thanos / kathikon / 461b7cf58c62cb4318e3f18453beb2d8dd4a4cc0-PR-30

93%

main: 89%

Pull #30

github

What failed
All 6 failures were in Kathikon.IntegrationTest on Elixir 1.18 / OTP 27, each hitting {:error, :timeout} from TestSupport.await_state/3 after ~10s:

insert executes a successful job
insert retries failed jobs
insert discards after max attempts
sleep defers then completes…
scheduler promotes scheduled jobs
pruner removes old terminal jobs
Root cause
batch_test.exs and coverage_test.exs call Kathikon.pause_queue(:default) in setup. Those modules are async: false, but different async: false modules still run in parallel in ExUnit.

While batch/coverage tests had :default paused, integration tests were enqueueing on :default and waiting forever for jobs that could never run.

That’s why it looked flaky: it only failed when those modules overlapped on the slow 1.18 CI runner.

Fix applied
Added a dedicated :integration queue in config/test.exs
Moved all integration tests to queue: :integration
Added TestSupport.resume_all_queues!/0 and call it in integration setup
Bumped timeouts on backoff-heavy tests to 15s as a CI safety margin
Verified with mix test --max-cases 32 and repeated runs of integration + batch + coverage together — 0 failures.

The Dialyzer failure on that same run looks unrelated (passes locally); if it shows up again we can dig into that separately.

846 of 910 new or added lines in 21 files covered. (92.97%)

980 of 1050 relevant lines covered (93.33%)

75.23 hits per line