fix(dsrepl): trigger election for new ra servers unconditionallly

Otherwise we might end up in a situation when there's no member online
yet at the time of the election trigger, and the election will never
happen.

apps/emqx_durable_storage/src/emqx_ds_replication_layer_shard.erl

@@ -341,29 +341,21 @@ start_shard(DB, Shard, #{replication_options := ReplicationOpts}) ->
                 log_init_args => LogOpts
             })
     end,
-    case Servers of
+    %% NOTE
-        [LocalServer | _] ->
+    %% Triggering election is necessary when a new consensus group is being brought up.
-            %% TODO
+    %% TODO
-            %% Not super robust, but we probably don't expect nodes to be down
+    %% It's probably a good idea to rebalance leaders across the cluster from time to
-            %% when we bring up a fresh consensus group. Triggering election
+    %% time. There's `ra:transfer_leadership/2` for that.
-            %% is not really required otherwise.
+    try Bootstrap andalso ra:trigger_election(LocalServer, _Timeout = 1_000) of
-            %% TODO
+        false ->
-            %% Ensure that doing that on node restart does not disrupt consensus.
+            ok;
-            %% Edit: looks like it doesn't, this could actually be quite useful
+        ok ->
-            %% to "steal" leadership from nodes that have too much leader load.
+            ok
-            %% TODO
+    catch
-            %% It doesn't really work that way. There's `ra:transfer_leadership/2`
+        %% TODO
-            %% for that.
+        %% Tolerating exceptions because server might be occupied with log replay for
-            try
+        %% a while.
-                ra:trigger_election(LocalServer, _Timeout = 1_000)
+        exit:{timeout, _} when not Bootstrap ->
-            catch
-                %% TODO
-                %% Tolerating exceptions because server might be occupied with log
-                %% replay for a while.
-                exit:{timeout, _} when not Bootstrap ->
-                    ok
-            end;
-        _ ->
             ok
     end.
 

apps/emqx_durable_storage/test/emqx_ds_replication_SUITE.erl

@@ -435,8 +435,8 @@ t_rebalance_offline_restarts(Config) ->
         erpc:multicall(Nodes, emqx_ds, open_db, [?DB, Opts])
     ),
     ?retry(
-        500,
+        1000,
-        10,
+        5,
         ?assertEqual([8 || _ <- Nodes], [n_shards_online(N, ?DB) || N <- Nodes])
     ),