otp深入浅出之: supervisor的simple_one_for_one

    supervisor是一种很常用的erlang pattern，各种资料比较多，但是网上各种例子和说明都主要是针对one_for_one的restart strategies，simple_one_for_one的资料比较少。本文通过分析supervisor源码，说明simple_one_for_one的使用方法与运行机制。


what  is the  supervisor ？

引用

A  behaviour  module  for  implementing  a  supervisor, a process which supervises other processes called child processes.

简单的说，实现了supervisor的进程可以负责监控其他子进程，supervisor的子进程有4种重启策略，

引用

     A supervisor can have one of the following restart strategies:

         * one_for_one  -  if  one  child  process  terminates  and  should be
           restarted, only that child process is affected.

         * one_for_all -  if  one  child  process  terminates  and  should  be
           restarted,  all  other  child processes are terminated and then all
           child processes are restarted.

         * rest_for_one - if  one  child  process  terminates  and  should  be
           restarted, the 'rest' of the child processes -- i.e. the child pro-
           cesses after the terminated child process in the start order -- are
           terminated.  Then  the  terminated child process and all child pro-
           cesses after it are restarted.

         * simple_one_for_one - a simplified one_for_one supervisor, where all
           child processes are dynamically added instances of the same process
           type, i.e. running the same code.

           The functions terminate_child/2, delete_child/2 and restart_child/2
           are  invalid  for  simple_one_for_one  supervisors  and will return
           {error, simple_one_for_one} if the specified supervisor  uses  this
           restart strategy.

其中simple_one_for_one用来创建一组动态子进程，一个使用了simple_one_for_one 策略的supervisor可以类比为factory pattern,最常见的例子就是socket 服务端接受新的连接请求以后，需要创建新的socket连接进程。

假设通过supervisor，在某些情况下动态创建一个container进程，每个container behaviour(gen_server) 作为进程存在。


container.erl代码:

-module(container).

-behaviour(gen_server).


-export([start_link/1]).
-export([stop/1]).
-export([setup/0]).

-export([init/1, handle_call/3, handle_cast/2, handle_info/2,
         terminate/2, code_change/3]).

-record(state, {name}).

setup()->
        supervisor:start_child(container_sup,["container1"]),
        supervisor:start_child(container_sup,["container2"]),
        supervisor:start_child(container_sup,["container3"]),
        supervisor:start_child(container_sup,["container4"]).
stop(ContainerName)->
        	gen_server:call(list_to_atom(ContainerName) , {stop } ).

start_link(Name) when is_list(Name) ->
    gen_server:start_link( {local, list_to_atom(Name)}, ?MODULE, Name, []).
    
init(Name) ->
    {ok, #state{name=Name}}.
    
handle_call( {stop } , From, State) ->
        { stop , normal , ok , State };
handle_call(_Request, _From, State) ->
    Reply = ok,
    {reply, Reply, State}.

handle_cast(_Msg, State) ->
    {noreply, State}.

handle_info(Info, State) ->
    {noreply, State}.

terminate(_Reason, _State) ->
    ok.

code_change(_OldVsn, State, _Extra) ->
    {ok, State}.

container_sup.erl代码：

-module(container_sup).

-behaviour(supervisor).

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

-define(SERVER, ?MODULE).

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

init([]) ->
    ContainerSpec = {container,{container,start_link,[]},
              transient,2000,worker,[container]},
    {ok,{{simple_one_for_one,10,100}, [ContainerSpec]}}.

1 how to start the container_sup ?

1.1 start  supervisor

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

1.2 supervisor:start_link回调gen_server:start_link

start_link(SupName, Mod, Args) ->
    gen_server:start_link(SupName, supervisor, {SupName, Mod, Args}, []).

1.3  gen_server:start_link最终回调supervisor:init(具体请参考http://uniseraph.iteye.com/blog/372838)

1.4  supervisor:init

init({SupName, Mod, Args}) ->
    process_flag(trap_exit, true),
    case Mod:init(Args) of
        {ok, {SupFlags, StartSpec}} ->
            case init_state(SupName, SupFlags, Mod, Args) of
                {ok, State} when ?is_simple(State) ->
                    init_dynamic(State, StartSpec);
                {ok, State} ->
                    init_children(State, StartSpec);
                Error ->
                    {stop, {supervisor_data, Error}}
            end;
        ignore ->
            ignore;
        Error ->
            {stop, {bad_return, {Mod, init, Error}}}
    end.

