Module containing functions for constructing and supervising pipelines.

Pipelines are units that make it possible to instantiate, link and manage elements and bins in convenient way (actually they should always be used inside a pipeline). Linking pipeline children together enables them to pass data to one another, and process it in different ways.

To create a pipeline, use the __using__/1 macro and implement callbacks of Membrane.Pipeline behaviour. For details on instantiating and linking children, see Membrane.ChildrenSpec.

starting-and-supervision

Starting and supervision

Pipeline can be started with start_link/2 or start/2 functions. They both return {:ok, supervisor_pid, pipeline_pid} in case of success, because the pipeline is always spawned under a dedicated supervisor. The supervisor never restarts the pipeline, but it makes sure that the pipeline and its children terminate properly. If the pipeline needs to be restarted in case of failure, it should be spawned under another supervisor with a proper strategy.

starting-under-a-supervision-tree

Starting under a supervision tree

The pipeline can be spawned under a supervision tree like any GenServer. Also, __using__/1 macro injects a child_spec/1 function. A simple scenario can look like:

defmodule MyPipeline do
  use Membrane.Pipeline

  def start_link(options) do
    Membrane.Pipeline.start_link(options, name: MyPipeline)
  end

  # ...
end

Supervisor.start_link([{MyPipeline, option: :value}], strategy: :one_for_one)
send(MyPipeline, :message)

starting-outside-of-a-supervision-tree

Starting outside of a supervision tree

When starting a pipeline outside a supervision tree and willing to interact with the pipeline by pid, pipeline_pid returned from start_link can be used, for example

{:ok, _supervisor_pid, pipeline_pid} = Membrane.Pipeline.start_link(MyPipeline, option: :value)
send(pipeline_pid, :message)

visualizing-pipeline-s-supervision-tree

Visualizing pipeline's supervision tree

Pipeline's internal supervision tree can be looked up with Applications tab of Erlang's Observer or with Livebook's Kino library. For debugging (and ONLY for debugging) purposes, you may use the following configuration:

  config :membrane_core, unsafely_name_processes_for_observer: [:components]

that makes the observer's process tree graph more readable by naming pipeline's descendants, for example: .