trunk/userspace-package-management/get-site-status.sh

#!/bin/bash

# FIXMELATER
delegation=dennis

mopup=0

while getopts m opt; do
    case $opt in
        m)
            mopup=1
            ;;
        ?)
            echo "ERROR: unknown option $opt."
            exit 1
            ;;
    esac
done
shift $(($OPTIND - 1))

# look up the resources in the data file

if [ ! -r resources.dat ] ; then
    echo "ERROR: Missing resource datafile resources.dat" >&2
    exit 1
fi

sites=`awk ' $1 !~ /^#/ { print $1 }' resources.dat`

pids=
for i in $sites ; do
    # create a work directory
    workdir=$i
    if [ ! -d $workdir ]; then
        mkdir $workdir
    fi
    ( cd $workdir
        endpoint=`awk '$1 == "'$i'" { print $2 }' ../resources.dat`
        jobfile=pkgsrc-status.jdl
        if [ ! -r $jobfile ]; then
            cat > $jobfile <<EOF
Executable = "pkgsrc-cmd.sh";
Arguments = "-d check";
Stdoutput = "stdout";
StdError = "stderror";
InputSandbox = "../pkgsrc-cmd.sh";
OutputSandbox = {"stdout","status.txt"};
EOF
        fi
        if grep ^https jids ; then
            echo "DEBUG: jids file already has jobs, skipping $i"
        else
            echo "DEBUG: glite-wms-job-submit -d $delegation -o jids -r $endpoint $jobfile"
            glite-wms-job-submit -d $delegation -o jids -r $endpoint $jobfile
            if [ $? -ne 0 ]; then
                echo "Error: Failed to submit job to $i" >&2
            fi
        fi
    ) &
    pids="$pids $!"
done

for i in $pids ; do
    wait $i
    if [ $? -ne 0 ] ; then
        echo "ERROR: job $i failed."
    fi
done

# wait and poll; systematically go through all the jid files and check their
# status. When a job has finished, fetch the results and clean up.

# $1 is job id
get_job_output() {
    jobhash=`echo $1 | sed -e 's,.*/,,'`

#    mkdir $jobdir || die "can't mkdir $jobdir"
    glite-wms-job-output --noint --nosubdir --logfile $jobhash.log --dir $jobhash $1
    if [ $? -ne 0 ] ; then
        echo "failed to get job output for $1" >&2
        echo "job output retrieval failure" >&3
        return 1
    fi
    # just append
    cat $jobhash/status.txt >> status
    cat $jobhash/stdout >> job.log
    rm -rf $jobhash
}

# $1 = jobid $2 = state
get_logging_and_clear() {
    # get the logging info for the job ...
    echo "job $1 is $2, getting logging and removing it"
    glite-wms-job-logging-info -o joblog-$jobhash.log \
        --noint -v 3 $1
    # ... then remove the jobid from the list.
    grep -v -F "$1" jids > jids.new && mv jids.new jids
}

waitlonger=1

while [ $waitlonger -eq 1 ]; do
    waitlonger=0
    for i in $sites ; do
            cd $i
            if test -f jids && grep -q '^https:' jids; then
                jobs=`grep -v '^#' jids`
                for j in $jobs ; do
                    jobhash=`echo $j | sed -e 's,.*/,,'`
                    rm -f jobstate
                    glite-wms-job-status --noint --logfile joblog -o jobstate $j > /dev/null 2>&1
                    if [ $? -ne 0 ]; then
                        echo "failed to retrieve job status for $j; skipping."
                        break
                    fi
                    # grep the status file for "Current Status"
                    state=`sed -n -e '/Current Status/  s/.*:\s*// p' jobstate`
                    case $state in
                        "Done"* )
                        # job is done, get the output
                            echo "job $j is done, getting output"
                            get_job_output $j
                            get_logging_and_clear $j $state
                            rm jobstate
                            ;;
                        Cleared* | Aborted* | Cancelled* ) 
                            get_logging_and_clear $j $state
                            rm jobstate
                            ;;
                        * )
                             # job still in queue.
                            waitlonger=1
                            ;;
                    esac
                done
            fi
            cd ..
    done
    sleep 10
done

