davisp commented on a change in pull request #470: Scheduling Replicator URL: https://github.com/apache/couchdb/pull/470#discussion_r110428830
########## File path: src/couch_replicator/src/couch_replicator_scheduler.erl ########## @@ -0,0 +1,1380 @@ +% 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(couch_replicator_scheduler). +-behaviour(gen_server). +-behaviour(config_listener). +-vsn(1). + +-include("couch_replicator_scheduler.hrl"). +-include("couch_replicator.hrl"). +-include("couch_replicator_api_wrap.hrl"). +-include_lib("couch/include/couch_db.hrl"). + +%% public api +-export([start_link/0, add_job/1, remove_job/1, reschedule/0]). +-export([rep_state/1, find_jobs_by_dbname/1, find_jobs_by_doc/2]). +-export([job_summary/2, health_threshold/0]). +-export([jobs/0, job/1]). + +%% gen_server callbacks +-export([init/1, terminate/2, code_change/3]). +-export([handle_call/3, handle_cast/2, handle_info/2]). +-export([format_status/2]). + +%% config_listener callback +-export([handle_config_change/5, handle_config_terminate/3]). + +%% types +-type event_type() :: added | started | stopped | {crashed, any()}. +-type event() :: {Type:: event_type(), When :: erlang:timestamp()}. +-type history() :: nonempty_list(event()). + +%% definitions +-define(MAX_BACKOFF_EXPONENT, 10). +-define(BACKOFF_INTERVAL_MICROS, 30 * 1000 * 1000). +-define(DEFAULT_HEALTH_THRESHOLD_SEC, 2 * 60). +-define(RELISTEN_DELAY, 5000). + +-define(DEFAULT_MAX_JOBS, 500). +-define(DEFAULT_MAX_CHURN, 20). +-define(DEFAULT_MAX_HISTORY, 20). +-define(DEFAULT_SCHEDULER_INTERVAL, 60000). +-record(state, {interval, timer, max_jobs, max_churn, max_history}). +-record(job, { + id :: job_id() | '$1' | '_', + rep :: #rep{} | '_', + pid :: undefined | pid() | '$1' | '_', + monitor :: undefined | reference() | '_', + history :: history() | '_'}). + +-record(stats_acc, { + now :: erlang:timestamp(), + pending_t = 0 :: non_neg_integer(), + running_t = 0 :: non_neg_integer(), + crashed_t = 0 :: non_neg_integer(), + pending_n = 0 :: non_neg_integer(), + running_n = 0 :: non_neg_integer(), + crashed_n = 0 :: non_neg_integer()}). + + +%% public functions + +-spec start_link() -> {ok, pid()} | ignore | {error, term()}. +start_link() -> + gen_server:start_link({local, ?MODULE}, ?MODULE, [], []). + + +-spec add_job(#rep{}) -> ok. +add_job(#rep{} = Rep) when Rep#rep.id /= undefined -> + Job = #job{ + id = Rep#rep.id, + rep = Rep, + history = [{added, os:timestamp()}]}, + gen_server:call(?MODULE, {add_job, Job}, infinity). + + +-spec remove_job(job_id()) -> ok. +remove_job(Id) -> + gen_server:call(?MODULE, {remove_job, Id}, infinity). + + +-spec reschedule() -> ok. +% Trigger a manual reschedule. Used for testing and/or ops. +reschedule() -> + gen_server:call(?MODULE, reschedule, infinity). + + +-spec rep_state(rep_id()) -> #rep{} | nil. +rep_state(RepId) -> + case (catch ets:lookup_element(?MODULE, RepId, #job.rep)) of + {'EXIT',{badarg, _}} -> + nil; + Rep -> + Rep + end. + + +-spec job_summary(job_id(), non_neg_integer()) -> [_] | nil. +job_summary(JobId, HealthThreshold) -> + case job_by_id(JobId) of + {ok, #job{pid = Pid, history = History, rep = Rep}} -> + ErrorCount = consecutive_crashes(History, HealthThreshold), + {State, Info} = case {Pid, ErrorCount} of + {undefined, 0} -> + {pending, null}; + {undefined, ErrorCount} when ErrorCount > 0 -> + [{{crashed, Error}, _When} | _] = History, + ErrMsg = couch_replicator_utils:rep_error_to_binary(Error), + {crashing, ErrMsg}; + {Pid, ErrorCount} when is_pid(Pid) -> + {running, null} + end, + [ + {source, iolist_to_binary(ejson_url(Rep#rep.source))}, + {target, iolist_to_binary(ejson_url(Rep#rep.target))}, + {state, State}, + {info, Info}, + {error_count, ErrorCount}, + {last_updated, last_updated(History)}, + {start_time, couch_replicator_utils:iso8601(Rep#rep.start_time)}, + {proxy, job_proxy_url(Rep#rep.source)} + ]; + {error, not_found} -> + nil % Job might have just completed + end. + + +job_proxy_url(#httpdb{proxy_url = ProxyUrl}) when is_list(ProxyUrl) -> + list_to_binary(couch_util:url_strip_password(ProxyUrl)); + +job_proxy_url(_Endpoint) -> + null. + + +-spec health_threshold() -> non_neg_integer(). +health_threshold() -> + config:get_integer("replicator", "health_threshold", + ?DEFAULT_HEALTH_THRESHOLD_SEC). + + +-spec find_jobs_by_dbname(binary()) -> list(#rep{}). +find_jobs_by_dbname(DbName) -> + Rep = #rep{db_name = DbName, _ = '_'}, + MatchSpec = #job{id = '$1', rep = Rep, _ = '_'}, + [RepId || [RepId] <- ets:match(?MODULE, MatchSpec)]. + + +-spec find_jobs_by_doc(binary(), binary()) -> list(#rep{}). +find_jobs_by_doc(DbName, DocId) -> + Rep = #rep{db_name = DbName, doc_id = DocId, _ = '_'}, + MatchSpec = #job{id = '$1', rep = Rep, _ = '_'}, + [RepId || [RepId] <- ets:match(?MODULE, MatchSpec)]. + + + + +%% gen_server functions + +init(_) -> + EtsOpts = [named_table, {read_concurrency, true}, {keypos, #job.id}], + ?MODULE = ets:new(?MODULE, EtsOpts), + ok = config:listen_for_changes(?MODULE, nil), + Interval = config:get_integer("replicator", "interval", ?DEFAULT_SCHEDULER_INTERVAL), + MaxJobs = config:get_integer("replicator", "max_jobs", ?DEFAULT_MAX_JOBS), + MaxChurn = config:get_integer("replicator", "max_churn", ?DEFAULT_MAX_CHURN), + MaxHistory = config:get_integer("replicator", "max_history", ?DEFAULT_MAX_HISTORY), + {ok, Timer} = timer:send_after(Interval, reschedule), + State = #state{ + interval = Interval, + max_jobs = MaxJobs, + max_churn = MaxChurn, + max_history = MaxHistory, + timer = Timer + }, + {ok, State}. + + +handle_call({add_job, Job}, _From, State) -> + ok = maybe_remove_job_int(Job#job.id, State), + true = add_job_int(Job), + ok = maybe_start_newly_added_job(Job, State), + couch_stats:increment_counter([couch_replicator, jobs, adds]), + TotalJobs = ets:info(?MODULE, size), + couch_stats:update_gauge([couch_replicator, jobs, total], TotalJobs), + {reply, ok, State}; + +handle_call({remove_job, Id}, _From, State) -> + ok = maybe_remove_job_int(Id, State), + {reply, ok, State}; + +handle_call(reschedule, _From, State) -> + ok = reschedule(State), + {reply, ok, State}; + +handle_call(_, _From, State) -> + {noreply, State}. + + +handle_cast({set_max_jobs, MaxJobs}, State) when is_integer(MaxJobs), MaxJobs >= 0 -> + couch_log:notice("~p: max_jobs set to ~B", [?MODULE, MaxJobs]), + {noreply, State#state{max_jobs = MaxJobs}}; + +handle_cast({set_max_churn, MaxChurn}, State) when is_integer(MaxChurn), MaxChurn > 0 -> + couch_log:notice("~p: max_churn set to ~B", [?MODULE, MaxChurn]), + {noreply, State#state{max_churn = MaxChurn}}; + +handle_cast({set_max_history, MaxHistory}, State) when is_integer(MaxHistory), MaxHistory > 0 -> + couch_log:notice("~p: max_history set to ~B", [?MODULE, MaxHistory]), + {noreply, State#state{max_history = MaxHistory}}; + +handle_cast({set_interval, Interval}, State) when is_integer(Interval), Interval > 