diff options
Diffstat (limited to 'lib/kernel/src/logger_h_common.erl')
| -rw-r--r-- | lib/kernel/src/logger_h_common.erl | 619 |
1 files changed, 373 insertions, 246 deletions
diff --git a/lib/kernel/src/logger_h_common.erl b/lib/kernel/src/logger_h_common.erl index 38ac7d8ffc..16946ff97c 100644 --- a/lib/kernel/src/logger_h_common.erl +++ b/lib/kernel/src/logger_h_common.erl @@ -18,26 +18,359 @@ %% %CopyrightEnd% %% -module(logger_h_common). +-behaviour(gen_server). -include("logger_h_common.hrl"). -include("logger_internal.hrl"). --export([log_to_binary/2, - check_common_config/1, - call_cast_or_drop/4, - check_load/1, - limit_burst/1, - kill_if_choked/5, - flush_log_events/0, - flush_log_events/1, - handler_exit/2, - set_restart_flag/2, - unset_restart_flag/2, - cancel_timer/1, - stop_or_restart/3, - overload_levels_ok/1, - error_notify/1, - info_notify/1]). +%% API +-export([filesync/2]). + +%% logger_olp callbacks +-export([init/1, handle_load/2, handle_call/3, handle_cast/2, handle_info/2, + terminate/2, code_change/3, notify/2, reset_state/1]). + +%% logger callbacks +-export([log/2, adding_handler/1, removing_handler/1, changing_config/3, + filter_config/1]). + +%% Library functions for handlers +-export([error_notify/1]). + +-define(OLP_KEYS,[sync_mode_qlen, + drop_mode_qlen, + flush_qlen, + burst_limit_enable, + burst_limit_max_count, + burst_limit_window_time, + overload_kill_enable, + overload_kill_qlen, + overload_kill_mem_size, + overload_kill_restart_after]). + +-define(COMMON_KEYS,[filesync_repeat_interval]). + +-define(READ_ONLY_KEYS,[olp]). + +%%%----------------------------------------------------------------- +%%% API + +%% This function is called by the logger_sup supervisor +filesync(Module, Name) -> + call(Module, Name, filesync). + +%%%----------------------------------------------------------------- +%%% Handler being added +adding_handler(#{id:=Name,module:=Module}=Config) -> + HConfig0 = maps:get(config, Config, #{}), + HandlerConfig0 = maps:without(?OLP_KEYS++?COMMON_KEYS,HConfig0), + case Module:check_config(Name,set,undefined,HandlerConfig0) of + {ok,HandlerConfig} -> + ModifiedCommon = maps:with(?COMMON_KEYS,HandlerConfig), + CommonConfig0 = maps:with(?COMMON_KEYS,HConfig0), + CommonConfig = maps:merge( + maps:merge(get_default_config(), CommonConfig0), + ModifiedCommon), + case check_config(CommonConfig) of + ok -> + HConfig = maps:merge(CommonConfig,HandlerConfig), + OlpOpts = maps:with(?OLP_KEYS,HConfig0), + start(OlpOpts, Config#{config => HConfig}); + {error,Faulty} -> + {error,{invalid_config,Module,Faulty}} + end; + Error -> + Error + end. + +%%%----------------------------------------------------------------- +%%% Handler being removed +removing_handler(#{id:=Name, module:=Module, config:=#{olp:=Olp}}) -> + case whereis(?name_to_reg_name(Module,Name)) of + undefined -> + ok; + _Pid -> + %% We don't want to do supervisor:terminate_child here + %% since we need to distinguish this explicit stop from a + %% system termination in order to avoid circular attempts + %% at removing the handler (implying deadlocks and + %% timeouts). + %% And we don't need to do supervisor:delete_child, since + %% the restart type is temporary, which means that the + %% child specification is automatically removed from the + %% supervisor when the process dies. + _ = logger_olp:stop(Olp), + ok + end. + +%%%----------------------------------------------------------------- +%%% Updating