tests-scan: re-enable cross-project 'job' creation #6066
Conversation
![github-advanced-security[bot]](https://avatars.githubusercontent.com/in/57789?s=60&v=4){kind=small}
|
Aside from blocking on the integration test cockpit-project/cockpituous#593 , I'd like to add a unit test here. Covering a cross-project test in the unit test should be quite simple, I'm happy to look at this tomorrow. |
|
Landed the cockpituous changes and retriggered the test run here. |
allisonkarlitskaya
force-pushed
the
cross-jobs
branch
from
0cf6f9d
to
f54c3bc
Compare
That's one for the mocking framework in cockpituous, and I guess it's not going to be super trivial to fix it, either... I start to think about reviving my |
|
I'll send a test fix to provide the missing gh API endpoint in a minute. In the meantime, the test fails like this now: |
job-runner needs that for cross-project testing [1]. A clean way would be to actually resolve "main" to a current SHA in the test, but as nothing actually cares, just rely on the `main` fallback for now -- nothing cares whether this is an actual SHA, as most things in git are happy with a "ref" -- and we don't use that to post statuses (and that would fail because the mock doesn't provide these paths). [1] cockpit-project/bots#6066
job-runner needs that for cross-project testing [1]. A clean way would be to actually resolve "main" to a current SHA in the test, but as nothing actually cares, just rely on the `main` fallback for now -- nothing cares whether this is an actual SHA, as most things in git are happy with a "ref" -- and we don't use that to post statuses (and that would fail because the mock doesn't provide these paths). [1] cockpit-project/bots#6066
|
@allisonkarlitskaya I added a unit test, adjusted it for re-enabling job (the FIXUP), and added two more commits for the fallout it detected. So while I rebased, I didn't touch your commits, all my changes are the ones authored by me -- for easier review. Obviously we should squash the fixup before we land. First test run to see where I screwed up, i.e. where the unit and integration tests disagree. |
There was a problem hiding this comment.
I can't approve (own PR).
One ask: can we also have a test for checking cross-project runs on the non-default branch? I really want to make sure that the logic is right there — it's the biggest part of this PR that I'm unsure about (mostly because of the impenetrable logic of tests-scan)
test/test_tests_scan.py
Outdated
| # mock time for predictable test name | ||
| @unittest.mock.patch("time.strftime", return_value="20240102-030405") | ||
| @unittest.mock.patch("task.distributed_queue.DistributedQueue") | ||
| def test_amqp_sha_pr_cross_project(self, mock_queue, _mock_strftime): |
There was a problem hiding this comment.
I wonder if I could convince you to add this as a pytest-style test. We're going to want to convert everything at some point anyway...
You're probably interested in the monkeypatch fixture, if you do that.
Good call -- I think that's wrong right now, it sets |
Make this a bit easier to read and tweak how we handle JSON formatting. Instead of special-casing the top-level, add a `default` handler which encodes all encountered objects via `__dict__`. We're going to introduce a new object soon.
We may also (some day) split this into a sub-object in the "job" object, as it appears on the wire, but for now let's not break protocol.
This is the same as the fields which allow us to specify what we've been calling "subject" up to this point, but refers exclusively to the thing that we checkout and run. The statuses (and any issues or anything else) will continue to be reported against the main Subject, as before. We copy our existing get_object() code from Cockpit to help us with this. This enables us to run cross-project tests, albeit still in an ad-hoc manner (but no worse than we had it before).
allisonkarlitskaya
force-pushed
the
cross-jobs
branch
from
8441223
to
f42c6bd
Compare
pyright noticed that we were making use of an uninitialized variable in this case. Make sure we return if the file is missing.
Before, if we got a sha given as part of the Job, we'd assume that we also got the target given to us and skip looking up anything on the PR (when it was set). Change that logic a bit: now, if the pull number is given, always do the lookup and use it as default values for both the sha and the target. It should almost never be necessary to specify the target from outside, anymore.
clone_url is a really a derived property of the repo and the forge. Store those in there, instead, and turn clone_url into a property.
We do the same thing for GET and POST, so re-use that code. We're about to reuse it some more, and a third copy is really just too many.
Add support for fetching file content from GitHub via the raw.githubusercontent.com interface.
In case the job didn't explicitly specify a container image to use, query the target repository for the `.cockpit-ci/container` file. Only fall back to the default container image if we get a 404. Also: add a message to the log about which container image we decided to use in the end.
job-runner is interested in the thing we call `github_context` here, so make sure we use it instead. Fixes #6065
We can do this inline, avoiding some messy conditionals. Adjust tests and use normal asserts in a couple places.
This validates the mechanics for make-checkout/tests-invoke and confirms commit c377eb8. Cover both "implicit default branch" and "explicitly specified branch" scenarios.
We're actually interested in options.repo, not job.repo. Job.repo is what gets called "context_project" elsewhere, and refers more to the command that we want to run than it does to the thing we're interested in testing (and definitely doesn't refer to the thing we report status on). This change doesn't actually do anything (since we currently ignore the case where job.repo != options.repo) but it's important that we get the semantics correct here, because it's about to start mattering. job.repo will be handled differently in the next commit.
If the test project isn't the same as the status repo (i.e. a cross-project test), then we need to read the container file from the actual test repo. This isn't yet covered by unit tests due to commit c377eb8, but the next commit will enable that again and check it.
We want to include the test repository for cross-project tests, which is `job.github_context`.
It's going to look this up on its own, now.
allisonkarlitskaya
force-pushed
the
cross-jobs
branch
from
f42c6bd
to
9c81530
Compare
job-runner can do this on its own now.
Use the new 'command-subject' support in job-runner to bring back cross-project support for creating 'Job' objects. Co-authored-by: Martin Pitt <martin@piware.de>
|
In your previous update, you accidentally squashed two commits together. I broke e7f0c9f ("tests-scan: re-enable cross-project 'job' creation") out again, and fixed the trivial ruff failure. I also ensured that test/run passes in the intermediate commits now. The net diff is just the ruff fix now. |
|
I retried the previously broken cross-project test in #6070 and it uses job-runner and correct container now 💯 |
This should be mostly right. I'm not 100% sure about this part, though:
+ "branch": job.ref,