defmodule Membrane.Pipeline do
@moduledoc """
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
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
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(__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 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
Pipeline's internal supervision tree can be looked up with Applications tab of Erlang's Stalker
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_stalker: [:components]
that makes the stalker's process tree graph more readable by naming pipeline's descendants, for example:
![Stalker graph](assets/images/stalker_graph.png).
"""
use Bunch
alias __MODULE__.{Action, CallbackContext}
alias Membrane.{Child, Pad, PipelineError}
require Membrane.Logger
require Membrane.Core.Message, as: Message
@typedoc """
Defines options that can be passed to `start/3` / `start_link/3` and received
in `c:handle_init/2` callback.
"""
@type pipeline_options :: any
@type name :: GenServer.name()
@type config :: [config_entry()]
@type config_entry :: {:name, name()}
@type on_start ::
{:ok, supervisor_pid :: pid, pipeline_pid :: pid}
| {:error, {:already_started, pid()} | term()}
@type state :: any()
@typedoc """
Defines return values from Pipeline callback functions.
## Return values
* `{[action], state}` - Return a list of actions that will be performed within the
pipeline. This can be used to start new children, or to send messages to specific children,
for example. Actions are a tuple of `{type, arguments}`, so may be written in the
form a keyword list. See `Membrane.Pipeline.Action` for more info.
"""
@type callback_return ::
{[Action.t()], state}
@doc """
Callback invoked on initialization of pipeline.
This callback is synchronous: the process which started the pipeline waits until `handle_init`
finishes. For that reason, it's important to do any long-lasting or complex work in `c:handle_setup/2`,
while `handle_init` should be used for things like parsing options, initializing state or spawning
children.
By default, it converts the `opts` to a map if they're a struct and sets them as the pipeline state.
"""
@callback handle_init(context :: CallbackContext.t(), options :: pipeline_options) ::
{[Action.common_actions()], state()}
@doc """
Callback invoked when pipeline is requested to terminate with `terminate/2`.
By default, it returns `t:Membrane.Pipeline.Action.terminate/0` with reason `:normal`.
"""
@callback handle_terminate_request(context :: CallbackContext.t(), state) ::
{[Action.common_actions()], state()}
@doc """
Callback invoked on pipeline startup, right after `c:handle_init/2`.
Any long-lasting or complex initialization should happen here.
By default, it does nothing.
"""
@callback handle_setup(
context :: CallbackContext.t(),
state
) ::
{[Action.common_actions()], state()}
@doc """
Callback invoked when pipeline switches the playback to `:playing`.
By default, it does nothing.
"""
@callback handle_playing(
context :: CallbackContext.t(),
state
) ::
{[Action.common_actions()], state()}
@doc """
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.
"""
@callback handle_child_pad_removed(
child :: Child.name(),
pad :: Pad.ref(),
context :: CallbackContext.t(),
state :: state
) :: {[Action.common_actions()], state()}
@doc """
Callback invoked when a notification comes in from a child.
By default, it ignores the notification.
"""
@callback handle_child_notification(
notification :: Membrane.ChildNotification.t(),
element :: Child.name(),
context :: CallbackContext.t(),
state
) :: {[Action.common_actions()], state()}
@doc """
Callback invoked when pipeline receives a message that is not recognized
as an internal membrane message.
Useful for receiving data sent from NIFs or other stuff.
By default, it logs and ignores the received message.
"""
@callback handle_info(
message :: any,
context :: CallbackContext.t(),
state
) ::
{[Action.common_actions()], state()}
@doc """
Callback invoked when a child element starts processing stream via given pad.
By default, it does nothing.
"""
@callback handle_element_start_of_stream(
child :: Child.name(),
pad :: Pad.ref(),
context :: CallbackContext.t(),
state
) :: {[Action.common_actions()], state()}
@doc """
Callback invoked when a child element finishes processing stream via given pad.
By default, it does nothing.
"""
@callback handle_element_end_of_stream(
child :: Child.name(),
pad :: Pad.ref(),
context :: CallbackContext.t(),
state
) :: {[Action.common_actions()], state()}
@doc """
Callback invoked when children of `Membrane.ChildrenSpec` are started.
By default, it does nothing.
"""
@callback handle_spec_started(
children :: [Child.name()],
context :: CallbackContext.t(),
state
) :: {[Action.common_actions()], state()}
@doc """
Callback invoked upon each timer tick. A timer can be started with `Membrane.Pipeline.Action.start_timer`
action.
