ports/105213: mail/p5-Mail-SpamAssassin: Dies occasionally: prefork: ordered...

[Larry Rosenman]

>Number:         105213
>Category:       ports
>Synopsis:       mail/p5-Mail-SpamAssassin: Dies occasionally: prefork: ordered...
>Confidential:   no
>Severity:       non-critical
>Priority:       high
>Responsible:    freebsd-ports-bugs
>State:          open
>Quarter:        
>Keywords:       
>Date-Required:
>Class:          change-request
>Submitter-Id:   current-users
>Arrival-Date:   Mon Nov 06 15:20:27 GMT 2006
>Closed-Date:
>Last-Modified:
>Originator:     Larry Rosenman
>Release:        FreeBSD 6.2-PRERELEASE amd64
>Organization:
LERCTR Consulting
>Environment:
System: FreeBSD thebighonker.lerctr.org 6.2-PRERELEASE FreeBSD 6.2-PRERELEASE #80: Sat Nov 4 20:57:14 CST 2006 root at thebighonker.lerctr.org:/usr/obj/usr/src/sys/THEBIGHONKER amd64


>Description:
spamd will occasionally die with "prefork: ordered child to accept, but child reported state '1'.

The SA guys have a bug opened (4594), and it's been going on for a while. 

They have a patch, which seems to fix it at:
http://issues.apache.org/SpamAssassin/attachment.cgi?id=3713&action=view
(I've added the patch below, but the link is better if the MTA's mangle
it).



Can we get this patch into the port?
>How-To-Repeat:
It's very spotty.  I can go weeks without seeing it, then see tons of it in
a short time. 

>Fix:

Index: lib/Mail/SpamAssassin/SpamdForkScaling.pm
===================================================================
--- lib/Mail/SpamAssassin/SpamdForkScaling.pm	(revision 462907)
+++ lib/Mail/SpamAssassin/SpamdForkScaling.pm	(working copy)
@@ -426,7 +426,13 @@
     dbg("prefork: ordered $kid to accept");
 
     # now wait for it to say it's done that
-    return $self->wait_for_child_to_accept($sock);
+    my $ret = $self->wait_for_child_to_accept($kid, $sock);
+    if ($ret) {
+      return $ret;
+    } else {
+      # retry with another child
+      return $self->order_idle_child_to_accept();
+    }
 
   }
   else {
@@ -436,10 +442,11 @@
 }
 
 sub wait_for_child_to_accept {
-  my ($self, $sock) = @_;
+  my ($self, $kid, $sock) = @_;
 
   while (1) {
     my $state = $self->read_one_message_from_child_socket($sock);
+
     if ($state == PFSTATE_BUSY) {
       return 1;     # 1 == success
     }
@@ -447,7 +454,12 @@
       return undef;
     }
     else {
-      die "prefork: ordered child to accept, but child reported state '$state'";
+      warn "prefork: ordered child $kid to accept, but they reported state '$state', killing rogue";
+      $self->child_error_kill($kid, $sock);
+      $self->adapt_num_children();
+      sleep 1;
+
+      return undef;
     }
   }
 }
@@ -479,6 +491,8 @@
 
 sub update_child_status_busy {
   my ($self) = @_;
+
+# if (rand 4 < 1) { $self->report_backchannel_socket("I".pack("N",$self->{pid})."\n");return; warn "TEST for bug 4594"; die; }
   # "B  b1 b2 b3 b4 \n "
   $self->report_backchannel_socket("B".pack("N",$self->{pid})."\n");
 }

>Release-Note:
>Audit-Trail:
>Unformatted: