123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304 |
- -module(syn_netsplits).
- -behaviour(gen_server).
- -export([start_link/0]).
- -export([init/1, handle_call/3, handle_cast/2, handle_info/2, terminate/2, code_change/3]).
- -export([get_processes_info_of_node/1]).
- -export([write_processes_info_to_node/2]).
- -export([conflicting_mode/1]).
- -record(state, {
- conflicting_mode = kill :: kill | send_message,
- message = undefined :: any()
- }).
- -include("syn.hrl").
- -spec start_link() -> {ok, pid()} | {error, any()}.
- start_link() ->
- Options = [],
- gen_server:start_link({local, ?MODULE}, ?MODULE, [], Options).
- -spec conflicting_mode(undefined | kill | {send_message, any()}) -> ok.
- conflicting_mode(undefined) -> ok;
- conflicting_mode(kill) -> gen_server:call(?MODULE, {conflicting_mode, kill});
- conflicting_mode({send_message, Message}) -> gen_server:call(?MODULE, {conflicting_mode, {send_message, Message}}).
- -spec init([]) ->
- {ok, #state{}} |
- {ok, #state{}, Timeout :: non_neg_integer()} |
- ignore |
- {stop, Reason :: any()}.
- init([]) ->
-
- process_flag(trap_exit, true),
-
- mnesia:subscribe(system),
-
- {ok, #state{}}.
- -spec handle_call(Request :: any(), From :: any(), #state{}) ->
- {reply, Reply :: any(), #state{}} |
- {reply, Reply :: any(), #state{}, Timeout :: non_neg_integer()} |
- {noreply, #state{}} |
- {noreply, #state{}, Timeout :: non_neg_integer()} |
- {stop, Reason :: any(), Reply :: any(), #state{}} |
- {stop, Reason :: any(), #state{}}.
- handle_call({conflicting_mode, kill}, _From, State) ->
- {reply, ok, State#state{
- conflicting_mode = kill,
- message = undefined
- }};
- handle_call({conflicting_mode, {send_message, Message}}, _From, State) ->
- {reply, ok, State#state{
- conflicting_mode = send_message,
- message = Message
- }};
- handle_call(Request, From, State) ->
- error_logger:warning_msg("Received from ~p an unknown call message: ~p~n", [Request, From]),
- {reply, undefined, State}.
- -spec handle_cast(Msg :: any(), #state{}) ->
- {noreply, #state{}} |
- {noreply, #state{}, Timeout :: non_neg_integer()} |
- {stop, Reason :: any(), #state{}}.
- handle_cast(Msg, State) ->
- error_logger:warning_msg("Received an unknown cast message: ~p~n", [Msg]),
- {noreply, State}.
- -spec handle_info(Info :: any(), #state{}) ->
- {noreply, #state{}} |
- {noreply, #state{}, Timeout :: non_neg_integer()} |
- {stop, Reason :: any(), #state{}}.
- handle_info({mnesia_system_event, {inconsistent_database, Context, Node}}, #state{
- conflicting_mode = ConflictingMode,
- message = Message
- } = State) ->
- error_logger:warning_msg("MNESIA signalled an inconsistent database on node: ~p with context: ~p, initiating automerge~n", [Node, Context]),
- automerge(Node, ConflictingMode, Message),
- {noreply, State};
- handle_info({mnesia_system_event, {mnesia_down, Node}}, State) when Node =/= node() ->
- error_logger:warning_msg("Received a MNESIA down event, removing all pids of node ~p~n", [Node]),
- delete_pids_of_disconnected_node(Node),
- {noreply, State};
- handle_info({mnesia_system_event, _MnesiaEvent}, State) ->
-
- {noreply, State};
- handle_info(Info, State) ->
- error_logger:warning_msg("Received an unknown info message: ~p~n", [Info]),
- {noreply, State}.
- -spec terminate(Reason :: any(), #state{}) -> terminated.
- terminate(Reason, _State) ->
- error_logger:info_msg("Terminating syn netsplits with reason: ~p~n", [Reason]),
- terminated.
- -spec code_change(OldVsn :: any(), #state{}, Extra :: any()) -> {ok, #state{}}.
- code_change(_OldVsn, State, _Extra) ->
- {ok, State}.
- -spec delete_pids_of_disconnected_node(Node :: atom()) -> pid().
- delete_pids_of_disconnected_node(Node) ->
-
- spawn(fun() ->
-
- MatchHead = #syn_processes_table{key = '$1', node = '$2', _ = '_'},
- Guard = {'=:=', '$2', Node},
- IdFormat = '$1',
-
- DelF = fun(Id) -> mnesia:dirty_delete({syn_processes_table, Id}) end,
- NodePids = mnesia:dirty_select(syn_processes_table, [{MatchHead, [Guard], [IdFormat]}]),
- lists:foreach(DelF, NodePids)
- end).
- -spec automerge(RemoteNode :: atom(), ConflictingMode :: kill | send_message, Message :: any()) -> ok.