1.4.1  设置进程模式，当处理子进程退出是supervisor会受到一个{'EXIT', Pid, Reason}消息

 process_flag(trap_exit, true),

1.4.2  回调container_sup:init获得container_sup的子进程规约
1.4.3  根据不同的resart strategies调用相应的init_xxx方法

 case init_state(SupName, SupFlags, Mod, Args) of
                {ok, State} when ?is_simple(State) ->
                    init_dynamic(State, StartSpec);
                {ok, State} ->
                    init_children(State, StartSpec);
                Error ->
                    {stop, {supervisor_data, Error}}
            end;

1.4.4  init_dynamic检查子进程规约的合法性,并不真正启动子进程

init_dynamic(State, [StartSpec]) ->
    case check_startspec([StartSpec]) of
        {ok, Children} ->
            {ok, State#state{children = Children}};
        Error ->
            {stop, {start_spec, Error}}
    end;
init_dynamic(_State, StartSpec) ->
    {stop, {bad_start_spec, StartSpec}}.

2 how to add a container processor ?
2.1

 supervisor:start_child(container_sup,["container1"]),

2.2 supervisor:start_child触发一个start_child事件，在supervisor中处理

handle_call({start_child, EArgs}, _From, State) when ?is_simple(State) ->
    #child{mfa = {M, F, A}} = hd(State#state.children),
    Args = A ++ EArgs,
    case do_start_child_i(M, F, Args) of
        {ok, Pid} ->
            NState = State#state{dynamics =
                                 ?DICT:store(Pid, Args, State#state.dynamics)},
            {reply, {ok, Pid}, NState};
        {ok, Pid, Extra} ->
            NState = State#state{dynamics =
                                 ?DICT:store(Pid, Args, State#state.dynamics)},
            {reply, {ok, Pid, Extra}, NState};
        What ->
            {reply, What, State}
    end;

2.3 do_start_child_i根据子进程配置规约回调container:start_link方法，并传入参数"container1"

2.4 记录子进程Pid

3 how to stop a container processor ?

引用

   The functions terminate_child/2, delete_child/2 and restart_child/2
           are  invalid  for  simple_one_for_one  supervisors  and will return
           {error, simple_one_for_one} if the specified supervisor  uses  this
           restart strategy.

3.1 container进程处理一个{stop｝消息

handle_call( {stop } , From, State) ->
        { stop , normal , ok , State };

3.2 container_sup受到一个{'EXIT', Pid, Reason}消息，在handle_info中处理

handle_info({'EXIT', Pid, Reason}, State) ->
    case restart_child(Pid, Reason, State) of
        {ok, State1} ->
            {noreply, State1};
        {shutdown, State1} ->
            {stop, shutdown, State1}
    end;

3.3 restart_child根据pid在state中查找并restart这个进程

restart_child(Pid, Reason, State) when ?is_simple(State) ->
    case ?DICT:find(Pid, State#state.dynamics) of
        {ok, Args} ->
            [Child] = State#state.children,
            RestartType = Child#child.restart_type,
            {M, F, _} = Child#child.mfa,
            NChild = Child#child{pid = Pid, mfa = {M, F, Args}},
            do_restart(RestartType, Reason, NChild, State);
        error ->
            {ok, State}
    end;

3.4 应用Reason=normal并且conainer是restart type是transient,所以这个container退出，不再重启。

do_restart(permanent, Reason, Child, State) ->
    report_error(child_terminated, Reason, Child, State#state.name),
    restart(Child, State);
do_restart(_, normal, Child, State) ->
    NState = state_del_child(Child, State),
    {ok, NState};
do_restart(_, shutdown, Child, State) ->
    NState = state_del_child(Child, State),
    {ok, NState};
do_restart(transient, Reason, Child, State) ->
    report_error(child_terminated, Reason, Child, State#state.name),
    restart(Child, State);
do_restart(temporary, Reason, Child, State) ->
    report_error(child_terminated, Reason, Child, State#state.name),
    NState = state_del_child(Child, State),
    {ok, NState}.

评论

1 楼 mryufeng 2009-08-12  

simple_one_to_one在网络程序里面用的挺多的 主要用于比如连接什么的管理。
3个作用：

1. otp体系 完整的出错信息管理
2. tree拥有节点的信息 可以方便遍历 自己不用写这部分代码了。
3. 进程关闭什么的可以推给tree来管理。