# fetch results, interpret and publish

1	dennisvd	1904	#!/bin/bash
2
3			# FIXMELATER
4			delegation=dennis
5
6			mopup=0
7
8			while getopts m opt; do
9			case $opt in
10			m)
11			mopup=1
12			;;
13			?)
14			echo "ERROR: unknown option $opt."
15			exit 1
16			;;
17			esac
18			done
19			shift $(($OPTIND - 1))
20
21			# look up the resources in the data file
22
23			if [ ! -r resources.dat ] ; then
24			echo "ERROR: Missing resource datafile resources.dat" >&2
25			exit 1
26			fi
27
28			sites=`awk ' $1 !~ /^#/ { print $1 }' resources.dat`
29
30			pids=
31			for i in $sites ; do
32			# create a work directory
33			workdir=$i
34			if [ ! -d $workdir ]; then
35			mkdir $workdir
36			fi
37			( cd $workdir
38			endpoint=`awk '$1 == "'$i'" { print $2 }' ../resources.dat`
39			jobfile=pkgsrc-status.jdl
40			if [ ! -r $jobfile ]; then
41			cat > $jobfile <<EOF
42			Executable = "pkgsrc-cmd.sh";
43			Arguments = "-d check";
44			Stdoutput = "stdout";
45			StdError = "stderror";
46			InputSandbox = "../pkgsrc-cmd.sh";
47			OutputSandbox = {"stdout","status.txt"};
48			EOF
49			fi
50			if grep ^https jids ; then
51			echo "DEBUG: jids file already has jobs, skipping $i"
52			else
53			echo "DEBUG: glite-wms-job-submit -d $delegation -o jids -r $endpoint $jobfile"
54			glite-wms-job-submit -d $delegation -o jids -r $endpoint $jobfile
55			if [ $? -ne 0 ]; then
56			echo "Error: Failed to submit job to $i" >&2
57			fi
58			fi
59			) &
60			pids="$pids $!"
61			done
62
63			for i in $pids ; do
64			wait $i
65			if [ $? -ne 0 ] ; then
66			echo "ERROR: job $i failed."
67			fi
68			done
69
70			# wait and poll; systematically go through all the jid files and check their
71			# status. When a job has finished, fetch the results and clean up.
72
73			# $1 is job id
74			get_job_output() {
75			jobhash=`echo $1 \| sed -e 's,.*/,,'`
76
77			# mkdir $jobdir \|\| die "can't mkdir $jobdir"
78			glite-wms-job-output --noint --nosubdir --logfile $jobhash.log --dir $jobhash $1
79			if [ $? -ne 0 ] ; then
80			echo "failed to get job output for $1" >&2
81			echo "job output retrieval failure" >&3
82			return 1
83			fi
84			# just append
85			cat $jobhash/status.txt >> status
86			cat $jobhash/stdout >> job.log
87			rm -rf $jobhash
88			}
89
90			# $1 = jobid $2 = state
91			get_logging_and_clear() {
92			# get the logging info for the job ...
93			echo "job $1 is $2, getting logging and removing it"
94			glite-wms-job-logging-info -o joblog-$jobhash.log \
95			--noint -v 3 $1
96			# ... then remove the jobid from the list.
97			grep -v -F "$1" jids > jids.new && mv jids.new jids
98			}
99
100			waitlonger=1
101
102			while [ $waitlonger -eq 1 ]; do
103			waitlonger=0
104			for i in $sites ; do
105			cd $i
106			if test -f jids && grep -q '^https:' jids; then
107			jobs=`grep -v '^#' jids`
108			for j in $jobs ; do
109			jobhash=`echo $j \| sed -e 's,.*/,,'`
110			rm -f jobstate
111			glite-wms-job-status --noint --logfile joblog -o jobstate $j > /dev/null 2>&1
112			if [ $? -ne 0 ]; then
113			echo "failed to retrieve job status for $j; skipping."
114			break
115			fi
116			# grep the status file for "Current Status"
117			state=`sed -n -e '/Current Status/ s/.:\s// p' jobstate`
118			case $state in
119			"Done"* )
120			# job is done, get the output
121			echo "job $j is done, getting output"
122			get_job_output $j
123			get_logging_and_clear $j $state
124			rm jobstate
125			;;
126			Cleared* \| Aborted* \| Cancelled* )
127			get_logging_and_clear $j $state
128			rm jobstate
129			;;
130			* )
131			# job still in queue.
132			waitlonger=1
133			;;
134			esac
135			done
136			fi
137			cd ..
138			done
139			sleep 10
140			done
141
142			# fetch results, interpret and publish
143
grid.support@nikhef.nl	ViewVC Help
Powered by ViewVC 1.1.28