Skip to content

Revert "Antalya 26.8 Backport of #126138 - Revert "Revert "Verify the server certificate hostname on outbound TLS connections""" - #2842

Merged
mkmkme merged 3 commits into
antalya-26.8from
revert-2671-backports/antalya-26.8/126138
Nov 19, 2026
Merged

mkmkme merged 3 commits into
antalya-26.8from
revert-2671-backports/antalya-26.8/126138

Conversation

@mkmkme

@mkmkme mkmkme commented Nov 13, 2026

Copy link
Copy Markdown
Collaborator

Reverts #2671 due to issues in CI

Changelog category (leave one):

  • Not for changelog (changelog entry is not required)

CI/CD Options

Exclude tests:

  • Fast test
  • Integration Tests
  • Stateless tests
  • Stateful tests
  • Unit tests
  • Performance tests
  • Aarch64 tests
  • All with ASAN
  • All with TSAN
  • All with MSAN
  • All with UBSAN
  • All with Coverage
  • All Regression
  • Disable CI Cache

Regression jobs to run:

  • Fast suites (mostly <1h)
  • Aggregate Functions (2h)
  • Alter (1.5h)
  • Benchmark (30m)
  • CAS (content-addressed storage; Antalya only)
  • ClickHouse Keeper (1h)
  • Iceberg (2h)
  • LDAP (1h)
  • OAuth (5m)
  • Parquet (1.5h)
  • RBAC (1.5h)
  • SSL Server (1h)
  • S3 (2h)
  • S3 Export (2h)
  • Swarms (30m)
  • Tiered Storage (2h)

…Verify the server certificate hostname on outbound TLS connections"""
@github-actions

github-actions Bot commented Nov 13, 2026 •

Copy link
Copy Markdown

Workflow [PR], commit [767c35b]

@mkmkme

mkmkme commented Nov 13, 2026

Copy link
Copy Markdown
Collaborator Author

@blau-ai

@blau-ai

blau-ai commented Nov 13, 2026

Copy link
Copy Markdown
Collaborator

CI triage for 10e50a4

There are 3 failing checks, and none of them come from this PR. 2 are pre-existing on antalya-26.8 and 1 is a stateless test that fails only under some randomized settings.

The PR head is base 03ee68973a plus this single revert, and the PR only touches TLS/SSLManager code, its config and its tests. The base-branch MasterCI run on 03ee68973a still contains #2671, and it fails both regression suites in the same scenarios. So the failures don't depend on #2671 being in or out.

Check Classification PR-caused?
Regression release s3_azure_1 Pre-existing on base No
Regression release swarms Pre-existing on base and other PRs No
Stateless (amd_debug, distributed plan, s3 storage, parallel): 03572_export_merge_tree_part_limits_and_table_functions Test breaks under some randomized MergeTree settings No

1. RegressionTestsRelease / S3 (azure, 1) / s3_azure_1

Failing scenarios:

✘ [ Fail ] '/s3/azure/part 1/disk/remote host filter' (16m 55s)   ClickHouse server is not healthy
✘ [ Fail ] '/s3/azure/part 1/disk/syntax' (5m 15s)                AssertionError

Server exceptions include URL "http://azurite:10000" is not allowed in configuration file, see <remote_url_allow_hosts> (that is the remote-host-filter scenario at work) and tlsv1 alert unknown ca on :9440.

Evidence that it's pre-existing: base antalya-26.8 MasterCI run 43337484940 on 03ee68973a fails s3_azure_1 on both x86_64 and aarch64. Its report shows the same two scenarios, disk/remote host filter ("ClickHouse server is not healthy") and disk/syntax (AssertionError), with nearly the same durations. The earlier base run 43246617309 on d038b99718 also fails s3_azure_1.
Next step: none for this PR. It's a separate base-branch problem.

2. RegressionTestsRelease / Swarms / swarms

✘ [ Fail ] '/swarms/feature/cluster discovery/check cluster discovery with wrong cluster name'
✘ [ Fail ] '/swarms/feature/node failure/cpu overload'
✘ [ Fail ] '/swarms/feature/node failure/cpu overload all swarm nodes'

Evidence that it's pre-existing: unrelated PR #2834 fails the same three scenarios (job). Swarms is also red on #2845 and #2846, and in base MasterCI runs 43337484940 and 43246617309 on both x86_64 and aarch64.
Next step: none for this PR.

3. Stateless 03572_export_merge_tree_part_limits_and_table_functions

Failed 1 test, 13358 passed.

 ---- Count files in big_destination_max_bytes, should be 5 (4 parquet, 1 commit)
-5
+4

The CI randomized-settings diagnosis settles it:

Re-running with the same randomized settings:  Runs: 5, Failed: 5
Re-running without randomized settings:         Runs: 5, Failed: 0, Passed: 5
Culprit MergeTree settings: --index_granularity 61962

The export_merge_tree_part_max_bytes_per_file=3500000 export can only start a new file at block boundaries. With very large granules the reader produces bigger blocks, so you get 3 parquet files instead of 4. This is a fragile test, not a regression: the PR doesn't touch MergeTree or export code. The same job passed on base run 43246617309.
Suggested fix (separate from this revert): pin the granularity on the source table in the test, for example

CREATE TABLE $big_table (...) Engine=MergeTree() ORDER BY id PARTITION BY year SETTINGS index_granularity = 8192

or add no-random-merge-tree-settings to the test's # Tags: line.


Bottom line: none of these failures block this revert. They are all on the antalya-26.8 side or in the test itself.

