You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
We have no CI coverage for genuinely long-running (hours-long) queries. The doc calls for such tests scheduled nightly, not per-commit.
Current state
The repo has exactly one scheduled workflow: .github/workflows/reyden-rest-nightly.yml (cron '0 2 * * *', lines 17-21). Everything else is per-PR/push/dispatch.
The longest test today, StatusPollerKeepsQueryAlive (csharp/test/E2E/StatementTests.cs:949-975), is a real 30-minute test (Task.Delay(30 min) over RANGE(30000000)) — but it is excluded from every CI run: Skip.If(TestConfiguration.IsCITesting, …) (line 955), and both the nightly (reyden-rest-nightly.yml:164) and per-PR e2e (e2e-tests.yml:232) set isCITesting: true. So no long query ever executes in CI.
There is no test-category convention — zero [Trait]/[Category] attributes in csharp/test. CI runs dotnet test with no --filter (ci/scripts/csharp_test_databricks_e2e.sh).
Introduce a long-running test category — either an xUnit [Trait("Category","LongRunning")] driven by dotnet test --filter, or a dedicated config flag analogous to IsCITesting (note: the existing Skip.If(IsCITesting) on StatusPollerKeepsQueryAlive must be inverted/relaxed so it can run in the long-running tier).
Add hours-long query tests targeting existing large data (e.g. main.tpcds_sf100_delta.*, already used by DriverResilienceTests.cs:98-135; larger scale factors available).
Wire them into reyden-rest-nightly.yml (or a sibling scheduled workflow).
Caveats to design around
GitHub-hosted runners have a 6-hour job limit; reyden-rest-nightly uses concurrency … cancel-in-progress: false.
dotnet test currently has no timeout override — hours-long tests need explicit per-test timeout handling.
Activity