fix: Force load of ?rlog_sync table

Author: ieQu1

src/mria_mnesia.erl

@@ -1,5 +1,5 @@
 %%--------------------------------------------------------------------
-%% Copyright (c) 2019-2023 EMQ Technologies Co., Ltd. All Rights Reserved.
+%% Copyright (c) 2019-2026 EMQ Technologies Co., Ltd. All Rights Reserved.
 %%
 %% Licensed under the Apache License, Version 2.0 (the "License");
 %% you may not use this file except in compliance with the License.
@@ -264,7 +264,7 @@ copy_table(Name, Storage) ->
 -spec wait_for_tables([mria:table()]) -> ok | {error, _Reason}.
 wait_for_tables(Tables) ->
     ?tp(mria_wait_for_tables, #{tables => Tables}),
-    case mnesia:wait_for_tables(Tables, 30000) of
+    case mnesia:wait_for_tables(Tables, 5_000) of
         ok ->
             ?tp(mria_wait_for_tables_done, #{result => ok}),
             ok;

src/mria_schema.erl

@@ -1,5 +1,5 @@
 %%--------------------------------------------------------------------
-%% Copyright (c) 2021-2023 EMQ Technologies Co., Ltd. All Rights Reserved.
+%% Copyright (c) 2021-2026 EMQ Technologies Co., Ltd. All Rights Reserved.
 %%
 %% Licensed under the Apache License, Version 2.0 (the "License");
 %% you may not use this file except in compliance with the License.
@@ -343,9 +343,9 @@ bootstrap() ->
     %% Ensure replicas are available before starting copy:
     %% If we've managed to sync only mnesia schema up to this point, `copy_table/2` may
     %% fail if other nodes suddenly become unavailable.
-    ok = mria_mnesia:wait_for_tables([?rlog_sync]),
     ok = mria_mnesia:copy_table(?rlog_sync, null_copies),
-    mria_mnesia:wait_for_tables([?schema, ?rlog_sync]),
+    %% Contents of this table are irrelevant, so force load:
+    force_load(?rlog_sync),
     %% Seed the table with the metadata:
     {atomic, _} = mnesia:transaction(fun mnesia:write/3, [?schema, MetaSpec, write], infinity),
     {atomic, _} = mnesia:transaction(fun mnesia:write/3, [?schema, RlogSyncSpec, write], infinity),
@@ -386,3 +386,12 @@ notify_change(Shard, Entry, Subscribers) ->
 -spec create_table(entry()) -> ok | _.
 create_table(#?schema{mnesia_table = Table, storage = Storage, config = Config}) ->
     mria_lib:ensure_tab(mnesia:create_table(Table, [{Storage, [node()]} | Config])).
+
+%% @doc Force load a table. Note: mnesia waits for table implicitly.
+force_load(Table) ->
+    case mnesia:force_load_table(Table) of
+        yes ->
+            ok;
+        Other ->
+            logger:error("Failed to force loading table ~p: ~p", [Table, Other])
+    end.

test/mria_SUITE.erl

@@ -1250,7 +1250,7 @@ t_full_cluster_restart(_) ->
            ok = ?ON(N1, mria:dirty_write(tab1, {tab1, foo, bar})),
            ok = ?ON(N2, mria:dirty_write(tab2, {tab2, foo, bar})),
            ok = ?ON(N3, mria:dirty_write(tab3, {tab3, foo, bar})),
-           %% Restart nodes simulataneously:
+           %% Restart nodes simultaneously:
            ?tp(notice, test_restart_nodes, #{}),
            [ok = slave:stop(N) || N <- Nodes],
            [mria_ct:start_slave(mria, CN) || CN <- Cluster],
@@ -1262,7 +1262,7 @@ t_full_cluster_restart(_) ->
        end,
        []).
 
-%% This function creates an an inter-dependent table load order that
+%% This function creates an inter-dependent table load order that
 %% simulates startup sequence of a release containing multiple OTP
 %% applications with complex dependencies:
 full_restart_load_chain() ->

test/mria_autoclean_SUITE.erl

@@ -43,14 +43,21 @@ t_autoclean(_) ->
     ?check_trace(
        begin
            MaxDownSecs = 10,
-           Cluster = mria_ct:cluster([core, core], []),
+           Cluster = mria_ct:cluster([core, core, replicant], mria_mnesia_test_util:common_env()),
            try
                %% Prepare cluster:
-               [N0, N1] = mria_ct:start_cluster(mria, Cluster),
+               [N0, N1, N3] = mria_ct:start_cluster(mria, Cluster),
                ?assertMatch(
-                   [N0, N1],
+                   [N0, N1, N3],
                    erpc:call(N0, mria, info, [running_nodes])
                ),
+               timer:sleep(1000),
+               %% Create a table to simulate more realistic conditions:
+               ok = erpc:call(N0, mria, create_table,
+                              [?MODULE, [ {type, set}
+                                        , {storage, disc_copies}
+                                        , {rlog_shard, ?MODULE}
+                                        ]]),
                %% Shut down one node:
                mria_ct:stop_slave(N1),
                %% Reconfigure autoclean in the runtime to make sure autoclean
@@ -60,7 +67,7 @@ t_autoclean(_) ->
                %% The remaining node must eventually drop the peer:
                ?assertMatch(
                    {ok, T} when T >= MaxDownSecs,
-                   wait_down(N0, N1, MaxDownSecs * 2)
+                   wait_down(N3, N1, MaxDownSecs * 2)
                )
            after
                mria_ct:teardown_cluster(Cluster)

test/mria_mnesia_test_util.erl

@@ -70,6 +70,5 @@ common_env() ->
     [ {mria, db_backend, rlog}
     , {mria, rlog_startup_shards, [test_shard]}
     , {mria, strict_mode, true}
-    , {mria, rpc_module, gen_rpc}
     , {mria, lb_poll_interval, 100}
     ].