@mkmkme

mkmkme commented Nov 19, 2026

Copy link
Copy Markdown
Collaborator Author

@blau-ai

@blau-ai

blau-ai commented Nov 19, 2026

Copy link
Copy Markdown
Collaborator

CI triage for 767c35b7

Verdict: 4 failing checks. None are caused by this PR. 3 regression suites fail the same way on another open PR, so they were already failing. 1 stateless test hit a memory limit in CAS/S3 code that this PR doesn't touch. It's probably flaky; I couldn't reproduce it on base. SQLLogic test and SQLStorm test are still running, so the PR gate is pending. All integration suites passed, including the TLS tests this PR changes (test_accept_invalid_certificate, test_ssl_cert_authentication, test_acme_tls, test_bind_host).

What this PR changes at runtime: it puts the openSSL.client.extendedVerification default back to false, which turns off client-side hostname checking. A revert can only make TLS checks less strict. So it can't cause a new TLS rejection, a memory overrun or a timeout.

For comparison I used #2913 (an unrelated Iceberg change on the same base), run 44418800189, attempt 1.

Check Failing test Class Caused by this PR?
Regression release swarms /swarms/feature/node failure/cpu overload (+ … all swarm nodes) already failing No. Same failures on #2913
Regression release s3_azure_1 /s3/azure/part 1/disk/remote host filter, …/disk/syntax already failing No. Same failures on #2913
Regression release cas_lightweight_delete_4 /lightweight delete/concurrent delete/AggregatingMergeTree/concurrent delete without overlap already failing No. Same failure on #2913
Stateless (amd_binary, cas s3 storage, parallel) 03770_min_columns_to_activate_adaptive_write_buffer probably flaky (memory headroom) No. Different code path

1. RegressionTestsRelease / Swarms: already failing

✘ [ Fail ] '/swarms/feature/node failure/cpu overload' (5m 3s)
✘ [ Fail ] '/swarms/feature/node failure/cpu overload all swarm nodes' (5m 7s)
Code: 209. DB::Exception: ... Timeout exceeded while reading from socket (socket (172.19.0.6:9000), receive timeout 300000 ms): While executing Remote. (SOCKET_TIMEOUT)

#2913 fails the same two scenarios with the same SOCKET_TIMEOUT on plain TCP port 9000 (job 135267307883). No TLS is involved. Next step: none for this PR. This is a stress-scenario timeout that should be tracked on the base branch.

2. RegressionTestsRelease / S3 (azure, 1): already failing

✘ [ Fail ] '/s3/azure/part 1/disk/remote host filter' (16m 54s)
    ClickHouse server is not healthy
<Error> Application: Code: 491. DB::Exception: URL "http://azurite:10000" is not allowed in configuration file, see <remote_url_allow_hosts>. (UNACCEPTABLE_URL)
✘ [ Fail ] '/s3/azure/part 1/disk/syntax' (5m 15s)

#2913 fails the same two scenarios with the same UNACCEPTABLE_URL startup error (job 135267315238). The server can't start because http://azurite:10000 isn't in remote_url_allow_hosts. syntax then fails because the server is still down. This looks related to #2905 (Do not check Azure disks from the server config against remote_url_allow_hosts), but that's my guess and I didn't verify it. The server log also has a tlsv1 alert unknown ca line. That is a CA rejection on the server's 9440 port, not a hostname check, and it isn't the failing assertion. Next step: none for this PR. Expect this to clear once #2905 (or a matching test update) lands.

3. RegressionTestsRelease / CASLightweightDelete (4): already failing

✘ [ Fail ] '/lightweight delete/concurrent delete/AggregatingMergeTree/concurrent delete without overlap' (36s 451ms)
    AssertionError

#2913 fails the same scenario (job 135267314553). Next step: none for this PR.

4. Stateless tests (amd_binary, cas s3 storage, parallel): 1 of 13,781 failed, probably flaky

FAIL 03770_min_columns_to_activate_adaptive_write_buffer
Code: 241. DB::Exception: Query memory limit exceeded: would use 50.78 MiB (attempt to allocate chunk of 2.00 MiB), maximum: 50.00 MiB. (MEMORY_LIMIT_EXCEEDED)
(in query: insert into metric_log_1 select * from generateRandom() limit 1 settings max_memory_usage = '50Mi' ...)
  11. src/IO/WriteBufferFromS3.cpp:115  WriteBufferFromS3::WriteBufferFromS3
  14. .../ContentAddressed/Backend/CasObjectStorageBackend.cpp:923  Cas::ObjectStorageBackend::publish
  19. .../ContentAddressed/ContentAddressedTransaction.cpp:1869  Cas::fanOutBlobUploads

The limit was exceeded by 0.78 MiB while allocating a 2 MiB S3 write buffer. That allocation happens inside the CAS parallel blob-upload fan-out (randomized --max_insert_threads 3). Nothing in that path goes through Poco SSL hostname checks, and this PR changes nothing else. I could not confirm it fails on base: the latest base MasterCI (0aa837e) and #2913 attempt 1 both passed this suite. So I'm calling it a narrow memory margin: the fixed 50 MiB budget plus several concurrent CAS upload buffers. That makes it a CAS-side flake, not something from this revert. Next step: safe to re-run. If it fails again, the fix belongs in the test or the CAS branch, not here: raise the test's max_memory_usage for CAS storage, or tag it to skip on CAS. I'll file that separately if you'd like.


Nothing in this PR needs fixing based on this CI run. The failures above don't block the revert, except that the CAS stateless job needs a re-run. If you want more confidence, the remaining step is to wait for SQLLogic/SQLStorm to finish.

@mkmkme mkmkme added the verified Approved for release label Nov 19, 2026
@mkmkme
mkmkme merged commit 0d5106d into antalya-26.8 Nov 19, 2026
384 of 394 checks passed
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

Projects

None yet

Development

Successfully merging this pull request may close these issues.

2 participants