Technical stack


Technology choices are done component by component, trying to use the right tool for the different needs following these criteria:

  • Focus on reliability, as Rudder is used on critical production infrastructures, this means:

    • For components we develop, choosing languages and tools allowing us to maintain, add features and refactor with confidence. In this regard, we prefer statically typed languages, with a focus on safety. Our general choices are Scala for backend works, Rust for system services and Elm for Web frontend.

    • Limit script languages usage to integration/glue code. They are hard to maintain over time and hard to deploy in heterogeneous environments.

    • Avoid NIH: use the existing tools that does the job well, especially for things out of our core business (web server, data storage, etc.). Use standard system components whenever possible, especially for the agent.

    • For components we integrate (programs or libraries), this means choosing a mature and reliable option, even more for the agent.

  • Agent and relay components need to be able to run on very light systems with low network requirements, allowing embedded use cases

  • Stay as simple as possible, particularly on the operating side. Users should not have to spend time managing the management software itself. We try to avoid technologies that are difficult to operate or require specific knowledge. This also means limiting the number of technologies in one given component.

  • Know and contribute to our technical environment. We are regular contributors to several open-source ecosystems used in Rudder.


Engine and user interface

The main component, the core of Rudder, is the application written is Scala. It manages the configuration to apply to the nodes, generates the policies and handles nodes reports and compliance computation, among other things. It’s also the interface with the users (through the Web UI or the REST API).


We have chosen Scala since Rudder’s early days, for its robustness and performance, as well as its wide ecosystem of libraries (thanks to its compatibility with Java).

The Web frontend used AngularJS historically, and we are progressively moving to Elm which has static typing and focuses on reliability.


The second server service is the relay, which handles communication with the managed machines. It can be installed without the webapp part (and then be used as a kind of smart proxy).


It is written in Rust (using the tokio asynchronous runtime) to allow great performance even on very small hardware while providing security, safety and maintainability.

We have a second Rust program in the making, that will be our configuration policies compiler.


The HTTP server used on the Rudder servers is Apache httpd, as it is widely available and mature. It does TLS and reverse-proxying for the webapp and the relay, and applies a first level of security filters.

We use python on the server for integration plugins, allowing easy discovery for contributors and using official python APIs for some of the integrated tools.


There are three types of data managed by the server:

  • All the configuration data (configuration policies, groups, etc.) are stored in a git repository. It provides traceability over the configuration history and interoperability with external SCM.

  • The data generated from the configuration (reports, etc.) is stored in PostgreSQL, which is mature and reliable, but also provide modern features (JSON support, etc.).

  • The hierarchical information about nodes (inventories, etc.) are stored in an embedded OpenLDAP server.


We use a common policies representation (soon a fully-fledged language, rudder-lang, with its own compiler), that is used on the server to generate configuration code for different backends depending on the target platform. These configuration backends are:

  • On Windows, DSC, the native management technology, plus Powershell for agent cli

  • On Unix systems, CFEngine. It is a fast and powerful low-level portable configuration engine. Various scripts and the cli are implemented in POSIX shell for maximal portability and limited dependencies.

All our agents embed FusionInventory as inventory tool, allowing precise collection of software and hardware information, and use curl and openssl to communicate with the server in HTTPS.

Curl logo

← Architecture Rudder components →