Starting with the TSDB, there's been yet more performance improvements to the postings logic. Of particular note if you have a regex like
a|b (such as Grafana produces for multi-select variables), it's now evaluated as lookups rather than a regex scan so
x=~"a" is now as efficient as
x="a". For the WAL "unknown series references" should be fully fixed now, a new segment is started on every head compaction which is useful for reducing disk space on low-volume instances, and there's an opt-in
--storage.tsdb.wal-compression flag to compress the WAL.
In Prometheus proper there's performance improvements in PromQL, and remote write. You can now use globs for rules files for unittests, like you can in
prometheus.yml. There's also a configuration option to talk to the Alertmanager via its v2 API.
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.