traces

Query OpenSearch traces using PPL to investigate GenAI agent invocations and tool executions by targeting OTEL spans in the otel-v1-apm-span-* indices via the OpenSearch PPL API. This surfaces latency, errors, and token usage for end-to-end trace debugging.

Analyzing slow traces and token usage patterns in OpenSearch is best done using PPL queries against otel-v1-apm-span-* and service-map data in otel-v2-apm-service-map-*. This identifies slow agent calls, tool execution failures, and cross-service dependencies.

Yes, you can use PPL to debug errors and latency across OTEL spans stored in OpenSearch. The Skill queries otel-v1-apm-span-* indices to summarize error spans and identify slow traces, providing performance and reliability insights for distributed workflows.

Yes, you need environment variables to query OpenSearch trace data with PPL. The standard curl-based workflow against the /_plugins/_ppl API endpoint requires basic auth using the OPENSEARCH_ENDPOINT, OPENSEARCH_USER, and OPENSEARCH_PASSWORD environment variables.

Querying service-map data in otel-v2-apm-service-map-* helps OpenSearch trace debugging by revealing cross-service dependencies and performance bottlenecks. It complements span data to provide a complete view of distributed workflow reliability and latency.

GenAI workflow traces fail due to tool execution errors or latency spikes across distributed services. Pinpoint these failures by querying OpenSearch trace data with PPL to surface agent invocation spans, tool execution errors, and slow traces across OTEL indices.