LlamaIndex Embedding Request Failure

Monitor embedding request success rate by tracking llama_index.embedding.requests alongside error metrics (derived from traces/logs). Alert when error rate >5% sustained or when llama_index.embedding.duration shows timeouts (P99 > 10s). During indexing, complete embedding failure (requests = 0 when documents are being processed) indicates critical failure.

1. Investigate: Check embedding API health and rate limits. Review error logs for API response codes (429 rate limit, 500 server error, timeout). Verify network connectivity to embedding service. 2. Diagnose: Identify if failure is transient (temporary API issue) or persistent (configuration error, auth failure). Check if batch size is overwhelming API. Review API quota/billing status. 3. Remediate: Implement exponential backoff retry for transient errors (max 3 retries with 2x backoff). Reduce batch size if API is rate-limiting. Switch to alternative embedding provider as fallback. For indexing failures: implement checkpoint/resume logic to avoid re-processing all documents. 4. Prevent: Set up embedding API health checks independent of application metrics. Configure alerts on embedding.requests = 0 during expected indexing windows. Implement circuit breaker pattern to fail fast when API is unhealthy. Dashboard embedding success rate (custom metric: successful requests / total attempts).