Skip to content

fix(ai): retry transient model stream failures - #424

Closed
prvnsmpth wants to merge 2 commits into
masterfrom
fix/openai-compat-error
Closed

fix(ai): retry transient model stream failures#424
prvnsmpth wants to merge 2 commits into
masterfrom
fix/openai-compat-error

Conversation

@prvnsmpth

Copy link
Copy Markdown
Contributor
  • Retry provider-confirmed transport failures once before substantive content is emitted, preventing transient disconnects from failing chats or duplicating output.
  • Classify retryability explicitly per provider while leaving HTTP status retries to SDKs, avoiding deterministic-error replays and retry amplification.
  • Persist usage from failed attempts and add unit and integration coverage for classification, recovery, and single-message persistence.

- flag transport-level provider errors explicitly
- retry once before substantive stream content and persist usage
- cover provider classification and stream recovery behavior
- cover Anthropic Bedrock connection failures
- recognize raw httpx stream failures alongside botocore errors
- add provider classification coverage
@prvnsmpth prvnsmpth closed this Aug 21, 2026
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant