0Pricing
Erlang OTP: Distributed & Fault-Tolerant Systems Programming · Lesson

Advanced Restart Strategies

Delve into one_for_one, one_for_all, and rest_for_one restart strategies, understanding their implications for fault tolerance.

Advanced Restart Strategies is a free Erlang OTP: Distributed & Fault-Tolerant Systems Programming lesson on CoddyKit — lesson 3 of 4. You can read the complete lesson below for free — then practise it hands-on in the browser with a built-in code editor and a 24/7 AI tutor. It is part of the Erlang OTP: Distributed & Fault-Tolerant Systems Programming learning path, one of 4 lessons in the course, and your progress syncs across the web and the CoddyKit app.

Intro to Restart Strategies

Welcome to a crucial topic in Erlang: restart strategies! These define how a supervisor reacts when one of its child processes crashes.

Understanding these strategies is key to building fault-tolerant and self-healing systems, which is a hallmark of Erlang/OTP.

Supervisors: The Fault Managers

Before diving in, let's quickly recap: a supervisor is a special process that monitors other processes (its children).

  • If a child process dies, the supervisor detects it.
  • Based on its configured restart strategy, the supervisor decides how to bring the system back to a healthy state.
  • This ensures your application can recover from individual process failures automatically.

`one_for_one`: Isolated Restarts

The one_for_one strategy is the simplest and most common. When a child process terminates:

  • Only the failing child process is restarted.
  • All other sibling processes remain unaffected and continue running.

This strategy is ideal for systems where child processes are largely independent of each other, such as individual client connections.

`one_for_one` in Action

Let's see one_for_one. We'll have a supervisor managing a single worker. When the worker crashes, only it restarts.

To run:
1. Compile: c(worker_gen). c(supervisor_one_for_one).
2. Start supervisor: supervisor_one_for_one:start_link().
3. Crash worker: worker_gen:crash(worker_1).
Observe the output in your Erlang shell.

-module(worker_gen).
-behaviour(gen_server).
-export([start_link/1, init/1, handle_call/3, terminate/2, crash/1]).

start_link(Id) ->
    gen_server:start_link({local, Id}, ?MODULE, Id, []).

init(Id) ->
    io:format("Worker ~p (~p) started.~n", [Id, self()]),
    {ok, Id}.

handle_call(crash, _From, State) ->
    exit(i_crashed),
    {reply, ok, State};
handle_call(_Req, _From, State) ->
    {reply, ok, State}.

terminate(_Reason, Id) ->
    io:format("Worker ~p (~p) terminated.~n", [Id, self()]).

crash(Id) ->
    gen_server:call(Id, crash).

-module(supervisor_one_for_one).
-behaviour(supervisor).
-export([start_link/0, init/1]).

start_link() ->
    supervisor:start_link({local, ?MODULE}, ?MODULE, []).

init([]) ->
    ChildSpec = #{
        id => worker_1,
        start => {worker_gen, start_link, [worker_1]},
        restart => permanent,
        shutdown => 5000,
        type => worker,
        modules => [worker_gen]
    },
    {ok, #{
        strategy => {one_for_one, 3, 5},
        children => [ChildSpec]
    }}.

`one_for_all`: All for One Failure

The one_for_all strategy takes a more drastic approach:

  • If any child process terminates, all other child processes are first terminated.
  • Then, all child processes (including the one that crashed) are restarted.

This is useful when your child processes are tightly coupled and require a consistent, synchronized state. A failure in one implies a need to reset the entire group.

`one_for_all` in Action

Here, a supervisor manages two workers. Crash one, and both will restart.

To run:
1. Compile: c(worker_gen). c(supervisor_one_for_all).
2. Start supervisor: supervisor_one_for_all:start_link().
3. Crash worker: worker_gen:crash(worker_A).
Notice how both worker_A and worker_B restart.

-module(worker_gen).
-behaviour(gen_server).
-export([start_link/1, init/1, handle_call/3, terminate/2, crash/1]).

start_link(Id) ->
    gen_server:start_link({local, Id}, ?MODULE, Id, []).

init(Id) ->
    io:format("Worker ~p (~p) started.~n", [Id, self()]),
    {ok, Id}.

handle_call(crash, _From, State) ->
    exit(i_crashed),
    {reply, ok, State};
handle_call(_Req, _From, State) ->
    {reply, ok, State}.

terminate(_Reason, Id) ->
    io:format("Worker ~p (~p) terminated.~n", [Id, self()]).

crash(Id) ->
    gen_server:call(Id, crash).

-module(supervisor_one_for_all).
-behaviour(supervisor).
-export([start_link/0, init/1]).

start_link() ->
    supervisor:start_link({local, ?MODULE}, ?MODULE, []).

init([]) ->
    Child1 = #{
        id => worker_A,
        start => {worker_gen, start_link, [worker_A]},
        restart => permanent, shutdown => 5000, type => worker, modules => [worker_gen]
    },
    Child2 = #{
        id => worker_B,
        start => {worker_gen, start_link, [worker_B]},
        restart => permanent, shutdown => 5000, type => worker, modules => [worker_gen]
    },
    {ok, #{
        strategy => {one_for_all, 3, 5},
        children => [Child1, Child2]
    }}.

