Prometheus Indexing Backlog
Symptoms
Section titled “Symptoms”Prometheus is taking a long time to index new time series, and thus newly appearing series take a while to be queryable.
Possible checks
Section titled “Possible checks”Did the number of time series increase recently?
Graph prometheus_local_storage_memory_series
to see.
How did the indexing queue develop? See prometheus_local_storage_indexing_queue_length
.
Is it going down? Was it just one temporary spike of many new metrics?
In general, any kind of load can contribute to an indexing backlog, but most of the times, this is caused by trying to handle too many time series in one Prometheus server.
Resolution
Section titled “Resolution”Reduce the load on the Prometheus server, especially the number of handled or changing time series.