Catch up

VERSION

-VERSION=00.08
+VERSION=00.10

WN-mjf.py

@@ -22,7 +22,7 @@ class mjf:
                                               'job_id'              : self.ErrorCode,
                                               'wall_limit_secs'     : self.ErrorCode,
                                               'cpu_limit_secs'      : self.ErrorCode,
-                                              'max_rss_bytes'       : self.ErrorCode,
+                                              'max_rss_bytes'       : self.WarningCode,
                                               'max_swap_bytes'      : self.WarningCode,
                                               'scratch_limit_bytes' : self.WarningCode }
                            }

epilogue.user

@@ -32,8 +32,18 @@
 #  POSSIBILITY OF SUCH DAMAGE.
 
 # Remove $JOBFEATURES directory for this job where the directory
-# is at /tmp/jobfeatures-$PBS_JOBID   ($1 is $PBS_JOBID)
+# is at $mjf_tmp_dir/mjf-$USER/jobfeatures-$PBS_JOBID ($1 is $PBS_JOBID)
 
-if [ "$1" != "" -a -d "/tmp/jobfeatures-$1" -a -O "/tmp/jobfeatures-$1" ] ; then
-  rm -Rf "/tmp/jobfeatures-$1" 
+mjf_tmp_dir=/tmp
+
+if [ -r /etc/sysconfig/mjf ] ; then
+ . /etc/sysconfig/mjf
+fi
+
+if [ -r /var/run/mjf ] ; then
+ . /etc/sysconfig/mjf
+fi
+
+if [ "$1" != "" -a -d "$mjf_tmp_dir/mjf-$USER/jobfeatures-$1" ] ; then
+  rm -Rf "$mjf_tmp_dir/mjf-$USER/jobfeatures-$1" 
 fi

mjf-torque.csh

+mjf_tmp_dir=/tmp
+if ( -r /etc/sysconfig/mjf ) then
+  source /var/sysconfig/mjf
+endif
+if ( -r /var/run/mjf ) then
+  source /var/run/mjf
+endif
 if ( -d /etc/machinefeatures ) then
   setenv MACHINEFEATURES /etc/machinefeatures
 endif
 test 
-if ( "$PBS_JOBID" != "" && -d "/tmp/jobfeatures-$PBS_JOBID" ) then
-  setenv JOBFEATURES "/tmp/jobfeatures-$PBS_JOBID"
+if ( "$PBS_JOBID" != "" && -d "$mjf_tmp_dir/mjf-$USER/jobfeatures-$PBS_JOBID" ) then
+  setenv JOBFEATURES "$mjf_tmp_dir/mjf_$USER/jobfeatures-$PBS_JOBID"
 endif

mjf-torque.sh

+mjf_tmp_dir=/tmp
+if [ -r /etc/sysconfig/mjf ] ; then
+ . /etc/sysconfig/mjf
+fi
+if [ -r /var/run/mjf ] ; then
+ . /etc/sysconfig/mjf
+fi
 if [ -d /etc/machinefeatures ] ; then
   export MACHINEFEATURES=/etc/machinefeatures
 fi
-if [ "$PBS_JOBID" != "" -a -d "/tmp/jobfeatures-$PBS_JOBID" ] ; then
-  export JOBFEATURES=/tmp/jobfeatures-$PBS_JOBID
+if [ "$PBS_JOBID" != "" -a -d "$mjf_tmp_dir/mjf-$USER/jobfeatures-$PBS_JOBID" ] ; then
+  export JOBFEATURES="$mjf_tmp_dir/mjf-$USER/jobfeatures-$PBS_JOBID"
 fi
+ 
\ No newline at end of file

prologue.user

@@ -35,20 +35,26 @@
 #  specification in HSF-TN-2016-02
 #
 #  This script creates the $JOBFEATURES directory at 
-#  /tmp/jobfeatures-$PBS_JOBID and attempts to populate it
+#  /tmp/mjf-$USER/jobfeatures-$PBS_JOBID and attempts to populate it
 #  from Torque/PBS information and from 
 #  $MACHINEFEATURES=/etc/machinefeatures
 #
 #  If the following variables are present in /var/run/mjf or 
-#  /etc/sysconfig/mjf then they are preferred:
+#  /etc/sysconfig/mjf then they are used as defaults if not given
+#  by Torque/PBS:
 #
-#  - hs06_job
 #  - allocated_cpu
 #  - wall_limit_secs
 #  - cpu_limit_secs
 #  - max_rss_bytes
 #  - max_swap_bytes
 #  - scratch_limit_bytes
+#  - hs06_job
+#
+# Values in /var/run/mjf are preferred over /etc/sysconfig/mjf.
+#
+# Additionally you can set mjf_tmp_dir in those files to use a 
+# directory other than /tmp for the mjf-$USER directories.
 #
 
 import os
@@ -57,21 +63,67 @@ import sys
 import stat
 import time
 
+os.umask(0022)
+
 try:
   job_id = sys.argv[1]
 except:
   print 'JOB ID not given on command line'
-  sys.exit(1)
+  sys.exit(11)
 
