Repository navigation
Akash sidecar has no deployment status route #435
Description
The depin sidecar’s Akash router only serves POST /deployments/create, POST /deployments/close and GET /deployments/:id/logs (server.ts), but the Python adapter calls GET /deployments/status/{id}:
- AkashAdapter.wait_for_ready polls it, gets a 404, sees state unknown every time and times out. provision_direct_node waits on it, so every Akash deploy fails at the readiness timeout, after the paid deployment was already created.
- AkashAdapter has no get_node_status, so the liveness worker always gets "unknown" for Akash. Dead Akash deployments aren’t detected, and once InferiaAI/InferiaLLM#422 lands, an already-closed deployment with a deprovision_failed marker is retried every 45s forever.
Activity
Priyanshu-u07 commentedon Oct 25, 2026
Closing as not planned: Akash is parked, see #439 for the details.
Correction to how this started in #422: Akash's close is a mock that always succeeds, so the retry-forever case described above doesn't happen. The real problem is that the Akash flow isn't implemented end to end (bids query a retired API version, the manifest is never sent, and close is a no-op), so fixing the status route alone wouldn't make Akash work.