The first big change is that
for state in alerts is now persisted across restarts, so brief restarts will no longer require your pending alerts to start again from scratch. There's also now throttling in sending alerts to the Alertmanager, so rather than send existing alerts on every single evaluation resends are now only sent at most once a minute which reduces load on the Alertmanager.
There's a new WAL implementation, which means that downgrades from 2.4.0 aren't possible. There's new APIs to access rules, alerts, and metrics metadata from targets.
There's been several improvements in service discovery. If there are identical SD configurations, they will only be instantiated once rather than for each instance of them - more efficient and useful for integrations that tend to run into throttling. There's also new metadata fields for Consul, EC2, and GCE, and Azure gained VMSS support.
There’s also a number of improvements, bug fixes and other changes, see the release notes for more detail.
Wondering when and how to upgrade to Prometheus 2.x? Contact us.