Router needs to park requests and wait for capacity

[thockin]

Currently if no worker is available, the router returns 500 immediately.

$ curl -X POST -H "Host: ctr6.actors.resources.substrate.ate.dev" -i http://localhost:8888/
HTTP/1.1 500 Internal Server Error
content-type: 
content-length: 96
date: Wed, 20 May 2026 22:46:44 GMT
server: envoy

error resuming actor ctr6: rpc error: code = FailedPrecondition desc = no free workers available

[bowei]

Should such a timeout be configurable or we make this an internal implementation defined wait time. This needs to be coupled with an autoscaling system otherwise no specific choice will make sense.

[thockin]

I think we can start with something dumb like 1 minute and then revisit with specific needs for configurability (per-request? per-actor? per-template? per-substrate?)

[omeryahud]

Hi guys, I took a shot at this here: #221