[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 | |
---|
[43] | 18 | my $filecmd = ''; |
---|
[45] | 19 | my $dir = ''; |
---|
| 20 | my $cmd = ''; |
---|
[43] | 21 | my $logtrace = ''; |
---|
[13] | 22 | my $verbose; |
---|
[34] | 23 | my $job_np = 1; |
---|
[32] | 24 | my $nodefile = $ENV{OAR_NODE_FILE} || ''; |
---|
| 25 | my $masterio; |
---|
[13] | 26 | my $switchio; |
---|
| 27 | my $help; |
---|
| 28 | my $oarsh = 'oarsh -q -T'; |
---|
| 29 | |
---|
| 30 | Getopt::Long::GetOptions( |
---|
[43] | 31 | 'filecmd=s' => \$filecmd, |
---|
[45] | 32 | 'dir=s' => \$dir, |
---|
| 33 | 'cmd=s' => \$cmd, |
---|
[43] | 34 | 'logtrace=s' => \$logtrace, |
---|
[32] | 35 | 'verbose' => \$verbose, |
---|
| 36 | 'help' => \$help, |
---|
| 37 | 'oarsh=s' => \$oarsh, |
---|
[34] | 38 | 'jobnp=i' => \$job_np, |
---|
[32] | 39 | 'nodefile=s' => \$nodefile, |
---|
| 40 | 'masterio=s' => \$masterio, |
---|
| 41 | 'switchio' => \$switchio, |
---|
[41] | 42 | ) || pod2usage(-verbose => 0); |
---|
| 43 | pod2usage(-verbose => 2) if $help; |
---|
[45] | 44 | pod2usage(-verbose => 2) if not ( |
---|
| 45 | (-e "$filecmd") |
---|
| 46 | or (-d "$dir" and $cmd ne '') |
---|
| 47 | ); |
---|
[13] | 48 | |
---|
[43] | 49 | # re-run, keep trace of job already done |
---|
[38] | 50 | my %state; |
---|
| 51 | my $log_h = IO::File->new(); |
---|
[45] | 52 | if (-e "$logtrace") { |
---|
[43] | 53 | $log_h->open("< $logtrace") |
---|
| 54 | or die "error: can't read log file: $!"; |
---|
[38] | 55 | while (<$log_h>) { |
---|
[45] | 56 | $state{$1} = 'start' if m/^start\s+job\s+([^\s]+)\s/; |
---|
| 57 | $state{$1} = 'end' if m/^end\s+job\s+([^\s]+)\s/; |
---|
[41] | 58 | } |
---|
[38] | 59 | $log_h->close(); |
---|
| 60 | } |
---|
[43] | 61 | if ($logtrace) { |
---|
| 62 | $log_h->open(">> $logtrace") |
---|
| 63 | or die "error: can't append log file $logtrace: $!"; |
---|
[40] | 64 | $log_h->autoflush; |
---|
[38] | 65 | $log_h = unblock $log_h; |
---|
| 66 | } |
---|
| 67 | |
---|
[43] | 68 | # job to run |
---|
[13] | 69 | my @job = (); |
---|
[45] | 70 | if (-e "$filecmd") { |
---|
| 71 | my $job_num = 0; |
---|
| 72 | open(JOB_LIST, '<', "$filecmd") or die "error: can't open job file $filecmd: $!"; |
---|
| 73 | while (<JOB_LIST>) { |
---|
| 74 | chomp; |
---|
| 75 | next if m/^#/; |
---|
| 76 | next if m/^\s*$/; |
---|
| 77 | $job_num++; |
---|
| 78 | push @job, { name => $job_num, cmd => "$_" }; |
---|
| 79 | } |
---|
| 80 | close JOB_LIST; |
---|
[13] | 81 | } |
---|
[45] | 82 | else { |
---|
| 83 | opendir(DIR, $dir) or die "error: can't open folder $dir: $!"; |
---|
| 84 | while (my $item = readdir(DIR)) { |
---|
| 85 | next if $item =~ m/^\./; |
---|
| 86 | next if $item =~ m/:/; |
---|
| 87 | next if $item =~ m/\.old$/; |
---|
| 88 | next if $item =~ m/\.sav$/; |
---|
| 89 | next if $item =~ m/\.bak$/; |
---|
| 90 | next if $item =~ m/\.no$/; |
---|
| 91 | next unless (-d "$dir/$item"); |
---|
| 92 | push @job, { name => $item, cmd => "( cd $dir/$item/; $cmd )" }; |
