Revert merging of targeted check and flaky check into one#102148
Revert merging of targeted check and flaky check into one#102148alexey-milovidov merged 3 commits intomasterfrom
Conversation
PR #99513 merged the stateless targeted check and flaky check into a single combined job. This was a bad decision because all flaky check options — including Thread Fuzzer — got applied to the targeted check, making it non-representative of the normal test environment. This reverts the merge by restoring: - Separate `Stateless tests (arm_asan_ubsan, targeted)` job that runs coverage-relevant tests 5 times without Thread Fuzzer - `Stateless tests (amd_binary, flaky check)` job that was removed - Removes the `Stateless tests (arm_asan_ubsan, flaky check)` job that replaced both Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
Workflow [PR], commit [0a2d468] Summary: ✅ AI ReviewSummaryThis PR restores separation between Findings❌ Blockers
ClickHouse Rules
Final Verdict
|
| tests_to_run = list(tests) if tests else tests | ||
|
|
||
| for cnt in range(run_sets_cnt): | ||
| if global_time_limit > 0 and run_sets_cnt > 1: |
There was a problem hiding this comment.
The time-limit guard for targeted checks can be bypassed when the initial budget is already exhausted.
global_time_limit is initialized before the loop, and the per-iteration recomputation/break check is gated by if global_time_limit > 0 and run_sets_cnt > 1:. If setup consumed the entire targeted budget (global_time_limit == 0 before the first iteration), this condition is skipped and run_tests is still invoked with global_time_limit=0 (interpreted as no timeout in run_tests).
That means we can start another full clickhouse-test run even after the soft limit is exhausted.
Please move the targeted-check recomputation + exhaustion check outside the global_time_limit > 0 gate (or explicitly break when is_targeted_check and global_time_limit <= 0) before calling run_tests.
|
Disabling thread fuzzer here: #102038 |
…ed-flaky-check # Conflicts: # .github/workflows/pull_request.yml
When `global_time_limit` is already 0 after initial setup (e.g. setup consumed the entire budget), the guard `if global_time_limit > 0` was never entered, allowing all subsequent test iterations to run without a time check. Change the loop guard to check on every iteration after the first (`cnt > 0`) regardless of the initial budget value, so that exhausted budgets are caught before starting extra iterations. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
…lickHouse#101994, ClickHouse#102158, ClickHouse#101952, ClickHouse#102148, ClickHouse#102008, ClickHouse#102010) # Conflicts: # src/Common/AsynchronousMetrics.cpp
…lickHouse#101994, ClickHouse#102158, ClickHouse#101952, ClickHouse#102148, ClickHouse#102008, ClickHouse#102010) # Conflicts: # src/Formats/FormatSettings.h
PR #99513 merged the stateless targeted check and flaky check into a single combined job. This caused all flaky check options — including Thread Fuzzer — to be applied to the targeted check, making it non-representative of the normal test environment.
This reverts the merge by restoring:
Stateless tests (arm_asan_ubsan, targeted)job that runs coverage-relevant tests 5 times without Thread FuzzerStateless tests (amd_binary, flaky check)job that was removedStateless tests (arm_asan_ubsan, flaky check)job that replaced bothChangelog category (leave one):
Changelog entry (a user-readable short description of the changes that goes into CHANGELOG.md):
...
Documentation entry for user-facing changes