supervisor.erl

来自「OTP是开放电信平台的简称」· ERL 代码 · 共 889 行 · 第 1/2 页

	    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;
restart_child(Pid, Reason, State) ->
    Children = State#state.children,
    case lists:keysearch(Pid, #child.pid, Children) of
	{value, Child} ->
	    RestartType = Child#child.restart_type,
	    do_restart(RestartType, Reason, Child, State);
	_ ->
	    {ok, State}
    end.

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}.

restart(Child, State) ->
    case add_restart(State) of
	{ok, NState} ->
	    restart(NState#state.strategy, Child, NState);
	{terminate, NState} ->
	    report_error(shutdown, reached_max_restart_intensity,
			 Child, State#state.name),
	    {shutdown, remove_child(Child, NState)}
    end.

restart(simple_one_for_one, Child, State) ->
    #child{mfa = {M, F, A}} = Child,
    Dynamics = ?DICT:erase(Child#child.pid, State#state.dynamics),
    case do_start_child_i(M, F, A) of
	{ok, Pid} ->
	    NState = State#state{dynamics = ?DICT:store(Pid, A, Dynamics)},
	    {ok, NState};
	{ok, Pid, _Extra} ->
	    NState = State#state{dynamics = ?DICT:store(Pid, A, Dynamics)},
	    {ok, NState};
	{error, Error} ->
	    report_error(start_error, Error, Child, State#state.name),
	    restart(Child, State)
    end;
restart(one_for_one, Child, State) ->
    case do_start_child(State#state.name, Child) of
	{ok, Pid} ->
	    NState = replace_child(Child#child{pid = Pid}, State),
	    {ok, NState};
	{ok, Pid, _Extra} ->
	    NState = replace_child(Child#child{pid = Pid}, State),
	    {ok, NState};
	{error, Reason} ->
	    report_error(start_error, Reason, Child, State#state.name),
	    restart(Child, State)
    end;
restart(rest_for_one, Child, State) ->
    {ChAfter, ChBefore} = split_child(Child#child.pid, State#state.children),
    ChAfter2 = terminate_children(ChAfter, State#state.name),
    case start_children(ChAfter2, State#state.name) of
	{ok, ChAfter3} ->
	    {ok, State#state{children = ChAfter3 ++ ChBefore}};
	{error, ChAfter3} ->
	    restart(Child, State#state{children = ChAfter3 ++ ChBefore})
    end;
restart(one_for_all, Child, State) ->
    Children1 = del_child(Child#child.pid, State#state.children),
    Children2 = terminate_children(Children1, State#state.name),
    case start_children(Children2, State#state.name) of
	{ok, NChs} ->
	    {ok, State#state{children = NChs}};
	{error, NChs} ->
	    restart(Child, State#state{children = NChs})
    end.

%%-----------------------------------------------------------------
%% Func: terminate_children/2
%% Args: Children = [#child] in termination order
%%       SupName = {local, atom()} | {global, atom()} | {pid(),Mod}
%% Returns: NChildren = [#child] in
%%          startup order (reversed termination order)
%%-----------------------------------------------------------------
terminate_children(Children, SupName) ->
    terminate_children(Children, SupName, []).

terminate_children([Child | Children], SupName, Res) ->
    NChild = do_terminate(Child, SupName),
    terminate_children(Children, SupName, [NChild | Res]);
terminate_children([], _SupName, Res) ->
    Res.

do_terminate(Child, SupName) when Child#child.pid =/= undefined ->
    case shutdown(Child#child.pid,
		  Child#child.shutdown) of
	ok ->
	    Child#child{pid = undefined};
	{error, OtherReason} ->
	    report_error(shutdown_error, OtherReason, Child, SupName),
	    Child#child{pid = undefined}
    end;
do_terminate(Child, _SupName) ->
    Child.