0 -> + couch_log:notice("~p: interval set to ~B", [?MODULE, Interval]), + {noreply, State#state{interval = Interval}}; + +handle_cast(_, State) -> + {noreply, State}. + + +handle_info(reschedule, State) -> + ok = reschedule(State), + {ok, cancel} = timer:cancel(State#state.timer), + {ok, Timer} = timer:send_after(State#state.interval, reschedule), + {noreply, State#state{timer = Timer}}; + +handle_info({'DOWN', _Ref, process, Pid, normal}, State) -> + {ok, Job} = job_by_pid(Pid), + couch_log:notice("~p: Job ~p completed normally", [?MODULE, Job#job.id]), + remove_job_int(Job), + update_running_jobs_stats(), + {noreply, State}; + +handle_info({'DOWN', _Ref, process, Pid, Reason}, State) -> + {ok, Job} = job_by_pid(Pid), + ok = handle_crashed_job(Job, Reason, State), + {noreply, State}; + +handle_info(restart_config_listener, State) -> + ok = config:listen_for_changes(?MODULE, nil), + {noreply, State}; + +handle_info(_, State) -> + {noreply, State}. + + +code_change(_OldVsn, State, _Extra) -> + {ok, State}. + + +terminate(_Reason, _State) -> + ok. + + +format_status(_Opt, [_PDict, State]) -> + [{max_jobs, State#state.max_jobs}, + {running_jobs, running_job_count()}, + {pending_jobs, pending_job_count()}]. + + +%% config listener functions + +handle_config_change("replicator", "max_jobs", V, _, S) -> + ok = gen_server:cast(?MODULE, {set_max_jobs, list_to_integer(V)}), + {ok, S}; + +handle_config_change("replicator", "max_churn", V, _, S) -> + ok = gen_server:cast(?MODULE, {set_max_churn, list_to_integer(V)}), + {ok, S}; + +handle_config_change("replicator", "interval", V, _, S) -> + ok = gen_server:cast(?MODULE, {set_interval, list_to_integer(V)}), + {ok, S}; + +handle_config_change("replicator", "max_history", V, _, S) -> + ok = gen_server:cast(?MODULE, {set_history, list_to_integer(V)}), + {ok, S}; + +handle_config_change(_, _, _, _, S) -> + {ok, S}. + + +handle_config_terminate(_, stop, _) -> + ok; + +handle_config_terminate(_, _, _) -> + Pid = whereis(?MODULE), + erlang:send_after(?RELISTEN_DELAY, Pid, restart_config_listener). + + +%% private functions + + +% Handle crashed jobs. Handling differs between transient and permanent jobs. +% Transient jobs are those posted to the _replicate endpoint. They don't have a +% db associated with them. When those jobs crash, they are not restarted. That +% is also consistent with behavior when the node they run on, crashed and they +% do not migrate to other nodes. Permanent jobs are those created from +% replicator documents. Those jobs, once they pass basic validation and end up +% in the scheduler will be retried indefinitely (with appropriate exponential +% backoffs). +-spec handle_crashed_job(#job{}, any(), #state{}) -> ok. +handle_crashed_job(#job{rep = #rep{db_name = null}} = Job, Reason, _State) -> + Msg = "~p : Transient job ~p failed, removing. Error: ~p", + ErrorBinary = couch_replicator_utils:rep_error_to_binary(Reason), + couch_log:error(Msg, [?MODULE, Job#job.id, ErrorBinary]), + remove_job_int(Job), + update_running_jobs_stats(), + ok; + +handle_crashed_job(Job, Reason, State) -> + ok = update_state_crashed(Job, Reason, State), + case couch_replicator_doc_processor:update_docs() of + true -> + couch_replicator_docs:update_error(Job#job.rep, Reason); + false -> + ok + end, + case ets:info(?MODULE, size) < State#state.max_jobs of + true -> + % Starting pending jobs is an O(TotalJobsCount) operation. Only do + % it if there is a