-jobfeaturesDir = '/tmp/jobfeatures-' + job_id
+# Values in /var/run/mjf take precedence
+try:
+  fromRun = open('/var/run/mjf','r').read()
+except:
+  fromRun = ''
+
+# Also look in persistent /etc/sysconfig/mjf
+try:
+  fromSysconfig = open('/etc/sysconfig/mjf','r').read()
+except:
+  fromSysconfig = ''  
+
+mjfTmpDir = '/tmp'
+
+if 'mjf_tmp_dir=' in fromRun + fromSysconfig:
+  try:
+    matchObject = re.search("^mjf_tmp_dir=(.*)", fromRun + '\n' + fromSysconfig)
+    mjfTmpDir = matchObject.group(1)
+  except:
+    pass
+
+# First make sure the mjfTmpDir/mjf-$USER directory exists
+try:
+  os.mkdir(mjfTmpDir + '/mjf-' + sys.argv[2])
+except:
+  # Ok if exists already
+  pass
+
+try: 
+  userDirStat = os.stat(mjfTmpDir + '/mjf-' + sys.argv[2])
+except:
+  print mjfTmpDir + '/mjf-' + sys.argv[2] + ' could not be created'
+  sys.exit(12)
+
+# Check it is owned by us
+if userDirStat.st_uid != os.getuid() or userDirStat.st_gid != os.getgid():
+  print mjfTmpDir + '/mjf-' + sys.argv[2] + ' has the wrong UID/GID'
+  sys.exit(13)
+
+# Ensure the right permissions
+os.chmod(mjfTmpDir + '/mjf-' + sys.argv[2], 
+         stat.S_IRWXU + stat.S_IRGRP + stat.S_IXGRP + stat.S_IROTH + stat.S_IXOTH)
+
+# Now the directory for this job
+jobfeaturesDir = mjfTmpDir + '/mjf-' + sys.argv[2] + '/jobfeatures-' + job_id
 
 try:
   os.mkdir(jobfeaturesDir)
 except:
-  print jobfeaturesDir + ' already exists? - not overwriting'
-  sys.exit(0)
-else:
-  os.chmod(jobfeaturesDir, stat.S_IRWXU + stat.S_IRGRP + stat.S_IXGRP + stat.S_IROTH + stat.S_IXOTH)
+  print 'Failed to create ' + jobfeaturesDir
+  sys.exit(14)
+#else:
+#  os.chmod(jobfeaturesDir, stat.S_IRWXU + stat.S_IRGRP + stat.S_IXGRP + stat.S_IROTH + stat.S_IXOTH)
 
 open(jobfeaturesDir + '/job_id', 'w').write(job_id)
 
@@ -81,6 +133,17 @@ jobfeatures['allocated_cpu'] = 1
 jobstart_secs = int(time.time())
 open(jobfeaturesDir + '/jobstart_secs', 'w').write(str(jobstart_secs))
 
+# Get any defaults for these (integer) values
+for key in ['allocated_cpu', 'wall_limit_secs', 'cpu_limit_secs',
+            'max_rss_bytes', 'max_swap_bytes', 'scratch_limit_bytes']:
+
+  if key + '=' in fromRun + fromSysconfig:
+    try:
+      matchObject = re.search(key + "=([0-9]*)", fromRun + '\n' + fromSysconfig)
+      jobfeatures[key] = int(matchObject.group(1))
+    except:
+      pass
+
 # Examine the 5th argument, for resource limits
 try:
   # if just 1 processor-per-node then "nodes=1", but if 8, say, then "nodes=1:ppn=8"
@@ -119,29 +182,10 @@ else:
     # Is this the right name??
     jobfeatures['max_rss_bytes'] = rssInt
 
-# Values in /var/run/mjf take precedence
-try:
-  fromRun = open('/var/run/mjf','r').read()
-except:
-  fromRun = ''
-
-# Also look in persistent /etc/sysconfig/mjf
-try:
-  fromSysconfig = open('/etc/sysconfig/mjf','r').read()
-except:
-  fromSysconfig = ''  
-
-# These are all integers so handle with a list
+# Write out if these have been set from files or prologue.user arguments
 for key in ['allocated_cpu', 'wall_limit_secs', 'cpu_limit_secs',
             'max_rss_bytes', 'max_swap_bytes', 'scratch_limit_bytes']:
 
-  if key + '=' in fromRun + fromSysconfig:
-    try:
-      matchObject = re.search(key + "=([0-9]*)", fromSysconfig + fromRun)
-      jobfeatures[key] = int(matchObject.group(1))
-    except:
-      pass
-
   if key in jobfeatures:
     open(jobfeaturesDir + '/' + key, 'w').write(str(jobfeatures[key]))
 
@@ -150,7 +194,7 @@ hs06_job = None
 
 if 'hs06_job=' in fromRun + fromSysconfig:
   try:
-    matchObject = re.search("hs06_job=([0-9.]*)", fromSysconfig + fromRun)
+    matchObject = re.search("hs06_job=([0-9.]*)", fromRun + '\n' + fromSysconfig)
     hs06_job = float(matchObject.group(1))
   except:
     pass