Retry is a recovery procedure that repeats a failed operation when defined conditions, such as a timeout or rate-limit error, make another attempt appropriate. section 1.3.6
Use retry as part of a reliable generative AI integration:
- Pair retry mechanisms with proper error handling to address temporary request failures. section 1.3.6
- Combine retries with usage monitoring, rate limiting, caching, and other API optimizations. section 1.3.6
- Check the resulting output rather than treating a successful retry as proof of correctness. section 1.3.6
A failed API request
An analytics integration encounters a timeout or rate-limit error. Retry handling gives the request another chance and helps maintain integration stability, while output inspection verifies the result. section 1.3.6
