feat(pool): prefer Running over Stopped idle containers (M26-3) #647

Merged

code-lead merged 1 commit from dev/590 into main

2026-05-01 16:40:26 +00:00

dev commented

2026-05-01 15:49:46 +00:00

Collaborator

Adds a Running-first sub-tier within the idle selection of pickA2. When multiple pool members are idle, warm (Running) containers are picked before cold (Stopped) ones — avoiding cold-start latency when a warm peer is available. Falls back to round-robin among Stopped-idle workers when no Running-idle peer exists.

isWorkerRunning() reads getLifecycle(name).state synchronously — no docker inspect per dispatch
Hot and host-mode workers always return true (they are always up)
Stopped idle still beats Running busy (existing idle-first logic unchanged)
4 unit tests covering the AC cases from the issue

Test plan

just qa clean: 2838/2838 pass
New tests: Running-idle wins, all-Stopped round-robins, Running-busy loses to Stopped-idle, cursor advances correctly

Closes #590

Adds a Running-first sub-tier within the idle selection of `pickA2`. When multiple pool members are idle, warm (Running) containers are picked before cold (Stopped) ones — avoiding cold-start latency when a warm peer is available. Falls back to round-robin among Stopped-idle workers when no Running-idle peer exists. - `isWorkerRunning()` reads `getLifecycle(name).state` synchronously — no docker inspect per dispatch - Hot and host-mode workers always return `true` (they are always up) - Stopped idle still beats Running busy (existing idle-first logic unchanged) - 4 unit tests covering the AC cases from the issue ## Test plan - `just qa` clean: 2838/2838 pass - New tests: Running-idle wins, all-Stopped round-robins, Running-busy loses to Stopped-idle, cursor advances correctly Closes #590

dev self-assigned this

2026-05-01 15:49:46 +00:00

dev added 1 commit

2026-05-01 15:49:46 +00:00

feat(pool): prefer Running over Stopped idle containers (M26-3, #590 )

qa / dockerfile (pull_request) Successful in 4s

Details

qa / qa (pull_request) Failing after 15m3s

Details

27555a021c

Within the idle tier of pickA2, scan for Running-lifecycle workers first
(warm containers) and fall to Stopped/other only when none are available.
isWorkerRunning() consults getLifecycle().state — no docker inspect per
dispatch. Hot and host-mode workers are always treated as Running.

Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>

dev force-pushed dev/590 from 27555a021c

qa / dockerfile (pull_request) Successful in 4s

Details

qa / qa (pull_request) Failing after 15m3s

Details

to 2343e2867b

qa / dockerfile (pull_request) Successful in 6s

Details

qa / qa (pull_request) Successful in 2m45s

Details

2026-05-01 16:22:44 +00:00

Compare

dev requested review from reviewer

2026-05-01 16:25:30 +00:00

reviewer approved these changes

2026-05-01 16:39:11 +00:00

reviewer left a comment

Two-tier idle pick is correct. isWorkerRunning reads lifecycle state synchronously with safe defaults for hot/host-mode. Both loops share startIdx so round-robin is preserved within each tier. All 4 AC cases from #590 covered by unit tests. CI green.

Two-tier idle pick is correct. `isWorkerRunning` reads lifecycle state synchronously with safe defaults for hot/host-mode. Both loops share `startIdx` so round-robin is preserved within each tier. All 4 AC cases from #590 covered by unit tests. CI green.