I wrote a long time ago ...
> After looking at the log files I think I've found a pattern: This
> happens every time when no article was cancelled within the last 60
> minutes.

That's actually peertimeout in inn.conf.

> Is it possible the socket connection is closed after such a timeout?
> I'd add an according socket re-creation in perl-nocem then.

The patch attached checks for a timeout and re-opens the socket if
required. Additionally, perl-nocem will switch to cancel_ctlinnd
in case cancel_nntp fails after sending the Message-ID.

    Christoph

--- /usr/lib/news/bin/perl-nocem	2006-09-26 20:39:17.000000000 +0200
+++ perl-nocem	2007-04-27 23:15:29.000000000 +0200
@@ -38,6 +38,8 @@
 my $use_syslog = 0;
 my $log_open = 0;
 my $nntp_open = 0;
+my $last_cancel = 0;
+my $socket_timeout = $inn::peertimeout - 100;
 
 my $logfile = $inn::pathlog . '/perl-nocem.log';
 
@@ -371,6 +373,11 @@
     my $ids = shift;
     my $r;
     
+    if ($nntp_open and time - $socket_timeout > $last_cancel) {
+	logmsg('Close socket for timeout');
+	close (NNTP);
+	$nntp_open = 0;
+    }
     if (not $nntp_open) {
         use Socket;
         if (not socket(NNTP, PF_UNIX, SOCK_STREAM, 0)) {
@@ -400,8 +407,10 @@
         if (($r = <NNTP>) !~ /^289/) {
             $r =~ s/\r\n$//;
             logmsg("cannot cancel $_: $r", 'error');
+            goto ERR;
         }
     }
+    $last_cancel = time;
     return;
 
 ERR:
@@ -463,7 +472,7 @@
 
 # lint food
 print $inn::pathrun.$inn::pathlog.$inn::pathetc.$inn::newsbin.$inn::pathbin
-    . $inn::pathtmp;
+    . $inn::pathtmp.$inn::syslog_facility.$inn::peertimeout;
 
 __END__