---|
| 93 | } |
---|
| 94 | closedir DIR; |
---|
| 95 | } |
---|
[13] | 96 | |
---|
[43] | 97 | # ressources available |
---|
[34] | 98 | my @ressources = (); |
---|
[41] | 99 | open(NODE_FILE, '<', "$nodefile") |
---|
[34] | 100 | or die "can't open $nodefile: $!"; |
---|
| 101 | while (<NODE_FILE>) { |
---|
| 102 | chomp; |
---|
| 103 | next if m/^#/; |
---|
| 104 | next if m/^\s*$/; |
---|
[41] | 105 | push @ressources, $_; |
---|
[34] | 106 | } |
---|
| 107 | close NODE_FILE; |
---|
| 108 | |
---|
| 109 | my $ressource_size = scalar(@ressources); |
---|
[43] | 110 | die "error: not enought ressources jobnp $job_np > ressources $ressource_size" |
---|
[41] | 111 | if $job_np > $ressource_size; |
---|
[34] | 112 | |
---|
| 113 | my $current_dir = getcwd(); |
---|
| 114 | |
---|
[32] | 115 | my $stderr = $ENV{OAR_STDERR} || ''; |
---|
[13] | 116 | $stderr =~ s/\.stderr$//; |
---|
[32] | 117 | $stderr = $masterio if $masterio; |
---|
| 118 | my $stdout = $ENV{OAR_STDOUT} || ''; |
---|
[13] | 119 | $stdout =~ s/\.stdout$//; |
---|
[32] | 120 | $stdout = $masterio if $masterio; |
---|
[13] | 121 | |
---|
| 122 | my $finished = new Coro::Signal; |
---|
| 123 | my $job_todo = new Coro::Semaphore 0; |
---|
[45] | 124 | my $job_name_maxlen; |
---|
| 125 | for (@job) { |
---|
| 126 | $job_todo->up; |
---|
| 127 | $job_name_maxlen = length($_->{name}) if length($_->{name}) > $job_name_maxlen; |
---|
| 128 | } |
---|
[13] | 129 | |
---|
[43] | 130 | # slice of ressources for parallel job |
---|
[13] | 131 | my $ressources = new Coro::Channel; |
---|
[34] | 132 | for my $slot (1 .. int($ressource_size / $job_np)) { |
---|
[41] | 133 | $ressources->put( |
---|
| 134 | join(',', |
---|
| 135 | @ressources[ (($slot - 1) * $job_np) .. (($slot * $job_np) - 1) ]) |
---|
| 136 | ); |
---|
[13] | 137 | } |
---|
| 138 | |
---|
| 139 | my %scheduled = (); |
---|
| 140 | |
---|
[43] | 141 | # OAR checkpoint and default signal SIGUSR2 |
---|
[39] | 142 | my $oar_checkpoint = new Coro::Semaphore 0; |
---|
[42] | 143 | $SIG{USR2} = sub { |
---|
| 144 | print "warning: receive checkpoint at " |
---|
| 145 | . time |
---|
| 146 | . ", no new job, just finishing running job\n" |
---|
| 147 | if $verbose; |
---|
| 148 | $oar_checkpoint->up(); |
---|
| 149 | }; |
---|
[39] | 150 | |
---|
[43] | 151 | # asynchrone start job block |
---|
[13] | 152 | async { |
---|
[43] | 153 | JOB: |
---|
[13] | 154 | for my $job (@job) { |
---|
[45] | 155 | my $job_name = $job->{name}; |
---|
| 156 | my $job_cmd = $job->{cmd}; |
---|
[38] | 157 | |
---|
[43] | 158 | # job has been already run ? |
---|
[45] | 159 | if (exists $state{$job_name}) { |
---|
| 160 | if ($state{$job_name} eq 'start') { |
---|
| 161 | print "warning: job $job_name was not clearly finished, relaunching...\n" |
---|
[41] | 162 | if $verbose; |
---|
| 163 | } |
---|
[45] | 164 | elsif ($state{$job_name} eq 'end') { |
---|
| 165 | delete $state{$job_name}; # free memory |
