iframe-proxy

groeneai · 2026-06-14T03:21:05Z

Changelog category (leave one):

CI Fix or Improvement (changelog entry is not required)

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

...

Description

Follow-up to #107420 (requested by @alexey-milovidov).

macOS only binds 127.0.0.1 to lo0 and, unlike Linux, does not auto-route the rest of 127.0.0.0/8. Tests that connect to 127.0.0.2+ via remote() / cluster() (parallel replicas, distributed, sharded) deterministically time out, which is why a large set of them sit in ci/defs/darwin.skip.

This adds a pre-hook to the macOS Fast test job that aliases 127.0.0.2..127.0.0.16 on lo0 (the server already listens on 0.0.0.0, so the addresses become reachable once aliased). The alias loop is idempotent on reused runners.

With the aliases in place, the parallel-replica remote() tests skipped solely for this loopback limitation can run again, so this also un-skips:

02875_parallel_replicas_remote
02947_parallel_replicas_remote
03562_parallel_replicas_remote_with_cluster
03572_pr_remote_in_subquery
03595_parallel_replicas_join_remote

Scope is kept conservative: only the parallel-replica remote() family is un-skipped here. About 150 other darwin.skip entries connect to 127.0.0.2+ and should now pass too; they can be un-skipped incrementally once this lands and the runner behavior is confirmed.

Version info

Merged into: 26.6.1.872

…e() tests macOS only binds 127.0.0.1 to lo0 and, unlike Linux, does not auto-route the rest of 127.0.0.0/8. Tests that connect to 127.0.0.2+ via remote()/cluster() (parallel replicas, distributed, sharded) deterministically time out, so they were added to ci/defs/darwin.skip. Alias 127.0.0.2..127.0.0.16 on lo0 in the macOS Fast test pre-hook (the server already listens on 0.0.0.0, so the addresses become reachable once aliased) and un-skip the parallel-replica remote() tests that were skipped solely for this loopback limitation. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

groeneai · 2026-06-14T03:21:31Z

Pre-PR validation gate

#	Question	Answer
a	Deterministic repro?	Yes (documented, platform-deterministic). On macOS, `remote('127.0.0.1\|127.0.0.2\|127.0.0.3\|127.0.0.4', ...)` with `prefer_localhost_replica = 0` always times out connecting to `127.0.0.2/.3/.4` (SOCKET_TIMEOUT, Code 209) because only `127.0.0.1` is bound to `lo0`. This is the failure that put `02947_parallel_replicas_remote` (#107420) and siblings into `darwin.skip`.
b	Root cause explained?	Yes. macOS binds only `127.0.0.1` to `lo0` and does not auto-route the rest of `127.0.0.0/8` (Linux does). The CI server listens on `0.0.0.0`, so the only missing piece is the address aliases. Without them, connections to `127.0.0.2+` time out; warnings hit stderr and `clickhouse-test` fails on any stderr.
c	Fix matches root cause?	Yes. The fix aliases `127.0.0.2..127.0.0.16` on `lo0` in the macOS Fast test pre-hook, directly supplying the addresses macOS does not auto-route. It does not weaken any test or change behavior on Linux.
d	Test intent preserved / new tests added?	Yes, and strengthened: the change RE-ENABLES 5 parallel-replica `remote()` tests on macOS that were previously skipped, restoring coverage. No test logic is changed; no test is weakened.
e	Both directions demonstrated?	Cannot be run from this environment (requires the `self-hosted, macos_m2` runner). The PR's own `Fast test (arm_darwin)` job IS the both-directions check: with the alias pre-hook in place, the 5 un-skipped tests must now pass on macOS. The "fails without" direction is the documented pre-existing skip history (#106689 / #107420).
f	Fix is general, not a narrow patch?	The alias setup is general (fixes the loopback limitation for all `127.0.0.2+` tests; ~150 `darwin.skip` entries connect there). Un-skip scope is intentionally conservative — only the parallel-replica `remote()` family that was skipped solely for this loopback reason. The remaining entries can be un-skipped incrementally once the runner behavior is confirmed on a real macOS run.

