Repository navigation
ci: refresh PostgreSQL shard weights from retained timings - #1041
1 commit merged into
Conversation
|
|
Exact head Full quality/graph attribution remains on #1043. The original failed #1036 attempt remains preserved; this does not qualify #1036's feature blobs. Workflow owns follow-through; General owns integration of contributor #1044, #1040 and this PR. No merge or repeated full run was performed. |



The PostgreSQL feature shards in #1036's original full run were estimated at about 489 seconds each but took 545–603 seconds. Shard 3 exceeded its 600-second bound. Increase each existing PostgreSQL file weight to the larger of its committed weight and the retained measured duration.
All MySQL weights, test files and deadlines are preserved. The source is run 41966046437, attempt 1, head 5d490bb, using PostgreSQL timing artifacts 13387763296, 13388255961, 13387479423 and 13387509427. All 156 files have unique retained timing entries, including the timed-out shard's 1,040 actual testcase records; its process result remains exit 124. This branch updates the 154 files already on main 98bd986. The two #1036 files keep their existing default until source integration.
Validation: the actual PHP shard verifier passes, representing all 154 main files exactly once across four shards for both backends. A data-only model of the splitter exactly reproduces all four original file manifests. Applying these weights to the measured 156-file corpus projects 572.97–577.65 seconds per shard, compared with 545.10–603.43 using old weights. These are estimates, not a new feature pass or a guaranteed margin. No hosted rerun was requested. The separate MySQL schema repair is draft #1040.