Set segment_entry_count per vhost and use a better default

The new default of 2048 was chosen based on various scenarios.
It provides much better memory usage when many queues are used
(allowing one host to go from 500 queues to 800+ queues) and
there seems to be none or negligible performance cost (< 1%)
for single queues.

Author: Loïc Hoguin

deps/rabbit/Makefile

@@ -119,12 +119,7 @@ define PROJECT_ENV
 	    %% interval at which connection/channel tracking executes post operations
 	    {tracking_execution_timeout, 15000},
 	    {stream_messages_soft_limit, 256},
-        {track_auth_attempt_source, false},
-        %% Number of entries per index segment.
-        %% This value can only be changed safely
-        %% on an empty node. Default calculated
-        %% as trunc(math:pow(2,?REL_SEQ_BITS))).
-        {queue_index_segment_entry_count, 16384}
+        {track_auth_attempt_source, false}
 	  ]
 endef
 

deps/rabbit/src/rabbit_queue_index.erl

@@ -24,6 +24,9 @@
          read_global_recovery_terms/1,
          cleanup_global_recovery_terms/0]).
 
+%% Used by rabbit_vhost to set the segment_entry_count.
+-export([all_queue_directory_names/1]).
+
 -define(CLEAN_FILENAME, "clean.dot").
 
 %%----------------------------------------------------------------------------
@@ -280,6 +283,8 @@ reset_state(#qistate{ queue_name     = Name,
                  on_sync_fun(), on_sync_fun()) -> qistate().
 
 init(#resource{ virtual_host = VHost } = Name, OnSyncFun, OnSyncMsgFun) ->
+    #{segment_entry_count := SegmentEntryCount} = rabbit_vhost:read_config(VHost),
+    put(segment_entry_count, SegmentEntryCount),
     VHostDir = rabbit_vhost:msg_store_dir_path(VHost),
     State = #qistate { dir = Dir } = blank_state(VHostDir, Name),
     false = rabbit_file:is_file(Dir), %% is_file == is file or dir
@@ -294,6 +299,8 @@ init(#resource{ virtual_host = VHost } = Name, OnSyncFun, OnSyncMsgFun) ->
 
 recover(#resource{ virtual_host = VHost } = Name, Terms, MsgStoreRecovered,
         ContainsCheckFun, OnSyncFun, OnSyncMsgFun) ->
+    #{segment_entry_count := SegmentEntryCount} = rabbit_vhost:read_config(VHost),
+    put(segment_entry_count, SegmentEntryCount),
     VHostDir = rabbit_vhost:msg_store_dir_path(VHost),
     State = blank_state(VHostDir, Name),
     State1 = State #qistate{on_sync     = OnSyncFun,
@@ -728,6 +735,9 @@ queue_index_walker_reader(QueueName, Gatherer) ->
     ok = gatherer:finish(Gatherer).
 
 scan_queue_segments(Fun, Acc, #resource{ virtual_host = VHost } = QueueName) ->
+    %% Set the segment_entry_count for this worker process.
+    #{segment_entry_count := SegmentEntryCount} = rabbit_vhost:read_config(VHost),
+    put(segment_entry_count, SegmentEntryCount),
     VHostDir = rabbit_vhost:msg_store_dir_path(VHost),
     scan_queue_segments(Fun, Acc, VHostDir, QueueName).
 
@@ -1167,9 +1177,7 @@ array_new(Default) ->
     array:new([{default, Default}, fixed, {size, segment_entry_count()}]).
 
 segment_entry_count() ->
-    {ok, SegmentEntryCount} =
-        application:get_env(rabbit, queue_index_segment_entry_count),
-    SegmentEntryCount.
+    get(segment_entry_count).
 
 bool_to_int(true ) -> 1;
 bool_to_int(false) -> 0.

deps/rabbit/src/rabbit_vhost.erl

@@ -10,12 +10,12 @@
 -include_lib("rabbit_common/include/rabbit.hrl").
 -include("vhost.hrl").
 
