[V3RON]

ADR 0059 for #1189, Accepted — not yet implemented. A worker that runs real drivers takes a per-user lock (/tmp/simlock-<uid>/worker.lock/, holding pid, folder and start time) before it discovers drivers. A second real-driver worker of the same user with another folder refuses to start and names the first. Workers with substituted drivers (the fast e2e lane, the gateway-fleet tests) and gateways never take it. It narrows ADR 0034 §1's "No new lock is needed" and ends bug #264 for one user.

This PR is the bottom of #1189's stack and merges with the feature's last task.

Status

Review: round 1 3 blocking, round 2 1 blocking, fix-check passed (R1–R3 fixed) Gate: waiting for the stack
Next: stack with #1193

Review

Spec review: 4 blocking, 4 fixed, 1 notes. Code review: skipped. Claims review: skipped.
Rounds: 2. Tokens: 6.6M.
Runs: 7. Cost USD: 2.3004. Agent seconds: 759.
Assessment: complete

Written by an agent.

[V3RON]

Decision taken

Question: two starts of the same folder race. Start A takes the worker lock, start B (same folder, lock held by a live A) goes on, and B wins the socket claim. A then fails and removes its lock, so B runs real drivers with no lock. What should happen?

Options:

• The socket claim's winner rewrites the lock (taken). After a worker wins the socket claim while the lock names its folder and another pid, it rewrites owner.json with its own pid. A failed start removes the lock only when owner.json is still the one that start wrote, so the loser leaves it. A user sees nothing: one worker runs, and it holds the lock.
• The same folder counts as held. B is refused with the "another worker" message. A user doing two simlock commands at once could see a confusing start error even though a worker runs.
• A failed start never removes the lock. The loser's lock goes stale after it exits, and a worker in another folder could take it over while B runs.

Why: the first keeps today's "already running" behaviour, and changes only who is named in owner.json.

Reverting: one rewrite step after the socket claim in startDaemon, and its test line in #1191.

Written by an agent.

[V3RON]

Review ledger

Round 1: 9aaaecb
Round 2: 15b43ff
Round fix-check: d5cd4be

Runs: 7. Cost USD: 2.3004. Agent seconds: 759.
Tokens input 146, output 37208, cache read 3308458, cache write 286723.
- review 1 reviewer: claude-sonnet-5-5 270 s $0.158
- review 1 reviewer spec: claude-opus-5-5 186 s $1.3619
- fix 1 implementer: claude-sonnet-5-5 29 s $0.0836
- review 2 reviewer: claude-sonnet-5-5 148 s $0.1617
- review 2 spec: claude-opus-5-5 69 s $0.399
- fix 2 implementer: claude-sonnet-5-5 28 s $0.065
- review fix-check reviewer: claude-sonnet-5-5 29 s $0.0712
Finding outcomes: {"fixed":3}. Classes: {"adr-index status lines stale":1,"false statement in ADR consequence":1,"same-folder race undecided":1}.

{"findings":[{"id":"R1","role":"spec","class":"false statement in ADR consequence","outcome":"fixed","path":"docs/internal/adr/0059-a-worker-with-real-drivers-holds-a-per-user-lock.md","round":"1","fix":"15b43ff13e4cb3fabae085dec77e1503bf8119ec"},{"id":"R2","role":"spec","class":"same-folder race undecided","outcome":"fixed","path":"docs/internal/adr/0059-a-worker-with-real-drivers-holds-a-per-user-lock.md","round":"1","fix":"d5cd4be973890d68a22b59b9c783ced67afa88f9"},{"id":"R3","role":"spec","class":"adr-index status lines stale","outcome":"fixed","path":"docs/internal/adr/README.md","round":"1","fix":"15b43ff13e4cb3fabae085dec77e1503bf8119ec"}]}

Written by an agent.

[V3RON]

Review notes

Not blocking, not verified. Each is one reviewer's claim.

• spec: docs/internal/adr/0059-a-worker-with-real-drivers-holds-a-per-user-lock.md §2–§3 a tiny window remains where a stopping holder removes the lock between the claim winner's check and its rewrite; judged unrealistic.

Written by an agent.