Repository navigation
Conversation
|
Normal successor The baseline artifact The successor records the full byte-sorted listed-test names with tee, retains that list and each selected shard file through the existing bounded summary, and enables pipefail before selection. Inert partial-output/exit-7 controls prove that neither shard runs a partial execution suite or final build when listing fails. All 48 focused part/artifact tests, the full isolation scan and whitespace/secret checks pass. The fresh changed-head ordinary run owns the next real pair-of-shards and complete-union/timing evidence; historical checks are not relabeled. |
|
|
The two CI pacing commits are now preserved by normal merge in active #1036 head The original changed-source run 42099885143 already supplies bounded measurement: MariaDB native parts succeeded in 12m33s and 11m07s, compared with the prior unsharded 22m28s. Both artifact ZIP hashes match their provider digests; complete inventories prove 728 listed tests, two disjoint selections of 364 and no omission. Full execution report retention for the -b part was still missing in this attempt and is corrected in #1036. Other pending/terminal workflow contexts are not claimed as acceptance or transferred to another head. #1036 also corrects the independent frozen-PHP SQLite cleanup contention discovered by its original failed queue group, while preserving exact assertions and all fixture coverage. That original failure remains failed. No manual rerun, new native experiment or publication effect is involved. |
|
The carried CI pacing source is now in accepted Server main through #1036, actual merge |
The unsharded MariaDB native execution lane was the final check holding Server #1036. Its runner steps took 22m28s; the later provider terminal-status delay is separate. Both MySQL labels already split the same execution suite across two parts.
Apply that existing two-part matrix to MariaDB, preserving every execution test once and keeping native database/TLS checks in their original part. Image pins, resource limits, test commands and two test threads are preserved. Retain the complete listed-test inventory and each shard's exact selection under the existing artifact limits. A failing list pipeline now refuses execution even if it emits partial output.
Validation: 48 part/artifact controls pass, including exact union/no-duplication/odd-list checks, real shell command selection with a fake Cargo, partial-output failure controls, byte-exact retained inventories and explicit truncation. The full repository isolation scan, whitespace and secret/private-path checks pass. Current ordinary CI will supply real shard results, selected-test inventories and timings before improvement is claimed. The original #1036 baseline artifact is authenticated and reports 727 passing tests plus one ignored test, but retains only a 64-KiB execution-log tail and no selection list; that limit is explicit.
#1036's original run and queue are untouched. This branch will refresh over its actual accepted main before integration. Related pacing record: #325 (comment) .