Hooray??! Fix the damn PULSE hangs by using infinity supervisor shutdown times

src/machi_flu_psup.erl

@@ -66,6 +66,9 @@
 -ifdef(PULSE).
 -compile({parse_transform, pulse_instrument}).
 -include_lib("pulse_otp/include/pulse_otp.hrl").
+-define(SHUTDOWN, infinity).
+-else.
+-define(SHUTDOWN, 5000).
 -endif.
 
 %% External API
@@ -80,7 +83,7 @@
 make_package_spec(FluName, TcpPort, DataDir, Props) ->
     {FluName, {machi_flu_psup, start_link,
                [FluName, TcpPort, DataDir, Props]},
-     permanent, 5000, supervisor, []}.
+     permanent, ?SHUTDOWN, supervisor, []}.
 
 start_flu_package(FluName, TcpPort, DataDir, Props) ->
     Spec = make_package_spec(FluName, TcpPort, DataDir, Props),
@@ -111,15 +114,15 @@ init([FluName, TcpPort, DataDir, Props0]) ->
     ProjSpec = {ProjRegName,
                {machi_projection_store, start_link,
                 [ProjRegName, DataDir, FluName]},
-               permanent, 5000, worker, []},
+               permanent, ?SHUTDOWN, worker, []},
     MgrSpec = {make_mgr_supname(FluName),
                {machi_chain_manager1, start_link,
                 [FluName, [], Props]},
-               permanent, 5000, worker, []},
+               permanent, ?SHUTDOWN, worker, []},
     FluSpec = {FluName,
                {machi_flu1, start_link,
                 [ [{FluName, TcpPort, DataDir}|Props] ]},
-               permanent, 5000, worker, []},
+               permanent, ?SHUTDOWN, worker, []},
     {ok, {SupFlags, [ProjSpec, MgrSpec, FluSpec]}}.
 
 make_p_regname(FluName) when is_atom(FluName) ->

src/machi_flu_sup.erl

@@ -33,6 +33,9 @@
 -ifdef(PULSE).
 -compile({parse_transform, pulse_instrument}).
 -include_lib("pulse_otp/include/pulse_otp.hrl").
+-define(SHUTDOWN, infinity).
+-else.
+-define(SHUTDOWN, 5000).
 -endif.
 
 %% API

src/machi_sup.erl

@@ -30,6 +30,9 @@
 -ifdef(PULSE).
 -compile({parse_transform, pulse_instrument}).
 -include_lib("pulse_otp/include/pulse_otp.hrl").
+-define(SHUTDOWN, infinity).
+-else.
+-define(SHUTDOWN, 5000).
 -endif.
 
 %% API
@@ -44,7 +47,11 @@ start_link() ->
     supervisor:start_link({local, ?SERVER}, ?MODULE, []).
 
 init([]) ->
-    erlang:display({flu_sup,self()}),
+    erlang:display({flu_sup,?LINE,self()}),
+    {_, Ps} = process_info(self(), links),
+    erlang:display({flu_sup,self(), links, Ps}),
+    [unlink(P) || P <- Ps],
+    erlang:display({flu_sup,?LINE,self()}),
     RestartStrategy = one_for_one,
     MaxRestarts = 1000,
     MaxSecondsBetweenRestarts = 3600,
@@ -52,7 +59,7 @@ init([]) ->
     SupFlags = {RestartStrategy, MaxRestarts, MaxSecondsBetweenRestarts},
 
     Restart = permanent,
-    Shutdown = 5000,
+    Shutdown = ?SHUTDOWN,
     Type = supervisor,
 
     ServerSup =

test/machi_chain_manager1_pulse.erl

@@ -307,14 +307,17 @@ dump_state() ->
                             [P || P <- Ps,
                                   P#projection_v1.epoch_number /= 0]
                         end} || {Name, Proxy} <- ProxiesDict],
-
+    ?V("~w,", [ [{X,whereis(X)} || X <- [machi_sup, machi_flu_sup, machi_partition_simulator]] ]),
+    ?V("~w,", [catch application:stop(machi)]),
+    [?V("~w,", [timer:sleep(10)]) || _ <- lists:seq(1,50)],
+    ?V("~w,", [ [{X,whereis(X)} || X <- [machi_sup, machi_flu_sup, machi_partition_simulator]] ]),
     ?V(")", []),
     Diag1 = Diag2 = "skip_diags",
     {Report, PrivProjs, lists:flatten([Diag1, Diag2])}
   catch XX:YY ->
         ?V("OUCH: ~p ~p @ ~p\n", [XX, YY, erlang:get_stacktrace()]),
         ?V("Exiting now to move to manual post-mortem....\n", []),
-        erlang:halt(0),
+        erlang:halt(66),
         false
   end.
 
@@ -489,14 +492,16 @@ erlang:display({hard,?MODULE,?LINE,self()}),
 
     (catch unlink(whereis(machi_partition_simulator))),
     [begin
-erlang:display({hard,?MODULE,?LINE,self(),X}),
          Pid = whereis(X),
-         %%%%%%DELME deadlock source? spawn(fun() -> ?QC_FMT("shutdown-~w,", [self()]), (catch X:stop()) end),
+erlang:display({hard,?MODULE,?LINE,self(),X,Pid}),
-         timer:sleep(50),
+okokokokokokwhaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa
-         timer:sleep(10),
+         %% %%%%%%DELME deadlock source? spawn(fun() -> ?QC_FMT("shutdown-~w,", [self()]), (catch X:stop()) end),
-         (catch exit(Pid, shutdown)),
+         %% timer:sleep(50),
-         timer:sleep(1),
+         %% timer:sleep(10),
-         (catch exit(Pid, kill))
+         %% (catch exit(Pid, shutdown)),
+         %% timer:sleep(1),
+         %% (catch exit(Pid, kill))
+     %% end || X <- [machi_partition_simulator] ],
      end || X <- [machi_partition_simulator, machi_flu_sup, machi_sup] ],
     timer:sleep(100),
     ok.