[13] | 1 | #!/usr/bin/perl |
---|
| 2 | # |
---|
| 3 | # 2011/11/27 gabriel |
---|
| 4 | |
---|
| 5 | use strict; |
---|
| 6 | |
---|
| 7 | use Getopt::Long(); |
---|
| 8 | use Pod::Usage; |
---|
| 9 | use Coro; |
---|
| 10 | use Coro::Semaphore; |
---|
| 11 | use Coro::Signal; |
---|
| 12 | use Coro::Channel; |
---|
| 13 | use Coro::Handle; |
---|
| 14 | use IO::File; |
---|
| 15 | use POSIX qw( WNOHANG WEXITSTATUS ); |
---|
[32] | 16 | use Cwd qw( getcwd ); |
---|
[13] | 17 | |
---|
| 18 | my $file = ''; |
---|
| 19 | my $verbose; |
---|
[32] | 20 | my $nodefile = $ENV{OAR_NODE_FILE} || ''; |
---|
| 21 | my $masterio; |
---|
[13] | 22 | my $switchio; |
---|
| 23 | my $help; |
---|
| 24 | my $oarsh = 'oarsh -q -T'; |
---|
| 25 | |
---|
| 26 | Getopt::Long::GetOptions( |
---|
[32] | 27 | 'file=s' => \$file, |
---|
| 28 | 'verbose' => \$verbose, |
---|
| 29 | 'help' => \$help, |
---|
| 30 | 'oarsh=s' => \$oarsh, |
---|
| 31 | 'nodefile=s' => \$nodefile, |
---|
| 32 | 'masterio=s' => \$masterio, |
---|
| 33 | 'switchio' => \$switchio, |
---|
[13] | 34 | ) || pod2usage( -verbose => 0 ); |
---|
| 35 | pod2usage( -verbose => 2 ) if $help; |
---|
[21] | 36 | pod2usage( -verbose => 2 ) if not -e $file; |
---|
[13] | 37 | |
---|
| 38 | my @job = (); |
---|
| 39 | open( JOB_LIST, '<', "$file" ) or die "can't open $file: $!"; |
---|
| 40 | while (<JOB_LIST>) { |
---|
| 41 | chomp; |
---|
| 42 | next if m/^#/; |
---|
[32] | 43 | next if m/^\s*$/; |
---|
| 44 | push @job, $_ ; |
---|
[13] | 45 | } |
---|
| 46 | close JOB_LIST; |
---|
| 47 | |
---|
[32] | 48 | my $stderr = $ENV{OAR_STDERR} || ''; |
---|
[13] | 49 | $stderr =~ s/\.stderr$//; |
---|
[32] | 50 | $stderr = $masterio if $masterio; |
---|
| 51 | my $stdout = $ENV{OAR_STDOUT} || ''; |
---|
[13] | 52 | $stdout =~ s/\.stdout$//; |
---|
[32] | 53 | $stdout = $masterio if $masterio; |
---|
[13] | 54 | |
---|
[32] | 55 | my $current_dir = getcwd(); |
---|
| 56 | |
---|
[13] | 57 | my $finished = new Coro::Signal; |
---|
| 58 | my $job_todo = new Coro::Semaphore 0; |
---|
| 59 | $job_todo->up for (@job); |
---|
| 60 | |
---|
| 61 | my $ressources = new Coro::Channel; |
---|
[32] | 62 | open( NODE_FILE, '<', "$nodefile" ) |
---|
| 63 | or die "can't open $nodefile: $!"; |
---|
[13] | 64 | while (<NODE_FILE>) { |
---|
| 65 | chomp; |
---|
[32] | 66 | next if m/^#/; |
---|
| 67 | next if m/^\s*$/; |
---|
[13] | 68 | $ressources->put($_); |
---|
| 69 | } |
---|
| 70 | close NODE_FILE; |
---|
| 71 | |
---|
| 72 | my $job_num = 0; |
---|
| 73 | my %scheduled = (); |
---|
| 74 | |
---|
| 75 | async { |
---|
| 76 | for my $job (@job) { |
---|
| 77 | my $node = $ressources->get; |
---|
| 78 | |
---|
| 79 | $job_num++; |
---|
| 80 | |
---|
| 81 | my $fh = IO::File->new(); |
---|
| 82 | my $job_pid = $fh->open("| $oarsh $node >/dev/null 2>&1") |
---|
| 83 | or die "don't start subjob: $!"; |
---|
| 84 | |
---|
| 85 | $fh->autoflush; |
---|
| 86 | $fh = unblock $fh; |
---|
| 87 | |