handler config +changing_config(SetOrUpdate, + #{id:=Name,config:=OldHConfig,module:=Module}, + NewConfig0) -> + NewHConfig0 = maps:get(config, NewConfig0, #{}), + NoHandlerKeys = ?OLP_KEYS++?COMMON_KEYS++?READ_ONLY_KEYS, + OldHandlerConfig = maps:without(NoHandlerKeys,OldHConfig), + NewHandlerConfig0 = maps:without(NoHandlerKeys,NewHConfig0), + case Module:check_config(Name, SetOrUpdate, + OldHandlerConfig,NewHandlerConfig0) of + {ok, NewHandlerConfig} -> + ModifiedCommon = maps:with(?COMMON_KEYS,NewHandlerConfig), + NewCommonConfig0 = maps:with(?COMMON_KEYS,NewHConfig0), + OldCommonConfig = maps:with(?COMMON_KEYS,OldHConfig), + CommonDefault = + case SetOrUpdate of + set -> + get_default_config(); + update -> + OldCommonConfig + end, + NewCommonConfig = maps:merge( + maps:merge(CommonDefault,NewCommonConfig0), + ModifiedCommon), + case check_config(NewCommonConfig) of + ok -> + OlpDefault = + case SetOrUpdate of + set -> + logger_olp:get_default_opts(); + update -> + maps:with(?OLP_KEYS,OldHConfig) + end, + Olp = maps:get(olp,OldHConfig), + NewOlpOpts = maps:merge(OlpDefault, + maps:with(?OLP_KEYS,NewHConfig0)), + case logger_olp:set_opts(Olp,NewOlpOpts) of + ok -> + logger_olp:cast(Olp, {config_changed, + NewCommonConfig, + NewHandlerConfig}), + ReadOnly = maps:with(?READ_ONLY_KEYS,OldHConfig), + NewHConfig = + maps:merge( + maps:merge( + maps:merge(NewCommonConfig,NewHandlerConfig), + ReadOnly), + NewOlpOpts), + NewConfig = NewConfig0#{config=>NewHConfig}, + {ok,NewConfig}; + Error -> + Error + end; + {error,Faulty} -> + {error,{invalid_config,Module,Faulty}} + end; + Error -> + Error + end. + +%%%----------------------------------------------------------------- +%%% Log a string or report +-spec log(LogEvent, Config) -> ok when + LogEvent :: logger:log_event(), + Config :: logger:handler_config(). + +log(LogEvent, Config = #{config := #{olp:=Olp}}) -> + %% if the handler has crashed, we must drop this event + %% and hope the handler restarts so we can try again + true = is_process_alive(logger_olp:get_pid(Olp)), + Bin = log_to_binary(LogEvent, Config), + logger_olp:load(Olp,Bin). + +%%%----------------------------------------------------------------- +%%% Remove internal fields from configuration +filter_config(#{config:=HConfig}=Config) -> + Config#{config=>maps:without(?READ_ONLY_KEYS,HConfig)}. + +%%%----------------------------------------------------------------- +%%% Start the handler process +%%% +%%% The process must always exist if the handler is registered with +%%% logger (and must not exist if the handler is not registered). +%%% +%%% The handler process is linked to logger_sup, which is part of the +%%% kernel application's supervision tree. +start(OlpOpts0, #{id := Name, module:=Module, config:=HConfig} = Config0) -> + RegName = ?name_to_reg_name(Module,Name), + ChildSpec = + #{id => Name, + start => {logger_olp, start_link, [RegName,?MODULE, + Config0, OlpOpts0]}, + restart => temporary, + shutdown => 2000, + type => worker, + modules => [?MODULE]}, + case supervisor:start_child(logger_sup, ChildSpec) of + {ok,Pid,Olp} -> + ok = logger_handler_watcher:register_handler(Name,Pid), + OlpOpts = logger_olp:get_opts(Olp), + {ok,Config0#{config=>(maps:merge(HConfig,OlpOpts))#{olp=>Olp}}}; + {error,{Reason,Ch}} when is_tuple(Ch), element(1,Ch)==child -> + {error,Reason}; + Error -> + Error + end. + +%%%=================================================================== +%%% gen_server