--export([recover/0, recover/1]).
+-export([recover/0, recover/1, read_config/1]).
 -export([add/2, add/4, delete/2, exists/1, with/2, with_user_and_vhost/3, assert/1, update/2,
          set_limits/2, vhost_cluster_state/1, is_running_on_all_nodes/1, await_running_on_all_nodes/2,
         list/0, count/0, list_names/0, all/0, parse_tags/1]).
 -export([info/1, info/2, info_all/0, info_all/1, info_all/2, info_all/3]).
--export([dir/1, msg_store_dir_path/1, msg_store_dir_wildcard/0]).
+-export([dir/1, msg_store_dir_path/1, msg_store_dir_wildcard/0, config_file_path/1]).
 -export([delete_storage/1]).
 -export([vhost_down/1]).
 -export([put_vhost/5]).
@@ -48,6 +48,7 @@ recover(VHost) ->
     VHostStubFile = filename:join(VHostDir, ".vhost"),
     ok = rabbit_file:ensure_dir(VHostStubFile),
     ok = file:write_file(VHostStubFile, VHost),
+    ok = ensure_config_file(VHost),
     {Recovered, Failed} = rabbit_amqqueue:recover(VHost),
     AllQs = Recovered ++ Failed,
     QNames = [amqqueue:get_name(Q) || Q <- AllQs],
@@ -57,6 +58,42 @@ recover(VHost) ->
     ok = rabbit_mirror_queue_misc:on_vhost_up(VHost),
     ok.
 
+ensure_config_file(VHost) ->
+    Path = config_file_path(VHost),
+    case filelib:is_regular(Path) of
+        %% The config file exists. Do nothing.
+        true ->
+            ok;
+        %% The config file does not exist.
+        %% Check if there are queues in this vhost.
+        false ->
+            QueueDirs = rabbit_queue_index:all_queue_directory_names(VHost),
+            SegmentEntryCount = case QueueDirs of
+                %% There are no queues. Write the configured value for
+                %% the segment entry count, or the new RabbitMQ default
+                %% introduced in v3.8.17. The new default provides much
+                %% better memory footprint when many queues are used.
+                [] ->
+                    application:get_env(rabbit, queue_index_segment_entry_count,
+                        2048);
+                %% There are queues already. Write the historic RabbitMQ
+                %% default of 16384 for forward compatibility. Historic
+                %% default calculated as trunc(math:pow(2,?REL_SEQ_BITS)).
+                _ ->
+                    16384
+            end,
+            rabbit_log:info("Setting segment_entry_count for vhost '~s' with ~b queues to '~b'",
+                            [VHost, length(QueueDirs), SegmentEntryCount]),
+            file:write_file(Path, io_lib:format(
+                "%% This file is auto-generated! Edit at your own risk!~n"
+                "{segment_entry_count, ~b}.",
+                [SegmentEntryCount]))
+    end.
+
+read_config(VHost) ->
+    {ok, Config} = file:consult(config_file_path(VHost)),
+    maps:from_list(Config).
+
 -define(INFO_KEYS, vhost:info_keys()).
 
 -spec parse_tags(binary() | string() | atom()) -> [atom()].
@@ -377,6 +414,10 @@ msg_store_dir_base() ->
     Dir = rabbit_mnesia:dir(),
     filename:join([Dir, "msg_stores", "vhosts"]).
 
+config_file_path(VHost) ->
+    VHostDir = msg_store_dir_path(VHost),
+    filename:join(VHostDir, ".config").
+
 -spec trim_tag(list() | binary() | atom()) -> atom().
 trim_tag(Val) ->
     rabbit_data_coercion:to_atom(string:trim(rabbit_data_coercion:to_list(Val))).

deps/rabbit/test/backing_queue_SUITE.erl

@@ -527,6 +527,7 @@ bq_queue_index(Config) ->
       ?MODULE, bq_queue_index1, [Config]).
 
 bq_queue_index1(_Config) ->
+    init_queue_index(),
     SegmentSize = rabbit_queue_index:next_segment_boundary(0),
     TwoSegs = SegmentSize + SegmentSize,
     MostOfASegment = trunc(SegmentSize*0.75),
@@ -708,6 +709,7 @@ bq_queue_recover(Config) ->
       ?MODULE, bq_queue_recover1, [Config]).
 
 bq_queue_recover1(Config) ->
+    init_queue_index(),
     Count = 2 * rabbit_queue_index:next_segment_boundary(0),
     QName0 = queue_name(Config, <<"bq_queue_recover-q">>),
     {new, Q} = rabbit_amqqueue:declare(QName0, true, false, [], none, <<"acting-user">>),
@@ -1315,6 +1317,13 @@ with_empty_test_queue(Fun) ->
     {0, 0, Qi} = init_test_queue(QName),
     rabbit_queue_index:delete_and_terminate(Fun(Qi, QName)).
 
+init_queue_index() ->
+    %% We must set the segment entry count in the process dictionary
+    %% for tests that call the queue index directly to have a correct
+    %% value.
+    put(segment_entry_count, 2048),
+    ok.
+
 restart_app() ->
     rabbit:stop(),
     rabbit:start().