Pebble is the default entrypoint for all ROCKs!

Similar to other well-known process managers such as supervisord, runit, or s6, Pebble is a service manager that enables the seamless orchestration of a collection of local service processes as an organised set. The main difference is that Pebble has been designed with custom-tailored features that significantly enhance the overall container experience, making it the ideal candidate for the container’s init process (also known as the entrypoint, with PID=1).

Multiple processes in a container?

Containers’ best practices advocate the separation of concerns and the adoption of a single service per container. With the introduction of Pebble as the ROCKs’ entrypoint, this principle is elevated to new heights:

if multiple processes rely on shared dependencies and are tightly coupled together (i.e. they serve a single purpose and cannot be executed independently), then the best practice entails orchestrating them within the same container, with Pebble as their manager.

This new notion addresses existing pain points arising from the excessive separation of concerns, which results in numerous container images whose entrypoints lack the ability to gracefully handle the underlying child processes. This is one of the main reasons behind the gradual shift in the best practices, as there is an increasing emphasis on adopting init processes such as tini, s6-overlay, or Pebble.

What to expect?

Pebble distinguishes itself from other similar tools (like tini and s6-overlay) by offering the following core features:

  • client-server model behind a single binary: Pebble is injected into ROCKs as a single binary which acts both as a daemon and a client to itself;

  • declarative service definition: the Pebble service processes (or simply Pebble services) are declaratively defined in YAML files called layers. Compared to imperative wrapper scripts (as suggested in the Docker documentation), this provides a much cleaner and less error-prone way to define the processes that should run inside the container.

  • services as first-class citizens: unlike wrapper scripts, Pebble treats services as manageable units with a defined lifecycle and service-specific definitions for health monitoring, inter-service dependencies, restart policies, and much more;

  • layering: Pebble can stack multiple layers (represented as YAML files) into a single Pebble plan where all services are defined. With this layering mechanism, existing services can be overridden or re-configured;

  • container-optimised init process: as a ROCK’s PID 1, Pebble acts as an init process and thus offers:

    • support for multiple child processes,

    • reaping and subreaping,

    • signal forwarding,

    • graceful shutdown,

    • log rotation,

    • run the Pebble daemon and client commands in a single operation;

  • consistent user experience: since every ROCK has Pebble as its entrypoint, a predictable and consistent user experience is guaranteed;

  • embedded utilities: regardless of the ROCK’s contents, Pebble offers a comprehensive suite of commands for inspecting and interacting with the container. These commands are especially useful for Chiselled ROCKs, as they encompass functionalities such as listing and deleting files, creating directories, and inspecting Pebble services, among others.

Creating services

Rockcraft follows the Pebble layer specification to the letter, with Pebble services defined in rockcraft.yaml. How to convert a popular entrypoint to a Pebble layer provides an example of how to convert a Docker entrypoint to a Pebble layer.