yonahd/krr

Prometheus-based Kubernetes Resource Recommendations

Robusta KRR

Prometheus-based Kubernetes Resource Recommendations
Usage docs »

Report Bug · Request Feature · Slack Channel

About The Project

Robusta KRR (Kubernetes Resource Recommender) is a CLI tool for optimizing resource allocation in Kubernetes clusters. It gathers pod usage data from Prometheus and recommends requests and limits for CPU and memory. This reduces costs and improves performance.

Features

• No Agent Required: Robusta KRR is a CLI tool that runs on your local machine. It does not require running Pods in your cluster.
• Prometheus Integration: Gather resource usage data using built-in Prometheus queries, with support for custom queries coming soon.
• Extensible Strategies: Easily create and use your own strategies for calculating resource recommendations.
• Future Support: Upcoming versions will support custom resources (e.g. GPUs) and custom metrics.

Resource Allocation Statistics

According to a recent Sysdig study, on average, Kubernetes clusters have:

• 69% unused CPU
• 18% unused memory

By right-sizing your containers with KRR, you can save an average of 69% on cloud costs.

How it works

Metrics Gathering

Robusta KRR uses the following Prometheus queries to gather usage data:

• CPU Usage:

  sum(node_namespace_pod_container:container_cpu_usage_seconds_total:sum_irate{namespace="{object.namespace}", pod="{pod}", container="{object.container}"})
  

• Memory Usage:

  sum(container_memory_working_set_bytes{job="kubelet", metrics_path="/metrics/cadvisor", image!="", namespace="{object.namespace}", pod="{pod}", container="{object.container}"})
  

Need to customize the metrics? Tell us and we'll add support.

Algorithm

By default, we use a simple strategy to calculate resource recommendations. It is calculated as follows (The exact numbers can be customized in CLI arguments):

• For CPU, we set a request at the 99th percentile with no limit. Meaning, in 99% of the cases, your CPU request will be sufficient. For the remaining 1%, we set no limit. This means your pod can burst and use any CPU available on the node - e.g. CPU that other pods requested but aren’t using right now.

• For memory, we take the maximum value over the past week and add a 5% buffer.

Prometheus connection

Find about how KRR tries to find the default prometheus to connect here.

Difference with Kubernetes VPA

Feature 🛠️	Robusta KRR 🚀	Kubernetes VPA 🌐
Resource Recommendations 💡	✅ CPU/Memory requests and limits	✅ CPU/Memory requests and limits
Installation Location 🌍	✅ Not required to be installed inside the cluster, can be used on your own device, connected to a cluster	❌ Must be installed inside the cluster
Workload Configuration 🔧	✅ No need to configure a VPA object for each workload	❌ Requires VPA object configuration for each workload
Immediate Results ⚡	✅ Gets results immediately (given Prometheus is running)	❌ Requires time to gather data and provide recommendations
Reporting 📊	✅ Detailed CLI Report, web UI in Robusta.dev	❌ Not supported
Extensibility 🔧	✅ Add your own strategies with few lines of Python	⚠️ Limited extensibility
Custom Metrics 📏	🔄 Support in future versions	❌ Not supported
Custom Resources 🎛️	🔄 Support in future versions (e.g., GPU)	❌ Not supported
Explainability 📖	🔄 Support in future versions (Robusta will send you additional graphs)	❌ Not supported
Autoscaling 🔀	🔄 Support in future versions	✅ Automatic application of recommendations

Robusta UI integration

If you are using Robusta SaaS, then KRR is integrated starting from v0.10.15. You can view all your recommendations (previous ones also), filter and sort them by either cluster, namespace or name.

More features (like seeing graphs, based on which recommendations were made) coming soon. Tell us what you need the most!

(back to top)

Getting Started

Installation

1. Make sure you have Python 3.9 (or greater) installed
2. Clone the repo:

git clone https://github.com/robusta-dev/krr

3. Navigate to the project root directory (cd ./krr)
4. Install requirements:

pip install -r requirements.txt

5. Run the tool:

python krr.py --help

(back to top)

Usage

Straightforward usage, to run the simple strategy:

python krr.py simple

If you want only specific namespaces (default and ingress-nginx):

python krr.py simple -n default -n ingress-nginx

By default krr will run in the current context. If you want to run it in a different context:

python krr.py simple -c my-cluster-1 -c my-cluster-2

If you want to get the output in JSON format (--logtostderr is required so no logs go to the result file):

python krr.py simple --logtostderr -f json > result.json

If you want to get the output in YAML format:

python krr.py simple --logtostderr -f yaml > result.yaml

If you want to see additional debug logs:

python krr.py simple -v

More specific information on Strategy Settings can be found using

python krr.py simple --help

(back to top)

Prometheus auto-discovery

By default, KRR will try to auto-discover the running Prometheus by scanning those labels:

"app=kube-prometheus-stack-prometheus"
"app=prometheus,component=server"
"app=prometheus-server"
"app=prometheus-operator-prometheus"
"app=prometheus-msteams"
"app=rancher-monitoring-prometheus"
"app=prometheus-prometheus"

If none of those labels result in finding Prometheus, you will get an error and will have to pass the working url explicitly (using the -p flag).

(back to top)

Example of using port-forward for Prometheus

If your prometheus is not auto-connecting, you can use kubectl port-forward for manually forwarding Prometheus.

For example, if you have a Prometheus Pod called kube-prometheus-st-prometheus-0, then run this command to port-forward it:

kubectl port-forward pod/kube-prometheus-st-prometheus-0 9090

Then, open another terminal and run krr in it, giving an explicit prometheus url:

python krr.py simple -p http://127.0.0.1:9090

(back to top)

Creating a Custom Strategy/Formatter

Look into the examples directory for examples on how to create a custom strategy/formatter.

(back to top)

Building

We are planning to use pyinstaller to build binaries for distribution. Right now you can build the binaries yourself, but we're not distributing them yet.

1. Install the project manually (see above)
2. Navigate to the project root directory
3. Install poetry (https://python-poetry.org/docs/#installing-with-the-official-installer)
4. Install requirements with dev dependencies:

poetry install --group dev

5. Build the binary:

poetry run pyinstaller krr.py

6. The binary will be located in the dist directory. Test that it works:

cd ./dist/krr
./krr --help

(back to top)

Testing

We use pytest to run tests.

1. Install the project manually (see above)
2. Navigate to the project root directory
3. Install poetry (https://python-poetry.org/docs/#installing-with-the-official-installer)
4. Install dev dependencies:

poetry install --group dev

5. Install robusta_krr as editable dependency:

pip install -e .

6. Run the tests:

poetry run pytest

(back to top)

Contributing

Contributions are what make the open source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

If you have a suggestion that would make this better, please fork the repo and create a pull request. You can also simply open an issue with the tag "enhancement". Don't forget to give the project a star! Thanks again!

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

(back to top)

License

Distributed under the MIT License. See LICENSE.txt for more information.

(back to top)

Contact

If you have any questions, feel free to contact support@robusta.dev

Project Link: https://github.com/robusta-dev/krr

(back to top)