%% ------------------------------------------------------------------- %% %% Copyright (c) 2013 Basho Technologies, Inc. All Rights Reserved. %% %% This file is provided to you under the Apache License, %% Version 2.0 (the "License"); you may not use this file %% except in compliance with the License. You may obtain %% a copy of the License at %% %% http://www.apache.org/licenses/LICENSE-2.0 %% %% Unless required by applicable law or agreed to in writing, %% software distributed under the License is distributed on an %% "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY %% KIND, either express or implied. See the License for the %% specific language governing permissions and limitations %% under the License. %% %% ------------------------------------------------------------------- %% @doc %% This module implements a central storage manager for riak_ensemble. %% Previously, individual ensembles as well as the ensemble manager would %% independently save their own state to disk. However, such an approach %% scaled poorly as the number of independent ensembles increased. It was %% not uncommon to see thousands of synchronous writes issued to disk per %% second, overwhelming the I/O subsystem. To solve this issue, this storage %% manager was created. %% %% Rather than storing data independently, the storage manager combines the %% state from multiple ensembles as well as the ensemble manager into a %% single entity that is stored together in a single file. Since this file %% is now a critical single point of failure, the storage manager uses the %% new {@link riak_ensemble_save} logic to save this data to disk such that %% there are four redundant copies to recover from. %% %% This manager is also responsible for coalescing multiple writes together %% to reduce disk traffic. Individual writes are staged in an ETS table and %% then flushed to disk after a delay (eg. 50ms). %% %% There are two ways to save data to disk that are used by other components %% in riak_ensemble: synchronous and asynchronous. %% %% For synchronous writes components use the sequence: %% riak_ensemble_storage:put(Key, Data), %% riak_ensemble_storage:sync(). %% The sync() call than blocks until the data has successfully been written, %% to disk. %% %% For asynchronous writes, components simply use put() without sync(). The %% data will then be written to disk either when another component calls sync, %% or after next storage manager tick (eg. every 5 seconds). %% -module(riak_ensemble_storage). -behaviour(gen_server). %% API -export([start_link/0]). -export([get/1, put/2, sync/0]). %% gen_server callbacks -export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]). -define(ETS, ets_riak_ensemble_storage). -define(SYNC_DELAY, riak_ensemble_config:storage_delay()). -define(TICK, riak_ensemble_config:storage_tick()). -type gen_server_from() :: any(). -record(state, {savefile :: file:filename(), waiting :: [gen_server_from()], previous :: binary(), timer :: reference()}). -type state() :: #state{}. %%%=================================================================== %%% API %%%=================================================================== start_link() -> gen_server:start_link({local, ?MODULE}, ?MODULE, [], []). -spec sync() -> ok. sync() -> gen_server:call(?MODULE, sync, infinity). -spec put(term(), term()) -> true. put(Key, Value) -> ets:insert(?ETS, {Key, Value}). -spec get(term()) -> {ok, term()} | not_found. get(Key) -> try Value = ets:lookup_element(?ETS, Key, 2), {ok, Value} catch _:_ -> %% Retry through the server in case data is being loaded gen_server:call(?MODULE, {get, Key}, infinity) end. %%%=================================================================== %%% gen_server callbacks %%%=================================================================== init([]) -> {ok, Root} = application:get_env(riak_ensemble, data_root), File = filename:join([Root, "ensembles", "ensemble_facts"]), _ = ets:new(?ETS, [named_table, public, {read_concurrency, true}, {write_concurrency, true}]), case riak_ensemble_save:read(File) of {ok, Bin} -> Existing = binary_to_term(Bin), true = ets:insert(?ETS, Existing); _ -> ok end, schedule_tick(), {ok, #state{savefile=File, waiting=[], timer=undefined}}. handle_call({get, Key}, _From, State) -> Reply = case ets:lookup(?ETS, Key) of [{_, Value}] -> {ok, Value}; _ -> not_found end, {reply, Reply, State}; handle_call(sync, From, State=#state{waiting=Waiting}) -> Waiting2 = [From|Waiting], State2 = maybe_schedule_sync(State), State3 = State2#state{waiting=Waiting2}, {noreply, State3}; handle_call(_Request, _From, State) -> {reply, ok, State}. handle_cast(_Msg, State) -> {noreply, State}. handle_info(tick, State) -> State2 = tick(State), schedule_tick(), {noreply, State2}; handle_info(do_sync, State) -> {noreply, do_sync(State)}; handle_info(_Info, State) -> {noreply, State}. terminate(_Reason, _State) -> ok. code_change(_OldVsn, State, _Extra) -> {ok, State}. %%%=================================================================== %%% Internal functions %%%=================================================================== -spec schedule_tick() -> ok. schedule_tick() -> _ = erlang:send_after(?TICK, self(), tick), ok. -spec tick(state()) -> state(). tick(State) -> State2 = maybe_schedule_sync(State), State2. -spec maybe_schedule_sync(state()) -> state(). maybe_schedule_sync(State=#state{timer=undefined}) -> Timer = erlang:send_after(?SYNC_DELAY, self(), do_sync), State#state{timer=Timer}; maybe_schedule_sync(State) -> State. -spec do_sync(state()) -> state(). do_sync(State=#state{savefile=File, waiting=Waiting, previous=PrevData}) -> Data = term_to_binary(ets:tab2list(?ETS)), case Data of PrevData -> ok; _ -> ok = riak_ensemble_save:write(File, Data) end, _ = [gen_server:reply(From, ok) || From <- Waiting], State#state{waiting=[], timer=undefined, previous=Data}.