Version 0.7

Makefile

@@ -31,7 +31,9 @@
 
 include VERSION
 
-INSTALL_FILES=VERSION prologue.user epilogue.user mjf.init mjf.sh mjf.csh mjf-get-total-cpu.torque
+INSTALL_FILES=VERSION prologue.user epilogue.user mjf.init \
+              mjf-torque.sh mjf-torque.csh mjf-get-total-cpu.torque \
+              mjf-htcondor.sh mjf-htcondor.csh
 
 TGZ_FILES=$(INSTALL_FILES) Makefile mjf-torque.spec mjf-htcondor.spec README
 
@@ -47,10 +49,12 @@ install: $(INSTALL_FILES)
                  $(RPM_BUILD_ROOT)/etc/profile.d
 	cp mjf.init \
            $(RPM_BUILD_ROOT)/etc/rc.d/init.d/mjf
-	cp mjf.sh mjf.csh \
-           $(RPM_BUILD_ROOT)/etc/profile.d
-	
+
 torque-install: $(INSTALL_FILES) install
+	cp mjf-torque.sh \
+           $(RPM_BUILD_ROOT)/etc/profile.d/mjf.sh
+	cp mjf-torque.csh \
+           $(RPM_BUILD_ROOT)/etc/profile.d/mjf.csh
 	mkdir -p $(RPM_BUILD_ROOT)/var/lib/torque/mom_priv
 	mkdir -p $(RPM_BUILD_ROOT)/usr/bin 
 	cp prologue.user epilogue.user \
@@ -58,6 +62,7 @@ torque-install: $(INSTALL_FILES) install
 	cp mjf-get-total-cpu.torque \
            $(RPM_BUILD_ROOT)/usr/bin/mjf-get-total-cpu
 	
+	
 torque-rpm: mjf-scripts.tgz
 	rm -Rf RPMTMP
 	mkdir -p RPMTMP/SOURCES RPMTMP/SPECS RPMTMP/BUILD \
@@ -68,6 +73,10 @@ torque-rpm: mjf-scripts.tgz
 	  --buildroot $(shell pwd)/RPMTMP/BUILDROOT mjf-torque.spec
 
 htcondor-install: $(INSTALL_FILES) install
+	cp mjf-htcondor.sh \
+           $(RPM_BUILD_ROOT)/etc/profile.d/mjf.sh
+	cp mjf-htcondor.csh \
+           $(RPM_BUILD_ROOT)/etc/profile.d/mjf.csh
 	
