OTP Design Principles

User's Guide

Version 10.5

Chapters

5 Supervisor Behaviour

This section should be read with the supervisor(3) manual page in STDLIB, where all details about the supervisor behaviour is given.

5.1  Supervision Principles

A supervisor is responsible for starting, stopping, and monitoring its child processes. The basic idea of a supervisor is that it is to keep its child processes alive by restarting them when necessary.

Which child processes to start and monitor is specified by a list of child specifications. The child processes are started in the order specified by this list, and terminated in the reversed order.

5.2  Example

The callback module for a supervisor starting the server from gen_server Behaviour can look as follows:

-module(ch_sup).
-behaviour(supervisor).

-export([start_link/0]).
-export([init/1]).

start_link() ->
    supervisor:start_link(ch_sup, []).

init(_Args) ->
    SupFlags = #{strategy => one_for_one, intensity => 1, period => 5},
    ChildSpecs = [#{id => ch3,
                    start => {ch3, start_link, []},
                    restart => permanent,
                    shutdown => brutal_kill,
                    type => worker,
                    modules => [cg3]}],
    {ok, {SupFlags, ChildSpecs}}.

The SupFlags variable in the return value from init/1 represents the supervisor flags.

The ChildSpecs variable in the return value from init/1 is a list of child specifications.

5.3  Supervisor Flags

This is the type definition for the supervisor flags:

sup_flags() = #{strategy => strategy(),         % optional
                intensity => non_neg_integer(), % optional
                period => pos_integer()}        % optional
    strategy() = one_for_all
               | one_for_one
               | rest_for_one
               | simple_one_for_one

5.4  Restart Strategy

The restart strategy is specified by the strategy key in the supervisor flags map returned by the callback function init:

SupFlags = #{strategy => Strategy, ...}

The strategy key is optional in this map. If it is not given, it defaults to one_for_one.

one_for_one

If a child process terminates, only that process is restarted.

IMAGE MISSING

Figure 5.1:   One_For_One Supervision

one_for_all

If a child process terminates, all other child processes are terminated, and then all child processes, including the terminated one, are restarted.

IMAGE MISSING

Figure 5.2:   One_For_All Supervision

rest_for_one

If a child process terminates, the rest of the child processes (that is, the child processes after the terminated process in start order) are terminated. Then the terminated child process and the rest of the child processes are restarted.

simple_one_for_one

See simple-one-for-one supervisors.

5.5  Maximum Restart Intensity

The supervisors have a built-in mechanism to limit the number of restarts which can occur in a given time interval. This is specified by the two keys intensity and period in the supervisor flags map returned by the callback function init:

SupFlags = #{intensity => MaxR, period => MaxT, ...}

If more than MaxR number of restarts occur in the last MaxT seconds, the supervisor terminates all the child processes and then itself. The termination reason for the supervisor itself in that case will be shutdown.

When the supervisor terminates, then the next higher-level supervisor takes some action. It either restarts the terminated supervisor or terminates itself.

The intention of the restart mechanism is to prevent a situation where a process repeatedly dies for the same reason, only to be restarted again.

The keys intensity and period are optional in the supervisor flags map. If they are not given, they default to 1 and 5, respectively.

Tuning the intensity and period

The default values are 1 restart per 5 seconds. This was chosen to be safe for most systems, even with deep supervision hierarchies, but you will probably want to tune the settings for your particular use case.

First, the intensity decides how big bursts of restarts you want to tolerate. For example, you might want to accept a burst of at most 5 or 10 attempts, even within the same second, if it results in a successful restart.

Second, you need to consider the sustained failure rate, if crashes keep happening but not often enough to make the supervisor give up. If you set intensity to 10 and set the period as low as 1, the supervisor will allow child processes to keep restarting up to 10 times per second, forever, filling your logs with crash reports until someone intervenes manually.

You should therefore set the period to be long enough that you can accept that the supervisor keeps going at that rate. For example, if you have picked an intensity value of 5, then setting the period to 30 seconds will give you at most one restart per 6 seconds for any longer period of time, which means that your logs won't fill up too quickly, and you will have a chance to observe the failures and apply a fix.

These choices depend a lot on your problem domain. If you don't have real time monitoring and ability to fix problems quickly, for example in an embedded system, you might want to accept at most one restart per minute before the supervisor should give up and escalate to the next level to try to clear the error automatically. On the other hand, if it is more important that you keep trying even at a high failure rate, you might want a sustained rate of as much as 1-2 restarts per second.