relatively small number of jobs. Otherwise + % scheduler could be blocked if there is a cascade of lots failing + % jobs in a row. + start_pending_jobs(State), + update_running_jobs_stats(), + ok; + false -> + ok + end. + + +% Attempt to start a newly added job. First quickly check if total jobs +% already exceed max jobs, then do a more expensive check which runs a +% select (an O(n) operation) to check pending jobs specifically. +-spec maybe_start_newly_added_job(#job{}, #state{}) -> ok. +maybe_start_newly_added_job(Job, State) -> + MaxJobs = State#state.max_jobs, + TotalJobs = ets:info(?MODULE, size), + case TotalJobs < MaxJobs andalso running_job_count() < MaxJobs of + true -> + start_job_int(Job, State), + update_running_jobs_stats(), + ok; + false -> + ok + end. + +% Return up to a given number of oldest, not recently crashed jobs. Try to be +% memory efficient and use ets:foldl to accumulate jobs. +-spec pending_jobs(non_neg_integer()) -> [#job{}]. +pending_jobs(0) -> + % Handle this case as user could set max_churn to 0. If this is passed to + % other function clause it will crash as gb_sets:largest assumes set is not + % empty. + []; + +pending_jobs(Count) when is_integer(Count), Count > 0 -> + Set0 = gb_sets:new(), % [{LastStart, Job},...] + Now = os:timestamp(), + Acc0 = {Set0, Now, Count, health_threshold()}, + {Set1, _, _, _} = ets:foldl(fun pending_fold/2, Acc0, ?MODULE), + [Job || {_Started, Job} <- gb_sets:to_list(Set1)]. + + +pending_fold(Job, {Set, Now, Count, HealthThreshold}) -> + Set1 = case {not_recently_crashed(Job, Now, HealthThreshold), + gb_sets:size(Set) >= Count} of + {true, true} -> + % Job is healthy but already reached accumulated limit, so might + % have to replace one of the accumulated jobs + pending_maybe_replace(Job, Set); + {true, false} -> + % Job is healthy and we haven't reached the limit, so add job + % to accumulator + gb_sets:add_element({last_started(Job), Job}, Set); + {false, _} -> + % This jobs is not healthy (has crashed too recently), so skip it. + Set + end, + {Set1, Now, Count, HealthThreshold}. + + +% Replace Job in the accumulator if it is older than youngest job there. +pending_maybe_replace(Job, Set) -> + Started = last_started(Job), + {Youngest, YoungestJob} = gb_sets:largest(Set), + case Started < Youngest of + true -> + Set1 = gb_sets:delete({Youngest, YoungestJob}, Set), + gb_sets:add_element({Started, Job}, Set1); + false -> + Set + end. + + +start_jobs(Count, State) -> + [start_job_int(Job, State) || Job <- pending_jobs(Count)], + ok. + + +-spec stop_jobs(non_neg_integer(), boolean(), #state{}) -> non_neg_integer(). +stop_jobs(Count, IsContinuous, State) -> + Running0 = running_jobs(), + ContinuousPred = fun(Job) -> is_continuous(Job) =:= IsContinuous end, + Running1 = lists:filter(ContinuousPred, Running0), + Running2 = lists:sort(fun oldest_job_first/2, Running1), + Running3 = lists:sublist(Running2, Count), + length([stop_job_int(Job, State) || Job <- Running3]). + + +oldest_job_first(#job{} = A, #job{} = B) -> + last_started(A) =< last_started(B). + + +not_recently_crashed(#job{history = History}, Now, HealthThreshold) -> + case History of + [{added, _When}] -> + true; + [{stopped, _When} | _] -> + true; + _ -> + LatestCrashT = latest_crash_timestamp(History), + CrashCount = consecutive_crashes(History, HealthThreshold), + timer:now_diff(Now, LatestCrashT) >= backoff_micros(CrashCount) + end. + + +% Count consecutive