Rather than doing this manually, you can add a dependency on fedfind; we already use it in the image uploader so there's examples there, but you can basically do:

from fedfind import release as ff_release
ffrel = ff_release.get_release(cid=compose_id)

and that object will have the release, version, etc

I don't love the idea of the read-append-upload flow. There's the issue of a concurrent test clobbering the file, of course, but we don't currently run more than a test at a time. The other issue is what happens if we decide to reorganize the test results (e.g. put them in folders by release, for example)? Or if we prune old (more than say, 6 months old) results?

It'd be nicer to list the keys in the bucket and build up a fresh list each time.

Updated the PR with the suggested changes to use fedfind and to build compose by listing keys in the bucket.

I assume this is because of old compose IDs getting garbage collected? I would really prefer not regexing things if we don't have to.

Why don't we write out the test results in a structure that makes it easy to parse out the info you want? e.g. a directory structure like:
{distro}/{version}/{date}/{build}/{arch}/. That makes it an "at write time" concern and we can be reasonable sure the compose still exists.

Also, we probably don't need to dig around the dictionary for the details, ffrel.type indicates if it's a nightly, ffrel.respin has the respin number, and ffrel.compose is the date.

Updated as suggested on the review comments and have a directory structure while uploading the results into the storage container.

Everything looks good to me, thanks!