---|
[41] | 166 | $job_todo->down; |
---|
[45] | 167 | print "warning: job $job_name already run\n" if $verbose; |
---|
[41] | 168 | cede; |
---|
[43] | 169 | next JOB; |
---|
[41] | 170 | } |
---|
| 171 | } |
---|
[40] | 172 | |
---|
[43] | 173 | # take job ressource |
---|
[36] | 174 | my $job_ressource = $ressources->get; |
---|
[13] | 175 | |
---|
[43] | 176 | # no more launch job when OAR checkpointing |
---|
| 177 | last JOB if $oar_checkpoint->count() > 0; |
---|
[39] | 178 | |
---|
[36] | 179 | my ($node_connect) = split ',', $job_ressource; |
---|
[41] | 180 | my $fh = IO::File->new(); |
---|
[34] | 181 | my $job_pid = $fh->open("| $oarsh $node_connect >/dev/null 2>&1") |
---|
[43] | 182 | or die "error: can't start subjob: $!"; |
---|
[13] | 183 | |
---|
| 184 | $fh->autoflush; |
---|
| 185 | $fh = unblock $fh; |
---|
| 186 | |
---|
[41] | 187 | $scheduled{$job_pid} = { |
---|
| 188 | fh => $fh, |
---|
| 189 | node_connect => $node_connect, |
---|
| 190 | ressource => $job_ressource, |
---|
[45] | 191 | name => $job_name |
---|
[41] | 192 | }; |
---|
[13] | 193 | |
---|
[45] | 194 | my $msg = sprintf "start job %${job_name_maxlen}s / %5i at %s on node %s\n", |
---|
| 195 | $job_name, $job_pid, time, $job_ressource; |
---|
[43] | 196 | $log_h->print($msg) if $logtrace; |
---|
[42] | 197 | print($msg) if $verbose; |
---|
[13] | 198 | |
---|
[41] | 199 | my ($job_stdout, $job_stderr); |
---|
[45] | 200 | $job_stdout = "> $stdout-$job_name.stdout" if $stdout ne '' and $switchio; |
---|
| 201 | $job_stderr = "2> $stderr-$job_name.stderr" if $stderr ne '' and $switchio; |
---|
[13] | 202 | |
---|
[45] | 203 | my $job_nodefile = "/tmp/oar-parexec-$ENV{LOGNAME}-$job_name"; |
---|
[34] | 204 | |
---|
[43] | 205 | # set job environment, run it and clean |
---|
[34] | 206 | if ($job_np > 1) { |
---|
[36] | 207 | $fh->print("printf \"" |
---|
[41] | 208 | . join('\n', split(',', $job_ressource,)) |
---|
| 209 | . "\" > $job_nodefile\n"); |
---|
[37] | 210 | $fh->print("OAR_NODE_FILE=$job_nodefile\n"); |
---|
[34] | 211 | $fh->print("OAR_NP=$job_np\n"); |
---|
[37] | 212 | $fh->print("export OAR_NODE_FILE\n"); |
---|
[34] | 213 | $fh->print("export OAR_NP\n"); |
---|
| 214 | $fh->print("unset OAR_MSG_NODEFILE\n"); |
---|
| 215 | } |
---|
[32] | 216 | $fh->print("cd $current_dir\n"); |
---|
[45] | 217 | $fh->print("$job_cmd $job_stdout $job_stderr\n"); |
---|
[34] | 218 | $fh->print("rm -f $job_nodefile\n") if $job_np > 1; |
---|
[13] | 219 | $fh->print("exit\n"); |
---|
| 220 | cede; |
---|
| 221 | } |
---|
| 222 | } |
---|
| 223 | |
---|
[43] | 224 | # asynchrone end job block |
---|
[13] | 225 | async { |
---|
| 226 | while () { |
---|
[41] | 227 | for my $job_pid (keys %scheduled) { |
---|
[43] | 228 | # non blocking PID test |
---|
[41] | 229 | if (waitpid($job_pid, WNOHANG)) { |
---|
[45] | 230 | my $msg = sprintf "end job %${job_name_maxlen}s / %5i at %s on node %s\n", |
---|
| 231 | $scheduled{$job_pid}->{name}, |
---|
