Are there any chances for easy fix?

[makkes]

This is an example message from GitLab:

remote: 
remote: ========================================================================
remote: 
remote: ERROR: The project you were looking for could not be found or you don't have permission to view it.

remote: 
remote: ========================================================================
remote: 
fatal: Could not read from remote repository.

The issue is that go-git only returns the first line of the message returned from the remote and in this case this line isn't very useful

[darkowlzz]

Hi, I started looking into it earlier and then got into something else. Getting back to it, looks like we have all the necessary clues about the problem above. In our go-git wrapper, we have a function that's specifically made to handle such errors goGitError().
Wrapping the code that returns the above error with goGitError() would provide a little better error.
But reading the current implementation of goGitError() I think it was initially written to be used for push operations only, maybe for image-automation-controller. Now that we have this in a get remote head call, it'll be better to change the error to something more appropriate.

[makkes]

In this case (GitLab apparently unavailable) the message "check git secret has write access" would be plain misleading. The remote: message can potentially be caused by all kinds of issues.

I would much rather improve (and rename) the checkNotFoundError message to catch more of the most common errors.

[pjbgf]

@makkes the handling of error descriptions from the server for http is slightly better than for ssh, as it try to provide that information back to the user. I would be more than happy to review a PR to improve that for ssh upstream.

[gecube]

Colleagues, any progress? Is there anything that I could test?

[makkes]

I created a draft PR to mitigate this situation in go-git. /cc @pjbgf

[makkes]

go-git PR has been merged so next up is bumping the go-git version as soon as a new version has been released.