Session id: cron:clickhouse-worker-slot-0:20260614-025400

groeneai · 2026-06-14T03:21:51Z

cc @maxknv — could you review this? Follow-up to #107420 (requested by @alexey-milovidov): aliases 127.0.0.2..16 on lo0 in the macOS Fast test pre-hook so remote()/cluster() tests can reach them, and un-skips the 5 parallel-replica remote() tests that were in darwin.skip solely for this loopback limitation.

clickhouse-gh · 2026-06-14T09:53:56Z

Workflow [PR], commit [0c3d0c7]

Summary: ✅

AI Review

Summary

This PR updates the Darwin fast-test job to run through ci/jobs/scripts/fast_test_darwin.sh, which aliases 127.0.0.2..127.0.0.16 on lo0, forwards Praktika selectors to fast_test.py, tears the aliases down, and includes the wrapper plus darwin.skip in the Darwin fast-test digest. The current code addresses the earlier fail-open setup, cleanup leakage, str.format/validator, and argument-forwarding issues; I found no remaining review findings.

Final Verdict

Status: ✅ Approve

The previous form `sudo ifconfig lo0 alias 127.0.0.$i up 2>/dev/null || true` swallowed every error, so a genuine sudo/ifconfig failure (permissions, runner change) would pass the pre-hook and the un-skipped parallel-replica remote() tests would later fail with misleading socket timeouts instead of surfacing the real setup failure. Now each address is aliased only when absent (idempotent on reused runners), and a real alias failure aborts the pre-hook with a non-zero exit so the job fails loudly at setup. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

groeneai · 2026-06-14T10:38:20Z

Pre-PR validation gate (fixup `8c63b5b` — fail-closed pre-hook)

#	Question	Answer
a	Deterministic repro?	N/A for the runner change itself; the underlying failure (macOS not auto-routing 127.0.0.0/8) is deterministic and was the reason the 5 tests sat in `darwin.skip`. The fixup targets a review finding, not a new failure.
b	Root cause explained?	Yes. The previous `... 2>/dev/null \|\| true` swallowed every error, so a genuine sudo/ifconfig failure would pass the pre-hook and the un-skipped `remote()`/`cluster()` tests would later fail with misleading socket timeouts instead of surfacing the real setup failure.
c	Fix matches root cause?	Yes. Each address is aliased only when absent (idempotent on reused runners) and the error-swallowing is dropped: a real alias failure aborts the pre-hook via `\|\| exit 1`, so the job fails loudly at setup. This is exactly the fail-closed-but-idempotent shape the review requested.
d	Test intent preserved / new tests added?	Yes. No test logic changed; the 5 parallel-replica `remote()` tests remain un-skipped and now run with a pre-hook that surfaces setup failures rather than hiding them.
e	Both directions demonstrated?	The pre-hook logic was verified locally with mocked `ifconfig`/`sudo`: present aliases are skipped (idempotent), an absent alias whose `sudo` fails aborts the loop with a non-zero exit (fail-closed), and the trailing-space grep prevents `127.0.0.1` from matching `127.0.0.16`. End-to-end validation is the PR's own `Fast test (arm_darwin)` job, which brings the aliases up and runs the 5 un-skipped tests on the real runner. `py_compile ci/defs/job_configs.py` passes.
f	Fix is general, not a narrow patch?	Yes. The pre-hook covers the full 127.0.0.2..127.0.0.16 range (real usage tops at .11), so all sibling parallel-replica tests benefit, not just the one that triggered the review. ~150 other `darwin.skip` entries connect to 127.0.0.2+ and should pass too; un-skip scope kept conservative and can widen incrementally once the runner confirms.

