When making a request to stream in a thread that uses a large tool call, subsequent stream requests do not stream. Instead they hang and then the whole stream is delivered at once after 30 seconds.
This looks like an internal issue, anyone experience this too?
Hi @danielsiemers ,
Could you please provide bit more details to confirm whether it’s a known streaming issue vs infra buffering, could you share:
- Agent Server / LangGraph SDK versions.
- Exact streaming endpoint used (threads.stream, runs.stream, or join_stream) and stream_mode value.
- Whether this happens only after large tool calls, and approximate tool payload size.
- Whether you’re behind a proxy/load balancer (nginx, Cloudflare, ALB), and any SSE buffering/timeout settings.
This will narrow down root cause.