From 79ce4791a326b15bec80e1a3870136548419a212 Mon Sep 17 00:00:00 2001 From: James Wheare Date: Tue, 25 Sep 2012 15:23:55 +0100 Subject: mnesia: Use chained send_after instead of send_interval timer:send_interval behaves badly when resuming from sleep on some platforms. For example, if I sleep for 10 minutes, and have a send_interval running once per minute, when I resume, 10 messages will be sent immediately, eliminating the benefit of only running the work periodically. This is admittedly a separate bug with send_interval, but the workaround is straightforward, and also protects from messages piling up in the queue when the work takes longer than the interval. This patch fixes piled up error reports on resume from sleep: ** WARNING ** Mnesia is overloaded: {dump_log, write_threshold} You'll still be warned if mnesia is overloaded, just not repeatedly. Additionally, erlang:send_after is more efficient than using the timer module equivalent [1] [1] http://www.erlang.org/doc/efficiency_guide/commoncaveats.html#id57251 --- lib/mnesia/src/mnesia_controller.erl | 15 ++++++-- lib/mnesia/src/mnesia_recover.erl | 66 ++++++++++++++++++++---------------- lib/mnesia/src/mnesia_tm.erl | 3 +- 3 files changed, 51 insertions(+), 33 deletions(-) diff --git a/lib/mnesia/src/mnesia_controller.erl b/lib/mnesia/src/mnesia_controller.erl index d488a33d67..ec67d9ec12 100644 --- a/lib/mnesia/src/mnesia_controller.erl +++ b/lib/mnesia/src/mnesia_controller.erl @@ -593,6 +593,12 @@ multicall(Nodes, Msg) -> {PatchedGood, Bad}. %% Make the replies look like rpc:multicalls.. %% rpc:multicall(Nodes, ?MODULE, call, [Msg]). +next_async_dump_log() -> + Interval = mnesia_monitor:get_env(dump_log_time_threshold), + Msg = {next_async_dump_log, time_threshold}, + Ref = erlang:send_after(Interval, self(), Msg), + Ref. + %%%---------------------------------------------------------------------- %%% Callback functions from gen_server %%%---------------------------------------------------------------------- @@ -614,9 +620,7 @@ init([Parent]) -> mnesia_lib:unset(original_nodes), mnesia_recover:connect_nodes(Diff), - Interval = mnesia_monitor:get_env(dump_log_time_threshold), - Msg = {async_dump_log, time_threshold}, - {ok, Ref} = timer:send_interval(Interval, Msg), + Ref = next_async_dump_log(), mnesia_dumper:start_regulator(), Empty = gb_trees:empty(), @@ -1121,6 +1125,11 @@ handle_sync_tabs([], _From) -> %% {stop, Reason, State} (terminate/2 is called) %%---------------------------------------------------------------------- +handle_info({next_async_dump_log, InitBy}, State) -> + async_dump_log(InitBy), + Ref = next_async_dump_log(), + noreply(State#state{dump_log_timer_ref=Ref}); + handle_info({async_dump_log, InitBy}, State) -> Worker = #dump_log{initiated_by = InitBy}, State2 = add_worker(Worker, State), diff --git a/lib/mnesia/src/mnesia_recover.erl b/lib/mnesia/src/mnesia_recover.erl index 4750291a10..b64f428f15 100644 --- a/lib/mnesia/src/mnesia_recover.erl +++ b/lib/mnesia/src/mnesia_recover.erl @@ -45,7 +45,8 @@ note_log_decision/2, outcome/2, start/0, - start_garb/0, + next_garb/0, + next_check_overload/0, still_pending/1, sync_trans_tid_serial/1, sync/0, @@ -91,10 +92,38 @@ start() -> init() -> call(init). -start_garb() -> +next_garb() -> Pid = whereis(mnesia_recover), - {ok, _} = timer:send_interval(timer:minutes(2), Pid, garb_decisions), - {ok, _} = timer:send_interval(timer:seconds(10), Pid, check_overload). + erlang:send_after(timer:minutes(2), Pid, garb_decisions). + +next_check_overload() -> + Pid = whereis(mnesia_recover), + erlang:send_after(timer:seconds(10), Pid, check_overload). + + +do_check_overload(S) -> + %% Time to check if mnesia_tm is overloaded + case whereis(mnesia_tm) of + Pid when is_pid(Pid) -> + Threshold = 100, + Prev = S#state.tm_queue_len, + {message_queue_len, Len} = + process_info(Pid, message_queue_len), + if + Len > Threshold, Prev > Threshold -> + What = {mnesia_tm, message_queue_len, [Prev, Len]}, + mnesia_lib:report_system_event({mnesia_overload, What}), + mnesia_lib:overload_set(mnesia_tm, true), + S#state{tm_queue_len = 0}; + Len > Threshold -> + S#state{tm_queue_len = Len}; + true -> + mnesia_lib:overload_set(mnesia_tm, false), + S#state{tm_queue_len = 0} + end; + undefined -> + S + end. allow_garb() -> cast(allow_garb). @@ -853,34 +882,13 @@ handle_info({connect_nodes, Ns, From}, State) -> handle_call({connect_nodes,Ns},From,State); handle_info(check_overload, S) -> - %% Time to check if mnesia_tm is overloaded - case whereis(mnesia_tm) of - Pid when is_pid(Pid) -> - - Threshold = 100, - Prev = S#state.tm_queue_len, - {message_queue_len, Len} = - process_info(Pid, message_queue_len), - if - Len > Threshold, Prev > Threshold -> - What = {mnesia_tm, message_queue_len, [Prev, Len]}, - mnesia_lib:report_system_event({mnesia_overload, What}), - mnesia_lib:overload_set(mnesia_tm, true), - {noreply, S#state{tm_queue_len = 0}}; - - Len > Threshold -> - {noreply, S#state{tm_queue_len = Len}}; - - true -> - mnesia_lib:overload_set(mnesia_tm, false), - {noreply, S#state{tm_queue_len = 0}} - end; - undefined -> - {noreply, S} - end; + State2 = do_check_overload(S), + next_check_overload(), + {noreply, State2}; handle_info(garb_decisions, State) -> do_garb_decisions(), + next_garb(), {noreply, State}; handle_info({force_decision, Tid}, State) -> diff --git a/lib/mnesia/src/mnesia_tm.erl b/lib/mnesia/src/mnesia_tm.erl index 0af7f55c06..b5b14ac05b 100644 --- a/lib/mnesia/src/mnesia_tm.erl +++ b/lib/mnesia/src/mnesia_tm.erl @@ -103,7 +103,8 @@ init(Parent) -> end, mnesia_schema:purge_tmp_files(), - mnesia_recover:start_garb(), + mnesia_recover:next_garb(), + mnesia_recover:next_check_overload(), ?eval_debug_fun({?MODULE, init}, [{nodes, AllOthers}]), -- cgit v1.2.3