[Bug 5388] socket errors in spamd, both ipv4 and ipv6

[bugzilla-daemon]

http://issues.apache.org/SpamAssassin/show_bug.cgi?id=5388





------- Additional Comments From [EMAIL PROTECTED]  2007-03-27 12:12 -------
hmm.

@@ -178,10 +178,11 @@
   foreach my $p (@pids) {
     if ($l > $p) { $l = $p };
   }
-  $self->{lowest_idle_pid} = $l;
+  $self->{lowest_idle_pid} = $l if defined $l;
 
+  # make sure not to auto-vivify lowest_idle_pid, so the parent knows no child
has ever gone idle.
   dbg("prefork: new lowest idle kid: ".
-            ($self->{lowest_idle_pid} ? $self->{lowest_idle_pid} : 'none'));
+            ((exists $self->{lowest_idle_pid} && $self->{lowest_idle_pid}) ?
$self->{lowest_idle_pid} : 'none'));
 }

If lowest_idle_pid is set to a value, then this is called and no child is idle,
the *previous* value will *be left intact*.  that's a bad idea, since that
child probably isn't idle anymore....



@@ -296,7 +297,8 @@
     if (!$self->order_idle_child_to_accept()) {
       # dbg("prefork: no idle kids, noting overloaded");
       # there are no idle kids!  we're overloaded, mark that
-      $self->{overloaded} = 1;
+      # Don't mark the server as overloaded if no child has ever gone idle.
+      $self->{overloaded} = 1 if exists $self->{lowest_idle_pid};
     }
     return;
   }

So if the spamd server accepts a connection before any of the initial
children reports "idle", it starts another child?  I'm inclined
to consider that a feature, to be honest ;)  if load is that high,
another child is probably going to be needed.


@@ -532,7 +534,7 @@
     chomp $line;
     if (index ($line, "P") == 0) {  # string starts with "P" = ping
       dbg("prefork: periodic ping from spamd parent");
-      next;
+      return PF_PING_ORDER;
     }
     if (index ($line, "A") == 0) {  # string starts with "A" = accept
       return PFORDER_ACCEPT;
Index: spamd/spamd.raw
===================================================================
--- spamd/spamd.raw     (revision 522978)
+++ spamd/spamd.raw     (working copy)
@@ -1059,7 +1059,12 @@
       if ($scaling) {
         $scaling->update_child_status_idle();
         $orders = $scaling->wait_for_orders(); # and sleep...
-
+        
+        if ($orders eq PF_PING_ORDER) {
+          $i--; #don't count this connection against the max; it's just a ping.
+          next;
+        }
+        
         if ($orders != PFORDER_ACCEPT) {
           info("spamd: unknown order: $orders");
         }

I don't get this either.  AFAICS this just defers the skipping of the 
ping request, up a level in the function call tree, and a few no-ops
later.


@@ -1165,7 +1170,7 @@
   }
 
   # Bah!
-  if ( !$client ) {
+  if ( !$client || !$client->connected() ) {
 
     # this can happen when interrupted by SIGCHLD on Solaris,
     # perl 5.8.0, and some other platforms with -m.

OK, on the other hand, I'm happy to apply this change!   It's in trunk now.





------- You are receiving this mail because: -------
You are the assignee for the bug, or are watching the assignee.