`rest_for_one`: Cascade Restarts

The rest_for_one strategy offers a middle ground:

  • If a child process terminates, it and all subsequent children (those defined after it in the supervisor's child list) are terminated.
  • Then, the failed child and all subsequent children are restarted.
  • Children defined before the failed child are left untouched.

This is useful when processes have sequential dependencies, where a failure in an earlier stage might invalidate the state of later stages.

`rest_for_one` in Action

We have three workers: X, Y, Z. If Y crashes, Y and Z restart, but X remains active.

To run:
1. Compile: c(worker_gen). c(supervisor_rest_for_one).
2. Start supervisor: supervisor_rest_for_one:start_link().
3. Crash worker: worker_gen:crash(worker_Y).
See worker_X stay running, while worker_Y and worker_Z restart.

-module(worker_gen).
-behaviour(gen_server).
-export([start_link/1, init/1, handle_call/3, terminate/2, crash/1]).

start_link(Id) ->
    gen_server:start_link({local, Id}, ?MODULE, Id, []).

init(Id) ->
    io:format("Worker ~p (~p) started.~n", [Id, self()]),
    {ok, Id}.

handle_call(crash, _From, State) ->
    exit(i_crashed),
    {reply, ok, State};
handle_call(_Req, _From, State) ->
    {reply, ok, State}.

terminate(_Reason, Id) ->
    io:format("Worker ~p (~p) terminated.~n", [Id, self()]).

crash(Id) ->
    gen_server:call(Id, crash).

-module(supervisor_rest_for_one).
-behaviour(supervisor).
-export([start_link/0, init/1]).

start_link() ->
    supervisor:start_link({local, ?MODULE}, ?MODULE, []).

init([]) ->
    ChildA = #{
        id => worker_X,
        start => {worker_gen, start_link, [worker_X]},
        restart => permanent, shutdown => 5000, type => worker, modules => [worker_gen]
    },
    ChildB = #{
        id => worker_Y,
        start => {worker_gen, start_link, [worker_Y]},
        restart => permanent, shutdown => 5000, type => worker, modules => [worker_gen]
    },
    ChildC = #{
        id => worker_Z,
        start => {worker_gen, start_link, [worker_Z]},
        restart => permanent, shutdown => 5000, type => worker, modules => [worker_gen]
    },
    {ok, #{
        strategy => {rest_for_one, 3, 5},
        children => [ChildA, ChildB, ChildC]
    }}.

Selecting the Best Strategy

Choosing the right strategy depends on your application's architecture and process dependencies:

  • one_for_one: Use for independent processes, like individual client connections or request handlers.
  • one_for_all: Best for tightly coupled processes that must always be in a consistent state together (e.g., a group of processes managing a single resource).
  • rest_for_one: Suitable for sequential pipelines or layered systems where a failure in an earlier stage affects subsequent stages.

Test Your Knowledge

You are building a system where a primary worker fetches data, and two secondary workers process different aspects of that data. If the primary worker fails, the secondary workers cannot continue with stale data and must also restart. If a secondary worker fails, the others are unaffected. Which strategy is most appropriate for the primary worker and its dependent secondary workers?

Recap: Mastering Restarts

You've explored Erlang's powerful restart strategies:

  • one_for_one: Restarts only the failed child, keeping others running.
  • one_for_all: Restarts all children if any one fails, ensuring full consistency.
  • rest_for_one: Restarts the failed child and all subsequent children in the list.

These strategies are fundamental to building robust, self-healing applications in Erlang, allowing your system to recover from failures gracefully.

Frequently asked questions

Is the “Advanced Restart Strategies” lesson free?

Yes — the full text of “Advanced Restart Strategies” is free to read here on the web, and the Erlang OTP: Distributed & Fault-Tolerant Systems Programming course includes 4 lessons in total. To practise it interactively (a built-in code editor and a 24/7 AI tutor) and unlock the rest of the Erlang OTP: Distributed & Fault-Tolerant Systems Programming course, upgrade to CoddyKit PRO.

What will I learn in “Advanced Restart Strategies”?

Delve into one_for_one, one_for_all, and rest_for_one restart strategies, understanding their implications for fault tolerance. You practise Erlang OTP: Distributed & Fault-Tolerant Systems Programming with hands-on code you run directly in the browser, and a 24/7 AI tutor answers your questions as you work through the lesson.

Do I need any experience to start Erlang OTP: Distributed & Fault-Tolerant Systems Programming?

No prior experience is required. Erlang OTP: Distributed & Fault-Tolerant Systems Programming on CoddyKit is structured for beginners through advanced learners; this is — lesson 3 of 4, so you can start here or from the beginning and move at your own pace.

How long does the “Advanced Restart Strategies” lesson take?

Most CoddyKit lessons take about 5–10 minutes. Each one is bite-sized and interactive, so you make steady progress and pick up exactly where you left off across the web and the app.

Can I write and run code in this Erlang OTP: Distributed & Fault-Tolerant Systems Programming lesson?

Yes. Every Erlang OTP: Distributed & Fault-Tolerant Systems Programming lesson includes a built-in code editor, so you write and run real code right in your browser and get instant AI feedback — no local setup required.

All lessons in this course

  1. Complex Supervision Trees
  2. Dynamic Process Management
  3. Advanced Restart Strategies
  4. Supervisor Bridges & Mixed Process Hierarchies
← Back to Erlang OTP: Distributed & Fault-Tolerant Systems Programming