The release of Mimir 3.0 marks a major milestone for Grafana Labs. This version introduces a redesigned architecture that significantly improves performance, reliability and operational efficiency. As a result, organizations dealing with massive volumes of metrics now gain a more predictable and scalable backend for their observability stacks.
How Mimir 3.0 transforms ingestion and querying
Earlier versions of Mimir relied on a single component to manage both reads and writes. However, this approach caused contention under heavy load. To solve this, Mimir 3.0 separates ingestion and querying into two independent paths. Moreover, Grafana added Apache Kafka as an asynchronous buffer between them.
Because of this decoupling, ingestion no longer slows down during complex queries. Similarly, queries remain stable even when write traffic spikes. Consequently, the system handles unpredictable workloads with much greater resilience.
Why the new decoupled model in Mimir 3.0 matters
The redesigned flow adds what Grafana calls an “ingest storage” layer. This improvement prevents query surges from affecting write operations and vice versa. Furthermore, internal testing showed that the risk of outages in the read path dropped substantially. In earlier versions, random ingester failures often cascaded into broader system issues. In contrast, Mimir 3.0 isolates these failures and contains them early.
This architectural shift reflects lessons learned from real production clusters, including extremely large environments such as CERN.
Mimir 3.0 adopts MQE as the default query engine
Another major upgrade in Mimir 3.0 is the transition to the Mimir Query Engine (MQE). Previously introduced in version 2.17, MQE replaces the traditional Prometheus query model. Instead of loading data in bulk, MQE processes samples in a streaming fashion. As a result, memory usage becomes far more predictable.
Additionally, Grafana reports up to a 92% reduction in peak memory consumption during queries. This allows large clusters to remain stable under pressure. Furthermore, the engine stays fully PromQL compatible, meaning users do not need to change their existing dashboards or queries.
Performance improvements across large deployments
The benefits of Mimir 3.0 extend beyond the query path. Grafana observed that large clusters now use up to 15% fewer resources. Meanwhile, they also deliver faster responses and improved reliability. These gains come from the combination of the decoupled architecture and the streaming query engine working together.
Consequently, the new release is far more cost-effective for organizations processing billions of active time series.
Mimir 3.0 focuses on reliability, speed and cost efficiency
Grafana shaped the design of this version around three strategic goals:
- Higher reliability through clear separation of concerns
- Better performance via streaming query execution
- Lower costs thanks to more efficient resource usage
In addition, these improvements stem directly from real-world usage patterns observed in long-running Mimir clusters.
How to approach the upgrade to Mimir 3.0
Because Mimir 3.0 introduces deep architectural changes, Grafana recommends a cautious, staged upgrade. First, teams should deploy a second Mimir cluster next to the existing one. Then, they must configure write clients to send data to both clusters. Finally, they can redirect read clients to the new environment.
Moreover, organizations will need to update Helm or Jsonnet configurations for both clusters. This process ensures a smoother migration without risking downtime.
Mimir 3.0 availability for cloud and self-hosted environments
The updated version is already integrated into Grafana Cloud Metrics. For self-hosted users, the documentation includes detailed upgrade guides and release notes. These resources help teams transition to Mimir 3.0 more confidently and with fewer risks.
Alternatives to consider alongside Mimir 3.0
Organizations exploring time series storage have several strong alternatives. For example, Prometheus remains a popular single-node option with robust ecosystem support. InfluxDB offers high performance for real-time analytics and IoT workloads. Meanwhile, TimescaleDB appeals to teams that prefer SQL and PostgreSQL compatibility.
Furthermore, cloud-native choices such as Amazon Timestream or Google Cloud Monitoring remove operational overhead. On the other hand, Thanos extends Prometheus with long-term storage and global querying. Each option comes with trade-offs in scalability, performance and maintenance complexity.
Conclusion
With Mimir 3.0, Grafana Labs delivers a more resilient and scalable foundation for metrics storage. The decoupled architecture, combined with a streaming query engine, significantly improves performance while reducing operational costs. As observability workloads continue to grow, Mimir offers organizations a faster, more reliable and more efficient path forward.
Read also
Join the discussion in our Facebook community.