1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
|
%% The contents of this file are subject to the Mozilla Public License
%% Version 1.1 (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.mozilla.org/MPL/
%%
%% Software distributed under the License is distributed on an "AS IS"
%% basis, WITHOUT WARRANTY OF ANY KIND, either express or implied. See
%% the License for the specific language governing rights and
%% limitations under the License.
%%
%% The Original Code is RabbitMQ.
%%
%% The Initial Developer of the Original Code is VMware, Inc.
%% Copyright (c) 2007-2011 VMware, Inc. All rights reserved.
%%
%% This module handles the node-wide memory statistics.
%% It receives statistics from all queues, counts the desired
%% queue length (in seconds), and sends this information back to
%% queues.
-module(rabbit_memory_monitor).
-behaviour(gen_server2).
-export([start_link/0, update/0, register/2, deregister/1,
report_ram_duration/2, stop/0]).
-export([init/1, handle_call/3, handle_cast/2, handle_info/2,
terminate/2, code_change/3]).
-record(process, {pid, reported, sent, callback, monitor}).
-record(state, {timer, %% 'internal_update' timer
queue_durations, %% ets #process
queue_duration_sum, %% sum of all queue_durations
queue_duration_count, %% number of elements in sum
memory_limit, %% how much memory we intend to use
desired_duration %% the desired queue duration
}).
-define(SERVER, ?MODULE).
-define(DEFAULT_UPDATE_INTERVAL, 2500).
-define(TABLE_NAME, ?MODULE).
%% Because we have a feedback loop here, we need to ensure that we
%% have some space for when the queues don't quite respond as fast as
%% we would like, or when there is buffering going on in other parts
%% of the system. In short, we aim to stay some distance away from
%% when the memory alarms will go off, which cause backpressure (of
%% some sort) on producers. Note that all other Thresholds are
%% relative to this scaling.
-define(MEMORY_LIMIT_SCALING, 0.4).
-define(LIMIT_THRESHOLD, 0.5). %% don't limit queues when mem use is < this
%% If all queues are pushed to disk (duration 0), then the sum of
%% their reported lengths will be 0. If memory then becomes available,
%% unless we manually intervene, the sum will remain 0, and the queues
%% will never get a non-zero duration. Thus when the mem use is <
%% SUM_INC_THRESHOLD, increase the sum artificially by SUM_INC_AMOUNT.
-define(SUM_INC_THRESHOLD, 0.95).
-define(SUM_INC_AMOUNT, 1.0).
%% If user disabled vm_memory_monitor, let's assume 1GB of memory we can use.
-define(MEMORY_SIZE_FOR_DISABLED_VMM, 1073741824).
-define(EPSILON, 0.000001). %% less than this and we clamp to 0
%%----------------------------------------------------------------------------
-ifdef(use_specs).
-spec(start_link/0 :: () -> rabbit_types:ok_pid_or_error()).
-spec(update/0 :: () -> 'ok').
-spec(register/2 :: (pid(), {atom(),atom(),[any()]}) -> 'ok').
-spec(deregister/1 :: (pid()) -> 'ok').
-spec(report_ram_duration/2 ::
(pid(), float() | 'infinity') -> number() | 'infinity').
-spec(stop/0 :: () -> 'ok').
-endif.
%%----------------------------------------------------------------------------
%% Public API
%%----------------------------------------------------------------------------
start_link() ->
gen_server2:start_link({local, ?SERVER}, ?MODULE, [], []).
update() ->
gen_server2:cast(?SERVER, update).
register(Pid, MFA = {_M, _F, _A}) ->
gen_server2:call(?SERVER, {register, Pid, MFA}, infinity).
deregister(Pid) ->
gen_server2:cast(?SERVER, {deregister, Pid}).
report_ram_duration(Pid, QueueDuration) ->
gen_server2:call(?SERVER,
{report_ram_duration, Pid, QueueDuration}, infinity).
stop() ->
gen_server2:cast(?SERVER, stop).
%%----------------------------------------------------------------------------
%% Gen_server callbacks
%%----------------------------------------------------------------------------
init([]) ->
MemoryLimit = trunc(?MEMORY_LIMIT_SCALING *
(try
vm_memory_monitor:get_memory_limit()
catch
exit:{noproc, _} -> ?MEMORY_SIZE_FOR_DISABLED_VMM
end)),
{ok, TRef} = timer:apply_interval(?DEFAULT_UPDATE_INTERVAL,
?SERVER, update, []),
Ets = ets:new(?TABLE_NAME, [set, private, {keypos, #process.pid}]),
{ok, internal_update(
#state { timer = TRef,
queue_durations = Ets,
queue_duration_sum = 0.0,
queue_duration_count = 0,
memory_limit = MemoryLimit,
desired_duration = infinity })}.