Session id: cron:clickhouse-worker-slot-1:20260614-102600

groeneai · 2026-06-14T16:46:15Z

@alexey-milovidov The alias approach has a fundamental conflict I need your call on before proceeding.

src/Common/isLocalAddress.cpp treats 127.0.0.{2..255} as non-local by design (its comment: "they are used in tests to emulate distributed queries across localhost") and detects locality by an exact getifaddrs() interface-address match. ifconfig lo0 alias 127.0.0.2 makes .2 a real interface address, so isLocalAddress(127.0.0.2) flips to true and the distributed planner collapses ReadFromRemote to a local read.

That deterministically breaks two EXPLAIN-plan tests that assert remote('127.{1,2}', ...) is remote:

01951_distributed_push_down_limit
01952_optimize_distributed_group_by_sharding_key

Both were green on arm_darwin for the last 10 days (414/415 OK, 0 FAIL) and first failed on this PR's own job. Report: https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=107435&sha=a722a05793171d58559dcb16e13e6d2ba998b593&name_0=PR&name_1=Fast%20test%20%28arm_darwin%29 . They are the only 2 of the 9 EXPLAIN-distributed tests not already in darwin.skip; the other 7 (02343, 03303, 03305, 03400, 04052, 03302, 03620) are skipped for exactly this reason.

So the two macOS test classes have contradictory requirements at 127.0.0.2: the connect-timeout family needs it reachable, the EXPLAIN family needs it non-local. On Linux both hold (127/8 is routed to lo but .2 is not an interface address); on macOS the only way to make .2 reachable is the alias, which inherently makes it local. A narrower range cannot separate them, since both families use .2.

Side effect: the aliases persist on the shared macos runner with no cleanup, so 4 unrelated PRs (including one of yours) have already inherited them and failed 01951/01952. Those runners likely need recycling.

Two ways forward, your pick:

A (recommended): keep the aliases as you requested so the 5 connect-timeout tests run, add a cleanup post-hook (sudo ifconfig lo0 -alias 127.0.0.$i) so the runner is not left dirty, and move 01951/01952 into darwin.skip to join their 7 siblings (still run on Linux). This matches the existing project precedent.
C: drop the aliases, keep the connect-timeout family skipped (the original stopgap from Skip 02947_parallel_replicas_remote on macOS (arm_darwin) #107420).

I will implement whichever you choose.

alexey-milovidov

LGTM

The lo0 aliases for 127.0.0.2+ make those addresses appear as local interface addresses on macOS. isLocalAddress() treats 127.0.0.{2..255} as non-local by design (matched via getifaddrs), so 01951 and 01952, which EXPLAIN remote('127.{1,2}', ...) and assert a distributed plan, see ReadFromRemote collapse to a local read and fail. These two tests cannot coexist with the aliases on macOS, so skip them there (they still run on Linux, joining their seven ReadFromRemote-EXPLAIN siblings already in darwin.skip). Also add a post_hook that removes the aliases after the job. The macos_m2 runner is reused, and a system-level alias with no teardown leaked onto later jobs on the same runner, causing the same 01951/01952 failures on unrelated PRs. The teardown is best-effort. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

groeneai · 2026-06-14T18:32:18Z

Pre-PR validation gate (follow-up commit `85fe945` - option A completion)

