Make link sup sup startup idempotent

By handling `{error, already_started}` which can be seen in the wild
due to distributed race conditions.

Per discussion with @lukebakken.

2 files changed, 30 insertions, 4 deletions

diff --git a/deps/rabbitmq_federation/src/rabbit_federation_exchange_link_sup_sup.erl b/deps/rabbitmq_federation/src/rabbit_federation_exchange_link_sup_sup.erl
index efcb8eaff1..6c415b8432 100644
--- a/deps/rabbitmq_federation/src/rabbit_federation_exchange_link_sup_sup.erl
+++ b/deps/rabbitmq_federation/src/rabbit_federation_exchange_link_sup_sup.erl
@@ -42,7 +42,12 @@ start_child(X) ->
            {id(X), {rabbit_federation_link_sup, start_link, [X]},
             transient, ?SUPERVISOR_WAIT, supervisor,
             [rabbit_federation_link_sup]}) of
-        {ok, _Pid}             -> ok;
+        {ok, _Pid}               -> ok;
+        {error, already_started} ->
+          #exchange{name = ExchangeName} = X,
+          rabbit_log_federation:debug("Federation link for exchange ~p was already started",
+                                      [rabbit_misc:rs(ExchangeName)]),
+          ok;
         %% A link returned {stop, gone}, the link_sup shut down, that's OK.
         {error, {shutdown, _}} -> ok
     end.
@@ -53,7 +58,15 @@ adjust(Reason) ->
     ok.
 
 stop_child(X) ->
-    ok = mirrored_supervisor:terminate_child(?SUPERVISOR, id(X)),
+    case mirrored_supervisor:terminate_child(?SUPERVISOR, id(X)) of
+      ok -> ok;
+      {error, Err} ->
+        #exchange{name = ExchangeName} = X,
+        rabbit_log_federation:warning(
+          "Attempt to stop a federation link for exchange ~p failed: ~p",
+          [rabbit_misc:rs(ExchangeName), Err]),
+        ok
+    end,
     ok = mirrored_supervisor:delete_child(?SUPERVISOR, id(X)).
 
 %%----------------------------------------------------------------------------
diff --git a/deps/rabbitmq_federation/src/rabbit_federation_queue_link_sup_sup.erl b/deps/rabbitmq_federation/src/rabbit_federation_queue_link_sup_sup.erl
index faa94abc3b..9cb0780b60 100644
--- a/deps/rabbitmq_federation/src/rabbit_federation_queue_link_sup_sup.erl
+++ b/deps/rabbitmq_federation/src/rabbit_federation_queue_link_sup_sup.erl
@@ -41,7 +41,12 @@ start_child(Q) ->
            {id(Q), {rabbit_federation_link_sup, start_link, [Q]},
             transient, ?SUPERVISOR_WAIT, supervisor,
             [rabbit_federation_link_sup]}) of
-        {ok, _Pid}             -> ok;
+        {ok, _Pid}               -> ok;
+        {error, already_started} ->
+          QueueName = amqqueue:get_name(Q),
+          rabbit_log_federation:warning("Federation link for queue ~p was already started",
+                                        [rabbit_misc:rs(QueueName)]),
+          ok;
         %% A link returned {stop, gone}, the link_sup shut down, that's OK.
         {error, {shutdown, _}} -> ok
     end.
@@ -52,7 +57,15 @@ adjust(Reason) ->
     ok.
 
 stop_child(Q) ->
-    ok = supervisor2:terminate_child(?SUPERVISOR, id(Q)),
+    case supervisor2:terminate_child(?SUPERVISOR, id(Q)) of
+      ok -> ok;
+      {error, Err} ->
+        QueueName = amqqueue:get_name(Q),
+        rabbit_log_federation:warning(
+          "Attempt to stop a federation link for queue ~p failed: ~p",
+          [rabbit_misc:rs(QueueName), Err]),
+        ok
+    end,
     ok = supervisor2:delete_child(?SUPERVISOR, id(Q)).
 
 %%----------------------------------------------------------------------------