---|
| 88 | $scheduled{$job_pid} = { fh => $fh, node => $node, num => $job_num }; |
---|
| 89 | |
---|
| 90 | printf "start job %5i / %5i on node %s at %s\n", |
---|
| 91 | $job_num, $job_pid, $node, time |
---|
| 92 | if $verbose; |
---|
| 93 | |
---|
| 94 | my ( $job_stdout, $job_stderr ); |
---|
| 95 | $job_stdout = "> $stdout-$job_num.stdout" if $stdout ne '' and $switchio; |
---|
| 96 | $job_stderr = "2> $stderr-$job_num.stderr" if $stderr ne '' and $switchio; |
---|
| 97 | |
---|
[32] | 98 | $fh->print("cd $current_dir\n"); |
---|
[13] | 99 | $fh->print("$job $job_stdout $job_stderr\n"); |
---|
| 100 | $fh->print("exit\n"); |
---|
| 101 | cede; |
---|
| 102 | } |
---|
| 103 | } |
---|
| 104 | |
---|
| 105 | async { |
---|
| 106 | while () { |
---|
| 107 | for my $job_pid ( keys %scheduled ) { |
---|
| 108 | if ( waitpid( $job_pid, WNOHANG ) ) { |
---|
| 109 | printf "end job %5i / %5i on node %s at %s\n", |
---|
| 110 | $scheduled{$job_pid}->{num}, |
---|
| 111 | $job_pid, $scheduled{$job_pid}->{node}, time |
---|
| 112 | if $verbose; |
---|
| 113 | close $scheduled{$job_pid}->{fh}; |
---|
| 114 | $ressources->put( $scheduled{$job_pid}->{node} ); |
---|
| 115 | $job_todo->down; |
---|
| 116 | delete $scheduled{$job_pid}; |
---|
| 117 | } |
---|
| 118 | cede; |
---|
| 119 | } |
---|
| 120 | |
---|
| 121 | $finished->send if $job_todo->count == 0; |
---|
| 122 | cede; |
---|
| 123 | } |
---|
| 124 | } |
---|
| 125 | |
---|
| 126 | cede; |
---|
| 127 | |
---|
| 128 | $finished->wait; |
---|
| 129 | |
---|
| 130 | __END__ |
---|
| 131 | |
---|
| 132 | =head1 NAME |
---|
| 133 | |
---|
| 134 | oar-parexec - parallel execute lot of small job |
---|
| 135 | |
---|
| 136 | =head1 SYNOPSIS |
---|
| 137 | |
---|
[32] | 138 | oar-parexec --file filecommand [--verbose] [--nodefile filenode] [--masterio basefileio] [--switchio] [--oarsh sssh] |
---|
[13] | 139 | oar-parexec --help |
---|
| 140 | |
---|
[32] | 141 | =head1 DESCRIPTION |
---|
| 142 | |
---|
| 143 | C<oar-parexec> execute lot of small job.in parallel inside a cluster. |
---|
| 144 | Number of parallel job at one time cannot excede core number in the node file. |
---|
| 145 | C<oar-parexec> is easier to use inside an OAR job environment |
---|
| 146 | which define automatically theses strategics parameters... |
---|
| 147 | |
---|
| 148 | Option C<--file> is the only mandatory one. |
---|
| 149 | |
---|
| 150 | Small job will be launch in the same folder as the master job. |
---|
| 151 | |
---|
| 152 | |
---|
[13] | 153 | =head1 OPTIONS |
---|
| 154 | |
---|
[32] | 155 | =over 12 |
---|
[13] | 156 | |
---|
[32] | 157 | =item B<-f|--file filecommand> |
---|
[13] | 158 | |
---|
[32] | 159 | File name which content job list. |
---|
[13] | 160 | |
---|
[32] | 161 | =item B<-v|--verbose> |
---|
[13] | 162 | |
---|
[32] | 163 | =item B<-n|nodefile filenode> |
---|
[13] | 164 | |
---|
[32] | 165 | File name that list all the node to launch job. |
---|
