First version

nagios/README

+CMS-specific Nagios files can be found here.
+
+HOW TO BUILD THE CMS PROBE RPM
+------------------------------
+Run the following commands:
+
+export NAG_TMP=`mktemp -d`
+export NAG_DEST=/afs/cern.ch/cms/LCG/SAM/RPM
+cd $NAG_TMP
+cvs co CMSSAM
+svn co svn+ssh://svn.cern.ch/reps/it-es-vos/trunk/sam-vo-probes
+export NAG_SCRIPTS=/afs/cern.ch/cms/LCG/SAM/scripts
+$NAG_SCRIPTS/sam2nagios.sh
+
+Edit the spec file to change version number and add description.
+Run the following commands:
+
+$NAG_SCRIPTS/make_rpm.sh <version>
+cd
+rm -rf $NAG_TMP

nagios/grid-monitoring-probes-org.cms.spec

+%define site org.cms
+%define dir %{_libexecdir}/grid-monitoring/probes
+%define conf /usr/lib/perl5/vendor_perl/5.8.5/NCG/LocalMetrics
+%define ncg /etc/ncg
+
+Summary: WLCG Compliant Probes from %{site}
+Name: grid-monitoring-probes-org.cms
+Version: 0.1.41
+Release: 1%{?dist}
+
+License: GPL
+Group: Network/Monitoring
+Source0: %{name}-%{version}.tgz
+BuildRoot: %{_tmppath}/%{name}-%{version}-root
+Requires: python >= 2.4
+Requires: python-GridMon >= 1.1.3
+Requires: python-ldap
+Requires: python-suds >= 0.3.5
+AutoReqProv: no
+BuildArch: noarch
+
+%description
+Contains the CMS-specific probes and metrics.
+
+%prep
+%setup -q
+
+%build
+
+%install
+%{__rm} -rf %{buildroot}
+install --directory %{buildroot}/%{dir}
+install --directory %{buildroot}/%{conf}
+install --directory %{buildroot}/%{ncg}
+install --directory %{buildroot}/etc/cron.d
+%{__cp} -rpf %{site} %{buildroot}%{dir}
+%{__cp} -rpf config/Hash_local.pm %{buildroot}%{conf}
+%{__cp} -rpf config/cms_glexec %{buildroot}/etc/cron.d
+
+%clean
+%{__rm} -rf %{buildroot}
+
+%files
+%defattr(-,root,root,-)
+%{dir}/%{site}
+%{conf}/Hash_local.pm
+/etc/cron.d/cms_glexec
+
+%changelog
+* Thu Aug 23 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.41-1.
+- Fixed bug in mc test causing the test to always pass
+- converted errors to warnings in glexec test
+* Wed Aug 22 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.40-1.
+- mc test now uses WMCore code
+* Tue Aug 21 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.39-1.
+- Removed SL5 check from analysis and frontier test
+* Thu Jul 5 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.38-1.
+- Made basic samtest-run compliant and changed SCRAM_ARCH
+- Fixed squid exception for RAL in fetch-from-web
+* Mon Jul 2 2012 Duncan Ralph <Duncan.Kelley.Ralph@cern.ch> 0.1.37-1.
+- Updated swinst test
+* Wed Jun 15 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.36-1.
+- Changed analysis test to use CMSSW_5_3_1; updated swinst test
+* Tue May 15 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.35-1.
+- Allow glexec test to succeed wen only the GID changes
+* Wed May 9 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.34-1.
+- Increased verbosity of Nagios on WN to 3
+- Removed calls to unwanted commands in glprobe.sh; added sleep in analysis test to delay its execution
+* Wed Apr 25 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.33-1.
+- Fixed bug in RPM (Hash_local.pm and cms_glexec being links)
+* Thu Apr 16 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.32-1.
+- Set --timeout-job-discard to 84600 and --timeout-wnjob-global to 1800 for JobState; updated glexec probe
+* Wed Mar 14 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.31-1.
+- Updated basic, CMSSW_frontier.sh and swinst
+* Tue Feb 14 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.30-1.
+- Updated swinst
+* Thu Jan 26 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.29-1.
+- Fixed bug in Hash_local.pm