callbacks +%%%=================================================================== + +init(#{id := Name, module := Module, config := HConfig}) -> + process_flag(trap_exit, true), + + ?init_test_hooks(), + + case Module:init(Name, HConfig) of + {ok,HState} -> + %% Storing common config in state to avoid copying + %% (sending) the config data for each log message + CommonConfig = maps:with(?COMMON_KEYS,HConfig), + State = CommonConfig#{id => Name, + module => Module, + ctrl_sync_count => + ?CONTROLLER_SYNC_INTERVAL, + last_op => sync, + handler_state => HState}, + State1 = set_repeated_filesync(State), + {ok,State1}; + Error -> + Error + end. + +%% This is the log event. +handle_load(Bin, #{id:=Name, + module:=Module, + handler_state:=HandlerState, + ctrl_sync_count := CtrlSync}=State) -> + if CtrlSync==0 -> + {_,HS1} = Module:write(Name, sync, Bin, HandlerState), + State#{handler_state => HS1, + ctrl_sync_count => ?CONTROLLER_SYNC_INTERVAL, + last_op=>write}; + true -> + {_,HS1} = Module:write(Name, async, Bin, HandlerState), + State#{handler_state => HS1, + ctrl_sync_count => CtrlSync-1, + last_op=>write} + end. + +handle_call(filesync, _From, State = #{id := Name, + module := Module, + handler_state := HandlerState}) -> + {Result,HandlerState1} = Module:filesync(Name,sync,HandlerState), + {reply, Result, State#{handler_state=>HandlerState1, last_op=>sync}}. + +%% If FILESYNC_REPEAT_INTERVAL is set to a millisec value, this +%% clause gets called repeatedly by the handler. In order to +%% guarantee that a filesync *always* happens after the last log +%% event, the repeat operation must be active! +handle_cast(repeated_filesync,State = #{filesync_repeat_interval := no_repeat}) -> + %% This clause handles a race condition which may occur when + %% config changes filesync_repeat_interval from an integer value + %% to no_repeat. + {noreply,State}; +handle_cast(repeated_filesync, + State = #{id := Name, + module := Module, + handler_state := HandlerState, + last_op := LastOp}) -> + State1 = + if LastOp == sync -> + State; + true -> + {_,HS} = Module:filesync(Name, async, HandlerState), + State#{handler_state => HS, last_op => sync} + end, + {noreply,set_repeated_filesync(State1)}; +handle_cast({config_changed, CommonConfig, HConfig}, + State = #{id := Name, + module := Module, + handler_state := HandlerState, + filesync_repeat_interval := OldFSyncInt}) -> + State1 = + case maps:get(filesync_repeat_interval,CommonConfig) of + OldFSyncInt -> + State; + FSyncInt -> + set_repeated_filesync( + cancel_repeated_filesync( + State#{filesync_repeat_interval=>FSyncInt})) + end, + HS = try Module:config_changed(Name, HConfig, HandlerState) + catch error:undef -> HandlerState + end, + {noreply, State1#{handler_state => HS}}. + +handle_info(Info, #{id := Name, module := Module, + handler_state := HandlerState} = State) -> + {noreply,State#{handler_state => Module:handle_info(Name,Info,HandlerState)}}. + +terminate(overloaded=Reason, #{id:=Name}=State) -> + _ = log_handler_info(Name,"Handler ~p overloaded and stopping",[Name],State), + do_terminate(Reason,State), + ConfigResult = logger:get_handler_config(Name), + case ConfigResult of + {ok,#{module:=Module}=HConfig0} -> + spawn(fun() -> logger:remove_handler(Name) end), + HConfig = try Module:filter_config(HConfig0) + catch _:_ -> HConfig0 + end, + {ok,fun() -> logger:add_handler(Name,Module,HConfig) end}; + Error -> + error_notify({Name,restart_impossible,Error}), + Error + end; +terminate(Reason, State) -> + do_terminate(Reason, State). + +do_terminate(Reason, State = #{id := Name, + module := Module, + handler_state := HandlerState}) -> + _ = cancel_repeated_filesync(State), + _ = Module:terminate(Name, Reason, HandlerState), + ok. + +code_change(_OldVsn, State, _Extra) -> + {ok, State}. + +reset_state(#{id:=Name, module:=Module, handler_state:=HandlerState} = State) -> + State#{handler_state=>Module:reset_state(Name, HandlerState)}. + +%%%----------------------------------------------------------------- +%%% Internal functions +call(Module, Name, Op) when is_atom(Name) -> + case logger_olp:call(?name_to_reg_name(Module,Name), Op) of + {error,busy} -> {error,handler_busy}; + Other -> Other + end; +call(_, Name, Op) -> + {error,{badarg,{Op,[Name]}}}. + +notify({mode_change,Mode0,Mode1},#{id:=Name}=State) -> + log_handler_info(Name,"Handler ~p switched from ~p to ~p mode", + [Name,Mode0,Mode1], State); +notify({flushed,Flushed},#{id:=Name}=State) -> + log_handler_info(Name, "Handler ~p flushed ~w log events", + [Name,Flushed], State); +notify(restart,#{id:=Name}=State) -> + log_handler_info(Name, "Handler ~p restarted", [Name], State); +notify(idle,#{id:=Name,module:=Module,handler_state:=HandlerState}=State) -> + {_,HS} = Module:filesync(Name,async,HandlerState), + State#{handler_state=>HS, last_op=>sync}. + +log_handler_info(Name, Format, Args, #{module:=Module, + handler_state:=HandlerState}=State) -> + Config = + case logger:get_handler_config(Name) of + {ok,Conf} -> Conf; + _ -> #{formatter=>{?DEFAULT_FORMATTER,?DEFAULT_FORMAT_CONFIG}} + end, + Meta = #{time=>logger:timestamp()}, + Bin = log_to_binary(#{level => notice, + msg => {Format,Args}, + meta => Meta}, Config), + {_,HS} = Module:write(Name, async, Bin, HandlerState), + State#{handler_state=>HS, last_op=>write}. %%%----------------------------------------------------------------- %%% Convert log data on any form to binary @@ -94,243 +427,37 @@ string_to_binary(String) -> throw(Error) end. - %%%----------------------------------------------------------------- %%% Check that the configuration term is valid -check_common_config({mode_tab,_Tid}) -> - valid; -check_common_config({handler_pid,Pid}) when is_pid(Pid) -> - valid; - -check_common_config({sync_mode_qlen,N}) when is_integer(N) -> - valid; -check_common_config({drop_mode_qlen,N}) when is_integer(N) -> - valid; -check_common_config({flush_qlen,N}) when is_integer(N) -> - valid; - -check_common_config({burst_limit_enable,Bool}) when Bool == true; - Bool == false -> - valid; -check_common_config({burst_limit_max_count,N}) when is_integer(N) -> - valid; -check_common_config({burst_limit_window_time,N}) when is_integer(N) -> - valid; - -check_common_config({overload_kill_enable,Bool}) when Bool == true; - Bool == false -> - valid; -check_common_config({overload_kill_qlen,N}) when is_integer(N) -> - valid; -check_common_config({overload_kill_mem_size,N}) when is_integer(N) -> - valid; -check_common_config({overload_kill_restart_after,NorA}) when is_integer(NorA); - NorA == infinity -> - valid; - -check_common_config({filesync_repeat_interval,NorA}) when is_integer(NorA); - NorA == no_repeat -> - valid; -check_common_config(_) -> - invalid. - - -%%%----------------------------------------------------------------- -%%% Overload Protection -call_cast_or_drop(_Name, HandlerPid, ModeTab, Bin) -> - %% If the handler process is