pathcl/pyrra

Making SLOs with Prometheus manageable, accessible, and easy to use for everyone!

Features

Support for Kubernetes, Docker, and filesystem
Alerting: Generates 4 Multi Burn Rate Alerts with different severity
Page listing all Service Level Objectives
- All columns sortable
- Sorted by remaining error budget to see worst ones quickly
- Tool-tips when hovering for extra context
Page with details for a Service Level Objective
- Objective, Availability, Error Budget highlighted as 3 most important numbers
- Graph to see how the error budget develops over time
- Time range picker to change graphs
- Request, Errors, Duration (RED) graphs for the underlying service
- Multi Burn Rate Alerts overview table
Caching of Prometheus query results
Thanos: Disabling of partial responses and downsampling to 5m and 1h
OpenAPI generated API

Feedback & Support

If you have any feedback, please open a discussion in the GitHub Discussions of this project.
We would love to learn what you think!

Acknowledgements

@aditya-konarde, @brancz, @cbrgm, @codesome, @ekeih, @guusvw, @jzelinskie, @kakkoyun, @lilic, @markusressel, @morremeyer, @mxinden, @numbleroot, @paulfantom, @RiRa12621, @tboerger, and Maria Franke.

While we were working on Pyrra in private these amazing people helped us with a look of feedback and some even took an extra hour for a in-depth testing! Thank you all so much!

Additionally, @metalmatze would like to thank Polar Signals for allowing us to work on this project in his 20% time.

Demo

Check out our live demo on demo.pyrra.dev!

Feel free to give it a try there!

Installation

There are pre-build container images available:

docker pull ghcr.io/pyrra-dev/pyrra:v0.3.0

While running Pyrra on its own works there won't be any SLO configured nor will there be any data from a Prometheus to work with.

Therefore, you can find a docker-compose example in examples/docker-compose.
This stack comes with Pyrra and Prometheus pre-configured, as well as some SLOs.

Tech Stack

Client: TypeScript with React, Bootstrap, and uPlot.

Server: Go with libraries such as: chi, ristretto, xxhash, client-go.

OpenAPI generated API with server (Go) and clients (Go & TypeScript).

Run Locally

You need to have Go and Node installed.

Clone the project

git clone https://github.com/pyrra-dev/pyrra.git

Go to the project directory

cd pyrra

Install dependencies

make install

Build the UI and compile the Go binaries

make

Run the API and UI

Run the API binary in one terminal

./bin/api

Note: the API assumes a Prometheus is running on localhost:9090 and a backend on localhost:9444) by default. Check ./bin/api --help flag for the parameters to change those.

Run a Kubernetes or filesystem backend

Run the filesystem binary in another terminal

./bin/filesystem

Or run the Kubernetes binary in the other terminal

./bin/kubernetes

Note: This binary tries to run against your default Kubernetes context. Use the -kubeconfig flag to change for another kubeconfig

Running the UI standalone

Run the Node server to work on the UI itself

cd ui
npm run start

Note: This still needs the API and one of the backends to really work.

Most likely you need to update the window.PUBLIC_API constant in ui/public/index.html.

-    <script>window.PUBLIC_API = '/'</script>
+    <script>window.PUBLIC_API = 'http://localhost:9099/'</script>

Roadmap

Best to check the Projects board and if you cannot find what you're looking for feel free to open an issue!

Contributing

Contributions are always welcome!

See contributing.md for ways to get started.

Please adhere to this project's code of conduct.

Maintainers

Name	Area	GitHub	Twitter	Company
Nadine Vehling	UX/UI	@nadinevehling	@nadinevehling	Grafana Labs
Matthias Loibl	Engineering	@metalmatze	@metalmatze	Polar Signals

We are mostly maintaining Pyrra in our free time.

FAQ

Why not use Grafana in this particular use case?

Right now we could have used Grafana indeed. In upcoming releases, we plan to add more interactive features to give you better context when coming up with new SLOs. This is something we couldn't do with Grafana.

Do I still need Grafana?

Yes, Grafana is an amazing data visualization tool for Prometheus metrics. You can create your own custom dashboards and dive a lot deeper into each component while debugging.

Does it work with Thanos too?

Yes, in fact I've been developing this against my little Thanos cluster most of the time.
The queries even dynamically add headers for downsampling and disable partial responses.

How many instances should I deploy?

It depends on the topology of your infrastructure, however, we think that alerting should still happen within each individual Prometheus and therefore running one instance with one Prometheus (pair) makes the most sense. Pyrra itself only needs one instance per Prometheus (pair).

Why don't you support more complex SLOs?

For now, we try to accomplish an easy-to-setup workflow for the most common SLOs. It is still possible to write these more complex SLOs manually and deploy them to Prometheus along those generated. You can base more complex SLOs on the output of one SLO from this tool.

Why is the objective target a string not a float?

Kubebuilder doesn't support floats in CRDs...
Therefore, we need to pass it as string and internally convert it from string to float64.

Here are some related projects:

pathcl / pyrra