+* Wed Jan 18 2012 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.28-1.
+- All WN tests (apart mc) now generate a good HTML
+* Wed Dec 14 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.27-1.
+- Removed hr.srce.SRM2-CertLifetime from the tests to run
+* Wed Nov 30 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.26-1.
+- Removed slc4 architecture from swinst test
+* Wed Nov 16 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.25-1.
+- Changed mc test to support fallback
+* Wed Nov 2 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.24-1.
+- Added support for ARC-CE in Hash_local.pm
+- New version of swinst
+- NCG cron disabled by default
+* Tue Oct 4 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.23-1.
+- NCG configuration supports OSG services
+* Tue Sep 27 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.22-1.
+- Fixed bug in SRM probe causing tests to fail when a cached PFN is used that was not properly deleted by a previous test
+* Tue Sep 13 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.21-1.
+- Fixed small bug in analysis test output
+* Tue Sep 13 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.20-1.
+- Updated analysis, swinst tests to new CMSSW version
+* Fri Jul 29 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.19-1.
+- Fixed bug in new Frontier test
+* Wed Jul 27 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.18-1.
+- Updated Frontier test
+- Added timeouts to lcg-del in mc test
+* Mon Jun 22 2011  Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.17-1.
+- Added new glexec test
+- New SRM probe with correct behaviour for the GetPFNFromTFC test
+- Modified Hash_local.pm to expose org.cms.glexec.WN-gLExec for CREAMCE
+* Tue May 31 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.16-1.
+- Added glexec test
+* Tue May 24 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.15-1.
+- Added white/black list of sites
+* Fri May 6 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.13-1.
+- Timeouts for jobs lengthened to 12 h
+- Timeouts for SRM lengthened to 30 minutes
+* Tue Apr 11 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.12-1.
+- Updated tests.
+* Tue Mar 29 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.11-1.
+- Updated tests, increased SRMAll timeout, fixed missing WN conf.
+* Thu Mar 24 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.10-1.
+- Updated tests and configuration to run SRM tests only once.
+* Mon Mar 14 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.9-1.
+-  Updated tests.
+* Mon Feb 21 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.8-1.
+-  Added configuration files.
+* Fri Feb 18 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.7-1.
+-  Added configuration files.
+* Fri Feb 18 2011 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.6-1.
+-  Added new version of srmvometrics.py.
+* Thu Dec 9 2010 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.5-1.
+-  Added new version of srmvometrics.py.
+* Tue Nov 9 2010 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.4-1.
+-  Added custom variables to sam-generic-wn in services.cfg.
+* Wed Aug 25 2010 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.3-1.
+-  Upgraded to latest version of SRM probe.
+* Tue Aug 3 2010 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.2-1.
+-  Fixed bug in SRM probe.
+* Fri Jul 23 2010 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.1-1.
+-  Updated the CMS tests and included them all, including SRM.
+* Mon Feb 08 2010 Andrea Sciaba <Andrea.Sciaba@cern.ch> 0.1.0-1.
+-  First version with some WN tests.