| 166 | By defaut, it's define automatically by OAR via |
---|
| 167 | environment variable C<OAR_NODE_FILE>. |
---|
[13] | 168 | |
---|
[32] | 169 | For example, if you want to use 6 core on your cluster node, |
---|
| 170 | you need to put 6 times the hostname node in this file, |
---|
| 171 | one per line... |
---|
| 172 | It's a very common file in MPI process ! |
---|
[13] | 173 | |
---|
[32] | 174 | =item B<-m|--masterio basefileio> |
---|
[13] | 175 | |
---|
[32] | 176 | The C<basefileio> will be use in place of environment variable |
---|
| 177 | C<OAR_STDOUT> and C<OAR_STDERR> (without extension) to build the base name of the small job standart output |
---|
| 178 | (only when option C<swithio> is activated). |
---|
[13] | 179 | |
---|
[32] | 180 | =item B<-s|--switchio> |
---|
[21] | 181 | |
---|
[32] | 182 | Each small job will have it's own output STDOUT and STDERR |
---|
| 183 | base on master OAR job with C<JOB_NUM> inside |
---|
| 184 | (or base on C<basefileio> if option C<masterio>). |
---|
| 185 | Example : |
---|
[21] | 186 | |
---|
[32] | 187 | OAR.151524.stdout -> OAR.151524-JOB_NUM.stdout |
---|
[21] | 188 | |
---|
[32] | 189 | where 151524 here is the master C<OAR_JOB_ID> |
---|
| 190 | and C<JOB_NUM> is the small job nnumber. |
---|
[21] | 191 | |
---|
[32] | 192 | =item B<-o|-oarsh command> |
---|
| 193 | |
---|
| 194 | Command use to launch a shell on a node. |
---|
| 195 | By default |
---|
| 196 | |
---|
| 197 | oarsh -q -T |
---|
| 198 | |
---|
| 199 | =item B<-h|--help> |
---|
| 200 | |
---|
| 201 | =back |
---|
| 202 | |
---|
| 203 | |
---|
| 204 | =head1 EXAMPLE |
---|
| 205 | |
---|
[21] | 206 | Content for the job file (option C<--file>) could have: |
---|
| 207 | |
---|
[13] | 208 | - empty line |
---|
| 209 | - comment line begin with # |
---|
| 210 | - valid shell command |
---|
| 211 | |
---|
| 212 | Example where F<$HOME/test/subjob1.sh> is a shell script (executable). |
---|
| 213 | |
---|
| 214 | $HOME/test/subjob1.sh |
---|
| 215 | $HOME/test/subjob2.sh |
---|
| 216 | $HOME/test/subjob3.sh |
---|
| 217 | $HOME/test/subjob4.sh |
---|
[32] | 218 | ... |
---|
[13] | 219 | $HOME/test/subjob38.sh |
---|
| 220 | $HOME/test/subjob39.sh |
---|
| 221 | $HOME/test/subjob40.sh |
---|
| 222 | |
---|
| 223 | These jobs could be launch by |
---|
| 224 | |
---|
| 225 | oarsub -n test -l /core=6,walltime=00:35:00 "oar-parexec -f ./subjob.list.txt" |
---|
| 226 | |
---|
[28] | 227 | |
---|
[21] | 228 | =head1 SEE ALSO |
---|
| 229 | |
---|
| 230 | oar-dispatch, mpilauncher |
---|
| 231 | |
---|
| 232 | |
---|
[13] | 233 | =head1 AUTHORS |
---|
| 234 | |
---|
[21] | 235 | Written by Gabriel Moreau, Grenoble - France |
---|
[13] | 236 | |
---|
[21] | 237 | |
---|
| 238 | =head1 LICENSE AND COPYRIGHT |
---|
| 239 | |
---|
| 240 | GPL version 2 or later and Perl equivalent |
---|
| 241 | |
---|
[28] | 242 | Copyright (C) 2011 Gabriel Moreau / LEGI - CNRS UMR 5519 - France |
---|
[21] | 243 | |
---|