Periodic checkin. Some bug fixes, feature additions, etc.. things seem to

be working well right now.

pelab/bgmon/automanage.pl

@@ -141,7 +141,7 @@ while(1)
 
     outputErrors();
 
-    sleep( 60 );
+    sleep( 60*5 );
     $f_firsttime = 0;
 }
 
@@ -195,7 +195,7 @@ sub choosenodes
 	if( $bestnode ne "NONE" &&
 	    !defined $intersitenodes{$site} )
 	{
-	    print "SECTION 1: adding $bestnode at $site\n";
+	    print time()." SECTION 1: adding $bestnode at $site\n";
 	    # ** This section handles when a site is seen for the 1st time
 
 	    #set new node to represent this site
@@ -205,7 +205,7 @@ sub choosenodes
 	}
 	elsif( ("NONE" eq $bestnode) && defined $intersitenodes{$site} )
 	{
-	    print "SECTION 2: removing tests to $site / ".
+	    print time()." SECTION 2: removing tests to $site / ".
 		"$intersitenodes{$site} \n";
 	    # ** This section handles when a site has no nodes available
 
@@ -221,7 +221,7 @@ sub choosenodes
 	       #&& isnodeinconstrset($bestnode) 
 	       )
 	{
-	    print "SECTION 3: node change at $site from ".
+	    print time()." SECTION 3: node change at $site from ".
 		"$intersitenodes{$site} to $bestnode\n";
 	    # ** This section handles when a "bestnode" at a site changes
 	    
@@ -341,18 +341,9 @@ sub choosebestnode($)
 	    $flag_siteIncluded = 1;
 #	    print "SETTING SITEINCLUDED=1 for $node at $site\n";
 	}
-        #this command acts like a bgmon "ping" - used to 
-        #determine if bgmon running correctly
-=pod
-        my %cmd = ( expid    => $expid,
-		    cmdtype  => "EDIT",
-		    dstnode  => "NOADDR",
-		    testtype => "bw",
-		    testper  => 0,
-		    limitTime=> 0 );
-=cut
+
 	if( $allnodes{$node}{free} == 1 && isnodeinconstrset($node) ){
-	    print "choosebestnode: considering $node\n";
+#	    print "choosebestnode: considering $node\n";
 #	    print "choosing best node for site $site\n";
 	    #first time thru loop...
 	    if( $bestnode eq "NONE" ){
@@ -492,7 +483,7 @@ sub stoppairtest($$)
     foreach my $testtype (@testtypes){
 	edittest($srcnode, $destnode, 0, $testtype);
     }
-    print "stopping pair tests from $srcnode to $destnode\n";
+#    print "stopping pair tests from $srcnode to $destnode\n";
 }
 
 #

pelab/bgmon/libwanetmon.pm

@@ -10,7 +10,6 @@ use IO::Select;
 require Exporter;
 
 @ISA    = "Exporter";
-
 @EXPORT = qw ( 
 	       %deadnodes
 	       %ERRID
@@ -26,16 +25,24 @@ require Exporter;
 	       );
 
 
+# These errors define specifics of when a measurement value cannot be
+# reported due to some error in the network or at the remote end.
+our %ERRID;
+$ERRID{timeout} = -1;
+$ERRID{ttlexceed} = -2; # was an error for "ping", but is not seen in fping.
+$ERRID{unknown} = -3;   # general error, which cannot be classified into others
+$ERRID{unknownhost} = -4;
+$ERRID{ICMPunreachable} = -5;  #used for all ICMP errors (see fping.c for strs)
+$ERRID{iperfHostUnreachable} = -6;  #for iperf: error flagged by: write1 failed: Broken pipe
+
+our %deadnodes;
+
 my $socket;
 my $sel = IO::Select->new();
 my $port;
 my $expid;
-my %deadnodes;
 
-my %ERRID;
-$ERRID{timeout} = -1;
-$ERRID{ttlexceed} = -2;
-$ERRID{unknown} = -3;
+
 
 
 sub setcmdport($)
@@ -114,12 +121,10 @@ sub sendcmd($$)
 					 PeerAddr => $node,
 					 Timeout  => 1 );
 	if( defined $socket ){
-	    $sel->add($socket);
+#	    $sel->add($socket);
 	    print $socket "$sercmd\n";
-	    #todo: wait for ack;
-	    # timeout period?
 	    $sel->add($socket);
-	    my ($ready) = $sel->can_read(1);
+	    my ($ready) = $sel->can_read(2);  #timeout (seconds) (?)
 	    if( defined($ready) && $ready eq $socket ){
 		my $ack = <$ready>;
 #		chomp $ack;

