Prometheus 2.18 dabbles with Jaeger while waiting for OpenTelemetry dust to settle


Prometheus – the cloud native ecosystem’s monitoring and alerting system of choice – is now available in version 2.18, marking the project’s first steps towards proper tracing support, although it’s more for internal purposes at this point.

The new release includes an experimental implementation of support for CNCF tracing project Jaeger, among other things. Discussions about adding ways to work with tracing backends to Prometheus have been running since 2018 at least, so it’s only fitting that version 2.18 of the platform finally addresses this. 

However, the new accession is only meant to solve some team needs, such as investigating query performance, for now, and isn’t intended to be part of the toolkit in the long run, which is also apparent by the lack of feature documentation. 

Instead, the team is currently looking into supporting OpenTelemetry, as this would make working with a wider variety of tools an option. Until this becomes a reality, it could however still take a while. After all, the OpenTracing/OpenCensus merger that marked the beginning of OpenTelemetry was only announced about a year ago, and the Prometheus team understandably would prefer for “the dust to settle” before integrating it into their project.

In addition, Prometheus has reduced the size of its write-ahead log, from circa 6 hours of WAL to around 3 hours. This is meant to “halve startup time for those who are I/O bound” but could also “increase the checkpoint site for those with certain churn patterns” though by less than what the change is saving. 

Federation was changed to use only local time series database TSDB and ignore remote reads. There’s now also an architecture meta label for EC2 instances and rule_evaluations_total and rule_evaluation_failures_total got a rule_group label.

The team also fixed some remote write bugs as well as some UI issues. More details are available via the project repository.

Prometheus was initially developed at SoundCloud in 2012. Four years later, the toolkit became the second project to be hosted by the Cloud Native Computing Foundation, following in container orchestrator Kubernetes’ footsteps. It continued that trajectory by graduating the foundation’s incubator in 2018, a good five month after K8s succeeded in doing so.