payload/rpm-ostree: Copy stderr to our output #6019
Conversation
The rpm-ostree container deployment path can fail for many reasons from networking to details in mount setup. What we really want is a proper API with progress out of bootc/ostree; I will work on that at some point. In the meantime though, just capture stderr on failure and include it in the payload installation error so people don't have to dig into `program.log` which is very obscure. Signed-off-by: Colin Walters <[email protected]>
cgwalters
had a problem deploying
to
gh-cockpituous
GitHub Actions
Failure
|
I think this will work but trying to test it out locally with |
| # For now we just use anaconda's stdout. | ||
| # Also TODO: Add a progress/status API to bootc and use it here | ||
| # https://github.com/containers/bootc/issues/522 | ||
| proc = startProgram(args, stdout=None, stderr=subprocess.PIPE) |
There was a problem hiding this comment.
You can pass to execWithRedirect the stderr and stdout parameters and achieve the same.
There was a problem hiding this comment.
And for printing the error in the PayloadInstallationError it should be enough to catch the exception
There was a problem hiding this comment.
Are you sure? I didn't actually try it but just reading the code for _run_program it sure looks to me like err_string (i.e. the result of stderr, which is what we want to capture) never ends up being included in a thrown exception (which is what we want to show to the user).
I am fine with any solution that results in stderr from this process being shown to the user, to be clear.
There was a problem hiding this comment.
it's part of the output_string that it's returned (stderr is redirected for that method by default to stdout):
https://github.com/rhinstaller/anaconda/blob/master/pyanaconda/core/util.py#L192
| args | ||
| ) | ||
| # TODO refactor things so we can log stdout to the program log or journal. | ||
| # For now we just use anaconda's stdout. |
There was a problem hiding this comment.
If you can make it log to Journal, it should be captured even now when something goes wrong - Journal is the de-facto primary log sink for Anaconda these days & what we gather and ask users to attach to bugs. :)
Sure, it would be better if Anaconda got distinct event/callback when something happens, but just making sure everything ends up in Journal would help already. :)
The rpm-ostree container deployment path can fail for many reasons from networking to details in mount setup.
What we really want is a proper API with progress out of bootc/ostree; I will work on that at some point.
In the meantime though, just capture stderr on failure and include it in the payload installation error so people don't have to dig into
program.logwhich is very obscure.