From 13b55d1010e23e6fa2f11ac5ba273b7bc0be2aef Mon Sep 17 00:00:00 2001
From: Serge Tupchii <serge.tupchii@protonmail.com>
Date: Tue, 18 Jun 2024 19:19:46 +0300
Subject: [PATCH] fix: apply autoheal once majority of (core) nodes are alive
 (reachable)

---
 src/mria_autoheal.erl        |  7 ++++++-
 test/mria_autoheal_SUITE.erl | 26 ++++++++++++++++++++++++++
 2 files changed, 32 insertions(+), 1 deletion(-)

diff --git a/src/mria_autoheal.erl b/src/mria_autoheal.erl
index 0a9b214..247c71c 100644
--- a/src/mria_autoheal.erl
+++ b/src/mria_autoheal.erl
@@ -77,7 +77,7 @@ handle_msg({report_partition, Node}, Autoheal = #autoheal{delay = Delay, timer =
 handle_msg(Msg = {create_splitview, Node}, Autoheal = #autoheal{delay = Delay, timer = TRef})
   when Node =:= node() ->
     ensure_cancel_timer(TRef),
-    case mria_membership:is_all_alive() of
+    case is_majority_alive() of
         true ->
             Nodes = mria_mnesia:db_nodes(),
             RPCResult = erpc:multicall(Nodes, mria_mnesia, running_nodes, []),
@@ -160,3 +160,8 @@ ensure_cancel_timer(undefined) ->
     ok;
 ensure_cancel_timer(TRef) ->
     catch erlang:cancel_timer(TRef).
+
+is_majority_alive() ->
+    All = mria_mnesia:cluster_nodes(all),
+    NotAliveLen = length(All -- [node() | nodes()]),
+    NotAliveLen < (length(All) div 2).
diff --git a/test/mria_autoheal_SUITE.erl b/test/mria_autoheal_SUITE.erl
index 7b2d373..7c77e63 100644
--- a/test/mria_autoheal_SUITE.erl
+++ b/test/mria_autoheal_SUITE.erl
@@ -87,6 +87,32 @@ t_autoheal_with_replicants(Config) when is_list(Config) ->
        end,
        [fun ?MODULE:prop_callbacks/1]).
 
+t_autoheal_majority_reachable(Config) when is_list(Config) ->
+    Cluster = mria_ct:cluster([core, core, core, core, core], [{mria, cluster_autoheal, 200}]),
+    ?check_trace(
+       #{timetrap => 25000},
+       try
+           Nodes = [N1, N2, N3, N4, N5] = mria_ct:start_cluster(mria, Cluster),
+           %% Simulate netsplit
+           true = rpc:cast(N4, erlang, disconnect_node, [N1]),
+           true = rpc:cast(N5, erlang, disconnect_node, [N1]),
+           ok = mria_ct:stop_slave(N5),
+           ok = timer:sleep(1000),
+           AliveMajorityNodes = [N1, N2, N3, N4],
+           %% Wait for autoheal, it should happen automatically:
+           ?retry(1000, 20,
+                  begin
+                      ?assertMatch({AliveMajorityNodes, [N5]}, view(N1)),
+                      ?assertMatch({AliveMajorityNodes, [N5]}, view(N2)),
+                      ?assertMatch({AliveMajorityNodes, [N5]}, view(N3)),
+                      ?assertMatch({AliveMajorityNodes, [N5]}, view(N4))
+                  end),
+           Nodes
+       after
+           ok = mria_ct:teardown_cluster(lists:sublist(Cluster, 4))
+       end,
+       [fun ?MODULE:prop_callbacks/1]).
+
 todo_t_reboot_rejoin(Config) when is_list(Config) -> %% FIXME: Flaky and somewhat broken, disable for now
     CommonEnv = [ {mria, cluster_autoheal, 200}
                 , {mria, db_backend, rlog}