Properly detect and report failures while trying to start slothd, and remove...

Properly detect and report failures while trying to start slothd, and remove the lock file before trying to start it if it isn't running.

sensors/slothd/sdisrunning.in

@@ -68,7 +68,7 @@ sub check {
     my $ssh="sshtb -q";
     my $node = shift;
     my $cmd1 = "ps auxwww | grep slothd | grep -v grep";
-    my $cmd2 = "/etc/testbed/slothd -f";
+    my $cmd2 = "'rm -f /tmp/.sdpid ; /etc/testbed/slothd -f'";
     # Run an ssh command in a child process, protected by an alarm to
     # ensure that the ssh is not hung up forever if the machine is in
     # some funky state.
@@ -81,24 +81,23 @@ sub check {
 	#print "$syspid - Alarm set.\n";
         waitpid($syspid, 0);
 	my $rv = $?;
-	#print "$syspid - Done waiting. Got '$rv'\n";
+	if ($rv > 255) { $rv /= 256; }
+	#print "$syspid - Done waiting. Got '$rv' ",$rv/256," ",$rv%256,"\n";
         alarm 0;
 	if ($rv == 15) { 
-	    #print "Node is wedged.\n"; 
-	    $str="unreachable"; 
-	} elsif ($rv == 256) { 
-	    #print "Node is not running sshd.\n"; 
-	    $str="SSH not available"; 
-	} elsif ($rv == 512) { 
-	    $str="not running, couldn't start slothd"; 
+	    $str="operation timed out"; 
+	#} elsif ($rv == 256) { 
+	#    $str="SSH not available"; 
 	} elsif ($rv == 0) {
 	    $str="running";
 	} elsif ($rv == 1) {
 	    $str="not running";
 	} elsif ($rv == 2) {
 	    $str="not running, started";
+	} elsif ($rv == 3) { 
+	    $str="not running, couldn't start slothd"; 
 	} else {
-	    $str="I don't know what happened...$rv";
+	    $str="I don't know what happened...returned $rv";
 	}
     } else {
 	# child
@@ -110,12 +109,34 @@ sub check {
 	} else {
 	    #print "not running. ";
 	    if ($opts{"s"}) {
+		my $f = "/tmp/tmp-$node-$$";
+		open (STDERR, "> $f") || 
+		  warn("Couldn't open $f: $!\n");
 		$str = `sudo $ssh $node $cmd2`;
+		my $rv = $?;
+		if ($rv > 255) { $rv /= 256; }
+		#print "\n(start=$rv)\n";
+		#system("ls -l $f");
+		if ( -s "$f") {
+		    $rv=127;
+		    print "Error starting slothd:\n";
+		    system("cat $f");
+		}
+		system("rm $f");
+		#print "(start=$rv)\n";
 		if ($str) {
-		    #print "(start returned '$str') ";
+		    print "(start returned '$str') ";
+		}
+		if ($rv==1) {
+		    # success
+		    exit(2);
+		} elsif ($rv==127) {
+		    # error trying to run slothd
+		    exit(3);
+		} else {
+		    # I dunno...
+		    exit($rv);
 		}
-		#print "started...";
-		exit(2);
 	    }
 	    #print "\n";
 	    exit(1);