View Source Membrane.Pipeline behaviour (Membrane Core v1.1.1)

A behaviour module for implementing pipelines.

Membrane.Pipeline contains the callbacks and functions for constructing and supervising pipelines. Pipelines facilitate the convenient instantiation, linking, and management of elements and bins.
Linking pipeline children together enables them to pass and process data.

To create a pipeline, use use Membrane.Pipeline and implement callbacks of Membrane.Pipeline behaviour. See Membrane.ChildrenSpec for details on instantiating and linking children.

Starting and supervision

Start a pipeline with start_link/2 or start/2. Pipelines always spawn under a dedicated supervisor, so in the case of success, either function will return {:ok, supervisor_pid, pipeline_pid} .

The supervisor never restarts the pipeline, but it does ensure that the pipeline and its children terminate properly. If the pipeline needs to be restarted, it should be spawned under a different supervisor with the appropriate strategy.

Starting under a supervision tree

A pipeline can be spawned under a supervision tree like any other GenServer.
use Membrane.Pipeline injects a child_spec/1 function. A simple scenario could look like this:

defmodule MyPipeline do
  use Membrane.Pipeline

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

  # ...
end

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

Starting outside of a supervision tree

When starting a pipeline outside a supervision tree, use the pipeline_pid pid to interact with the pipeline. A simple scenario could look like this:

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

Visualizing the supervision tree

Use the Applications tab in Erlang's Observer GUI (or the Kino library in Livebook) to visualize a pipeline's internal supervision tree. Use the following configuration for debugging purposes only:

  config :membrane_core, unsafely_name_processes_for_stalker: [:components]

This improves the readability of the Observer's process tree graph by naming the pipeline descendants, as demonstrated here:

Observer graph.

Summary

Types

Defines return values from Pipeline callback functions.

List of configurations used by start/3 and start_link/3.

Defines configuration value used by the start/3 and start_link/3.

The Pipeline name

Defines the return value of the start/3 and start_link/3."

Defines options passed to the start/3 and start_link/3 and subsequently received in the handle_init/2 callback.

The pipeline state.

Callbacks

Callback invoked when the pipeline is called using a synchronous call.

Callback invoked when a notification comes in from a child.

Callback invoked when a child removes its pad.

Callback invoked when a child enters playing playback.

Callback invoked when a child completes its setup.

Callback invoked when a crash group crashes.

Callback invoked when a child element finishes processing a stream via the given pad.

Callback invoked when a child element starts processing a stream via the given pad.

Callback invoked when the pipeline receives a message that is not recognized as an internal Membrane message.

Callback invoked on initialization of the pipeline.

Callback invoked when the pipeline switches the playback to :playing. By default, it does nothing.

Callback invoked on pipeline startup, right after handle_init/2.

This callback is deprecated since v1.1.0.

Callback invoked when the pipeline is requested to terminate with terminate/2. By default, it returns Membrane.Pipeline.Action.terminate/0 with reason :normal.

Callback invoked upon each timer tick. A timer can be started with Membrane.Pipeline.Action.start_timer action.

Functions

Brings all the stuff necessary to implement a pipeline.

Calls the pipeline with a message.

Returns list of pipeline PIDs currently running on the current node.

Returns list of pipeline PIDs currently running on the passed node.
Compare to list_pipelines/0.

Checks whether the module is a pipeline.

Starts the pipeline outside a supervision tree. Compare to start_link/3.

Starts the pipeline based on the given module and links it to the current process.

Terminates the pipeline.

Types

@type callback_return() :: {[Membrane.Pipeline.Action.t()], state()}

Defines return values from Pipeline callback functions.

Return values

  • {[action], state} - Returns a list of actions that will be performed within the pipeline, e.g., starting new children, sending messages to specific children, etc. Actions are tuples of {type, arguments}, so they can be expressed as a keyword list. See Membrane.Pipeline.Action for more info.
@type config() :: [config_entry()]

List of configurations used by start/3 and start_link/3.

@type config_entry() :: {:name, name()}

Defines configuration value used by the start/3 and start_link/3.

@type name() :: GenServer.name()

The Pipeline name

@type on_start() ::
  {:ok, supervisor_pid :: pid(), pipeline_pid :: pid()}
  | {:error, {:already_started, pid()} | term()}

Defines the return value of the start/3 and start_link/3."

@type pipeline_options() :: any()

Defines options passed to the start/3 and start_link/3 and subsequently received in the handle_init/2 callback.

@type state() :: any()

The pipeline state.

Callbacks

Link to this callback

handle_call(message, context, state)

View Source (optional)
@callback handle_call(
  message :: any(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) ::
  {[
     Membrane.Pipeline.Action.common_actions()
     | Membrane.Pipeline.Action.reply()
   ], state()}

Callback invoked when the pipeline is called using a synchronous call.

Context passed to this callback contains an additional field :from. By default, it does nothing.

Link to this callback

handle_child_notification(notification, element, context, state)

View Source (optional)
@callback handle_child_notification(
  notification :: Membrane.ChildNotification.t(),
  element :: Membrane.Child.name(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when a notification comes in from a child.

By default, it ignores the notification.

Link to this callback

handle_child_pad_removed(child, pad, context, state)

View Source (optional)
@callback handle_child_pad_removed(
  child :: Membrane.Child.name(),
  pad :: Membrane.Pad.ref(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state :: state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when a child removes its pad.

The callback won't be invoked, when you have initiated the pad removal, eg. when you have returned t:Membrane.Pipeline.Action.remove_link() action which made one of your children's pads be removed. By default, it does nothing.

Link to this callback

handle_child_playing(child, context, state)

View Source (optional)
@callback handle_child_playing(
  child :: Membrane.Child.name(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when a child enters playing playback.

