Task #11059: Use CEP4 nodes of the "cobalt" reservation on CEP4, if it exists.

RTCP/Cobalt/GPUProc/src/scripts/cobalt_functions.sh

@@ -53,10 +53,21 @@ function read_cluster_model {
     CEP4)
       HEADNODE=head01.cep4.control.lofar
       SLURM_PARTITION=cpu
-      COMPUTENODES="`ssh $HEADNODE sinfo --responding --states=idle,mixed,alloc --format=%n.cep4.infiniband.lofar,%T --noheader --partition=$SLURM_PARTITION --sort=N | fgrep -v ,draining | cut -f1 -d,`"
+      SLURM_RESERVATION=cobalt
+      # Get the nodes in the cobalt reservation. The reservation must be active.
+      RESVNODES=$(ssh $HEADNODE scontrol show res -o $SLURM_RESERVATION  | perl -n -e 'm/Nodes=(.*?) .*State=ACTIVE/ ? print STDOUT $1 : die "No active cobalt reservation found"')
+      if [ $? -eq 0 ]; then
+        echo "Active reservation '$SLURM_RESERVATION' found, get online nodes in the reservation"
+        SINFO_FLAGS="--responding -n $RESVNODES"
+      else 
+        echo "WARNING: No reservation '$SLURM_RESERVATION' found, defaulting to all online nodes in partition '$SLURM_PARTITION'"
+        SINFO_FLAGS="--responding --states=idle,mixed,alloc --partition=$SLURM_PARTITION"
+      fi
+      COMPUTENODES="$(ssh $HEADNODE sinfo --format=%n.cep4.infiniband.lofar,%T --noheader --sort=N $SINFO_FLAGS | fgrep -v ,draining | cut -f1 -d,)"
+      # OLD COMPUTENODES="`ssh $HEADNODE sinfo --responding --states=idle,mixed,alloc --format=%n.cep4.infiniband.lofar,%T --noheader --partition=$SLURM_PARTITION --sort=N | fgrep -v ,draining | cut -f1 -d,`"
       if [ -z "$COMPUTENODES" ]; then
         echo "ERROR: Could not obtain list of available CEP4 nodes. Defaulting to all."
-        COMPUTENODES="`seq -f "cpu%02.0f.cep4.infiniband.lofar" 1 50`"
+        COMPUTENODES="`seq -f "cpu%02.0f.cep4.infiniband.lofar" 1 47`"
       fi
 
       GLOBALFS_DIR=/data
@@ -73,8 +84,8 @@ function read_cluster_model {
       SLURM_PARTITION=lofarobs  # NOTE: sinfo (without -a) only displays this partition for members of the lofarsys group (+ slurm,root)
       COMPUTENODES="`ssh $HEADNODE sinfo --responding --states=idle,mixed,alloc --format=%n-ib.dragnet.infiniband.lofar,%T --noheader --partition=$SLURM_PARTITION --sort=N | fgrep -v ,draining | cut -f1 -d,`"
       if [ -z "$COMPUTENODES" ]; then
-        echo "ERROR: Could not obtain list of available DRAGNET nodes. Defaulting to drg01 - drg23 -ib.dragnet.infiniband.lofar"
-        COMPUTENODES=`seq -f drg%02.0f-ib.dragnet.infiniband.lofar 1 23`
+        echo "ERROR: Could not obtain list of available DRAGNET nodes. Defaulting to drg01 - drg20 -ib.dragnet.infiniband.lofar"
+        COMPUTENODES="`seq -f "drg%02-ib.dragnet.infiniband.lofar" 1 20`"
       fi
       COMPUTENODES=$(echo $COMPUTENODES | sed -e s/dragproc-ib.dragnet.infiniband.lofar/dragproc-10g.online.lofar/g)  # dragproc has no infiniband i/f, so use 10g