lexmachinainc / munin-plugin-elasticsearch Goto Github PK

elasticsearch_gc_time currently tracks the average total gc time for the young and old generations. Another useful metric may be the average individual gc time for these which is the current metric divided by collection_count.

Currently cache quality tracking just shows the size in bytes of:

• fielddata "field_size"
• query_cache "filter_size"
• request_cache "request_cache_size"

Add related metrics which tell about the effectiveness of the caches via normalized metrics like:

• "hit rate" hits/total
• "miss rate" misses/total
• hit rate / miss rate
• eviction counts and/or rates

These would need to be scaled differently from measurements of bytes currently shown - maybe just another panel, i.e., elasticsearch_cache_rates.

Panels don't clearly tell whether a metric is specific to a node or is summed across all nodes to represent the cluster as a whole: sometimes both make sense (though putting all on one, the scales would often be blown out for larger clusters).

Additionally, cluster-level rollups can be harvested directly vs. manually as is done now.

Examples:

• Elasticsearch documents - node or cluster?
• Elasticsearch cache - just switched this from cluster to node
• Elasticsearch indexes - currently node
• Elasticsearch index operations - currently node

(Elasticsearch cluster shards is cluster-level)