Prometheus 2.17.0 is now out, following on from 2.16.0 with many fixes and improvements.

The big change in this release is that isolation has been added to the tsdb. This means that queries will no longer see partial scrapes, avoiding a number of race conditions such as with histograms. This does use a bit more memory and CPU, but it is worth it not to produce incorrect results. Remote read to multiple endpoints is now done concurrently.

There were changes to the PromQL parser under the covers. The only user-visible aspect of that should be that certain keywords should now also work as bare metric names - but on the other hand I doubt anyone had a metric called count_values.

Consul, EC2, OpenStack, and Kubernetes service discovery gained some new metadata. Kubernetes SD also gained the ability to pre-filter targets where there was such a massive number that relabelling alone isn't practical.

There’s also a number of improvements, bug fixes and other changes, see the release notes for more detail.


Wondering how to upgrade? Contact us.