[42] | 232 | $job_pid, time, $scheduled{$job_pid}->{ressource}; |
---|
[43] | 233 | $log_h->print($msg) if $logtrace; |
---|
[42] | 234 | print($msg) if $verbose; |
---|
[13] | 235 | close $scheduled{$job_pid}->{fh}; |
---|
[43] | 236 | # leave ressources for another job |
---|
[41] | 237 | $ressources->put($scheduled{$job_pid}->{ressource}); |
---|
[13] | 238 | $job_todo->down; |
---|
| 239 | delete $scheduled{$job_pid}; |
---|
| 240 | } |
---|
| 241 | cede; |
---|
| 242 | } |
---|
| 243 | |
---|
[43] | 244 | # checkpointing ! just finishing running job and quit |
---|
[42] | 245 | $finished->send if $oar_checkpoint->count() > 0 and scalar(keys(%scheduled)) == 0; |
---|
[39] | 246 | |
---|
[42] | 247 | $finished->send if $job_todo->count() == 0; |
---|
[13] | 248 | cede; |
---|
| 249 | } |
---|
| 250 | } |
---|
| 251 | |
---|
| 252 | cede; |
---|
| 253 | |
---|
[43] | 254 | # all job have been done |
---|
[13] | 255 | $finished->wait; |
---|
| 256 | |
---|
[43] | 257 | # close log trace file |
---|
| 258 | $log_h->close() if $logtrace; |
---|
[38] | 259 | |
---|
[13] | 260 | __END__ |
---|
| 261 | |
---|
| 262 | =head1 NAME |
---|
| 263 | |
---|
[44] | 264 | oar-parexec - parallel execution of many small job |
---|
[13] | 265 | |
---|
| 266 | =head1 SYNOPSIS |
---|
| 267 | |
---|
[45] | 268 | oar-parexec --filecmd filecommand [--logtrace tracefile] [--verbose] \ |
---|
| 269 | [--jobnp integer] [--nodefile filenode] [--masterio basefileio] [--switchio] [--oarsh sssh] |
---|
| 270 | oar-parexec --dir foldertoitemize --cmd commandtolaunch [--logtrace tracefile] [--verbose] \ |
---|
| 271 | [--jobnp integer] [--nodefile filenode] [--masterio basefileio] [--switchio] [--oarsh sssh] |
---|
[13] | 272 | oar-parexec --help |
---|
| 273 | |
---|
[32] | 274 | =head1 DESCRIPTION |
---|
| 275 | |
---|
[44] | 276 | C<oar-parexec> can execute lot of small job in parallel inside a cluster. |
---|
| 277 | Number of parallel job at one time cannot exceed the number of core define in the node file |
---|
[32] | 278 | C<oar-parexec> is easier to use inside an OAR job environment |
---|
[44] | 279 | which define automatically these strategics parameters... |
---|
| 280 | However, it can be used outside OAR. |
---|
[32] | 281 | |
---|
[45] | 282 | Option C<--filecmd> or C<--dir> and C<--cmd> are the only mandatory parameters. |
---|
[32] | 283 | |
---|
| 284 | Small job will be launch in the same folder as the master job. |
---|
[44] | 285 | Two environment variable are defined for each small job |
---|
[37] | 286 | and only in case of parallel small job (option C<--jobnp> > 1). |
---|
[32] | 287 | |
---|
[34] | 288 | OAR_NODE_FILE - file that list node for parallel computing |
---|
| 289 | OAR_NP - number of processor affected |
---|
[32] | 290 | |
---|
[44] | 291 | The file define by OAR_NODE_FILE is created in /tmp |
---|
| 292 | on the node before launching the small job |
---|
| 293 | and this file will be delete after job complete. |
---|