Avoiding common mistakes:

  • Do not forget to consider the burst rate. If you set intensity to 1 and period to 6, it gives the same sustained error rate as 5/30 or 10/60, but will not allow even 2 restart attempts in quick succession. This is probably not what you wanted.

  • Do not set the period to a very high value if you want to tolerate bursts. If you set intensity to 5 and period to 3600 (one hour), the supervisor will allow a short burst of 5 restarts, but then gives up if it sees another single restart almost an hour later. You probably want to regard those crashes as separate incidents, so setting the period to 5 or 10 minutes will be more reasonable.

  • If your application has multiple levels of supervision, then do not simply set the restart intensities to the same values on all levels. Keep in mind that the total number of restarts (before the top level supervisor gives up and terminates the application) will be the product of the intensity values of all the supervisors above the failing child process.

    For example, if the top level allows 10 restarts, and the next level also allows 10, a crashing child below that level will be restarted 100 times, which is probably excessive. Allowing at most 3 restarts for the top level supervisor might be a better choice in this case.

5.6  Child Specification

The type definition for a child specification is as follows:

child_spec() = #{id => child_id(),       % mandatory
                 start => mfargs(),      % mandatory
                 restart => restart(),   % optional
                 shutdown => shutdown(), % optional
                 type => worker(),       % optional
                 modules => modules()}   % optional
    child_id() = term()
    mfargs() = {M :: module(), F :: atom(), A :: [term()]}
    modules() = [module()] | dynamic
    restart() = permanent | transient | temporary
    shutdown() = brutal_kill | timeout()
    worker() = worker | supervisor
  • id is used to identify the child specification internally by the supervisor.

    The id key is mandatory.

    Note that this identifier occasionally has been called "name". As far as possible, the terms "identifier" or "id" are now used but in order to keep backwards compatibility, some occurences of "name" can still be found, for example in error messages.

  • start defines the function call used to start the child process. It is a module-function-arguments tuple used as apply(M, F, A).

    It is to be (or result in) a call to any of the following:

    • supervisor:start_link
    • gen_server:start_link
    • gen_statem:start_link
    • gen_event:start_link
    • A function compliant with these functions. For details, see the supervisor(3) manual page.

    The start key is mandatory.

  • restart defines when a terminated child process is to be restarted.

    • A permanent child process is always restarted.
    • A temporary child process is never restarted (not even when the supervisor restart strategy is rest_for_one or one_for_all and a sibling death causes the temporary process to be terminated).
    • A transient child process is restarted only if it terminates abnormally, that is, with an exit reason other than normal, shutdown, or {shutdown,Term}.

    The restart key is optional. If it is not given, the default value permanent will be used.

  • shutdown defines how a child process is to be terminated.

    • brutal_kill means that the child process is unconditionally terminated using exit(Child, kill).
    • An integer time-out value means that the supervisor tells the child process to terminate by calling exit(Child, shutdown) and then waits for an exit signal back. If no exit signal is received within the specified time, the child process is unconditionally terminated using exit(Child, kill).
    • If the child process is another supervisor, it must be set to infinity to give the subtree enough time to shut down. It is also allowed to set it to infinity, if the child process is a worker. See the warning below:
    Warning

    Setting the shutdown time to anything other than infinity for a child of type supervisor can cause a race condition where the child in question unlinks its own children, but fails to terminate them before it is killed.

    Be careful when setting the shutdown time to infinity when the child process is a worker. Because, in this situation, the termination of the supervision tree depends on the child process; it must be implemented in a safe way and its cleanup procedure must always return.

    The shutdown key is optional. If it is not given, and the child is of type worker, the default value 5000 will be used; if the child is of type supervisor, the default value infinity will be used.

  • type specifies if the child process is a supervisor or a worker.

    The type key is optional. If it is not given, the default value worker will be used.

  • modules are to be a list with one element [Module], where Module is the name of the callback module, if the child process is a supervisor, gen_server, gen_statem. If the child process is a gen_event, the value shall be dynamic.

    This information is used by the release handler during upgrades and downgrades, see Release Handling.

    The modules key is optional. If it is not given, it defaults to [M], where M comes from the child's start {M,F,A}.

Example: The child specification to start the server ch3 in the previous example look as follows:

#{id => ch3,
  start => {ch3, start_link, []},
  restart => permanent,
  shutdown => brutal_kill,
  type => worker,
  modules => [ch3]}

or simplified, relying on the default values:

#{id => ch3,
  start => {ch3, start_link, []}
  shutdown => brutal_kill}

Example: A child specification to start the event manager from the chapter about gen_event:

#{id => error_man,
  start => {gen_event, start_link, [{local, error_man}]},
  modules => dynamic}

Both server and event manager are registered processes which can be expected to be always accessible. Thus they are specified to be permanent.

ch3 does not need to do any cleaning up before termination. Thus, no shutdown time is needed, but brutal_kill is sufficient. error_man can need some time for the event handlers to clean up, thus the shutdown time is set to 5000 ms (which is the default value).

Example: A child specification to start another supervisor:

#{id => sup,
  start => {sup, start_link, []},
  restart => transient,
  type => supervisor} % will cause default shutdown=>infinity

5.7  Starting a Supervisor

In the previous example, the supervisor is started by calling ch_sup:start_link():

start_link() ->
    supervisor:start_link(ch_sup, []).

ch_sup:start_link calls function supervisor:start_link/2, which spawns and links to a new process, a supervisor.

  • The first argument, ch_sup, is the name of the callback module, that is, the module where the init callback function is located.
  • The second argument, [], is a term that is passed as is to the callback function init. Here, init does not need any indata and ignores the argument.

In this case, the supervisor is not registered. Instead its pid must be used. A name can be specified by calling supervisor:start_link({local, Name}, Module, Args) or supervisor:start_link({global, Name}, Module, Args).

The new supervisor process calls the callback function ch_sup:init([]). init shall return {ok, {SupFlags, ChildSpecs}}:

init(_Args) ->
    SupFlags = #{},
    ChildSpecs = [#{id => ch3,
                    start => {ch3, start_link, []},
                    shutdown => brutal_kill}],
    {ok, {SupFlags, ChildSpecs}}.

The supervisor then starts all its child processes according to the child specifications in the start specification. In this case there is one child process, ch3.

supervisor:start_link is synchronous. It does not return until all child processes have been started.

5.8  Adding a Child Process

In addition to the static supervision tree, dynamic child processes can be added to an existing supervisor with the following call:

supervisor:start_child(Sup, ChildSpec)

Sup is the pid, or name, of the supervisor. ChildSpec is a child specification.

Child processes added using start_child/2 behave in the same way as the other child processes, with the an important exception: if a supervisor dies and is recreated, then all child processes that were dynamically added to the supervisor are lost.

5.9  Stopping a Child Process

Any child process, static or dynamic, can be stopped in accordance with the shutdown specification:

supervisor:terminate_child(Sup, Id)

The child specification for a stopped child process is deleted with the following call:

supervisor:delete_child(Sup, Id)

Sup is the pid, or name, of the supervisor. Id is the value associated with the id key in the child specification.

As with dynamically added child processes, the effects of deleting a static child process is lost if the supervisor itself restarts.

5.10  Simplified one_for_one Supervisors

A supervisor with restart strategy simple_one_for_one is a simplified one_for_one supervisor, where all child processes are dynamically added instances of the same process.

The following is an example of a callback module for a simple_one_for_one supervisor:

-module(simple_sup).
-behaviour(supervisor).

-export([start_link/0]).
-export([init/1]).

start_link() ->
    supervisor:start_link(simple_sup, []).

init(_Args) ->
    SupFlags = #{strategy => simple_one_for_one,
                 intensity => 0,
                 period => 1},
    ChildSpecs = [#{id => call,
                    start => {call, start_link, []},
                    shutdown => brutal_kill}],
    {ok, {SupFlags, ChildSpecs}}.

When started, the supervisor does not start any child processes. Instead, all child processes are added dynamically by calling:

supervisor:start_child(Sup, List)

Sup is the pid, or name, of the supervisor. List is an arbitrary list of terms, which are added to the list of arguments specified in the child specification. If the start function is specified as {M, F, A}, the child process is started by calling apply(M, F, A++List).

For example, adding a child to simple_sup above:

supervisor:start_child(Pid, [id1])

The result is that the child process is started by calling apply(call, start_link, []++[id1]), or actually:

call:start_link(id1)

A child under a simple_one_for_one supervisor can be terminated with the following:

supervisor:terminate_child(Sup, Pid)

Sup is the pid, or name, of the supervisor and Pid is the pid of the child.

Because a simple_one_for_one supervisor can have many children, it shuts them all down asynchronously. This means that the children will do their cleanup in parallel and therefore the order in which they are stopped is not defined.

5.11  Stopping

Since the supervisor is part of a supervision tree, it is automatically terminated by its supervisor. When asked to shut down, it terminates all child processes in reversed start order according to the respective shutdown specifications, and then terminates itself.