DaniAkash e77031025b fix(container): poll readiness probe within descriptor budget on start ...

ManagedContainer.start was firing the subclass `readinessProbe()`
exactly once, the moment containerd reported the container as Up.
For OpenClaw this raced the Node.js gateway's HTTP listener bind —
containerd flips status as soon as the entrypoint process spawns, but
the Express server takes a few hundred ms to start serving /readyz.
Single-shot probe → unlucky → state='errored' with
"Readiness probe failed after container reached running state".

Pre-refactor (dev branch) didn't hit this because openclaw used a
two-phase flow: `runtime.startGateway` (no probe) then
`service.waitForReady` (polled /readyz for 30s). When the new
runtime architecture folded openclaw under ManagedContainer, the
polling was lost.

Bring it into the base class: `ManagedContainer.start` now polls
`readinessProbe()` within `descriptor.readinessProbe.timeoutMs` at
`intervalMs` cadence. Deterministic probes (Hermes' `--version` exec)
succeed on the first call and exit immediately — no extra latency.
HTTP probes get the full budget they need.

Also stops misapplying `descriptor.readinessProbe` to the containerd
"Up" wait (which only takes ~50ms anyway — defaults are fine).

2026-05-11 20:41:43 +05:30

..

browseros

Revert "refactor(openclaw): lock no-auth gateway, bump image, delete token pl…" (#953)

2026-05-07 01:49:50 +05:30

browseros-agent

fix(container): poll readiness probe within descriptor budget on start

2026-05-11 20:41:43 +05:30