"""
@callback handle_tick(
timer_id :: any,
context :: CallbackContext.t(),
state
) :: {[Action.common_actions()], state()}
@doc """
Callback invoked when crash of the crash group happens.
Context passed to this callback contains 2 additional fields: `:members` and `:crash_initiator`.
By default, it does nothing.
"""
@callback handle_crash_group_down(
group_name :: Child.group(),
context :: CallbackContext.t(),
state
) :: {[Action.common_actions()], state()}
@doc """
Callback invoked when pipeline is called using a synchronous call.
Context passed to this callback contains additional field `:from`.
By default, it does nothing.
"""
@callback handle_call(
message :: any,
context :: CallbackContext.t(),
state
) ::
{[Action.common_actions() | Action.reply()], state()}
@optional_callbacks handle_init: 2,
handle_setup: 2,
handle_playing: 2,
handle_info: 3,
handle_spec_started: 3,
handle_element_start_of_stream: 4,
handle_element_end_of_stream: 4,
handle_child_notification: 4,
handle_tick: 3,
handle_crash_group_down: 3,
handle_call: 3,
handle_terminate_request: 2,
handle_child_pad_removed: 4
@doc """
Starts the Pipeline based on given module and links it to the current
process.
Pipeline options are passed to module's `c: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.
"""
@spec start_link(module, pipeline_options, config) :: on_start
def start_link(module, pipeline_options \\ nil, process_options \\ []),
do: do_start(:start_link, module, pipeline_options, process_options)
@doc """
Does the same as `start_link/3` but starts process outside of supervision tree.
"""
@spec start(module, pipeline_options, config) :: on_start
def start(module, pipeline_options \\ nil, process_options \\ []),
do: do_start(:start, module, pipeline_options, process_options)
defp do_start(method, module, pipeline_options, process_options) do
if module |> pipeline? do
Membrane.Logger.debug("""
Pipeline start link: module: #{inspect(module)},
pipeline options: #{inspect(pipeline_options)},
process options: #{inspect(process_options)}
""")
name =
case Keyword.fetch(process_options, :name) do
{:ok, name} when is_atom(name) -> Atom.to_string(name)
_other -> nil
end
|> case do
"Elixir." <> module -> module
name -> name
end
Membrane.Core.Pipeline.Supervisor.run(
method,
name,
&GenServer.start_link(
Membrane.Core.Pipeline,
%{
name: name,
module: module,
options: pipeline_options,
subprocess_supervisor: &1
},
process_options
)
)
else
Membrane.Logger.error("""
Cannot start pipeline, passed module #{inspect(module)} is not a Membrane Pipeline.
Make sure that given module is the right one and it uses Membrane.Pipeline
""")
{:error, {:not_pipeline, module}}
end
end
@doc """
Terminates the pipeline.
Accepts three options:
* `asynchronous?` - if set to `true`, pipline termination won't be blocking and
will be executed in the process, which pid is returned as 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` - tells how much time (ms) to wait for pipeline to get gracefully
terminated. Defaults to 5000.
* `force?` - if set to `true` and pipeline is still alive after `timeout`,
pipeline will be killed using `Process.exit/2` with reason `:kill`, and function
will return `{:error, :timeout}`. If set to `false` and pipeline is still alive
after `timeout`, function will raise an error. Defaults to `false`.
Returns:
* `{:ok, pid}` - if option `asynchronous?: true` was passed.
* `:ok` - if pipeline was gracefully terminated within `timeout`.
* `{:error, :timeout}` - if pipeline was killed after a `timeout`.
"""
@spec terminate(pipeline :: pid,
timeout: timeout(),
force?: boolean(),
asynchronous?: boolean()
) ::
:ok | {:ok, pid()} | {:error, :timeout}
def terminate(pipeline, opts \\ []) do
[asynchronous?: asynchronous?] ++ opts =
Keyword.validate!(opts,
asynchronous?: false,
force?: false,
timeout: 5000
)
|> Enum.sort()
if asynchronous? do
Task.start(__MODULE__, :do_terminate, [pipeline, opts])
else
do_terminate(pipeline, opts)
end
end
@doc false
@spec do_terminate(pipeline :: pid, timeout: timeout(), force?: boolean()) ::
:ok | {:error, :timeout}
def do_terminate(pipeline, opts) do
timeout = Keyword.get(opts, :timeout)
force? = Keyword.get(opts, :force?)
ref = Process.monitor(pipeline)
Message.send(pipeline, :terminate)
receive do
{:DOWN, ^ref, _process, _pid, _reason} ->
:ok
after
timeout ->
if force? do
Process.exit(pipeline, :kill)
{:error, :timeout}
else
raise PipelineError, """
Pipeline #{inspect(pipeline)} hasn't terminated within given timeout (#{inspect(timeout)} ms).