pelab/bgmon/opsrecv.pl

@@ -43,22 +43,24 @@ my $duplicateKillThresh = 5; #kill node if this many duplicates occur per index
 $| = 1;
 
 sub usage {
-	print "Usage: $0 [-p receiveport] [-a sendport] [-e pid/eid] [-d] [-i]\n";
+	print "Usage: $0 [-p receiveport] [-a sendport] [-e pid/eid]".
+	    " [-d debuglevel] [-i]\n";
 	return 1;
 }
 
 my $debug    = 0;
 my $impotent = 0;
+my $debug = 0;
 my ($port, $sendport, $expid);
 my %opt = ();
-if (!getopts("p:a:e:dih", \%opt)) {
+if (!getopts("p:a:e:d:ih", \%opt)) {
     exit &usage;
 }
 if ($opt{p}) { $port = $opt{p}; } else { $port = 5051; }
 if ($opt{a}) { $sendport = $opt{a}; } else { $sendport = 5050; }
 if ($opt{h}) { exit &usage; }
 if ($opt{e}) { $expid = $opt{e}; } else { $expid = "none"; }
-if ($opt{d}) { $debug = 1; }
+if ($opt{d}) { $debug = $opt{d}; } else { $debug = 0; }
 if ($opt{i}) { $impotent = 1; } 
 
 if (@ARGV !=0) { exit &usage; }
@@ -133,13 +135,13 @@ sub handleincomingmsgs()
 	    return;
 	}
 
-	print "\n";
+	print "\n" if( $debug > 1 );
 	print("linksrc=$linksrc\n".
 	      "linkdest=$linkdest\n".
 	      "testtype =$testtype\n".
 	      "result=$result\n".
 	      "index=$index\n".
-	      "tstamp=$tstamp\n");
+	      "tstamp=$tstamp\n") if( $debug > 1 );
 
 	if( defined $linksrc ){
 	    my $socket_snd = 
@@ -153,7 +155,7 @@ sub handleincomingmsgs()
 	    if( defined %ack && defined $socket_snd ){
 		my $ack_serial = serialize_hash( \%ack );
 		$socket_snd->send($ack_serial);
-		print "**SENT ACK**\n";	    
+		print "**SENT ACK**\n" if( $debug > 1 );
 
 		if( !defined $lasttimestamp{$linksrc}{$index} ||
 		    $tstamp ne $lasttimestamp{$linksrc}{$index} )
@@ -163,16 +165,19 @@ sub handleincomingmsgs()
 				 testtype  => $testtype,
 				 result    => $result,
 				 tstamp    => $tstamp );
-		    #decrement duplicatecnt for corresponding result index
+		    #clear duplicatecnt for corresponding result index
 		    if( defined($duplicatecnt{$linksrc}{$index}) ){
-			$duplicatecnt{$linksrc}{$index}--
-			    if( $duplicatecnt{$linksrc}{$index} > 0 );
-		    }else{
-			$duplicatecnt{$linksrc}{$index} = 0;
+			delete $duplicatecnt{$linksrc}{$index};
 		    }
 		}else{
-		    print "++++++duplicate data\n";
-		    
+		    print("++++++duplicate data\n".
+		          "linksrc=$linksrc\n".
+			  "linkdest=$linkdest\n".
+			  "testtype =$testtype\n".
+			  "result=$result\n".
+			  "index=$index\n".
+			  "tstamp=$tstamp\n") if( $debug > 0);
+
 		    #increment duplicatecnt for this src and index number
 		    if( defined($duplicatecnt{$linksrc}{$index}) ){
 			$duplicatecnt{$linksrc}{$index}++;
@@ -181,7 +186,8 @@ sub handleincomingmsgs()
 			    > $duplicateKillThresh )
 			{
 			    killnode($linksrc);
-			    print "KILLING OFF BGMON at $linksrc\n";
+			    print "KILLING OFF BGMON at $linksrc".
+				" for index $index\n" if( $debug > 0 );
 			    delete $duplicatecnt{$linksrc};
 			}
 		    }else{
@@ -297,7 +303,7 @@ sub SendBatchedInserts()
 	DBQueryWarn($insertions)
 	    if (!$impotent);
 	print "$insertions\n"
-	    if ($debug);
+	    if ($debug > 2);
 	$lastinsert = time();
     }
     $batchsize  = 0;