carlos / platform Public

Exec-child Docs promotion leaves prior process serving until scoped restart open

Observed with current CLI built from platformdf7d1d25dd2d069ed77a6ed11996a31baae079ce, supported console API, bds/docs exec-child instances. Main9871ba0 process-stamped binary promoted edge; channel/header advanced but /api/version remained dev from499d200. After configuring apps build-probe /api/version, re-deploy correctly timed out45s. Platform logs showed the route waking499d200 after new promotion. Supported carlos restart --app docs --host docs.bds.oncarlos.com made process return exact9871ba0f776fba35723fdd0d6aa60ef599ebfab3.

Repeated on isolated overnight-docs.bds.oncarlos.com: canary/overnight-release advanced5986df4->2480bae but process continued5986df4423362ca110e6e285dc0ac5747fa38108 until scoped restart requested. The restart API reports asleep on console box (actual exec-child runs elsewhere); its narrow selector is retained. No SSH or infrastructure bypass used. Please investigate channel convergence/process cycling for running exec-child instances. Header-only deploy success masked this until process probe configured. Root continues supported scoped restart + exact process verification for authorized releases.

Repeated on Home isolated credential host overnight-home-id.bds.oncarlos.com, bds/home canary/overnight-home: promotion651e5b3→4013745 updated header but API body stayed651; process-probe deploy watch timed out correctly. Paired directory host already ran401. Scoped supported restart (host plus matching overnight-home environment, dry-run one route) initially reports asleep on console; remote platform logs show restarting/stopping/exit0 at23:05:35UTC. API body and header both401 on both hosts at23:06:06UTC. No second restart or infrastructure workaround needed; allow asynchronous remote restart to complete before deciding it failed. Receipt /tmp/oficina-home-combined-review/canary-version.json.