By default, it does nothing.

Link to this callback

handle_child_setup_completed(child, context, state)

View Source (optional)
@callback handle_child_setup_completed(
  child :: Membrane.Child.name(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when a child completes its setup.

By default, it does nothing.

Link to this callback

handle_crash_group_down(group_name, context, state)

View Source (optional)
@callback handle_crash_group_down(
  group_name :: Membrane.Child.group(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when a crash group crashes.

Context passed to this callback contains 2 additional fields: :members and :crash_initiator. By default, it does nothing.

Link to this callback

handle_element_end_of_stream(child, pad, context, state)

View Source (optional)
@callback handle_element_end_of_stream(
  child :: Membrane.Child.name(),
  pad :: Membrane.Pad.ref(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when a child element finishes processing a stream via the given pad.

By default, it does nothing.

Link to this callback

handle_element_start_of_stream(child, pad, context, state)

View Source (optional)
@callback handle_element_start_of_stream(
  child :: Membrane.Child.name(),
  pad :: Membrane.Pad.ref(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when a child element starts processing a stream via the given pad.

By default, it does nothing.

Link to this callback

handle_info(message, context, state)

View Source (optional)
@callback handle_info(
  message :: any(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when the pipeline receives a message that is not recognized as an internal Membrane message.

Useful for receiving data sent from NIFs or other external sources. By default, it logs and ignores the received message.

Link to this callback

handle_init(context, options)

View Source (optional)
@callback handle_init(
  context :: Membrane.Pipeline.CallbackContext.t(),
  options :: pipeline_options()
) ::
  {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked on initialization of the pipeline.

This callback is synchronous: the process that started the pipeline waits until handle_init finishes, so it's important to do any long-lasting or complex work in handle_setup/2. handle_init should be used for things, like parsing options, initializing state, or spawning children. By default, handle_init converts opts to a map if they're a struct and sets them as the pipeline state.

Link to this callback

handle_playing(context, state)

View Source (optional)
@callback handle_playing(
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when the pipeline switches the playback to :playing. By default, it does nothing.

Link to this callback

handle_setup(context, state)

View Source (optional)
@callback handle_setup(
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked on pipeline startup, right after handle_init/2.

Any long-lasting or complex initialization should happen here. By default, it does nothing.

Link to this callback

handle_spec_started(children, context, state)

View Source (optional)
@callback handle_spec_started(
  children :: [Membrane.Child.name()],
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

This callback is deprecated since v1.1.0.

Callback invoked when children of Membrane.ChildrenSpec are started.

It is invoked, only if pipeline module contains its definition. Otherwise, nothing happens.

Link to this callback

handle_terminate_request(context, state)

View Source (optional)
@callback handle_terminate_request(
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) ::
  {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked when the pipeline is requested to terminate with terminate/2. By default, it returns Membrane.Pipeline.Action.terminate/0 with reason :normal.

Link to this callback

handle_tick(timer_id, context, state)

View Source (optional)
@callback handle_tick(
  timer_id :: any(),
  context :: Membrane.Pipeline.CallbackContext.t(),
  state()
) :: {[Membrane.Pipeline.Action.common_actions()], state()}

Callback invoked upon each timer tick. A timer can be started with Membrane.Pipeline.Action.start_timer action.

Functions

Link to this macro

__using__(options)

View Source (macro)

Brings all the stuff necessary to implement a pipeline.

Options:

Link to this function

call(pipeline, message, timeout \\ 5000)

View Source
@spec call(pid(), any(), timeout()) :: term()

Calls the pipeline with a message.

Returns the result of the pipeline call.

@spec list_pipelines() :: [pid()]

Returns list of pipeline PIDs currently running on the current node.

Use for debugging only.

@spec list_pipelines(node()) :: [pid()]

Returns list of pipeline PIDs currently running on the passed node.
Compare to list_pipelines/0.

@spec pipeline?(module()) :: boolean()

Checks whether the module is a pipeline.

Link to this function

start(module, pipeline_options \\ nil, process_options \\ [])

View Source
@spec start(module(), pipeline_options(), config()) :: on_start()

Starts the pipeline outside a supervision tree. Compare to start_link/3.

Link to this function

start_link(module, pipeline_options \\ nil, process_options \\ [])

View Source
@spec start_link(module(), pipeline_options(), config()) :: on_start()

Starts the pipeline based on the given module and links it to the current process.

Pipeline options are passed to the handle_init/2 callback. Note that this function returns {:ok, supervisor_pid, pipeline_pid} in case of success. Check the 'Starting and supervision' section of the moduledoc for details.

Link to this function

terminate(pipeline, opts \\ [])

View Source
@spec terminate(pipeline :: pid(),
  timeout: timeout(),
  force?: boolean(),
  asynchronous?: boolean()
) ::
  :ok | {:ok, pid()} | {:error, :timeout}

Terminates the pipeline.

Accepts three options:

  • asynchronous? - if set to true, pipeline termination won't be blocking and will be executed in the process whose pid is returned as a function result. If set to false, pipeline termination will be blocking and will be executed in the process that called this function. Defaults to false.
  • timeout - specifies how much time (ms) to wait for the pipeline to gracefully terminate. Defaults to 5000.
  • force? - determines how to handle a pipeline still alive after timeout. If set to true, Process.exit/2 kills the pipeline with reason :kill and returns {:error, :timeout}. If set to false, it raises an error. Defaults to false.

Returns:

  • {:ok, pid} - option asynchronous?: true was passed.
  • :ok - pipeline gracefully terminated within timeout.
  • {:error, :timeout} - pipeline was killed after timeout.