[34] | 294 | C<oar-parexec> is a simple script, |
---|
| 295 | OAR_NODE_FILE will not be deleted in case of crash of the master job. |
---|
| 296 | |
---|
[37] | 297 | OAR define other variable that are equivalent to OAR_NODE_FILE: |
---|
| 298 | OAR_NODEFILE, OAR_FILE_NODES, OAR_RESOURCE_FILE... |
---|
| 299 | You can use in your script the OAR original file ressources |
---|
| 300 | by using these variable if you need it. |
---|
| 301 | |
---|
[34] | 302 | |
---|
[13] | 303 | =head1 OPTIONS |
---|
| 304 | |
---|
[32] | 305 | =over 12 |
---|
[13] | 306 | |
---|
[45] | 307 | =item B<-f|--filecmd filecommand> |
---|
[13] | 308 | |
---|
[32] | 309 | File name which content job list. |
---|
[45] | 310 | For the JOB_NAME definition, |
---|
| 311 | the first valid job in the list will have the number 1 and so on... |
---|
[13] | 312 | |
---|
[45] | 313 | =item B<-d|--dir foldertoitemize> |
---|
| 314 | |
---|
| 315 | Command C<--cmd> will be launch in all sub-folder of this master folder. |
---|
| 316 | Files in this folder will be ignored. |
---|
| 317 | Sub-folder name which begin with '.' |
---|
| 318 | or finish with '.old', '.sav', '.bak', '.no' will either be ignored... |
---|
| 319 | |
---|
| 320 | The JOB_NAME is simply the Sub-folder name. |
---|
| 321 | |
---|
| 322 | =item B<-c|--cmd commandtolaunch> |
---|
| 323 | |
---|
| 324 | Command (and argument to it) tha will be launch in all sub-folder |
---|
| 325 | parameter folfer C<--dir> |
---|
| 326 | |
---|
[43] | 327 | =item B<-l|--logtrace tracefile> |
---|
| 328 | |
---|
| 329 | File which log and trace running job. |
---|
[44] | 330 | In case of running the same master command (after crash for example), |
---|
| 331 | only job that are not mark as done will be run again. |
---|
| 332 | Be careful, job mark as running (start but not finish) will be run again. |
---|
[45] | 333 | Tracing is base on the JOB_NAME between multiple run. |
---|
[43] | 334 | |
---|
| 335 | This option is very usefull in case of crash |
---|
| 336 | but also for checkpointing and idempotent OAR job. |
---|
| 337 | |
---|
[32] | 338 | =item B<-v|--verbose> |
---|
[13] | 339 | |
---|
[34] | 340 | =item B<-j|--jobnp integer> |
---|
[13] | 341 | |
---|
[34] | 342 | Number of processor to allocated for each small job. |
---|
| 343 | 1 by default. |
---|
| 344 | |
---|
| 345 | =item B<-n|--nodefile filenode> |
---|
| 346 | |
---|
[44] | 347 | File name that list all the node where job could be launch. |
---|
[32] | 348 | By defaut, it's define automatically by OAR via |
---|
| 349 | environment variable C<OAR_NODE_FILE>. |
---|
[13] | 350 | |
---|
[32] | 351 | For example, if you want to use 6 core on your cluster node, |
---|
| 352 | you need to put 6 times the hostname node in this file, |
---|
| 353 | one per line... |
---|
| 354 | It's a very common file in MPI process ! |
---|
[13] | 355 | |
---|
[32] | 356 | =item B<-m|--masterio basefileio> |
---|
[13] | 357 | |
---|