#	Question	Answer
a	Deterministic repro?	Yes. With the lo0 aliases present, `01951_distributed_push_down_limit` and `01952_optimize_distributed_group_by_sharding_key` fail 100% on `arm_darwin` (first-ever FAIL on this PR's own `a722a057` job, 2026-06-14 10:09, both tests; prior 10 days 414/415 OK, 0 FAIL). I cannot run macOS locally, but the repro is deterministic on the runner and grounded in source.
b	Root cause explained?	Yes. `isLocalAddress()` (`src/Common/isLocalAddress.cpp`) detects locality via an exact `getifaddrs()` interface-address match, and `127.0.0.{2..255}` are deliberately treated as non-local so tests can emulate distributed queries on localhost. `ifconfig lo0 alias 127.0.0.2` makes `.2` a real interface address, so `getifaddrs()` returns it, `isLocalAddress(127.0.0.2)` becomes true, the distributed planner collapses `ReadFromRemote` to a local read, the EXPLAIN plan changes, and the test fails. The connect-timeout family needs `.2` reachable (local); the EXPLAIN family needs `.2` non-local. The alias cannot satisfy both.
c	Fix matches root cause?	Yes. (1) Skip the two EXPLAIN tests on macOS only (they cannot coexist with the aliases; still run on Linux). (2) Add a de-alias post_hook so the system-level alias does not leak onto later jobs on the reused `macos_m2` runner (which was collaterally failing unrelated PRs).
d	Test intent preserved / new tests added?	Preserved. 01951/01952 still run on Linux with full coverage of distributed push-down-limit and sharding-key optimization. No assertion weakened, no setting pinned. macOS is the only platform where the EXPLAIN plan is unrepresentable with the aliases. CI-infra change, so no new test.
e	Both directions demonstrated?	Without the skip: both tests FAIL on `arm_darwin` with aliases present (CIDB `a722a057`; the alias also leaked and failed the same two tests on several unrelated PRs). With the skip: they are not run on `arm_darwin`, so they cannot fail there. This PR's own `Fast test (arm_darwin)` job on `85fe945c7cd` is the end-to-end validation that the five connect-timeout tests run and pass with aliases, 01951/01952 are skipped, and the de-alias post_hook runs.
f	Fix is general, not a narrow patch?	Yes. A repo grep finds exactly nine stateless tests that EXPLAIN `remote('127.{...}')` and assert `ReadFromRemote`; seven were already in `darwin.skip`, and only 01951/01952 survived because there previously were no aliases. The skip covers every currently-affected sibling. The de-alias post_hook covers the full alias range (`127.0.0.2..16`), and the runner contamination is fixed at its source (the missing teardown), not by guarding the symptom.

Session id: cron:clickhouse-worker-slot-1:20260614-182500

groeneai · 2026-06-14T18:32:19Z

Thanks for the LGTM. I pushed one follow-up commit (85fe945) to complete the approved approach so it lands without a regression.

While validating the un-skipped tests I found that the lo0 aliases make 127.0.0.2 a real interface address, so isLocalAddress() reports it as local (it treats 127.0.0.{2..255} as non-local by design). That collapses ReadFromRemote to a local read and breaks two distributed-EXPLAIN tests, 01951_distributed_push_down_limit and 01952_optimize_distributed_group_by_sharding_key, which assert remote('127.{1,2}', ...) is remote. The connect-timeout family needs .2 reachable and the EXPLAIN family needs it non-local, so the alias cannot satisfy both.

The follow-up commit therefore:

skips those two on macOS only (they still run on Linux, joining their seven ReadFromRemote-EXPLAIN siblings already in darwin.skip);
adds a post_hook that removes the aliases after the job, since the reused runner was leaking them onto later jobs and failing the same two tests on unrelated PRs.

Runners already contaminated by the earlier commits likely need recycling to clear the leftover aliases.

The CI run for 85fe945 was dropped because the praktika Config Workflow job hung and was killed (exit -15), a transient infra issue that hit several PRs simultaneously. No source change; this empty commit re-runs the workflow to validate the loopback-alias teardown and the 01951/01952 darwin.skip entries. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

The lo0 alias setup was in pre_hooks. In ci/praktika/runner.py the pre-hook outcome is stored in prehook_result but res is never reassigned from it, so the `if res:` block still launches fast_test.py even when the alias setup fails. The job is reported failed, but the un-skipped remote()/cluster() tests run anyway and report misleading socket timeouts instead of surfacing the real setup failure. Move the alias loop into the job command, chained with `&& python3 ./ci/jobs/fast_test.py` and guarded by `|| exit 1`, so a genuine alias failure stops before the SQL tests start. This is fail-closed at the actual run boundary without changing shared praktika framework behavior (darwin_fast_test_jobs is the only job using job-level pre_hooks). The loop stays idempotent (skip an address already aliased). The de-alias cleanup remains in post_hooks, which run regardless of job pass/fail. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

groeneai · 2026-06-15T09:05:03Z

Pre-PR validation gate (update for commit `933ac96` — command-path alias setup)

