action #99594: Fix flaky coverage - t/lib/OpenQA/Test/FullstackUtils.pm size:M - openQA Project (public) - openSUSE Project Management Tool

Actions

Copy link

action #99594

closed

Fix flaky coverage - t/lib/OpenQA/Test/FullstackUtils.pm size:M

Added by okurz about 3 years ago. Updated about 3 years ago.

Status:

Resolved

Priority:

High

Assignee:

tinita

Category:

Regressions/Crashes

Target version:

Ready

Start date:

2021-10-01

Due date:

2021-10-27

% Done:

Estimated time:

Description

Observation¶

Flaky coverage in https://github.com/os-autoinst/openQA/pull/4256#issuecomment-932021197. No logs available.

Acceptance criteria¶

AC1: t/lib/OpenQA/Test/FullstackUtils.pm is not showing up in codecov reports for multiple PRs with unrelated changes

Suggestions¶

Find an occurence and preserve the results
Try full-stack.t or scheduler-full.t

Related issues 1 (0 open — 1 closed)

Actions

Copy link

Updated by okurz about 3 years ago

Copied to action #99597: Fix flaky coverage - lib/OpenQA/Worker/WebUIConnection.pm size:M added

Actions

Copy link

Updated by livdywan about 3 years ago

Subject changed from Fix flaky coverage - t/lib/OpenQA/Test/FullstackUtils.pm to Fix flaky coverage - t/lib/OpenQA/Test/FullstackUtils.pm size:M
Description updated (diff)
Status changed from New to Workable

Actions

Copy link

Updated by tinita about 3 years ago

Status changed from Workable to Feedback
Assignee set to tinita

Monitoring new pull requests to catch a new occurrence of this coverage decrease, in order to save the codecov.io report and see which code actually wasn't covered.

Actions

Copy link

Updated by tinita about 3 years ago

I found a merged PR where the coverage decreased by 5%: https://github.com/os-autoinst/openQA/pull/4257

That's why we regularly see an increase of 5% now

https://app.codecov.io/gh/os-autoinst/openQA/compare/4257/changes
The lines marked with --- lost coverage:

     92 sub _bail_with_result_panel_contents {
---  93     my ($result_panel_contents) = @_;
---  94     diag("full result panel contents:\n$result_panel_contents");
---  95     javascript_console_has_no_warnings_or_errors;
     96     BAIL_OUT "Expected result not found";    # uncoverable statement
     97 }
...
    113         if ($looking_for_result && $status_text =~ qr/Result: (.*) finished/) {
--- 114             diag("stopped waiting for '$result_panel', result turned out to be '$1'");
--- 115             return _bail_with_result_panel_contents $status_text;
    116         }

Actions

Copy link

Updated by tinita about 3 years ago

Status changed from Feedback to In Progress

I identified the problem: for certain tests like t/full-stack.t, we retry the test.

When the test fails, the _bail_with_result_panel_contents will be called and recorded as covered.

When the test succeeds in one of the retries, _bail_with_result_panel_contents is never called, but it is still recorded, as Devel::Cover keeps coverage from previous tests.

This way we see a subroutine as covered which is only called when a test fails.

One solution would be to just delete the cover_db folder in tools/retry, but that would also remove coverage data from previous other tests which we want to keep.

Actions

Copy link

Updated by tinita about 3 years ago

https://github.com/os-autoinst/openQA/pull/4294 Mark uncoverable bailout code paths - merged

Actions

Copy link

Updated by livdywan about 3 years ago

tinita wrote:

One solution would be to just delete the cover_db folder in tools/retry, but that would also remove coverage data from previous other tests which we want to keep.

Here's an idea. We already define COVERDB_SUFFIX, and we can see that in tools/retry. So we can delete the correct db after an iteration. It doesn't seem 100% clean, but not sure how else we could perform such cleanup.

https://github.com/os-autoinst/openQA/pull/4295

Actions

Copy link