[32] | 358 | The C<basefileio> will be use in place of environment variable |
---|
| 359 | C<OAR_STDOUT> and C<OAR_STDERR> (without extension) to build the base name of the small job standart output |
---|
[37] | 360 | (only use when option C<swithio> is activated). |
---|
[13] | 361 | |
---|
[32] | 362 | =item B<-s|--switchio> |
---|
[21] | 363 | |
---|
[32] | 364 | Each small job will have it's own output STDOUT and STDERR |
---|
[45] | 365 | base on master OAR job with C<JOB_NAME> inside |
---|
[32] | 366 | (or base on C<basefileio> if option C<masterio>). |
---|
| 367 | Example : |
---|
[21] | 368 | |
---|
[45] | 369 | OAR.151524.stdout -> OAR.151524-JOB_NAME.stdout |
---|
[21] | 370 | |
---|
[32] | 371 | where 151524 here is the master C<OAR_JOB_ID> |
---|
[45] | 372 | and C<JOB_NAME> is the small job name. |
---|
[21] | 373 | |
---|
[32] | 374 | =item B<-o|-oarsh command> |
---|
| 375 | |
---|
| 376 | Command use to launch a shell on a node. |
---|
| 377 | By default |
---|
| 378 | |
---|
[44] | 379 | oarsh -q -T |
---|
[32] | 380 | |
---|
[44] | 381 | Change it to C<ssh> if you are not using an OAR cluster... |
---|
| 382 | |
---|
[32] | 383 | =item B<-h|--help> |
---|
| 384 | |
---|
| 385 | =back |
---|
| 386 | |
---|
| 387 | |
---|
| 388 | =head1 EXAMPLE |
---|
| 389 | |
---|
[44] | 390 | =head2 Simple list of sequential job |
---|
| 391 | |
---|
[43] | 392 | Content for the job file command (option C<--filecmd>) could have: |
---|
[21] | 393 | |
---|
[13] | 394 | - empty line |
---|
| 395 | - comment line begin with # |
---|
| 396 | - valid shell command |
---|
| 397 | |
---|
| 398 | Example where F<$HOME/test/subjob1.sh> is a shell script (executable). |
---|
| 399 | |
---|
| 400 | $HOME/test/subjob1.sh |
---|
| 401 | $HOME/test/subjob2.sh |
---|
| 402 | $HOME/test/subjob3.sh |
---|
| 403 | $HOME/test/subjob4.sh |
---|
[32] | 404 | ... |
---|
[13] | 405 | $HOME/test/subjob38.sh |
---|
| 406 | $HOME/test/subjob39.sh |
---|
| 407 | $HOME/test/subjob40.sh |
---|
| 408 | |
---|
[44] | 409 | These jobs could be launch by: |
---|
[13] | 410 | |
---|
[44] | 411 | oarsub -n test -l /core=6,walltime=04:00:00 "oar-parexec -f ./subjob.list.txt" |
---|
[13] | 412 | |
---|
[44] | 413 | =head2 Parallel job |
---|
[28] | 414 | |
---|
[44] | 415 | You need to put the number of core each small job need with option C<--jobnp>. |
---|
| 416 | If your job is build on OpenMP or MPI, |
---|
| 417 | you can use OAR_NP and OAR_NODE_FILE variables to configure them. |
---|
| 418 | On OAR cluster, you need to use C<oarsh> or a wrapper like C<oar-envsh> |
---|
| 419 | for connexion between node instead of C<ssh>. |
---|
| 420 | |
---|
| 421 | Example with parallel small job on 2 core: |
---|
| 422 | |
---|
| 423 | oarsub -n test -l /core=6,walltime=04:00:00 "oar-parexec -j 2 -f ./subjob.list.txt" |
---|
| 424 | |
---|
| 425 | =head2 Tracing and master crash |
---|
| 426 | |
---|
| 427 | If the master node crash after hours of calculus, everything is lost ? |
---|
