davisp commented on a change in pull request #610: Optimize ddoc cache URL: https://github.com/apache/couchdb/pull/610#discussion_r126491136
########## File path: src/ddoc_cache/src/ddoc_cache_entry.erl ########## @@ -0,0 +1,332 @@ +% Licensed 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. + +-module(ddoc_cache_entry). +-behaviour(gen_server). + + +-export([ + dbname/1, + ddocid/1, + recover/1, + insert/2, + + start_link/2, + shutdown/1, + open/2, + accessed/1, + refresh/1 +]). + +-export([ + init/1, + terminate/2, + handle_call/3, + handle_cast/2, + handle_info/2, + code_change/3 +]). + +-export([ + do_open/1 +]). + + +-include("ddoc_cache.hrl"). + + +-record(st, { + key, + val, + opener, + waiters, + ts, + accessed +}). + + +dbname({Mod, Arg}) -> + Mod:dbname(Arg). + + +ddocid({Mod, Arg}) -> + Mod:ddocid(Arg). + + +recover({Mod, Arg}) -> + Mod:recover(Arg). + + +insert({Mod, Arg}, Value) -> + Mod:insert(Arg, Value). + + +start_link(Key, Default) -> + Pid = proc_lib:spawn_link(?MODULE, init, [{Key, Default}]), + {ok, Pid}. + + +shutdown(Pid) -> + ok = gen_server:call(Pid, shutdown). + + +open(Pid, Key) -> + try + Resp = gen_server:call(Pid, open), + case Resp of + {open_ok, Val} -> + Val; + {open_error, {T, R, S}} -> + erlang:raise(T, R, S) + end + catch exit:_ -> + % Its possible that this process was evicted just + % before we tried talking to it. Just fallback + % to a standard recovery + recover(Key) + end. + + +accessed(Pid) -> + gen_server:cast(Pid, accessed). + + +refresh(Pid) -> + gen_server:cast(Pid, force_refresh). + + +init({Key, undefined}) -> + true = ets:update_element(?CACHE, Key, {#entry.pid, self()}), + St = #st{ + key = Key, + opener = spawn_opener(Key), + waiters = [], + accessed = 1 + }, + ?EVENT(started, Key), + gen_server:enter_loop(?MODULE, [], St); + +init({Key, Default}) -> + Updates = [ + {#entry.val, Default}, + {#entry.pid, self()} + ], + NewTs = os:timestamp(), + true = ets:update_element(?CACHE, Key, Updates), + true = ets:insert(?LRU, {{NewTs, Key, self()}}), + St = #st{ + key = Key, + val = {open_ok, {ok, Default}}, + opener = start_timer(), + waiters = undefined, + ts = NewTs, + accessed = 1 + }, + ?EVENT(default_started, Key), + gen_server:enter_loop(?MODULE, [], St). + + +terminate(_Reason, St) -> + #st{ + key = Key, + opener = Pid, + ts = Ts + } = St, + % We may have already deleted our cache entry + % during shutdown + Pattern = #entry{key = Key, pid = self(), _ = '_'}, + CacheMSpec = [{Pattern, [], [true]}], + true = ets:select_delete(?CACHE, CacheMSpec) < 2, + % We may have already deleted our LRU entry + % during shutdown + if Ts == undefined -> ok; true -> + LruMSpec = [{{{Ts, Key, self()}}, [], [true]}], + true = ets:select_delete(?LRU, LruMSpec) < 2 + end, + % Blow away any current opener if it exists + if not is_pid(Pid) -> ok; true -> + catch exit(Pid, kill) + end, + ok. + + +handle_call(open, From, #st{val = undefined} = St) -> + NewSt = St#st{ + waiters = [From | St#st.waiters] + }, + {noreply, NewSt}; + +handle_call(open, _From, St) -> + {reply, St#st.val, St}; + +handle_call(shutdown, _From, St) -> + remove_from_cache(St), + {stop, normal, ok, St}; + +handle_call(Msg, _From, St) -> + {stop, {bad_call, Msg}, {bad_call, Msg}, St}. + + +handle_cast(accessed, St) -> Review comment: The thing to note here is that we're *not* updating the LRU for every message. There's a drain_accessed call that cleans out the message queue of all accessed messages that are in the mailbox. This means that whatever you see in the message queue at any given point results in a single update_lru call. However, not sending the accessed message in the first place might be the better approach. I'm torn between checking message_queue_len on the entry pid vs adding an accessed count to the `#entry{}` record every ?REFRESH_TIMEOUT period that clients would use to calculate a probability of sending a message. ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
