TemperFS actor ask timeout (5s) causes session failures under normal load

[rita-aga]

Summary

TemperFS file read/write operations fail with actor dispatch failed: ask timeout after 5s even with a single concurrent agent session. This causes session failures that require manual intervention (Fail + Retry on parent jobs).

Impact

Agent sessions that use TemperFS (which is all of them — conversation writes, file reads, embodiment writes) randomly fail mid-execution. On Railway production (openpaw-production.up.railway.app):

• 11 concurrent sessions: 8 of 11 failed with actor timeout
• 3 concurrent sessions: 2 of 3 failed
• 1 concurrent session: Still fails intermittently (observed at turn 5 with a single session running)

This makes the agent pipeline unreliable — jobs require manual babysitting with Fail/Retry cycles.

Error

TemperFS write failed (HTTP 409): {"error":{"code":"ActionRejected","message":"actor dispatch failed: ask timeout after 5s"}}

Also seen as:

failed to write REPL state file (HTTP 409): {"error":{"code":"ActionRejected","message":"actor dispatch failed: ask timeout after 5s"}}

Questions

1. Is the 5s actor ask timeout configurable? Should it be higher for production?
2. Is the TemperFS actor single-threaded? Can it handle backpressure (queue requests instead of failing them)?
3. Could Turso latency be causing the actor to block on I/O, making it unable to respond within 5s?
4. Should there be automatic retry logic in the session's TemperFS client before surfacing the error?

Environment

• Railway deployment, single service
• Docker image: ghcr.io/nerdsane/openpaw:edge
• Storage: Turso (embedded SQLite)
• Observed with both concurrent and single-session workloads