Changeset 512 for trunk/jobmond

Timestamp:

03/07/08 17:58:33 (15 years ago)

Author:

bastiaans

Message:

web/addons/job_monarch/templates/overview.tpl,
web/addons/job_monarch/overview.php,
jobmond/jobmond.py:

• added Craig West's support for offline/down PBS nodes

File:

• trunk/jobmond/jobmond.py (modified) (2 diffs)

trunk/jobmond/jobmond.py

@@ -462 +462 @@
                 """Submit job info list"""
 
+                global BATCH_API
+
                 self.dp.multicastGmetric( 'MONARCH-HEARTBEAT', str( int( int( self.cur_time ) + int( self.timeoffset ) ) ) )
 
@@ -483 +485 @@
                 self.dp.multicastGmetric( 'MONARCH-RJ', str( running_jobs ), 'uint32', 'jobs' )
                 self.dp.multicastGmetric( 'MONARCH-QJ', str( queued_jobs ), 'uint32', 'jobs' )
+
+                # Report down/offline nodes in batch (PBS only ATM)
+                #
+                if BATCH_API == 'pbs':
+
+                        downed_nodes    = list()
+                        offline_nodes   = list()
+                
+                        l               = ['state']
+                
+                        for name, node in self.pq.getnodes().items():
+
+                                if ( node[ 'state' ].find( "down" ) != -1 ):
+
+                                        downed_nodes.append( name )
+
+                                if ( node[ 'state' ].find( "offline" ) != -1 ):
+
+                                        offline_nodes.append( name )
+
+                        self.dp.multicastGmetric( 'MONARCH-DOWN'   , str( downed_nodes ),  'uint32', 'jobs' )
+                        self.dp.multicastGmetric( 'MONARCH-OFFLINE', str( offline_nodes ), 'uint32', 'jobs' )
 
                 # Now let's spread the knowledge