handle_call({report_ram_duration, Pid, QueueDuration}, From,
State = #state { queue_duration_sum = Sum,
queue_duration_count = Count,
queue_durations = Durations,
desired_duration = SendDuration }) ->
[Proc = #process { reported = PrevQueueDuration }] =
ets:lookup(Durations, Pid),
gen_server2:reply(From, SendDuration),
{Sum1, Count1} =
case {PrevQueueDuration, QueueDuration} of
{infinity, infinity} -> {Sum, Count};
{infinity, _} -> {Sum + QueueDuration, Count + 1};
{_, infinity} -> {Sum - PrevQueueDuration, Count - 1};
{_, _} -> {Sum - PrevQueueDuration + QueueDuration,
Count}
end,
true = ets:insert(Durations, Proc #process { reported = QueueDuration,
sent = SendDuration }),
{noreply, State #state { queue_duration_sum = zero_clamp(Sum1),
queue_duration_count = Count1 }};
handle_call({register, Pid, MFA}, _From,
State = #state { queue_durations = Durations }) ->
MRef = erlang:monitor(process, Pid),
true = ets:insert(Durations, #process { pid = Pid, reported = infinity,
sent = infinity, callback = MFA,
monitor = MRef }),
{reply, ok, State};
handle_call(_Request, _From, State) ->
{noreply, State}.
handle_cast(update, State) ->
{noreply, internal_update(State)};
handle_cast({deregister, Pid}, State) ->
{noreply, internal_deregister(Pid, true, State)};
handle_cast(stop, State) ->
{stop, normal, State};
handle_cast(_Request, State) ->
{noreply, State}.
handle_info({'DOWN', _MRef, process, Pid, _Reason}, State) ->
{noreply, internal_deregister(Pid, false, State)};
handle_info(_Info, State) ->
{noreply, State}.
terminate(_Reason, #state { timer = TRef }) ->
timer:cancel(TRef),
ok.
code_change(_OldVsn, State, _Extra) ->
{ok, State}.
%%----------------------------------------------------------------------------
%% Internal functions
%%----------------------------------------------------------------------------
zero_clamp(Sum) ->
case Sum < ?EPSILON of
true -> 0.0;
false -> Sum
end.
internal_deregister(Pid, Demonitor,
State = #state { queue_duration_sum = Sum,
queue_duration_count = Count,
queue_durations = Durations }) ->
case ets:lookup(Durations, Pid) of
[] -> State;
[#process { reported = PrevQueueDuration, monitor = MRef }] ->
true = case Demonitor of
true -> erlang:demonitor(MRef);
false -> true
end,
{Sum1, Count1} =
case PrevQueueDuration of
infinity -> {Sum, Count};
_ -> {zero_clamp(Sum - PrevQueueDuration),
Count - 1}
end,
true = ets:delete(Durations, Pid),
State #state { queue_duration_sum = Sum1,
queue_duration_count = Count1 }
end.
internal_update(State = #state { memory_limit = Limit,
queue_durations = Durations,
desired_duration = DesiredDurationAvg,
queue_duration_sum = Sum,
queue_duration_count = Count }) ->
MemoryRatio = erlang:memory(total) / Limit,
DesiredDurationAvg1 =
case MemoryRatio < ?LIMIT_THRESHOLD orelse Count == 0 of
true ->
infinity;
false ->
Sum1 = case MemoryRatio < ?SUM_INC_THRESHOLD of
true -> Sum + ?SUM_INC_AMOUNT;
false -> Sum
end,
(Sum1 / Count) / MemoryRatio
end,
State1 = State #state { desired_duration = DesiredDurationAvg1 },
%% only inform queues immediately if the desired duration has
%% decreased
case DesiredDurationAvg1 == infinity orelse
(DesiredDurationAvg /= infinity andalso
DesiredDurationAvg1 >= DesiredDurationAvg) of
true ->
ok;
false ->
true =
ets:foldl(
fun (Proc = #process { reported = QueueDuration,
sent = PrevSendDuration,
callback = {M, F, A} }, true) ->
case (case {QueueDuration, PrevSendDuration} of
{infinity, infinity} ->
true;
{infinity, D} ->
DesiredDurationAvg1 < D;
{D, infinity} ->
DesiredDurationAvg1 < D;
{D1, D2} ->
DesiredDurationAvg1 <
lists:min([D1,D2])
end) of
true ->
ok = erlang:apply(
M, F, A ++ [DesiredDurationAvg1]),
ets:insert(
Durations,
Proc #process {sent = DesiredDurationAvg1});
false ->
true
end
end, true, Durations)
end,
State1.
|
