Version 2.32 of the Prometheus project managed by the Cloud Native Computing Foundation (CNCF) has been released. In addition to some bug fixes and improvements, fundamental innovations have also been incorporated into the update of the monitoring system – including an operating mode designed for edge use cases, in which Prometheus does not occupy any local memory blocks, but cannot be queried locally either.
Without a local database in Edge or IoT scenarios
The basis is the agent mode integrated in the Prometheus binaries, which deals with
--enable-feature=agent can be activated and enables easier horizontal scalability for the inclusion of metric data. The monitoring software changes to an operating mode that is optimized for Edge and IoT application scenarios with limited or less powerful resources, in which writing is only carried out remotely (remote write). The usual pull model for retrieving metrics by a Prometheus server via the corresponding endpoints is still used.
While the specification of the Prometheus Remote Write API is officially still in the review stage, large parts of the surrounding ecosystem have already adapted the remote write protocol as the standard for the export of metrics – Cortex, Thanos, OpenTelemetry and Amazon’s cloud services, Google, Grafana or Logz.io are already using it.
Closer integration of Prometheus and Grafana
The Grafana Agent introduced by Grafana in 2020 laid the foundations for the agent mode that is now integrated in Prometheus 2.32. The Grafana team had to a variant of the Prometheus code that has been reduced for use in the cloud has been adaptedwhich waived local storage resources and in particular all functions associated with the TSDB – including queries, recording rules and warnings. In the medium term, the code for both agents should be merged in a common basis and also be available to other projects such as Thanos.
In the current release, the Prometheus team has improved the local database for time series (Time Series Data Base – TSDB) in addition to some bug fixes. Under Windows, the database can now also be used on systems with ARM64 architecture. To optimize queries, unnecessary sorting processes within the TSDB will no longer be necessary.