_per_cpu values to accomodate mixed job sizes

README

@@ -33,6 +33,12 @@ total number of logical processors), but are overridden in turn when
 per-job values can be determined from the batch system (eg the number 
 of logical processors allocated to this job.)
 
+The values cpu_limit_secs_per_cpu, max_rss_bytes_per_cpu, 
+max_swap_bytes_per_cpu, and scratch_limit_bytes_per_cpu can be set in
+either mjf file to cause the scripts to calculate the corresponding 
+per-job value (eg cpu_limit_secs) using $JOBFEATURES/allocated_cpu (which
+will be determined from the batch system if available, otherwise 1.)
+
 If you know the HS06 of the worker node, you can also include a line like
 hs06=99.99  which will be picked up when populating /etc/machinefeatures/
 (you can force updates after changing that file with  service mjf start  

make-jobfeatures.htcondor

@@ -52,10 +52,19 @@
 #  - hs06_job
 #  - db12_job
 #
-# Values in /var/run/mjf are preferred over /etc/sysconfig/mjf.
+#  Values in /var/run/mjf are preferred over /etc/sysconfig/mjf.
 #
-# Additionally you can set mjf_tmp_dir in those files to use a 
-# directory other than /tmp for the mjf-$USER directories.
+#  The following per-cpu values in either mjf file will be used
+#  to calculate the corresponding per-job values if not given in
+#  either file or obtainable from PBS/Torque:
+# 
+#  - cpu_limit_secs_per_cpu
+#  - max_rss_bytes_per_cpu
+#  - max_swap_bytes_per_cpu
+#  - scratch_limit_bytes_per_cpu
+#
+#  Additionally you can set mjf_tmp_dir in those files to use a 
+#  directory other than /tmp for the mjf-$USER directories.
 #
 
 import os
@@ -162,13 +171,27 @@ for key in ['allocated_cpu', 'wall_limit_secs', 'cpu_limit_secs',
     except:
       pass
 
-# Examine the Job Ad, for resource limits
+# Examine the Job Ad, for number of CPUs allocated
 try:
   cpusMatchObject = re.search('CpusProvisioned *= *([0-9 +*/()-]*)', fromJobAd)
   jobfeatures['allocated_cpu'] = int(eval(cpusMatchObject.group(1)))
 except:
   pass
 
+# Look for any per-cpu values given in either file
+for key in ['cpu_limit_secs', 'max_rss_bytes', 
+            'max_swap_bytes', 'scratch_limit']:
+
+  if key + '_per_cpu=' in fromRun + fromSysconfig:
+    try:
+      matchObject = re.search(key + "_per_cpu=([0-9]*)", fromRun + '\n' + fromSysconfig)
+
+      if key not in jobfeatures:
+        jobfeatures[key] = int(matchObject.group(1)) * jobfeatures['allocated_cpu']
+    except:
+      pass
+
+# Examine the Job Ad, for resource limits
 try:
   wallMatchObject = re.search('MaxJobRetirementTime *= *([0-9 +*/()-]*)', fromMachineAd)
   jobfeatures['wall_limit_secs'] = int(eval(wallMatchObject.group(1)))

prologue.user

@@ -52,10 +52,19 @@
 #  - hs06_job
 #  - db12_job
 #
-# Values in /var/run/mjf are preferred over /etc/sysconfig/mjf.
+#  Values in /var/run/mjf are preferred over /etc/sysconfig/mjf.
 #
-# Additionally you can set mjf_tmp_dir in those files to use a 
-# directory other than /tmp for the mjf-$USER directories.
+#  The following per-cpu values in either mjf file will be used
+#  to calculate the corresponding per-job values if not given in
+#  either file or obtainable from PBS/Torque:
+# 
+#  - cpu_limit_secs_per_cpu
+#  - max_rss_bytes_per_cpu
+#  - max_swap_bytes_per_cpu
+#  - scratch_limit_bytes_per_cpu
+#
+#  Additionally you can set mjf_tmp_dir in those files to use a 
+#  directory other than /tmp for the mjf-$USER directories.
 #
 
 import os
@@ -153,6 +162,19 @@ try:
 except:
   pass
 
+# Look for any per-cpu values given in either file
+for key in ['cpu_limit_secs', 'max_rss_bytes', 
+            'max_swap_bytes', 'scratch_limit']:
+
+  if key + '_per_cpu=' in fromRun + fromSysconfig:
+    try:
+      matchObject = re.search(key + "_per_cpu=([0-9]*)", fromRun + '\n' + fromSysconfig)
+
+      if key not in jobfeatures:
+        jobfeatures[key] = int(matchObject.group(1)) * jobfeatures['allocated_cpu']
+    except:
+      pass
+
 try:
   wallMatchObject = re.search('walltime=([0-9]*):([0-9]*):([0-9]*)', sys.argv[5])
   jobfeatures['wall_limit_secs'] = int(wallMatchObject.group(1)) * 3600 + int(wallMatchObject.group(2)) * 60 + int(wallMatchObject.group(3))