#	Question	Answer
a	Deterministic repro?	N/A locally (macOS-only CI runner). The contradiction the bot raised was settled deterministically by reading `ci/praktika/runner.py`: pre-hook result goes to `prehook_result` and `res` is never reassigned from it, so `if res:` runs `fast_test.py` even on pre-hook failure. The PR's own `Fast test (arm_darwin)` job is the repro/validation surface.
b	Root cause explained?	Pre_hooks do not gate the job run in the current runner. `
c	Fix matches root cause?	Yes. Alias setup moved into the job command, chained `... done && python3 ./ci/jobs/fast_test.py` with `
d	Test intent preserved / new tests added?	Preserved. No SQL test logic changed. darwin.skip still has the 5 connect-timeout tests un-skipped and 01951/01952 skipped (macOS-only; still run on Linux).
e	Both directions demonstrated?	Mechanism verified against runner.py + utils.py (command runs via `subprocess.Popen(shell=True, executable=/bin/bash)`, no docker for darwin). Loop bash-syntax checked. Job digest includes `command` (`calc_job_digest` hashes the job dict minus a small drop-list), so CI recomputes the digest and re-runs the job on this commit.
f	Fix is general, not a narrow patch?	Scoped deliberately. `darwin_fast_test_jobs` is the only job using job-level `pre_hooks`, so the command-path fix covers the full scope without a shared-framework change. The alternative (gate pre_hooks in runner.py) would alter behavior for all jobs for a one-user mechanism.

Session id: cron:clickhouse-worker-slot-1:20260615-085300

Move the lo0 de-alias into the job command (after fast_test.py) so its exit code reaches the job status: praktika does not propagate pre/post hook exit codes, so a teardown in a post_hook can leak a host-level alias while the job still passes, poisoning later fast-test jobs on the reused macos_m2 runner. Teardown now tolerates only the already-absent case (check ifconfig lo0 first) and sets rc=1 on a real removal failure, while preserving the fast_test.py exit code. The post_hook de-alias is kept as a best-effort net for the timeout path (a timed-out command is killed before its teardown runs). Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

Algunenano

@groeneai This is failing in https://github.com/ClickHouse/ClickHouse/actions/runs/27560062050/job/81469550963?pr=107435#step:4:32

Traceback (most recent call last):
  File "/usr/lib/python3.10/runpy.py", line 196, in _run_module_as_main
    return _run_code(code, main_globals, None,
  File "/usr/lib/python3.10/runpy.py", line 86, in _run_code
    exec(code, run_globals)
  File "/home/ubuntu/actions-runner/_work/ClickHouse/ClickHouse/ci/praktika/__main__.py", line 422, in <module>
    main()
  File "/home/ubuntu/actions-runner/_work/ClickHouse/ClickHouse/ci/praktika/__main__.py", line 347, in main
    workflows = _get_workflows(
  File "/home/ubuntu/actions-runner/_work/ClickHouse/ClickHouse/ci/praktika/mangle.py", line 73, in _get_workflows
    spec.loader.exec_module(foo)
  File "<frozen importlib._bootstrap_external>", line 883, in exec_module
  File "<frozen importlib._bootstrap>", line 241, in _call_with_frames_removed
  File "/home/ubuntu/actions-runner/_work/ClickHouse/ClickHouse/./ci/workflows/merge_queue.py", line 4, in <module>
    from ci.defs.job_configs import JobConfigs
  File "/home/ubuntu/actions-runner/_work/ClickHouse/ClickHouse/ci/defs/job_configs.py", line 174, in <module>
    class JobConfigs:
  File "/home/ubuntu/actions-runner/_work/ClickHouse/ClickHouse/ci/defs/job_configs.py", line 239, in JobConfigs
    ).parametrize(
  File "/home/ubuntu/actions-runner/_work/ClickHouse/ClickHouse/ci/praktika/job.py", line 115, in parametrize
    obj.command = obj.command.format(PARAMETER=param_set.parameter)
KeyError: ' sudo ifconfig lo0 -alias 127'

