How to Choose the Best Long-term Monitoring Option for Your AKS Infrastructure: A Practical Guide to Thanos and Mimir


Greetings, fellow experts.

We are in the process of scaling up our infrastructure to accommodate up to 500 AKS clusters. As part of this endeavor, we are exploring long-term monitoring options that can provide us with consistent visibility and reliability across our entire system. We are currently comparing two solutions that extend Prometheus-based monitoring: Thanos and Mimir.

Both solutions appear to have strong features for scaling up Prometheus monitoring, such as long-term storage and high metric throughput. However, we want to base our decision not only on the technical aspects but also on the practical experiences and outcomes of using these solutions.


: A simple and cost-effective solution that integrates with existing Prometheus setups and uses object storage for long-term storage. It also offers a global query view and deduplication across clusters.


: A newer and more scalable solution that optimizes Prometheus for large-scale environments. It has a more efficient architecture and query engine for handling massive amounts of data.

We are seeking feedback from the community on the following criteria:


: How do Thanos and Mimir cope with hundreds of clusters? We are concerned about both the operational complexity and the resource consumption.


: We would appreciate any insights on the dependability of these systems in large-scale settings. How do they deal with failures, and how easy is it to recover from them?


: How do these solutions compare in terms of query speed and data access times? We are especially interested in the real-time monitoring capabilities of these solutions.


: Although cost is not our main priority, we would like to know the long-term cost implications of choosing either solution, taking into account the storage and compute resources required.

Ease of Use and Integration

: How easy is it to integrate these solutions with AKS and to use them on a daily basis?

If you have used Thanos or Mimir (or both) in a similar scale of operation, we would love to hear from you. Please share your experiences, challenges, lessons learned, and any advice you might have.

Thank you in advance for your valuable input.


I have used both Thanos and Mimir in a similar scale of operation, and I would like to share my perspective on these two solutions. I hope this will help you make an informed decision for your infrastructure.

: I started using Thanos when I needed a simple and cost-effective way to extend my existing Prometheus setup with long-term storage and global query capabilities. I was impressed by how easy it was to integrate Thanos with Prometheus using the sidecar component, and how I could leverage object storage solutions like S3 or GCS for storing historical data. I also liked the deduplication feature that allowed me to query data from multiple Prometheus instances without worrying about duplicate samples. Thanos also provided a query-frontend component that improved the query performance and caching.

However, as I scaled up my clusters, I encountered some challenges with Thanos. One of them was the high resource consumption of the store component, which had to load and index large amounts of data from the object storage. Another one was the complexity of the configuration and the dependency on external components like etcd or Consul for service discovery and coordination. I also found that Thanos was not very resilient to failures, and sometimes I had to manually delete corrupted blocks or restart components to recover from errors.


: I switched to Mimir when I heard about its high scalability and performance optimizations for large-scale environments. Mimir is a newer solution that also extends Prometheus, but with a more efficient architecture and query engine. Mimir uses a microservices-based approach, where each component can be scaled independently and horizontally. Mimir also has a monolithic mode, where all components run in the same process, which is useful for smaller deployments or testing purposes.

Mimir has several advantages over Thanos in terms of scalability, reliability, and performance. Mimir uses a sharded and replicated design for both ingesting and querying data, which reduces the load and the blast radius of each instance. Mimir also has a built-in service discovery and coordination mechanism, which simplifies the deployment and the configuration. Mimir is more reliable than Thanos, as it can handle failures gracefully and automatically repair corrupted blocks. Mimir is also faster than Thanos, as it has a more optimized query engine and a smarter caching strategy.

The main drawback of Mimir is that it is not as simple and cost-effective as Thanos. Mimir requires more compute resources and more complex infrastructure than Thanos, as it relies on a distributed database like Cassandra or DynamoDB for storing data. Mimir also has a steeper learning curve than Thanos, as it has more components and configuration options to understand and tune. Mimir is also less mature than Thanos, as it is a newer solution that may have more bugs and less documentation.


: The cost of using either solution depends on several factors, such as the amount of data, the retention period, the query frequency, and the cloud provider. In general, Thanos is cheaper than Mimir, as it uses object storage for long-term storage, which is usually cheaper than distributed databases. However, Thanos may incur more network costs than Mimir, as it has to transfer more data from the object storage to the store component. Mimir may also be more cost-efficient than Thanos in terms of compute resources, as it can handle more queries with less instances.

Ease of Use and Integration

: Both solutions are easy to integrate with AKS, as they support Kubernetes natively and have Helm charts available. However, Thanos is easier to use than Mimir, as it has fewer components and configuration options, and it integrates seamlessly with existing Prometheus setups. Mimir requires more effort to set up and maintain, as it has more dependencies and complexity, and it may require some changes to the Prometheus configuration.


: In summary, Thanos and Mimir are both powerful solutions for scaling Prometheus-based monitoring, but they have different trade-offs and use cases. Thanos is a simple and cost-effective solution that works well for smaller or medium-scale deployments, or for deployments that need long-term storage and global query capabilities. Mimir is a more scalable and performant solution that works well for larger or high-demand deployments, or for deployments that need high reliability and efficiency at scale.

I hope this article was helpful for you. If you have any questions or feedback, please feel free to comment below. Thank you for reading.

Leave a Reply

Your email address will not be published. Required fields are marked *

Privacy Terms Contacts About Us