crashes. A crash happens when there is a `crashed` event +% within a short period of time (configurable) after any other event. It could +% be `crashed, started` for jobs crashing quickly after starting, `crashed, +% crashed`, `crashed, stopped` if job repeatedly failed to start +% being stopped. Or it could be `crashed, added` if it crashed immediately after +% being added during start. +% +% The end of the consecutive crashes ends when a crashed event is seen with +% the time delta between previous greater then the threshold. +-spec consecutive_crashes(history(), non_neg_integer()) -> non_neg_integer(). +consecutive_crashes(History, HealthThreshold) when is_list(History) -> + consecutive_crashes(History, HealthThreshold, 0). + + +-spec consecutive_crashes(history(), non_neg_integer(), non_neg_integer()) -> + non_neg_integer(). +consecutive_crashes([], _HealthThreashold, Count) -> + Count; + +consecutive_crashes([{{crashed, _}, CrashT}, {_, PrevT} = PrevEvent | Rest], + HealthThreshold, Count) -> + case timer:now_diff(CrashT, PrevT) > HealthThreshold * 1000000 of + true -> + Count; + false -> + consecutive_crashes([PrevEvent | Rest], HealthThreshold, Count + 1) + end; + +consecutive_crashes([{stopped, _}, {started, _} | _], _HealthThreshold, Count) -> + Count; + +consecutive_crashes([_ | Rest], HealthThreshold, Count) -> + consecutive_crashes(Rest, HealthThreshold, Count). + + +-spec latest_crash_timestamp(history()) -> erlang:timestamp(). +latest_crash_timestamp([]) -> + {0, 0, 0}; % Used to avoid special-casing "no crash" when doing now_diff + +latest_crash_timestamp([{{crashed, _Reason}, When} | _]) -> + When; + +latest_crash_timestamp([_Event | Rest]) -> + latest_crash_timestamp(Rest). + + +-spec backoff_micros(non_neg_integer()) -> non_neg_integer(). +backoff_micros(CrashCount) -> + BackoffExp = erlang:min(CrashCount - 1, ?MAX_BACKOFF_EXPONENT), + (1 bsl BackoffExp) * ?BACKOFF_INTERVAL_MICROS. + + +-spec add_job_int(#job{}) -> boolean(). +add_job_int(#job{} = Job) -> + ets:insert_new(?MODULE, Job). + + +-spec maybe_remove_job_int(job_id(), #state{}) -> ok. +maybe_remove_job_int(JobId, State) -> + case job_by_id(JobId) of + {ok, Job} -> + ok = stop_job_int(Job, State), + true = remove_job_int(Job), + couch_stats:increment_counter([couch_replicator, jobs, removes]), + TotalJobs = ets:info(?MODULE, size), + couch_stats:update_gauge([couch_replicator, jobs, total], TotalJobs), + update_running_jobs_stats(), + ok; + {error, not_found} -> + ok + end. + + +start_job_int(#job{pid = Pid}, _State) when Pid /= undefined -> + ok; + +start_job_int(#job{} = Job0, State) -> + Job = maybe_optimize_job_for_rate_limiting(Job0), + case couch_replicator_scheduler_sup:start_child(Job#job.rep) of + {ok, Child} -> + Ref = monitor(process, Child), + ok = update_state_started(Job, Child, Ref, State), + couch_log:notice("~p: Job ~p started as ~p", + [?MODULE, Job#job.id, Child]); + {error, {already_started, OtherPid}} when node(OtherPid) =:= node -> Review comment: Certainly you mean `node()` and not the atom `node` here and in the next case statement right? Also, am I missing where we're checking to see if a job exists on a different node somehow? I'm not seeing that logic anywhere (granted this is a fairly large PR I'm still processing). ---------------------------------------------------------------- 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: [email protected] With regards, Apache Git Services