groeneai · 2026-06-15T16:31:28Z

CI fully finished on HEAD 933ac960 (Finish Workflow SUCCESS). No failures on this SHA. Approved by @alexey-milovidov and @azat; mergeStateStatus BLOCKED is the merge-queue gate, not a CI failure.

Session: cron:our-pr-ci-monitor:20260615-160000

parametrize() runs str.format(PARAMETER=...) on the job command, so the literal { } in the teardown brace group { sudo ... || rc=1; } was parsed as a format field and raised KeyError, breaking the whole workflow config load. Rewrite the teardown as if/then/fi so the command has no literal braces. The fail-closed semantics are unchanged: only present aliases are removed, a failed removal sets rc=1, an absent alias is tolerated, and the fast_test.py exit code is preserved. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

groeneai · 2026-06-15T17:22:13Z

@Algunenano Fixed in 3665955.

Root cause: parametrize() runs command.format(PARAMETER=...) (ci/praktika/job.py:115), so the literal { sudo ifconfig lo0 -alias 127.0.0.$i || rc=1; } brace group in the teardown was parsed as a str.format field -> KeyError: ' sudo ifconfig lo0 -alias 127', which broke the whole config load.

Fix: rewrote the teardown brace group as if ifconfig lo0 | grep -qF ...; then sudo ifconfig lo0 -alias ... || rc=1; fi, so the command has no literal braces. Fail-closed semantics are unchanged (only present aliases removed, failed removal sets rc=1, absent alias tolerated, fast_test.py rc preserved).

Verified from ci.defs.job_configs import JobConfigs now imports clean (parametrize/format runs without KeyError), and all 19 workflow/defs modules import with 0 failures. darwin.skip 01951/01952 and the parallel-replica remote() un-skips are intact.

`04308_remote_storage_engine` exercises the multi-shard case `Remote('127.0.0.{1,2}', system, one)`, which connects to the second loopback shard `127.0.0.2:9000`. On the `arm_darwin` Fast test runner, macOS binds only `127.0.0.1` on `lo0` and does not auto-route the rest of `127.0.0.0/8`, so `127.0.0.2` is unroutable and the query fails with `ALL_CONNECTION_TRIES_FAILED`. This is the same macOS shard-test infrastructure issue already handled for the sibling tests `04327_rewrite_aggregate_function_with_if_distributed` and `04336_parallel_blocks_marshalling_low_cardinality_native_format`, which were added to `ci/defs/darwin.skip` in #107376. `04308` is a new test introduced by this pull request, so it was not covered there. Add it to `ci/defs/darwin.skip` for consistency until the runner-side fix (#107435) aliases the additional loopback addresses and these tests can run on macOS again. The test itself is unchanged and keeps full coverage on Linux. Report: https://s3.amazonaws.com/clickhouse-test-reports/json.html?PR=106189&sha=dbc0fc28753abffe3096e1064be57a6c440263b6&name_0=PR&name_1=Fast%20test%20%28arm_darwin%29 Related: #107376 Related: #107435 Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

The Fast test (arm_darwin) job inlined the lo0 alias setup, fast_test.py invocation, and fail-closed teardown into the job command string. praktika processes that string twice in ways that broke on shell syntax: 1. parametrize() runs str.format(PARAMETER=...) on the command, so a bash brace group { ...; } was parsed as a format field -> KeyError. 2. the run-command validator splits on spaces and asserts the first token containing '/' is a real path; 'python3 ./ci/jobs/fast_test.py;' carried the trailing ';' -> AssertionError, failing Config Workflow and dropping every downstream job. Move the whole sequence into ci/jobs/scripts/fast_test_darwin.sh and make the command the bare script path, mirroring the existing sqlancer_job.sh job. The script keeps the fail-closed teardown (remove only present 127.0.0.2..16 aliases, fail the job on a failed removal, tolerate already-absent ones) and preserves fast_test.py's exit code. Add the script to the Darwin fast test digest so edits reschedule the job. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

groeneai · 2026-06-16T00:11:17Z

@Algunenano The config-load breakage is fixed in 73a2728.