%%-----------------------------------------------------------------
%% Shutdowns a child. We must check the EXIT value 
%% of the child, because it might have died with another reason than
%% the wanted. In that case we want to report the error. We put a 
%% monitor on the child an check for the 'DOWN' message instead of 
%% checking for the 'EXIT' message, because if we check the 'EXIT' 
%% message a "naughty" child, who does unlink(Sup), could hang the 
%% supervisor. 
%% Returns: ok | {error, OtherReason}  (this should be reported)
%%-----------------------------------------------------------------
shutdown(Pid, brutal_kill) ->
  
    case monitor_child(Pid) of
	ok ->
	    exit(Pid, kill),
	    receive
		{'DOWN', _MRef, process, Pid, killed} ->
		    ok;
		{'DOWN', _MRef, process, Pid, OtherReason} ->
		    {error, OtherReason}
	    end;
	{error, Reason} ->      
	    {error, Reason}
    end;

shutdown(Pid, Time) ->
    
    case monitor_child(Pid) of
	ok ->
	    exit(Pid, shutdown), %% Try to shutdown gracefully
	    receive 
		{'DOWN', _MRef, process, Pid, shutdown} ->
		    ok;
		{'DOWN', _MRef, process, Pid, OtherReason} ->
		    {error, OtherReason}
	    after Time ->
		    exit(Pid, kill),  %% Force termination.
		    receive
			{'DOWN', _MRef, process, Pid, OtherReason} ->
			    {error, OtherReason}
		    end
	    end;
	{error, Reason} ->      
	    {error, Reason}
    end.

%% Help function to shutdown/2 switches from link to monitor approach
monitor_child(Pid) ->
    
    %% Do the monitor operation first so that if the child dies 
    %% before the monitoring is done causing a 'DOWN'-message with
    %% reason noproc, we will get the real reason in the 'EXIT'-message
    %% unless a naughty child has already done unlink...
    erlang:monitor(process, Pid),
    unlink(Pid),

    receive
	%% If the child dies before the unlik we must empty
	%% the mail-box of the 'EXIT'-message and the 'DOWN'-message.
	{'EXIT', Pid, Reason} -> 
	    receive 
		{'DOWN', _, process, Pid, _} ->
		    {error, Reason}
	    end
    after 0 -> 
	    %% If a naughty child did unlink and the child dies before
	    %% monitor the result will be that shutdown/2 receives a 
	    %% 'DOWN'-message with reason noproc.
	    %% If the child should die after the unlink there
	    %% will be a 'DOWN'-message with a correct reason
	    %% that will be handled in shutdown/2. 
	    ok   
    end.
    
   
%%-----------------------------------------------------------------
%% Child/State manipulating functions.
%%-----------------------------------------------------------------
state_del_child(#child{pid = Pid}, State) when ?is_simple(State) ->
    NDynamics = ?DICT:erase(Pid, State#state.dynamics),
    State#state{dynamics = NDynamics};
state_del_child(Child, State) ->
    NChildren = del_child(Child#child.name, State#state.children),
    State#state{children = NChildren}.

del_child(Name, [Ch|Chs]) when Ch#child.name =:= Name ->
    [Ch#child{pid = undefined} | Chs];
del_child(Pid, [Ch|Chs]) when Ch#child.pid =:= Pid ->
    [Ch#child{pid = undefined} | Chs];
del_child(Name, [Ch|Chs]) ->
    [Ch|del_child(Name, Chs)];
del_child(_, []) ->
    [].

%% Chs = [S4, S3, Ch, S1, S0]
%% Ret: {[S4, S3, Ch], [S1, S0]}
split_child(Name, Chs) ->
    split_child(Name, Chs, []).

split_child(Name, [Ch|Chs], After) when Ch#child.name =:= Name ->
    {lists:reverse([Ch#child{pid = undefined} | After]), Chs};
split_child(Pid, [Ch|Chs], After) when Ch#child.pid =:= Pid ->
    {lists:reverse([Ch#child{pid = undefined} | After]), Chs};
split_child(Name, [Ch|Chs], After) ->
    split_child(Name, Chs, [Ch | After]);
split_child(_, [], After) ->
    {lists:reverse(After), []}.

get_child(Name, State) ->
    lists:keysearch(Name, #child.name, State#state.children).
replace_child(Child, State) ->
    Chs = do_replace_child(Child, State#state.children),
    State#state{children = Chs}.

do_replace_child(Child, [Ch|Chs]) when Ch#child.name =:= Child#child.name ->
    [Child | Chs];
do_replace_child(Child, [Ch|Chs]) ->
    [Ch|do_replace_child(Child, Chs)].

remove_child(Child, State) ->
    Chs = lists:keydelete(Child#child.name, #child.name, State#state.children),
    State#state{children = Chs}.