| 428 | No, with option C<--logtrace>, |
---|
| 429 | it's possible to remember older result |
---|
| 430 | and not re-run these job the second and next time. |
---|
| 431 | |
---|
| 432 | oarsub -n test -l /core=6,walltime=04:00:00 "oar-parexec -f ./subjob.list.txt -l ./subjob.list.log" |
---|
| 433 | |
---|
| 434 | After a crash or an C<oardel> command, |
---|
| 435 | you can then re-run the same command that will end to execute the jobs in the list |
---|
| 436 | |
---|
| 437 | oarsub -n test -l /core=6,walltime=04:00:00 "oar-parexec -f ./subjob.list.txt -l ./subjob.list.log" |
---|
| 438 | |
---|
| 439 | C<logtrace> file are just plain file. |
---|
| 440 | We use the extension '.log' because these files are automatically |
---|
| 441 | eliminate from our backup system! |
---|
| 442 | |
---|
| 443 | =head2 Checkpointing and Idempotent |
---|
| 444 | |
---|
| 445 | C<oar-parexec> is compatible with the OAR checkpointing. |
---|
| 446 | Il you have 2000 small jobs that need 55h to be done on 6 cores, |
---|
| 447 | you can cut this in small parts. |
---|
| 448 | |
---|
| 449 | For this example, we suppose that each small job need about 10min... |
---|
| 450 | So, we send a checkpoint 12min before the end of the process |
---|
| 451 | to let C<oar-parexec> finish the jobs started. |
---|
| 452 | After being checkpointed, C<oar-parexec> do not start any new small job. |
---|
| 453 | |
---|
| 454 | oarsub -t idempotent -n test -l /core=6,walltime=04:00:00 --checkpoint 720 \ |
---|
| 455 | "oar-parexec -f ./subjob.list.txt -l ./subjob.list.log" |
---|
| 456 | |
---|
| 457 | After 3h48min, the OAR job will begin to stop launching new small job. |
---|
| 458 | When all running small job are finished, it's exit. |
---|
| 459 | But as the OAR job is type C<idempotent>, |
---|
| 460 | OAR will re-submit it as long as all small job are not executed... |
---|
| 461 | |
---|
| 462 | This way, we let other users a chance to use the cluster! |
---|
| 463 | |
---|
| 464 | In this last exemple, we use moldable OAR job with idempotent |
---|
| 465 | to reserve many core for a small time or a few cores for a long time: |
---|
| 466 | |
---|
| 467 | oarsub -t idempotent -n test \ |
---|
| 468 | -l /core=50,walltime=01:05:00 \ |
---|
| 469 | -l /core=6,walltime=04:00:00 \ |
---|
| 470 | --checkpoint 720 \ |
---|
| 471 | "oar-parexec -f ./subjob.list.txt -l ./subjob.list.log" |
---|
| 472 | |
---|
| 473 | |
---|
[21] | 474 | =head1 SEE ALSO |
---|
| 475 | |
---|
[44] | 476 | oar-dispatch, mpilauncher, |
---|
| 477 | orsh, oar-envsh, ssh |
---|
[21] | 478 | |
---|
| 479 | |
---|
[13] | 480 | =head1 AUTHORS |
---|
| 481 | |
---|
[21] | 482 | Written by Gabriel Moreau, Grenoble - France |
---|
[13] | 483 | |
---|
[21] | 484 | |
---|
| 485 | =head1 LICENSE AND COPYRIGHT |
---|
| 486 | |
---|
| 487 | GPL version 2 or later and Perl equivalent |
---|
| 488 | |
---|
[28] | 489 | Copyright (C) 2011 Gabriel Moreau / LEGI - CNRS UMR 5519 - France |
---|
[21] | 490 | |
---|