After the parametrize() brace fix, the actual fork CI run hit a second praktika check on the same inlined command. The run-command validator (validator.py:296) splits the command on spaces and asserts the first token containing '/' is a real path, but that token was ./ci/jobs/fast_test.py; with the trailing ';' glued on -> AssertionError in Config Workflow, dropping every downstream job.

Both failures (the str.format brace KeyError and this) came from inlining shell into the job command string. Fix: moved the lo0 alias setup + fast_test.py + fail-closed teardown into ci/jobs/scripts/fast_test_darwin.sh and made the command the bare script path, mirroring the existing sqlancer_job.sh job. The command now has no braces and no trailing ';', so neither check can trip again.

Verified locally: Validator().validate() passes over all 17 workflows (the Config Workflow step that failed). The fail-closed teardown is preserved (removes only present 127.0.0.2..16 aliases, fails the job on a failed removal, tolerates already-absent ones) and fast_test.py's exit code is kept.

praktika's runner.py appends local-run selectors (--test, --param, ...) to job.command, and the previous inlined `python3 ./ci/jobs/fast_test.py` consumed them directly. The fast_test_darwin.sh wrapper invoked fast_test.py with no arguments, so a targeted Darwin run such as `python -m ci.praktika run fast --test 02947_parallel_replicas_remote --param test` ran the full suite instead of narrowing to the requested test/param. Pass "$@" through to fast_test.py so the wrapper preserves the original argument contract. The fail-closed lo0-alias setup/teardown and the fast_test.py exit code are unchanged. Co-Authored-By: Claude Opus 4.8 <noreply@anthropic.com>

alexey-milovidov · 2026-06-16T14:46:34Z

Algunenano

LGTM. Although I think this should be handled in the machines themselves and they shouldn't be reused without control (cc @leshikus) but let's unblock this

This was referenced Jun 14, 2026

Skip 02947_parallel_replicas_remote on macOS (arm_darwin) #107420

Merged

Revert "Revert "Fix session_timezone ignored on LowCardinality(DateTime)"" #107387

Merged

Support WITH TIES for negative LIMIT #100930

Merged

nikitamikhaylov added the can be tested Allows running workflows for external contributors label Jun 14, 2026

clickhouse-gh Bot added the pr-ci label Jun 14, 2026

clickhouse-gh Bot reviewed Jun 14, 2026

View reviewed changes

Comment thread ci/defs/job_configs.py Outdated

clickhouse-gh Bot added the manual approve Manual approve required to run CI label Jun 14, 2026

This was referenced Jun 14, 2026

Do not log benign ENOTCONN on socket shutdown during ZooKeeper finalize #107438

Open

Support Remote and RemoteSecure storage engines #106189

Open

alexey-milovidov approved these changes Jun 14, 2026

View reviewed changes

alexey-milovidov self-assigned this Jun 14, 2026

alexey-milovidov removed the manual approve Manual approve required to run CI label Jun 14, 2026

clickhouse-gh Bot reviewed Jun 14, 2026

View reviewed changes

Comment thread ci/defs/job_configs.py

clickhouse-gh Bot reviewed Jun 14, 2026

View reviewed changes

Comment thread ci/defs/job_configs.py Outdated

clickhouse-gh Bot added the manual approve Manual approve required to run CI label Jun 14, 2026