%%-----------------------------------------------------------------
%% Func: init_state/4
%% Args: SupName = {local, atom()} | {global, atom()} | self
%%       Type = {Strategy, MaxIntensity, Period}
%%         Strategy = one_for_one | one_for_all | simple_one_for_one |
%%                    rest_for_one 
%%         MaxIntensity = integer()
%%         Period = integer()
%%       Mod :== atom()
%%       Arsg :== term()
%% Purpose: Check that Type is of correct type (!)
%% Returns: {ok, #state} | Error
%%-----------------------------------------------------------------
init_state(SupName, Type, Mod, Args) ->
    case catch init_state1(SupName, Type, Mod, Args) of
	{ok, State} ->
	    {ok, State};
	Error ->
	    Error
    end.

init_state1(SupName, {Strategy, MaxIntensity, Period}, Mod, Args) ->
    validStrategy(Strategy),
    validIntensity(MaxIntensity),
    validPeriod(Period),
    {ok, #state{name = supname(SupName,Mod),
	       strategy = Strategy,
	       intensity = MaxIntensity,
	       period = Period,
	       module = Mod,
	       args = Args}};
init_state1(_SupName, Type, _, _) ->
    {invalid_type, Type}.

validStrategy(simple_one_for_one) -> true;
validStrategy(one_for_one)        -> true;
validStrategy(one_for_all)        -> true;
validStrategy(rest_for_one)       -> true;
validStrategy(What)               -> throw({invalid_strategy, What}).

validIntensity(Max) when is_integer(Max),
                         Max >=  0 -> true;
validIntensity(What)              -> throw({invalid_intensity, What}).

validPeriod(Period) when is_integer(Period),
                         Period > 0 -> true;
validPeriod(What)                   -> throw({invalid_period, What}).

supname(self,Mod) -> {self(),Mod};
supname(N,_)      -> N.

%%% ------------------------------------------------------
%%% Check that the children start specification is valid.
%%% Shall be a six (6) tuple
%%%    {Name, Func, RestartType, Shutdown, ChildType, Modules}
%%% where Name is an atom
%%%       Func is {Mod, Fun, Args} == {atom, atom, list}
%%%       RestartType is permanent | temporary | transient
%%%       Shutdown = integer() | infinity | brutal_kill
%%%       ChildType = supervisor | worker
%%%       Modules = [atom()] | dynamic
%%% Returns: {ok, [#child]} | Error
%%% ------------------------------------------------------

check_startspec(Children) -> check_startspec(Children, []).

check_startspec([ChildSpec|T], Res) ->
    case check_childspec(ChildSpec) of
	{ok, Child} ->
	    case lists:keysearch(Child#child.name, #child.name, Res) of
		{value, _} -> {duplicate_child_name, Child#child.name};
		_ -> check_startspec(T, [Child | Res])
	    end;
	Error -> Error
    end;
check_startspec([], Res) ->
    {ok, lists:reverse(Res)}.

check_childspec({Name, Func, RestartType, Shutdown, ChildType, Mods}) ->
    catch check_childspec(Name, Func, RestartType, Shutdown, ChildType, Mods);
check_childspec(X) -> {invalid_child_spec, X}.

check_childspec(Name, Func, RestartType, Shutdown, ChildType, Mods) ->
    validName(Name),
    validFunc(Func),
    validRestartType(RestartType),
    validChildType(ChildType),
    validShutdown(Shutdown, ChildType),
    validMods(Mods),
    {ok, #child{name = Name, mfa = Func, restart_type = RestartType,
		shutdown = Shutdown, child_type = ChildType, modules = Mods}}.

