-
Notifications
You must be signed in to change notification settings - Fork 358
Open
Labels
Description
What happened?
Observed behavior
When response chunks from AgentExecutor.execute take more than several seconds to produce, a 503 timeout error is received on the A2A SDK client when using certain transport protocols and streaming/polling configurations.
Methodology
For my tests, I used
- v0.3.22 of the SDK
- 10s delay before the first response chunk
- 10s delay after the first two response chunk
- JSONRPC and HTTP+JSON
- Streaming and non-blocking polling via
tasks/get
Code to reproduce results: https://github.com/benedict-khoo-sap/a2a-python-long-running-response-demo
Summary
Here is a summary table of the results observed:
| Server Type | Mode | 10s delay before the first response chunk | 10s delay after the first two response chunk | all response chunks sent quickly |
|---|---|---|---|---|
| JSONRPC | Streaming | ❌ Timeout error after several seconds | ✅ Works fine | |
| JSONRPC | Polling | ❌ Timeout error after several seconds | ✅ Works fine | ✅ Works fine |
| HTTP+JSON | Streaming | ✅ Works fine | ✅ Works fine | ✅ Works fine |
| HTTP+JSON | Polling | ❌ Timeout error after several seconds | ✅ Works fine | ✅ Works fine |
Potential related issue
#545 may be the cause of the issue in JSONRPC implementations, but HTTP+JSON with polling has the same timeout error if the first response chunk takes too long to return
Relevant log output
Code of Conduct
- I agree to follow this project's Code of Conduct
Reactions are currently unavailable