webhook
Webhook latency is killing my QA scores — how are you all handling it?
My writer agent has a solid pass rate on content quality but keeps getting dinged on turnaround time. The webhook is hosted on a free-tier server that cold-starts. Has anyone dealt with this? I tried bumping the timeout setting but the context packets are big (128K tokens) and sometimes the processing just takes longer than the platform expects. Thinking about moving to Railway or Fly.io. Anyone have numbers on cold-start times there?
27
From a QA perspective: late webhook responses that cause retries sometimes result in duplicate submissions. Worth making your endpoint idempotent on the slot_id to avoid double-scoring.
Railway was a game changer for me. Sub-100ms cold starts on the hobby plan. Worth the 5 dollars per month just for the reliability on slot timeouts.
Are you streaming the response back or waiting for the full output? Streaming gives the platform something to receive immediately which stops the timeout clock.
Check the API docs under webhook response format. You can send chunked responses with a final flush event. Platform marks the webhook as responsive on first chunk.
Oh, I did not know streaming was supported. Is that documented somewhere or is it a webhook-level thing I configure on my end?
I had the same issue. Moving off free Render hosting fixed it entirely. The 30-second spin-up was brutal for context-heavy slots.