ERL-163: Spurious error report on node exit with TLS distribution

[OTP-Maintainer]

Original reporter: legoscia
Affected version: OTP-19.0
Fixed in version: OTP-19.2
Component: Not Specified
Migrated from: https://bugs.erlang.org/browse/ERL-163

---

When an Erlang node that has an active distribution connection using TLS exits, it prints the following error message:

{noformat}
2016-06-13 15:21:28 supervisor_report   
    supervisor: {local,ssl_connection_sup_dist}
    errorContext: shutdown_error
    reason: {shutdown,normal}
    offender: [{nb_children,1},{id,undefined},{mfargs,{tls_connection,start_link,[]}},{restart_type,temporary},{shutdown,4000},{child_type,worker}]
{noformat}

To reproduce, start two nodes configured for TLS distribution, and in one of them, type:

{noformat}
net_adm:ping(foo@localhost).
q().
{noformat}

I dug into this, and found two possible changes that make this error message go away. The first one makes the supervisor module ignore exit reasons of the form {{❴shutdown, _❵}} just like it ignores the exit reason {{shutdown}}, when terminating {{simple_one_for_one}} children:

{code:diff}
diff --git a/lib/stdlib/src/supervisor.erl b/lib/stdlib/src/supervisor.erl
index a594c66..1b33803 100644
--- a/lib/stdlib/src/supervisor.erl
+++ b/lib/stdlib/src/supervisor.erl
@@ -1081,6 +1081,9 @@ wait_dynamic_children(#child{restart_type=RType} = Child, Pids, Sz,
         {'DOWN', _MRef, process, Pid, shutdown} ->
             wait_dynamic_children(Child, ?SETS:del_element(Pid, Pids), Sz-1,
                                   TRef, EStack);
+        {'DOWN', _MRef, process, Pid, {shutdown, _}} ->
+            wait_dynamic_children(Child, ?SETS:del_element(Pid, Pids), Sz-1,
+                                  TRef, EStack);
 
         {'DOWN', _MRef, process, Pid, normal} when RType =/= permanent ->
             wait_dynamic_children(Child, ?SETS:del_element(Pid, Pids), Sz-1,

{code}

I couldn't find this spelled out in the documentation anywhere, but I seem to remember that a shutdown exit reason in a tuple is supposed to be equivalent to the plain atom.

The other possible fix is to make {{ssl_connection}} exit with just {{shutdown}}:

{code:diff}
diff --git a/lib/ssl/src/ssl_connection.erl b/lib/ssl/src/ssl_connection.erl
index 90e0810..16edefd 100644
--- a/lib/ssl/src/ssl_connection.erl
+++ b/lib/ssl/src/ssl_connection.erl
@@ -848,7 +848,7 @@ handle_call({close, _} = Close, From, StateName, State, Connection) ->
     %% Run terminate before returning so that the reuseaddr
     %% inet-option 
     Result = Connection:terminate(Close, StateName, State),
-    {stop_and_reply, {shutdown, normal},  
+    {stop_and_reply, shutdown,  
      {reply, From, Result}, State};
 handle_call({shutdown, How0}, From, _,
            #state{transport_cb = Transport,

{code}

[OTP-Maintainer]

ingela said:

I think this is a supervisor bug, {shutdown, Reason} should be handled the same a shutdown.
The function wait_dynamic_children should behave as do_restart.

[OTP-Maintainer]

ingela said:

I think this is MW issue.

[OTP-Maintainer]

legoscia said:

I submitted a pull request with the former fix: https://github.com/erlang/otp/pull/1158

[OTP-Maintainer]

siri said:

This is fixed by PR-1158