 htcondor-rpm: mjf-scripts.tgz
 	rm -Rf RPMTMP

WN-mjf.py

+#!/bin/env python
+
+import os, sys, urllib
+
+class mjf:
+
+    InfoCode, WarningCode, ErrorCode = range(3)
+        
+    def __init__(self):
+        # WarningCode etc values are what to report if that MJF key is absent
+        self.featuredict = { 
+                            'MACHINEFEATURES' : { 'hs06'        : self.WarningCode,
+                                                  'total_cpu'   : self.ErrorCode,
+                                                  'shutdowntime': self.InfoCode,
+                                                  'grace_secs'  : self.InfoCode },
+
+                            'JOBFEATURES' : { 'allocated_cpu'       : self.ErrorCode,
+                                              'hs06_job'            : self.WarningCode,
+                                              'shutdowntime_job'    : self.InfoCode,
+                                              'grace_secs_job'      : self.InfoCode,
+                                              'jobstart_secs'       : self.ErrorCode,
+                                              'job_id'              : self.ErrorCode,
+                                              'wall_limit_secs'     : self.ErrorCode,
+                                              'cpu_limit_secs'      : self.ErrorCode,
+                                              'max_rss_bytes'       : self.ErrorCode,
+                                              'max_swap_bytes'      : self.WarningCode,
+                                              'scratch_limit_bytes' : self.WarningCode }
+                           }
+        self.exitcode = 0
+        self.loglines = []
+        self.logdict = {'Info'    : self.InfoCode, 
+                        'Warning' : self.WarningCode, 
+                        'Error'   : self.ErrorCode}
+	self.logdictinv = {self.InfoCode: 'Info', self.WarningCode : 'Warning', self.ErrorCode : 'Error'}
+        self.reportline = 'Probe executed successfully'
+
+    def log(self, loglevel, logline):
+        if self.logdict[loglevel] > self.exitcode : self.reportline = logline
+        self.exitcode = max(self.exitcode, self.logdict[loglevel])
+        self.loglines.append(loglevel.ljust(8) + ': ' + logline)
+
+    def probe(self, featurevar):
+        self.log('Info', 'Processing %s' % featurevar)
+        proberoot = os.environ.get(featurevar)
+        if proberoot != None : 
+            self.log('Info', '%s=%s' % (featurevar, proberoot))
+            onefeaturefound = False
+            
+            for featureName in self.featuredict[featurevar]:
+                featureAbsentCode = self.featuredict[featurevar][featureName]
+                
+                featureValue = None
+                try: 
+                    featureValue = urllib.urlopen(proberoot + '/' + featureName).read()
+                    if not featureValue : 
+                      featureValue = 'None'
+
+                    while featureValue[-1] == '\n': 
+                      featureValue = featureValue[:-1]
+                    
+                    if not featureValue is None and featureValue != '':
+                        onefeaturefound = True
+                        self.log('Info', 'Key %s found with value %s' % (featureName, featureValue))
+                    else :
+                        featureValue = None
+                        self.log('Info', 'Key %s found but value is empty' % featureName)
+
+                except : 
+                    featureValue = None
+
+                if featureValue is None:
+                  self.log(self.logdictinv[featureAbsentCode], 'Key %s absent (or empty)' % featureName)
+
+            if not onefeaturefound: 
+              self.log('Error', 'Environment variable %s set but cannot find any valid keys in there' % featurevar)
+
+        else: 
+            self.log('Error', 'Environment variable %s not set' % featurevar)
+
+    def run(self):
+        map(lambda x: self.probe(x), self.featuredict.keys())
+        
+        sys.stdout.write('%s %s\n\n' % (self.logdictinv[self.exitcode], self.reportline))
+        map(lambda x: sys.stdout.write(x + '\n'), self.loglines)        
+        sys.exit(self.exitcode)
+
+if __name__ == '__main__' :
+    mjf().run()

epilogue.user

@@ -32,7 +32,7 @@
 #  POSSIBILITY OF SUCH DAMAGE.
 
 # Remove $JOBFEATURES directory for this job where the directory
-# is at $HOME/jobfeatures-$PBS_JOBID   ($1 is $PBS_JOBID)
+# is at $PBS_O_HOME/jobfeatures-$PBS_JOBID   ($1 is $PBS_JOBID)
 
 if [ "$1" != "" -a -d ~/jobfeatures-$1 ] ; then
   rm -Rf ~/jobfeatures-$1 

mjf-htcondor.csh

+if ( -d /etc/machinefeatures ) then
+  setenv MACHINEFEATURES /etc/machinefeatures
+endif

mjf-htcondor.sh

+if [ -d /etc/machinefeatures ] ; then
+  export MACHINEFEATURES=/etc/machinefeatures
+fi

mjf.csh → mjf-torque.csh

@@ -2,5 +2,5 @@ if ( -d /etc/machinefeatures ) then
   setenv MACHINEFEATURES /etc/machinefeatures
 endif
 if ( "$PBS_JOBID" != "" ) then
-  setenv JOBFEATURES $HOME/jobfeatures-$PBS_JOBID
+  setenv JOBFEATURES $PBS_O_HOME/jobfeatures-$PBS_JOBID
 endif

mjf.sh → mjf-torque.sh

@@ -2,5 +2,5 @@ if [ -d /etc/machinefeatures ] ; then
   export MACHINEFEATURES=/etc/machinefeatures
 fi
 if [ "$PBS_JOBID" != "" ] ; then
-  export JOBFEATURES=$HOME/jobfeatures-$PBS_JOBID
+  export JOBFEATURES=$PBS_O_HOME/jobfeatures-$PBS_JOBID
 fi

prologue.user

@@ -35,7 +35,7 @@
 #  specification in HSF-TN-2016-02
 #
 #  This script creates the $JOBFEATURES directory at 
-#  $HOME/jobfeatures-$PBS_JOBID and attempts to populate it
+#  $PBS_O_HOME/jobfeatures-$PBS_JOBID and attempts to populate it
 #  from Torque/PBS information and from 
 #  $MACHINEFEATURES=/etc/machinefeatures
 #
@@ -56,7 +56,7 @@ import re
 import sys
 import time
 
-homeDir = os.path.expanduser('~')
+homeDir = os.environ['PBS_O_HOME']
 
 job_id = sys.argv[1]