Changes to allow plab pnodes to check in normally (via the rusage/isalive

mechanism in the service sliver).

db/node_status.in

@@ -47,15 +47,14 @@ if (($UID != 0) && (!TBAdmin())) {
 #
 my $query_result =
     DBQueryFatal("SELECT n.node_id,ns.status,r.pid,nt.isremotenode, ".
-		 "       nt.isvirtnode, now() - ns.status_timestamp ".
+		 "       nt.isvirtnode, nt.isplabdslice".
+                 "now() - ns.status_timestamp ".
 		 " from nodes as n ".
 		 "left join node_types as nt on n.type=nt.type ".
 		 "left join node_status as ns on ns.node_id=n.node_id ".
 		 "left join reserved as r on r.node_id=n.node_id ".
 		 "where ".
-		 # We ignore plab physical nodes, because the physical nodes
-		 # do not do keepalive; handled by plab daemon instead.
-		 " n.type != 'pcplabphys' && ns.status != 'down' && " .
+		 " ns.status != 'down' && " .
 		 # Jailed and PLAB virtnodes report every 600 seconds.
 		 # Must be allocated to an experiment to be considerd.
 		 " ((nt.isvirtnode=1 && r.pid is not null && ".
@@ -68,7 +67,7 @@ my $query_result =
 		 "   ((now() - ns.status_timestamp) > 100)))");
 		 
 
-while (my ($node,$status,$pid,$remote,$isvirt,$timediff) =
+while (my ($node,$status,$pid,$remote,$isvirt,$isplab,$timediff) =
        $query_result->fetchrow_array) {
     my $newstatus = "down";
 
@@ -89,6 +88,20 @@ while (my ($node,$status,$pid,$remote,$isvirt,$timediff) =
 	    $newstatus = "possibly down";
 	}
     }
+
+    #
+    # If the plab management sliver hasn't checked in for a couple of
+    # hours, move the pnode to hwdown.
+    #
+    if ($isplab && !$isvirt) {
+        if ($timediff > 7200) {
+            MarkNodeDown($node);
+	    TBSetNodeLogEntry($node, $UID, TB_DEFAULT_NODELOGTYPE(),
+			      "'Moved to hwdown; ".
+			      "$node has not reported in for a long time.'");
+        }
+    }
+
     #
     # Repeat the time check to avoid dropping a node that just came up.
     #

tbsetup/plab/libplab.py.in

@@ -741,22 +741,26 @@ class Plab:
             print "Got known pnodes:"
             print known
 
+#
+# Disable up/down marking - we now do this normally via isalive from
+# the emulab service slice.  Eventually, this code should be removed.
+#
         # Mark known nodes that are not available as down and make sure
         # those that are available are marked as up
-        todown = []  # List of nodeid's
-        toup = []  # List of nodeid's
-        for ip in known.keys():
-            if not ip in avail:
-                todown.append(known[ip])
-            else:
-                toup.append(known[ip])
-        if verbose:
-            print "%d known Plab nodes not available" % len(todown)
-            print "%d known Plab nodes available" % len(toup)
-        self.__setVnodesStatus(todown, "down")
-        self.__setPnodesStatus(todown, "down")
-        self.__setVnodesStatus(toup, "up")
-        self.__setPnodesStatus(toup, "up")
+        #todown = []  # List of nodeid's
+        #toup = []  # List of nodeid's
+        #for ip in known.keys():
+        #    if not ip in avail:
+        #        todown.append(known[ip])
+        #    else:
+        #        toup.append(known[ip])
+        #if verbose:
+        #    print "%d known Plab nodes not available" % len(todown)
+        #    print "%d known Plab nodes available" % len(toup)
+        #self.__setVnodesStatus(todown, "down")
+        #self.__setPnodesStatus(todown, "down")
+        #self.__setVnodesStatus(toup, "up")
+        #self.__setPnodesStatus(toup, "up")
 
         # Add new nodes
         toadd = []  # List of IP's

tmcd/plab/rusaged

@@ -39,7 +39,7 @@ my $action	= "start";
 my $logname	= "$LOGDIR/rusaged.debug";
 my $pidfile	= "/var/run/emulab-rusaged.pid";
 my $debug	= 0;
-my $isalivewait = 300; # Seconds to wait.
+my $isalivewait = 60; # Seconds to wait.
 my $svcslice    = "utah_elab_svc";
 
 #

tmcd/tmcd.c

@@ -4889,9 +4889,8 @@ COMMAND_PROTOTYPE(dorusage)
 	 * See db/node_status script, which uses this info (timestamps)
 	 * to determine when nodes are down.
 	 *
-	 * XXX: Plab physnodes do not report at all; we copy the virtnode
-	 * info to the physnode. Do not update status field though; that
-	 * is handled from one of the plab daemons.
+	 * XXX: Plab physnode status is reported from the management slice.
+         *
 	 */
 	mydb_update("replace delayed into node_rusage "
 		    " (node_id, status_timestamp, "
@@ -4901,11 +4900,10 @@ COMMAND_PROTOTYPE(dorusage)
 
 	if (reqp->isplabdslice) {
 		mydb_update("replace delayed into node_status "
-			    " (node_id, status_timestamp, "
-			    "  load_1min, load_5min, load_15min, disk_used) "
-			    " values ('%s', now(), %f, %f, %f, %f)",
-			    reqp->pnodeid, la1, la5, la15, dused);
-	}
+			    " (node_id, status, status_timestamp "
+			    " values ('%s', 'up', now())",
+			    reqp->pnodeid);
+        }
 
 	/*
 	 * At some point, maybe what we will do is have the client