QuestionQ47

VerifiedEvaluating and deploying agentic workflows

You have deployed a custom, multistep financial agent on Cloud Run. The agent queries a Vector Search index, calls an external stock market API, and uses a Gemini LLM to synthesize an answer, in that order. In production, you find that responses have intermittent latency spikes. You need to identify which step in the agent’s execution workflow is causing the bottleneck.

What should you do?

Explanation

OpenTelemetry traces can capture individual spans for the Vector Search query, external API call, and Gemini interaction. Comparing their durations in Cloud Trace for slow requests identifies which operation contributes to the latency spikes. An overall average request latency cannot isolate the responsible step. (docs.cloud.google.com)

Learn more

Community Discussion

No comments yet. Be the first to start the discussion!