nagios/make_rpm.sh

+#!/bin/sh
+
+if [ -z "$NAG_TMP" ] ; then
+	echo "NAG_TMP undefined"
+	exit 1
+fi
+cd $NAG_TMP
+if [ -z "$NAG_SCRIPTS" ] ; then
+	echo "NAG_SCRIPTS undefined"
+	exit 1
+fi
+if [ -z "$NAG_DEST" ] ; then
+	echo "NAG_DEST undefined"
+	exit 1
+fi
+ver=$1
+if [ -z "$ver" ] ; then
+	echo "Specify version!"
+	exit 1
+fi
+specver=`grep 'Version:' $NAG_SCRIPTS/grid-monitoring-probes-org.cms.spec | cut -d' ' -f2`
+if [ "$specver" != $ver ] ; then
+	echo "Spec file not updated!"
+	exit 1
+fi
+rm -rf tmp/grid-monitoring-probes-org.cms-$ver
+mkdir -p tmp/grid-monitoring-probes-org.cms-$ver
+cp -a org.cms tmp/grid-monitoring-probes-org.cms-$ver
+cp -a config tmp/grid-monitoring-probes-org.cms-$ver
+cd tmp
+tar zcf grid-monitoring-probes-org.cms-$ver.tgz grid-monitoring-probes-org.cms-$ver
+mkdir -p $NAG_TMP/SOURCES
+mkdir -p $NAG_TMP/BUILD
+mkdir -p $NAG_TMP/RPMS
+mv grid-monitoring-probes-org.cms-$ver.tgz $NAG_TMP/SOURCES
+cd ..
+cat > .rpmmacros <<EOF
+%_topdir        $NAG_TMP
+%_dbpath        $NAG_TMP/db
+%_tmppath       $NAG_TMP/rpm-tmp
+%_sourcedir     $NAG_TMP/SOURCES
+EOF
+export HOME=`pwd`
+rpmbuild -bb $NAG_SCRIPTS/grid-monitoring-probes-org.cms.spec
+if [ $? -eq 0 ] ; then
+        cp -i $NAG_TMP/RPMS/noarch/grid-monitoring-probes-org.cms-$ver-1.noarch.rpm $NAG_DEST/
+fi
+rm -rf BUILD RPMS rpm-tmp SOURCES tmp

nagios/sam2nagios.sh

+#!/bin/bash
+
+BASEDIR=$NAG_TMP
+SAMDIR=$BASEDIR/CMSSAM
+NAGDIR=$BASEDIR
+NAGCONF=$SAMDIR/nagios/config
+rm -rf $NAGDIR/org.cms/*
+mkdir -p $NAGDIR/org.cms/wnjob/org.cms/etc/wn.d/org.cms
+cp -L $NAGCONF/services.cfg $NAGCONF/commands.cfg $NAGDIR/org.cms/wnjob/org.cms/etc/wn.d/org.cms/
+cp $NAGDIR/sam-vo-probes/srmvometrics.py $NAGDIR/org.cms/srmvometrics.py.ORIG
+sed -e 's/org.vos/org.cms/' $NAGDIR/org.cms/srmvometrics.py.ORIG > $NAGDIR/org.cms/srmvometrics.py
+chmod +x $NAGDIR/org.cms/srmvometrics.py
+rm -rf $NAGDIR/org.cms/srmvometrics.py.ORIG
+mkdir -p $NAGDIR/config
+cp -L $NAGCONF/Hash_local.pm -L $NAGCONF/cms_glexec $NAGDIR/config/
+srcdir=$SAMDIR
+dstdir=$NAGDIR/org.cms/wnjob/org.cms/probes/org.cms
+
+sensor='testjob'
+modules='FroNtier MonteCarlo testjob'
+for i in $modules ; do
+   cd $srcdir/SiteTests/$i
+   for j in `find . -not -regex '.*\(CVS\).*'` ; do
+      if [ -d $j ] ; then
+         mkdir -p $dstdir/$sensor/$j
+      else
+         cp $j $dstdir/$sensor/`dirname $j`
+      fi
+   done
+done
+
+cd $srcdir/nagios/
+dstdir=$NAGDIR/org.cms/wnjob/
+for j in `find org.cms.glexec -not -regex '.*\(CVS\).*'` ; do
+   if [ -d $j ] ; then
+      mkdir -p $dstdir/$j
+   else
+      cp $j $dstdir/`dirname $j`
+   fi
+done
+cd $BASEDIR