Repository navigation
Quarantine intermittent pool worker-start timing test - #5105
Conversation
Co-authored-by: Copilot App <223556219+Copilot@users.noreply.github.com>
Quarantine evidence, validation and incomplete discovery checkpoint - 2026-11-18Category-only scope and exact identityQuarantine the method // Assert.Equal: expected Running, actual WaitingToRun after the worker-start wait.
[Trait("category", "flaky")]Its body, conditional attribute, data provider, class-level The decisive CI population is Azure SQL, not Synapse:
Azure DevOps truncates connection-string display names before the MARS setting. Verified failures and separate passing controlsBoth builds are CI-SqlClient definition 1879,
Both failed children report exactly Test-specific eligibility and configuration comparisonThe unchanged source schedules a The held connection has already opened when this assertion fails. Both decisive runs identify stage The ConnectionPoolTest.cs blob is unchanged between the decisive SHA and upstream Job mapping, logs and retrieval limitationsBoth timelines have the unique job identifier
The build-wide 193040 Passed helper failed with Supplemental public REST reads were limited to the permitted individual run, Inspected interval, reconciliation and unfinished coverageNo complete tagging checkpoint exists in the live prior records read: Authenticated MCP rechecked all seven specified definition names/IDs as Enabled.
CI-SqlClient includes main 45, release/6.1 8, release/7.0 12, Authenticated build-helper failure discovery completed for main builds Two relevant run inventories were completely paginated in bounded 1,000-result
Both Completed runs report incompleteTests=0. Passed statistics retain and SUM Checkpoint NOT ADVANCED. Next discovery must rescan its full 14-day window, Other candidates, deduplication and release routing
No release branch qualifies for this new quarantine yet. Local category validation and execution limitationsRepository build.proj ordinary selection excludes Exact commands, run from this automation's isolated worktree: $p = 'src\Microsoft.Data.SqlClient\tests\ManualTests\Microsoft.Data.SqlClient.ManualTests.csproj'
$n = 'FullyQualifiedName=Microsoft.Data.SqlClient.ManualTesting.Tests.ConnectionPoolTest.MaxPoolWaitForConnectionTest'
dotnet test $p -f net10.0 --no-restore --list-tests --filter "$n&category=flaky" -v:q
dotnet restore $p -v:q
dotnet test $p -f net10.0 --no-restore --list-tests --filter "$n&category=flaky" -v:q
dotnet build $p -f net462 --no-restore -v:q
foreach ($f in 'net462', 'net10.0') {
dotnet test $p -f $f --no-build --no-restore --list-tests --filter "$n&category=flaky" -v:q
dotnet test $p -f $f --no-build --no-restore --list-tests --filter 'FullyQualifiedName~Microsoft.Data.SqlClient.ManualTesting.Tests.ConnectionPoolTest.&category!=failing&category!=flaky&category!=interactive' -v:q
dotnet test $p -f $f --no-build --no-restore --filter "$n&category=flaky" --logger 'console;verbosity=normal'
}
git diff --check
git diff HEAD^ HEAD --check
The loop presents the equivalent per-framework commands compactly; net10.0 commands Publication target and boundariesFresh upstream main resolves SqlClientNextVersion to 8.1.0-preview1, Published #5105, open and unmerged: The complete build inventory is established, but broader per-test discovery and |
There was a problem hiding this comment.
🟢 Approval recommended
No active review issues were identified for the change quarantining the intermittently failing connection-pool test.
No open issues 🎉
What changed in this PR
Quarantines MaxPoolWaitForConnectionTest following intermittent worker-start timing failures, without changing test logic or SqlClient behavior.
Changes:
- Adds a method-level
flakycategory covering all four MARS/pool-version cases on both target frameworks. - Adds a comment recording the observed
RunningversusWaitingToRunassertion failure.
| File | Description |
|---|---|
| src/Microsoft.Data.SqlClient/tests/ManualTests/SQL/ConnectionPoolTest/ConnectionPoolTest.cs | Marks the theory as flaky and documents the observed failure; test bodies and existing categories remain unchanged. |
Review effort: Balanced
Codecov Report✅ All modified and coverable lines are covered by tests. Additional details and impacted files@@ Coverage Diff @@
## main #5105 +/- ##
==========================================
+ Coverage 68.99% 72.41% +3.42%
==========================================
Files 314 315 +1
Lines 45857 71152 +25295
==========================================
+ Hits 31638 51527 +19889
- Misses 14219 19625 +5406
Flags with carried forward coverage won't be shown. Click here to find out more. ☔ View full report in Codecov by Harness. 🚀 New features to boost your workflow:
|
🤖 This pull request was created by a fully automated workflow.
Description
Quarantine
Microsoft.Data.SqlClient.ManualTesting.Tests.ConnectionPoolTest.MaxPoolWaitForConnectionTestwith a method-level[Trait("category", "flaky")]and an observed-error comment. This affects all four non-Synapse TCP cases: MARS false/true ×usePoolV2false/true, on net462 and net10.0. Test bodies, existing categories, APIs and pipelines are unchanged.083570c61341083570c61341083570c61341Both jobs use Windows 11 x64, net10.0 managed SNI, Azure SQL, Release/Project, Set 3, attempt 1 and image 56.0.0; allocations/times differ. The test assumes a queued worker starts within five 200-ms sleeps, then asserts its scheduler state without a readiness handshake. The observed failure is that test-specific deadline, not a SQL connection timeout. All four inputs share it.
Truncated CI names do not identify which MARS variant failed; the separate passing job covers the entire four-case population. No per-MARS failure attribution is claimed. Release-branch qualification remains incomplete; no hotfix label/backport requested. Fixes and probation are outside scope.
Issues
Historical #1465 adjusted this same worker-start wait. Open #5095 and #5073 quarantine different tests; #5094 repairs a different already-quarantined test. Complete collapsed evidence and run record, including exact case limitations, validation and checkpoint status.
Testing
Both target frameworks build; quarantine discovery selects this theory and ordinary discovery excludes it while retaining three sibling theories. Local data-provider initialization prevents individual-case enumeration. net462 focused execution failed before exercising the test (Northwind login failure); net10.0 skipped because configuration is missing. No local passing/stability coverage is claimed.
Result: one quarantined theory selected per framework; ordinary discovery retains three siblings and excludes this method.
git diff --checkandgit diff HEAD^ HEAD --checkpassed. The linked record retains build/execution commands, initial missing-assets error and restore result.The 518-build inventory covers 2026-11-04T14:02:42.087Z–2026-11-18T14:02:42.087Z; two run inventories reconcile 12,830 top-level results. Broader per-test discovery is incomplete after a build-wide MCP transport failure. Checkpoint not advanced.
Guidelines
Please review the contribution guidelines before submitting a pull request:
Contributing
Code of Conduct
Best Practices
Coding Style
Review Process
Tests added or updated (category and observed-error comment only).
Public API changes documented (not applicable).
Verified against customer repro (not applicable; local runtime validation blocked).
Ensure no breaking changes introduced.