Feature request: retry mechanism for model errors
- Dominant language
- Go
- Stars
- 8.8k
- Forks
- 1k
- Avg merge
- 3d 14h
- Merged PRs (30d)
- 93
Description
I am receiving frequent errors from Gemini models
```
error="workflow execution failed: failed to call model: Error 503, Message: This model is currently experiencing high demand. Spikes in demand are usually temporary. Please try again later., Status: UNAVAILABLE, Details: []"
```
I understand that the recommended solution is a retry mechanism ([link](https://discuss.ai.google.dev/t/handling-429-503-errors-from-the-gemini-api/124640))
> To handle these errors, we strongly recommend implementing backoff and retry (our SDKs support this through retry config).
However, the `adk-go` or `go-genai` packages do not appear to have built-in support or hooks for this. It would be great to have a configurable request retry mechanism to handle these cases.
Contributor guide
Assessment
This issue has not been assessed yet.