getting overloaded, the log event - %% will be synchronous instead of asynchronous (slows down the - %% logging tempo of a process doing lots of logging. If the - %% handler is choked, drop mode is set and no event will be sent. - try ?get_mode(ModeTab) of - async -> - gen_server:cast(HandlerPid, {log,Bin}); - sync -> - try gen_server:call(HandlerPid, {log,Bin}, ?DEFAULT_CALL_TIMEOUT) of - %% if return value from call == dropped, the - %% message has been flushed by handler and should - %% therefore not be counted as dropped in stats - ok -> ok; - dropped -> ok - catch - _:{timeout,_} -> - ?observe(_Name,{dropped,1}) - end; - drop -> - ?observe(_Name,{dropped,1}) - catch - %% if the ETS table doesn't exist (maybe because of a - %% handler restart), we can only drop the event - _:_ -> ?observe(_Name,{dropped,1}) - end, - ok. - -handler_exit(_Name, Reason) -> - exit(Reason). +check_config(Config) when is_map(Config) -> + check_common_config(maps:to_list(Config)). -set_restart_flag(Name, Module) -> - Flag = list_to_atom(lists:concat([Module,"_",Name,"_restarting"])), - spawn(fun() -> - register(Flag, self()), - timer:sleep(infinity) - end), +check_common_config([{filesync_repeat_interval,NorA}|Config]) + when is_integer(NorA); NorA == no_repeat -> + check_common_config(Config); +check_common_config([{Key,Value}|_]) -> + {error,#{Key=>Value}}; +check_common_config([]) -> ok. -unset_restart_flag(Name, Module) -> - Flag = list_to_atom(lists:concat([Module,"_",Name,"_restarting"])), - case whereis(Flag) of - undefined -> - false; - Pid -> - exit(Pid, kill), - true - end. +get_default_config() -> + #{filesync_repeat_interval => ?FILESYNC_REPEAT_INTERVAL}. -check_load(State = #{id:=_Name, mode_tab := ModeTab, mode := Mode, - sync_mode_qlen := SyncModeQLen, - drop_mode_qlen := DropModeQLen, - flush_qlen := FlushQLen}) -> - {_,Mem} = process_info(self(), memory), - ?observe(_Name,{max_mem,Mem}), - {_,QLen} = process_info(self(), message_queue_len), - ?observe(_Name,{max_qlen,QLen}), - %% When the handler process gets scheduled in, it's impossible - %% to predict the QLen. We could jump "up" arbitrarily from say - %% async to sync, async to drop, sync to flush, etc. However, when - %% the handler process manages the log events (without flushing), - %% one after the other, we will move "down" from drop to sync and - %% from sync to async. This way we don't risk getting stuck in - %% drop or sync mode with an empty mailbox. - {Mode1,_NewDrops,_NewFlushes} = - if - QLen >= FlushQLen -> - {flush, 0,1}; - QLen >= DropModeQLen -> - %% Note that drop mode will force log events to - %% be dropped on the client side (never sent get to - %% the handler). - IncDrops = if Mode == drop -> 0; true -> 1 end, - {?change_mode(ModeTab, Mode, drop), IncDrops,0}; - QLen >= SyncModeQLen -> - {?change_mode(ModeTab, Mode, sync), 0,0}; - true -> - {?change_mode(ModeTab, Mode, async), 0,0} - end, - State1 = ?update_other(drops,DROPS,_NewDrops,State), - {Mode1, QLen, Mem, - ?update_other(flushes,FLUSHES,_NewFlushes, - State1#{last_qlen => QLen})}. - -limit_burst(#{burst_limit_enable := false}) -> - {true,0,0}; -limit_burst(#{burst_win_ts := BurstWinT0, - burst_msg_count := BurstMsgCount, - burst_limit_window_time := BurstLimitWinTime, - burst_limit_max_count := BurstLimitMaxCnt}) -> - if (BurstMsgCount >= BurstLimitMaxCnt) -> - %% the limit for allowed messages has been reached - BurstWinT1 = ?timestamp(), - case ?diff_time(BurstWinT1,BurstWinT0) of - BurstCheckTime when BurstCheckTime < (BurstLimitWinTime*1000) -> - %% we're still within the burst time frame - {false,BurstWinT0,BurstMsgCount}; - _BurstCheckTime -> - %% burst time frame passed, reset counters - {true,BurstWinT1,0} - end; - true -> - %% the limit for allowed messages not yet reached - {true,BurstWinT0,BurstMsgCount+1} - end. - -kill_if_choked(Name, QLen, Mem, HandlerMod, - State = #{overload_kill_enable := KillIfOL, - overload_kill_qlen := OLKillQLen, - overload_kill_mem_size := OLKillMem}) -> - if KillIfOL andalso - ((QLen > OLKillQLen) orelse (Mem > OLKillMem)) -> - HandlerMod:log_handler_info(Name, - "Handler ~p overloaded and stopping", - [Name], State), - set_restart_flag(Name, HandlerMod), - handler_exit(Name, {shutdown,{overloaded,Name,QLen,Mem}}); - true -> - ok - end. +set_repeated_filesync(#{filesync_repeat_interval:=FSyncInt} = State) + when is_integer(FSyncInt) -> + {ok,TRef} = timer:apply_after(FSyncInt, gen_server, cast, + [self(),repeated_filesync]), + State#{rep_sync_tref=>TRef}; +set_repeated_filesync(State) -> + State. -flush_log_events() -> - flush_log_events(-1). - -flush_log_events(Limit) -> - process_flag(priority, high), - Flushed = flush_log_events(0, Limit), - process_flag(priority, normal), - Flushed. - -flush_log_events(Limit, Limit) -> - Limit; -flush_log_events(N, Limit) -> - %% flush log events but leave other events, such as - %% filesync, info and change_config, so that these - %% have a chance to be processed even under heavy load - receive - {'$gen_cast',{log,_}} -> - flush_log_events(N+1, Limit); - {'$gen_call',{Pid,MRef},{log,_}} -> - Pid ! {MRef, dropped}, - flush_log_events(N+1, Limit) - after - 0 -> N +cancel_repeated_filesync(State) -> + case maps:take(rep_sync_tref,State) of + {TRef,State1} -> + _ = timer:cancel(TRef), + State1; + error -> + State end. - -cancel_timer(TRef) when is_atom(TRef) -> ok; -cancel_timer(TRef) -> timer:cancel(TRef). - - -stop_or_restart(Name, {shutdown,Reason={overloaded,_Name,_QLen,_Mem}}, - #{overload_kill_restart_after := RestartAfter}) -> - %% If we're terminating because of an overload situation (see - %% logger_h_common:kill_if_choked/4), we need to remove the handler - %% and set a restart timer. A separate process must perform this - %% in order to avoid deadlock. - HandlerPid = self(), - ConfigResult = logger:get_handler_config(Name), - RemoveAndRestart = - fun() -> - MRef = erlang:monitor(process, HandlerPid), - receive - {'DOWN',MRef,_,_,_} -> - ok - after 30000 -> - error_notify(Reason), - exit(HandlerPid, kill) - end, - case ConfigResult of - {ok,#{module:=HMod}=HConfig} when is_integer(RestartAfter) -> - _ = logger:remove_handler(Name), - _ = timer:apply_after(RestartAfter, logger, add_handler, - [Name,HMod,HConfig]); - {ok,_} -> - _ = logger:remove_handler(Name); - {error,CfgReason} when is_integer(RestartAfter) -> - error_notify({Name,restart_impossible,CfgReason}); - {error,_} -> - ok - end - end, - spawn(RemoveAndRestart), - ok; -stop_or_restart(_Name, _Reason, _State) -> - ok. - -overload_levels_ok(HandlerConfig) -> - SMQL = maps:get(sync_mode_qlen, HandlerConfig, ?SYNC_MODE_QLEN), - DMQL = maps:get(drop_mode_qlen, HandlerConfig, ?DROP_MODE_QLEN), - FQL = maps:get(flush_qlen, HandlerConfig, ?FLUSH_QLEN), - (DMQL > 1) andalso (SMQL =< DMQL) andalso (DMQL =< FQL). - error_notify(Term) -> ?internal_log(error, Term). - -info_notify(Term) -> - ?internal_log(info, Term). |