Grafana has released version 3.0 of its open-source time-series database Mimir. A new architecture and a new query engine are intended to provide acceleration. Due to the significant changes, users should carefully plan their upgrade to the new version.
Mimir is a fork of the distributed time-series database Cortex, originally developed by Grafana Labs. Cortex is under the permissive Apache 2.0 license and is part of the Cloud Native Computing Foundation (CNCF), whereas Mimir, a fork further developed by Grafana Labs since 2022, uses the AGPL-3.0 license.
Decoupled Architecture
The architecture in the new Mimir release pursues a stronger separation of read and write paths, w…
Grafana has released version 3.0 of its open-source time-series database Mimir. A new architecture and a new query engine are intended to provide acceleration. Due to the significant changes, users should carefully plan their upgrade to the new version.
Mimir is a fork of the distributed time-series database Cortex, originally developed by Grafana Labs. Cortex is under the permissive Apache 2.0 license and is part of the Cloud Native Computing Foundation (CNCF), whereas Mimir, a fork further developed by Grafana Labs since 2022, uses the AGPL-3.0 license.
Decoupled Architecture
The architecture in the new Mimir release pursues a stronger separation of read and write paths, which, according to the announcement, contributes to increased speed. Since the ingester was part of both read and write paths in previous Mimir versions, high query loads could affect ingestion performance. Mimir 3.0 now utilizes the distributed event streaming platform Apache Kafka and uses it as an asynchronous buffer between ingestion and query, allowing each path to scale independently.
A dedicated blog post describes the new architecture in detail.
Mimir Query Engine
Mimir is designed for the long-term storage of Prometheus and OpenTelemetry metrics. Previously, Mimir used Prometheus’s PromQL Engine for evaluating queries. However, according to Grafana Labs, this could lead to unpredictable memory usage as the engine processes samples in bulk.
This is now contrasted by the new Mimir Query Engine, which is used as the default in version 3.0. It is intended to deliver faster and more memory-efficient performance; it processes queries using the streaming principle and only loads the necessary samples at each step.
Due to the significantly changed architecture, Grafana Labs recommends, before updating to Mimir 3.0, to consult the upgrade guide and the release notes.
(mai)
Don’t miss any news – follow us on Facebook, LinkedIn or Mastodon.
This article was originally published in German. It was translated with technical assistance and editorially reviewed before publication.