OpenSearch Cluster (elasticsearch_exporter)

Server-side view of an OpenSearch or Elasticsearch cluster: cluster health, search and indexing throughput, JVM and breaker saturation, per-index size and query rate, storage, and a node table. Verified against OpenSearch 3.5 with exporter v1.11.0.

Requires prometheus-community/elasticsearch_exporter with --es.indices and --es.indices_settings. No panel reads a metric from the shards, snapshots or mappings collectors, so nothing goes blank if you leave those off to control cardinality.

Four things it gets right that are easy to get wrong:

• The elasticsearch_indices_* families label the node in name, while elasticsearch_index_stats_* label the index in index. Cluster-wide rates come from the first, per-index rates from the second. Dashboards that treat name as an index show node counts under index labels.
• elasticsearch_indices_aliases is surfaced as an alias-to-index table. If you reindex into name-v2 and swap the alias, this is the only place that change is visible, and it is what ties a shift in document count or latency to the swap that caused it.
• Circuit breakers are shown against their own limit rather than as raw bytes, so the number is comparable across breakers and cluster sizes.
• Filesystem headroom uses available, not free; free counts space reserved for root, which the engine cannot use.

Thread-pool panels carry an unfiltered total alongside the per-pool breakdown, so a quiet cluster reads as quiet rather than as a missing metric. The per-pool series are filtered to non-zero pools to keep the legend short — a six-node cluster exposes roughly 84.

Select your Prometheus from the Data source picker; Cluster, Exporter target and Node narrow the view.