Failure: integration / test_mysql_dotnet_client — stale altinityinfra/mysql_dotnet_client:latest image on stable-25.8

Affected tests:

• tests/integration/test_mysql_protocol/test.py::test_mysql_dotnet_client

Affected jobs / branches:

• Integration tests (*_tsan, N/6) and other integration shards on stable-25.8 — 4 concurrent failures on run 28837216389, commit 29f8caa8140. Previous run passed.

Affected files (Altinity-fork only):

• tests/integration/test_mysql_protocol/test.py — missing the upstream cd /testapp && anchor.

Description

test_mysql_dotnet_client failed deterministically (same error, 4/4 shards) on stable-25.8 with:

Error: Received unknown response i for SSLRequest (expecting S or N)

Not caused by the two PRs merged since the last green run (Altinity/ClickHouse#2008 and #2009 — both CI/report-only). Not flaky either — the wording SSLRequest (expecting S or N) is PostgreSQL Npgsql handshake vocabulary, not MySQL, so the dotnet client is talking the wrong protocol against port 9001.

Root Cause

Upstream PR ClickHouse/ClickHouse#80785 added a PostgreSQL C# client to the same Dockerfile and left WORKDIR /pg_testapp as the last directive. That PR was later reverted in ClickHouse/ClickHouse#105615, but the published altinityinfra/mysql_dotnet_client:latest image on Docker Hub retained the wrong WORKDIR, and tests/integration/compose/docker_compose_mysql_dotnet_client.yml has no pull_policy: always. The test runs dotnet run -- ... without cding, so on runners that pulled the poisoned :latest, dotnet executes the Npgsql project against the MySQL port. The byte 'i' (0x69 = 105) is the length prefix of ClickHouse's MySQL handshake being misread by Npgsql.

That explains the flip: previous run's runners had the older cached image; the runners on run 28837216389 pulled the poisoned one.

Upstream fix ClickHouse/ClickHouse#105636 (commit 7f7a222b1a8, 2026-05-22) anchors the command to /testapp:

-  f"dotnet run -- --host {node.hostname} ..."
+  f"cd /testapp && dotnet run -- --host {node.hostname} ..."

Upstream backported it to 25.8.25.7. Altinity's stable-25.8 is currently based on v25.8.23.13-lts, so the fix has not reached us yet.

Suggested Solution

No dedicated cherry-pick planned — the fix will land automatically when stable-25.8 is rebased onto 25.8.28. Tracking here so the failure isn't re-triaged.

References

• Upstream fix: ClickHouse/ClickHouse#105636 — commit 7f7a222b1a8
• Culprit PR / revert: ClickHouse/ClickHouse#80785, ClickHouse/ClickHouse#105615
• Failing run: https://github.com/Altinity/ClickHouse/actions/runs/28837216389
• CI report: https://s3.amazonaws.com/altinity-build-artifacts/REFs/stable-25.8/29f8caa8140cbb64698641b11d83c74c3c7299f2/28837216389/ci_run_report.html#checks-fails

Start with tests/integration/test_mysql_protocol/test.py::test_mysql_dotnet_client and the related tests/integration/compose/docker_compose_mysql_dotnet_client.yml configuration. Check the command against the upstream fix; done means the test uses the intended client project and passes on stable-25.8 after the rebase onto 25.8.28.