Loading
Draft: chore(skills): describe the purger's counter-emit cap as derived (#559 stack: 2/2)
Summary
- Problem: #559 replaces the purger's fixed 64-slot counter-emit cap with a ceiling the composition root derives from the Redis pool,
max(1, redisPoolSize / 10). Therun-artifact-registryskill's section "The dispatch, and telling a shed from a late delta" still describes the fixed cap. - Impact: an engineer or agent driving a shed by hand reads arithmetic and an expected drops figure (36) that no longer hold for their pool size.
- Change: the section states the derived ceiling and where it is installed, presents the 64-slot arithmetic as the worked example of a 640-connection pool, and marks the recorded drive as measured under the old cap. It also drops the re-drive reference to
18c031fa2, which no longer resolves. - Downsides: the drive's figures were not re-measured under the derived ceiling.
Stack: this is part 2/2. It targets #559's branch and merges right after #559's MR (part 1/2). Wording follows the already-updated docs/dev/storage-accounting.md section "The emit is dispatched behind a shedding cap" on that branch.
Checklist
- Ran it: n/a, skill text only; nothing in
driver.shchanges. - Interfaces and data: none.
- Failure: n/a.
- Rollout: n/a.
- Hot path: n/a.
- Database: no.
- Security: no.
- ADRs: none; the change is skill text only and alters no behaviour.
- E2E: none; no observable behaviour changes.
- Simplicity: deleting the worked arithmetic was rejected: it still explains why a shed signals slow Redis long before a stall.
Related to #559