diff options
| -rw-r--r-- | src/rabbit_autoheal.erl | 21 |
1 files changed, 14 insertions, 7 deletions
diff --git a/src/rabbit_autoheal.erl b/src/rabbit_autoheal.erl index b5d64992..09e9aa6a 100644 --- a/src/rabbit_autoheal.erl +++ b/src/rabbit_autoheal.erl @@ -56,9 +56,17 @@ %% - we are the winner and are waiting for all losing nodes to stop %% before telling them they can restart %% +%% about_to_heal +%% - we are the leader, and have already assigned the winner and +%% losers. We are part of the losers and we wait for the winner_is +%% announcement. This leader-specific state differs from not_healing +%% (the state other losers are in), because the leader could still +%% receive request_start messages: those subsequent requests must be +%% ignored. +%% %% {leader_waiting, OutstandingStops} %% - we are the leader, and have already assigned the winner and losers. -%% We are neither but need to ignore further requests to autoheal. +%% We are neither but need to ignore further requests to autoheal. %% %% restarting %% - we are restarting. Of course the node monitor immediately dies @@ -130,14 +138,12 @@ handle_msg({request_start, Node}, " * Winner: ~p~n" " * Losers: ~p~n", [AllPartitions, Winner, Losers]), - Continue = fun(Msg) -> - handle_msg(Msg, not_healing, Partitions) - end, case node() =:= Winner of - true -> Continue({become_winner, Losers}); + true -> handle_msg({become_winner, Losers}, + not_healing, Partitions); false -> send(Winner, {become_winner, Losers}), %% [0] case lists:member(node(), Losers) of - true -> Continue({winner_is, Winner}); + true -> about_to_heal; false -> {leader_waiting, Losers} end end @@ -165,7 +171,8 @@ handle_msg({become_winner, Losers}, end; handle_msg({winner_is, Winner}, - not_healing, _Partitions) -> + State, _Partitions) + when State =:= not_healing orelse State =:= about_to_heal -> rabbit_log:warning( "Autoheal: we were selected to restart; winner is ~p~n", [Winner]), rabbit_node_monitor:run_outside_applications( |