From 64e6ecf91533fc179a889a74fd3dff3045257dfb Mon Sep 17 00:00:00 2001
From: Mikhail Uvarov <arcusfelis@gmail.com>
Date: Wed, 31 Jan 2024 22:13:05 +0100
Subject: [PATCH] Do ignore_cancel_check if table lists do not match

---
 src/cets_discovery.erl | 25 ++++++++++++++++++-------
 test/cets_SUITE.erl    |  4 ++--
 2 files changed, 20 insertions(+), 9 deletions(-)

diff --git a/src/cets_discovery.erl b/src/cets_discovery.erl
index 398f62ad..955ac7fb 100644
--- a/src/cets_discovery.erl
+++ b/src/cets_discovery.erl
@@ -319,8 +319,8 @@ handle_info({ping_result, Node, Result}, State) ->
     {noreply, handle_ping_result(Node, Result, State)};
 handle_info(enter_regular_phase, State) ->
     {noreply, State#{phase := regular}};
-handle_info({cancel_check, Node}, State) ->
-    {noreply, handle_cancel_check(Node, State)};
+handle_info({cancel_check, Node, Tables}, State) ->
+    {noreply, handle_cancel_check(Node, Tables, State)};
 handle_info(Msg, State) ->
     ?LOG_ERROR(#{what => unexpected_info, msg => Msg}),
     {noreply, State}.
@@ -348,17 +348,28 @@ handle_check(State = #{backend_module := Mod, backend_state := BackendState}) ->
     State#{get_nodes_status := running}.
 
 -spec send_cancel_check_to_other_nodes(state()) -> ok.
-send_cancel_check_to_other_nodes(State = #{nodes := Nodes}) ->
+send_cancel_check_to_other_nodes(State = #{nodes := Nodes, tables := Tables}) ->
     AliveNodes = lists:sort(nodes()),
     AliveKnownNodes = ordsets:intersection(Nodes, AliveNodes),
-    broadcast_to_other_nodes(AliveKnownNodes, {cancel_check, node()}, State).
+    broadcast_to_other_nodes(AliveKnownNodes, {cancel_check, node(), Tables}, State).
 
--spec handle_cancel_check(node(), state()) -> state().
-handle_cancel_check(_FromNode, State) ->
+-spec handle_cancel_check(node(), [cets:table_name()], state()) -> state().
+handle_cancel_check(_FromNode, Tables, State = #{tables := Tables}) ->
     %% Remote node asked us to skip our check to reduce
     %% resource usage and reduce a chance of a race condition.
     %% Start a new timeout for the next check.
-    schedule_check(State).
+    schedule_check(State);
+handle_cancel_check(FromNode, RemoteTables, State = #{tables := Tables}) ->
+    %% A corner case when some nodes have different table lists,
+    %% in this case all nodes need to call join for all tables.
+    ?LOG_INFO(#{
+        what => ignore_cancel_check,
+        text => <<"Table lists do not match, ignore cancel_check">>,
+        remote_tables => RemoteTables,
+        remote_node => FromNode,
+        tables => Tables
+    }),
+    State.
 
 -spec broadcast_to_other_nodes([node()], term(), state()) -> ok.
 broadcast_to_other_nodes([_ | _] = Nodes, Msg, #{name := Name}) ->
diff --git a/test/cets_SUITE.erl b/test/cets_SUITE.erl
index 8c3988ca..d4a6adfa 100644
--- a/test/cets_SUITE.erl
+++ b/test/cets_SUITE.erl
@@ -2834,7 +2834,7 @@ cancel_check_is_sent_after_check(Config) ->
     #{nodes := [Node1, Node2]} = cets_discovery:system_info(Disco),
     %% Force check, so we do not wait for long.
     Disco ! check,
-    receive_message({cancel_check, Node2}).
+    receive_message({cancel_check, Node2, []}).
 
 cancel_check_resets_check_timeout(Config) ->
     DiscoName = disco_name(Config),
@@ -2851,7 +2851,7 @@ cancel_check_resets_check_timeout(Config) ->
     %% Now next check would be not soon.
     %% So, we could be sure nobody would change timer_ref, interfering with our code.
     #{timer_ref := OldTRef} = cets_discovery:system_info(Disco),
-    Disco ! {cancel_check, Node1},
+    Disco ! {cancel_check, Node1, []},
     wait_for_different_timer_ref(Disco, OldTRef).
 
 %% Helper functions