- automerge(RemoteNode, ConflictingMode, Message) ->
- global:trans({{?MODULE, automerge}, self()},
- fun() ->
- error_logger:warning_msg("AUTOMERGE starting for remote node ~s (global lock is set)~n", [RemoteNode]),
- check_stitch(RemoteNode, ConflictingMode, Message),
- error_logger:warning_msg("AUTOMERGE done (global lock will be unset)~n")
- end).
- -spec check_stitch(RemoteNode :: atom(), ConflictingMode :: kill | send_message, Message :: any()) -> ok.
- check_stitch(RemoteNode, ConflictingMode, Message) ->
- case catch lists:member(RemoteNode, mnesia:system_info(running_db_nodes)) of
- true ->
- ok;
- false ->
- stitch(RemoteNode, ConflictingMode, Message),
- ok;
- Error ->
- error_logger:error_msg("Could not check if node is stiched: ~p~n", [Error])
- end.
- -spec stitch(RemoteNode :: atom(), ConflictingMode :: kill | send_message, Message :: any()) ->
- {'ok', any()} | {'error', any()}.
- stitch(RemoteNode, ConflictingMode, Message) ->
- mnesia_controller:connect_nodes(
- [RemoteNode],
- fun(MergeF) ->
- catch case MergeF([syn_processes_table]) of
- {merged, _, _} = Res ->
- stitch_tab(RemoteNode, ConflictingMode, Message),
- Res;
- Other ->
- Other
- end
- end).
- -spec stitch_tab(RemoteNode :: atom(), ConflictingMode :: kill | send_message, Message :: any()) -> ok.
- stitch_tab(RemoteNode, ConflictingMode, Message) ->
-
- RemoteProcessesInfo = rpc:call(RemoteNode, ?MODULE, get_processes_info_of_node, [RemoteNode]),
-
- LocalProcessesInfo = get_processes_info_of_node(node()),
-
- {LocalProcessesInfo1, RemoteProcessesInfo1} = purge_double_processes_from_local_node(
- LocalProcessesInfo,
- RemoteProcessesInfo,
- ConflictingMode,
- Message
- ),
-
- write_remote_processes_to_local(RemoteNode, RemoteProcessesInfo1),
- write_local_processes_to_remote(RemoteNode, LocalProcessesInfo1).
- -spec purge_double_processes_from_local_node(
- LocalProcessesInfo :: list(),
- RemoteProcessesInfo :: list(),
- ConflictingMode :: kill | send_message,
- Message :: any()
- ) ->
- {LocalProcessesInfo :: list(), RemoteProcessesInfo :: list()}.
- purge_double_processes_from_local_node(LocalProcessesInfo, RemoteProcessesInfo, ConflictingMode, Message) ->
-
- Tab = ets:new(syn_automerge_doubles_table, [set]),
-
- ets:insert(Tab, LocalProcessesInfo),
-
- F = fun({Key, _RemoteProcessPid}) ->
- case ets:lookup(Tab, Key) of
- [] -> ok;
- [{Key, LocalProcessPid}] ->
- error_logger:warning_msg("Found a double process for ~s, killing it on local node~n", [Key]),
-
- mnesia:dirty_delete(syn_processes_table, Key),
-
- ets:delete(Tab, Key),
-
- case ConflictingMode of
- kill -> exit(LocalProcessPid, kill);
- send_message -> LocalProcessPid ! Message
- end
- end
- end,
- lists:foreach(F, RemoteProcessesInfo),
-
- LocalProcessesInfo1 = ets:tab2list(Tab),
-
- ets:delete(Tab),
-
- {LocalProcessesInfo1, RemoteProcessesInfo}.
- -spec write_remote_processes_to_local(RemoteNode :: atom(), RemoteProcessesInfo :: list()) -> ok.
- write_remote_processes_to_local(RemoteNode, RemoteProcessesInfo) ->
- write_processes_info_to_node(RemoteNode, RemoteProcessesInfo).
- -spec write_local_processes_to_remote(RemoteNode :: atom(), LocalProcessesInfo :: list()) -> ok.
- write_local_processes_to_remote(RemoteNode, LocalProcessesInfo) ->
- ok = rpc:call(RemoteNode, ?MODULE, write_processes_info_to_node, [node(), LocalProcessesInfo]).
- -spec get_processes_info_of_node(Node :: atom()) -> list().
- get_processes_info_of_node(Node) ->
-
- MatchHead = #syn_processes_table{key = '$1', pid = '$2', node = '$3'},
- Guard = {'=:=', '$3', Node},
- ProcessInfoFormat = {{'$1', '$2'}},
-
- mnesia:dirty_select(syn_processes_table, [{MatchHead, [Guard], [ProcessInfoFormat]}]).
- -spec write_processes_info_to_node(Node :: atom(), ProcessesInfo :: list()) -> ok.
- write_processes_info_to_node(Node, ProcessesInfo) ->
- FWrite = fun({Key, ProcessPid}) ->
- mnesia:dirty_write(#syn_processes_table{
- key = Key,
- pid = ProcessPid,
- node = Node
- })
- end,
- lists:foreach(FWrite, ProcessesInfo).
|