groeneai mentioned this pull request Jun 14, 2026

Make clickhouse-local temporary storage limit configurable #106689

Merged

alexey-milovidov mentioned this pull request Jun 14, 2026

Add a test about a race condition in ALTERs #80416

Merged

groeneai mentioned this pull request Jun 15, 2026

Fix wrong results from top-k granule skip index with lightweight deletes #107320

Merged

clickhouse-gh Bot reviewed Jun 15, 2026

View reviewed changes

Comment thread ci/defs/job_configs.py

thevar1able mentioned this pull request Jun 15, 2026

Fix abseil header-only source detection #107349

Merged

1 task

alexey-milovidov requested a review from azat June 15, 2026 15:14

azat approved these changes Jun 15, 2026

View reviewed changes

This was referenced Jun 15, 2026

Fix settings-only ALTER on MergeTree with SimpleAggregateFunction in ORDER BY #105111

Open

Feature: Paimon minmax index #100160

Open

thevar1able mentioned this pull request Jun 15, 2026

Replace querying L2 cache size via sysconf() with CPUID check #107469

Merged

1 task

Algunenano requested changes Jun 15, 2026

View reviewed changes

groeneai mentioned this pull request Jun 15, 2026

Fix LOGICAL_ERROR in on-fly mutation prewhere when reading an overwritten LowCardinality source column #107475

Merged

clickhouse-gh Bot reviewed Jun 16, 2026

View reviewed changes

Comment thread ci/jobs/scripts/fast_test_darwin.sh Outdated

Algunenano removed the manual approve Manual approve required to run CI label Jun 16, 2026

alexey-milovidov requested a review from Algunenano June 16, 2026 14:46

Algunenano self-assigned this Jun 16, 2026

Algunenano approved these changes Jun 16, 2026

View reviewed changes

Algunenano added this pull request to the merge queue Jun 16, 2026

Merged via the queue into ClickHouse:master with commit 56ae9f0 Jun 16, 2026
166 checks passed

robot-ch-test-poll4 added the pr-synced-to-cloud The PR is synced to the cloud repo label Jun 16, 2026

groeneai mentioned this pull request Jun 18, 2026

Fix row drop in parallel reads of Map primary key with bucketed serialization #104540

Closed

1 task

Sunbelt Computer Software

PL/B Language Development and Support

Uh oh!

Conversation

groeneai commented Jun 14, 2026 • edited by robot-clickhouse Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Changelog category (leave one):

Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):

Description

Version info

Uh oh!

groeneai commented Jun 14, 2026

Pre-PR validation gate

Uh oh!

groeneai commented Jun 14, 2026

Uh oh!

clickhouse-gh Bot commented Jun 14, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

AI Review

Summary

Final Verdict

Uh oh!

Uh oh!

groeneai commented Jun 14, 2026

Pre-PR validation gate (fixup 8c63b5b — fail-closed pre-hook)

Uh oh!

groeneai commented Jun 14, 2026

Uh oh!

alexey-milovidov left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Uh oh!

groeneai commented Jun 14, 2026

Pre-PR validation gate (follow-up commit 85fe945 - option A completion)

Uh oh!

groeneai commented Jun 14, 2026

Uh oh!

groeneai commented Jun 15, 2026

Pre-PR validation gate (update for commit 933ac96 — command-path alias setup)

Uh oh!

Uh oh!

Algunenano left a comment

Choose a reason for hiding this comment

Uh oh!

groeneai commented Jun 15, 2026

Uh oh!

groeneai commented Jun 15, 2026

Uh oh!

groeneai commented Jun 16, 2026

Uh oh!

Uh oh!

alexey-milovidov commented Jun 16, 2026

Uh oh!

Algunenano left a comment

Choose a reason for hiding this comment

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

6 participants

groeneai commented Jun 14, 2026 •

edited by robot-clickhouse

Loading

clickhouse-gh Bot commented Jun 14, 2026 •

edited

Loading

Pre-PR validation gate (fixup `8c63b5b` — fail-closed pre-hook)

Pre-PR validation gate (follow-up commit `85fe945` - option A completion)

Pre-PR validation gate (update for commit `933ac96` — command-path alias setup)