← Back to guidesGuide

Diagnose OpenSearch vector-search hot shards before RAG latency spikes

A practical runbook to separate real shard pressure, skewed routing, and slow search symptoms without blaming the LLM by reflex.

Learn how to detect hot shards in OpenSearch vector indices, prove whether they affect RAG retrieval, and apply safe mitigations with Dev Tools, metrics, and canaries.

Created: July 6, 2026

Published: July 6, 2026

Estimated time45 min

LevelAdvanced

Before you startRead access to OpenSearch Dev Tools or equivalent API permissions.

PlatformsLinux / Docker

WhatsApp X LinkedIn

Linux

Diagnose from a shell with curl, jq, and safe OpenSearch endpoints.

curljqread-only OpenSearch credentials

Check shards and size by node

curl -s "$OS/_cat/shards/catalog-vectors-prod?v&s=store:desc"

Inspect search queue and rejections

curl -s "$OS/_cat/thread_pool/search?v&h=node_name,active,queue,rejected,completed"

Explain allocation for the suspicious shard

curl -s -XPOST "$OS/_cluster/allocation/explain" -H 'content-type: application/json' -d '{"index":"catalog-vectors-prod","shard":3,"primary":true}' | jq

Also available inDiagnosticar hot shards de búsqueda vectorial en OpenSearch antes de disparar la latencia RAG

Content locked

This guide requires both steps before full content is available.

○Click “Like” on this guide.
○Share on WhatsApp, X, LinkedIn, or copy the link.

Access is automatically unlocked as soon as both steps are completed.