ci(mutation): rebalance kernel-errors into eight shards

Closed
ST
2d ago
+22-13

What happened

The Oct 25 weekly sweep failed two jobs in one chain: Mutants (kernel-errors-4) was killed by the step's 22-minute timeout with Stryker at 41% (20→41% over the last minutes, remaining: ~5m and climbing), and Mutation score report then failed on the incomplete shard set (17/18) — the report lane's designed indictment of a dead shard, not a separate fault.

packages/kernel/src/errors.ts keeps growing (the resource-ownership error reasons #3887 landed alone pushed the module past 350 mutants, and its slow-test region runs ~28 s/mutant), so the shard holding that region can no longer finish inside the budget. The lane's own registry comment names the remedy: "a module big enough to outrun the lane's 30-minute budget declares more".

Fix

kernel-errors shards: 5 → 8. The slow region is diluted to ~10–12 minutes per bin, inside the budget with headroom. Everything the registry drives follows mechanically:

  • shardMatrix() and the weekly matrix gain kernel-errors-6/7/8; --expect-shards 18 → 21.
  • mutation-affected derives its matrix from the same registry and needs no edits; its step budget stays at the tested 22 minutes.
  • The pin tests that assert the matrix and the merged-shard requirement are updated to the new counts.

Verified: node --test scripts/mutation/*.test.ts — 70/70 pass.