diff options
| author | Simon MacMullen <simon@rabbitmq.com> | 2013-04-22 16:23:54 +0100 |
|---|---|---|
| committer | Simon MacMullen <simon@rabbitmq.com> | 2013-04-22 16:23:54 +0100 |
| commit | d35534e6a7a0a04fc2ed68061d9c2508864a0865 (patch) | |
| tree | b1f40193c6edefdd551d9f35a355c3402da8b1a4 | |
| parent | 8b7df16ae4b8014181c32b29db2cba53c35b2c24 (diff) | |
| download | rabbitmq-server-bug25471.tar.gz | |
Switch pause_minority mode to making its decisions entirely based on node-upness, not rabbit-application-upness and explain why.bug25471
| -rw-r--r-- | src/rabbit_node_monitor.erl | 20 |
1 files changed, 18 insertions, 2 deletions
diff --git a/src/rabbit_node_monitor.erl b/src/rabbit_node_monitor.erl index ca8e6dbd..7d844c72 100644 --- a/src/rabbit_node_monitor.erl +++ b/src/rabbit_node_monitor.erl @@ -200,6 +200,7 @@ init([]) -> %% writing out the cluster status files - bad things can then %% happen. process_flag(trap_exit, true), + net_kernel:monitor_nodes(true), {ok, _} = mnesia:subscribe(system), {ok, #state{monitors = pmon:new(), subscribers = pmon:new(), @@ -265,6 +266,10 @@ handle_info({'DOWN', _MRef, process, Pid, _Reason}, State = #state{subscribers = Subscribers}) -> {noreply, State#state{subscribers = pmon:erase(Pid, Subscribers)}}; +handle_info({nodedown, Node}, State) -> + ok = handle_dead_node(Node), + {noreply, State}; + handle_info({mnesia_system_event, {inconsistent_database, running_partitioned_network, Node}}, State = #state{partitions = Partitions, @@ -333,6 +338,18 @@ handle_dead_rabbit(Node) -> ok = rabbit_amqqueue:on_node_down(Node), ok = rabbit_alarm:on_node_down(Node), ok = rabbit_mnesia:on_node_down(Node), + ok. + +handle_dead_node(_Node) -> + %% In general in rabbit_node_monitor we care about whether the + %% rabbit application is up rather than the node; we do this so + %% that we can respond in the same way to "rabbitmqctl stop_app" + %% and "rabbitmqctl stop" as much as possible. + %% + %% However, for pause_minority mode we can't do this, since we + %% depend on looking at whether other nodes are up to decide + %% whether to come back up ourselves - if we decide that based on + %% the rabbit application we would go down and never come back. case application:get_env(rabbit, cluster_partition_handling) of {ok, pause_minority} -> case majority() of @@ -347,8 +364,7 @@ handle_dead_rabbit(Node) -> rabbit_log:warning("cluster_partition_handling ~p unrecognised, " "assuming 'ignore'~n", [Term]), ok - end, - ok. + end. await_cluster_recovery() -> rabbit_log:warning("Cluster minority status detected - awaiting recovery~n", |