validChildType(supervisor)  -> true;
validChildType(worker) -> true;
validChildType(What)  -> throw({invalid_child_type, What}).

validName(_Name) -> true. 

validFunc({M, F, A}) when is_atom(M), 
                          is_atom(F), 
                          is_list(A) -> true;
validFunc(Func)                      -> throw({invalid_mfa, Func}).

validRestartType(permanent)   -> true;
validRestartType(temporary)   -> true;
validRestartType(transient)   -> true;
validRestartType(RestartType) -> throw({invalid_restart_type, RestartType}).

validShutdown(Shutdown, _) 
  when is_integer(Shutdown), Shutdown > 0 -> true;
validShutdown(infinity, supervisor)    -> true;
validShutdown(brutal_kill, _)          -> true;
validShutdown(Shutdown, _)             -> throw({invalid_shutdown, Shutdown}).

validMods(dynamic) -> true;
validMods(Mods) when is_list(Mods) ->
    lists:foreach(fun(Mod) ->
		    if
			is_atom(Mod) -> ok;
			true -> throw({invalid_module, Mod})
		    end
		  end,
		  Mods);
validMods(Mods) -> throw({invalid_modules, Mods}).

%%% ------------------------------------------------------
%%% Add a new restart and calculate if the max restart
%%% intensity has been reached (in that case the supervisor
%%% shall terminate).
%%% All restarts accured inside the period amount of seconds
%%% are kept in the #state.restarts list.
%%% Returns: {ok, State'} | {terminate, State'}
%%% ------------------------------------------------------

add_restart(State) ->  
    I = State#state.intensity,
    P = State#state.period,
    R = State#state.restarts,
    Now = erlang:now(),
    R1 = add_restart([Now|R], Now, P),
    State1 = State#state{restarts = R1},
    case length(R1) of
	CurI when CurI  =< I ->
	    {ok, State1};
	_ ->
	    {terminate, State1}
    end.

add_restart([R|Restarts], Now, Period) ->
    case inPeriod(R, Now, Period) of
	true ->
	    [R|add_restart(Restarts, Now, Period)];
	_ ->
	    []
    end;
add_restart([], _, _) ->
    [].

inPeriod(Time, Now, Period) ->
    case difference(Time, Now) of
	T when T > Period ->
	    false;
	_ ->
	    true
    end.

%%
%% Time = {MegaSecs, Secs, MicroSecs} (NOTE: MicroSecs is ignored)
%% Calculate the time elapsed in seconds between two timestamps.
%% If MegaSecs is equal just subtract Secs.
%% Else calculate the Mega difference and add the Secs difference,
%% note that Secs difference can be negative, e.g.
%%      {827, 999999, 676} diff {828, 1, 653753} == > 2 secs.
%%
difference({TimeM, TimeS, _}, {CurM, CurS, _}) when CurM > TimeM ->
    ((CurM - TimeM) * 1000000) + (CurS - TimeS);
difference({_, TimeS, _}, {_, CurS, _}) ->
    CurS - TimeS.

%%% ------------------------------------------------------
%%% Error and progress reporting.
%%% ------------------------------------------------------

report_error(Error, Reason, Child, SupName) ->
    ErrorMsg = [{supervisor, SupName},
		{errorContext, Error},
		{reason, Reason},
		{offender, extract_child(Child)}],
    error_logger:error_report(supervisor_report, ErrorMsg).


extract_child(Child) ->
    [{pid, Child#child.pid},
     {name, Child#child.name},
     {mfa, Child#child.mfa},
     {restart_type, Child#child.restart_type},
     {shutdown, Child#child.shutdown},
     {child_type, Child#child.child_type}].

report_progress(Child, SupName) ->
    Progress = [{supervisor, SupName},
		{started, extract_child(Child)}],
    error_logger:info_report(progress, Progress).