If you want to kill it anyway, use `force?: true` option.
"""
end
end
end
@spec call(pid, any, timeout()) :: term()
def call(pipeline, message, timeout \\ 5000) do
GenServer.call(pipeline, message, timeout)
end
@doc """
Checks whether module is a pipeline.
"""
@spec pipeline?(module) :: boolean
def pipeline?(module) do
module |> Bunch.Module.check_behaviour(:membrane_pipeline?)
end
@doc """
Lists PIDs of all the pipelines currently running on the current node.
Use only for debugging purposes.
"""
@spec list_pipelines() :: [pid]
def list_pipelines() do
Process.list()
|> Enum.filter(fn pid ->
case Process.info(pid, :dictionary) do
{:dictionary, dictionary} -> List.keyfind(dictionary, :__membrane_pipeline__, 0)
nil -> false
end
end)
end
@doc """
Like `list_pipelines/0`, but allows to pass a node.
"""
@spec list_pipelines(node()) :: [pid]
def list_pipelines(node) do
:erpc.call(node, __MODULE__, :list_pipelines, [])
end
@doc """
Brings all the stuff necessary to implement a pipeline.
Options:
- `:bring_spec?` - if true (default) imports and aliases `Membrane.ChildrenSpec`
- `:bring_pad?` - if true (default) requires and aliases `Membrane.Pad`
"""
defmacro __using__(options) do
bring_spec =
if Keyword.get(options, :bring_spec?, true) do
quote do
import Membrane.ChildrenSpec
alias Membrane.ChildrenSpec
end
end
bring_pad =
if Keyword.get(options, :bring_pad?, true) do
quote do
require Membrane.Pad
alias Membrane.Pad
end
end
# credo:disable-for-next-line Credo.Check.Refactor.LongQuoteBlocks
quote do
alias unquote(__MODULE__)
require Membrane.Logger
@behaviour unquote(__MODULE__)
unquote(bring_spec)
unquote(bring_pad)
@doc """
Returns child specification for spawning under a supervisor
"""
# credo:disable-for-next-line Credo.Check.Readability.Specs
def child_spec(arg) do
%{
id: __MODULE__,
start: {__MODULE__, :start_link, [arg]},
type: :supervisor
}
end
@doc false
@spec membrane_pipeline?() :: true
def membrane_pipeline?, do: true
@impl true
def handle_init(_ctx, %_opt_struct{} = options),
do: {[], options |> Map.from_struct()}
@impl true
def handle_init(_ctx, options), do: {[], options}
@impl true
def handle_setup(_ctx, state), do: {[], state}
@impl true
def handle_playing(_ctx, state), do: {[], state}
@impl true
def handle_info(message, _ctx, state) do
Membrane.Logger.warning("""
Received message but no handle_info callback has been specified. Ignoring.
Message: #{inspect(message)}\
""")
{[], state}
end
@impl true
def handle_spec_started(new_children, _ctx, state), do: {[], state}
@impl true
def handle_element_start_of_stream(_element, _pad, _ctx, state), do: {[], state}
@impl true
def handle_element_end_of_stream(_element, _pad, _ctx, state), do: {[], state}
@impl true
def handle_child_notification(notification, element, _ctx, state), do: {[], state}
@impl true
def handle_crash_group_down(_group_name, _ctx, state), do: {[], state}
@impl true
def handle_call(message, _ctx, state), do: {[], state}
@impl true
def handle_terminate_request(_ctx, state), do: {[terminate: :normal], state}
defoverridable child_spec: 1,
handle_init: 2,
handle_setup: 2,
handle_playing: 2,
handle_info: 3,
handle_spec_started: 3,
handle_element_start_of_stream: 4,
handle_element_end_of_stream: 4,
handle_child_notification: 4,
handle_crash_group_down: 3,
handle_call: 3,
handle_terminate_request: 2
end
end
end