| 1 |
gutsche |
1.1 |
#!/usr/bin/env perl
|
| 2 |
|
|
|
| 3 |
|
|
##H This drop box agent initiates tasks on all published datasets.
|
| 4 |
|
|
##H A task is an application to run on a dataset at a particular site.
|
| 5 |
|
|
##H It keeps a record of tasks created and avoids submitting a task too
|
| 6 |
|
|
##H frequently.
|
| 7 |
|
|
##H
|
| 8 |
|
|
##H Usage:
|
| 9 |
|
|
##H
|
| 10 |
gutsche |
1.21 |
##H TaskSource
|
| 11 |
|
|
##H
|
| 12 |
|
|
##H -state agent state directory, including inbox
|
| 13 |
|
|
##H -next next agent to pass the drops to; can be given several times
|
| 14 |
|
|
##H -wait time to wait in seconds between work scans
|
| 15 |
|
|
##H -sec_between_inits seconds between submission of same dataset/tier combination
|
| 16 |
|
|
##H -siteconfig file with list of valid SE and their schedule limits
|
| 17 |
gutsche |
1.25 |
##H -dbsdlspath path to DBS/DLS discovery ascii files
|
| 18 |
gutsche |
1.21 |
##H -max-site-queue default value for maximum events scheduled per site
|
| 19 |
gutsche |
1.22 |
##H -max-jobs maximum number of jobs per project, -1 to deactivate
|
| 20 |
gutsche |
1.21 |
##H -scheduler scheduler
|
| 21 |
|
|
##H -events number of events per job
|
| 22 |
|
|
##H -pattern DBS discovery pattern
|
| 23 |
|
|
##H -validtiers perl string pattern with valid tiers (DIGI|RECO)
|
| 24 |
|
|
##H -simcfg CMSSW parameter-set for tier SIM
|
| 25 |
|
|
##H -simoutput putput (comma separated) for tier SIM
|
| 26 |
|
|
##H -digicfg CMSSW parameter-set for tier DIGI
|
| 27 |
|
|
##H -digioutput putput (comma separated) for tier DIGI
|
| 28 |
|
|
##H -recocfg CMSSW parameter-set for tier RECO
|
| 29 |
|
|
##H -recooutput putput (comma separated) for tier RECO
|
| 30 |
gutsche |
1.1 |
|
| 31 |
|
|
BEGIN {
|
| 32 |
|
|
use strict; use warnings; $^W=1;
|
| 33 |
|
|
our $me = $0; $me =~ s|.*/||;
|
| 34 |
|
|
our $home = $0; $home =~ s|/[^/]+$||; $home ||= "."; $home .= "/../PHEDEX/Toolkit/Common";
|
| 35 |
|
|
unshift(@INC, $home);
|
| 36 |
|
|
}
|
| 37 |
|
|
|
| 38 |
|
|
######################################################################
|
| 39 |
|
|
use UtilsHelp;
|
| 40 |
gutsche |
1.21 |
my %args = (WAITTIME => 600, SECS_PER_EVENT => 1., MAX_SITE_QUEUE => 10);
|
| 41 |
|
|
while (scalar @ARGV) {
|
| 42 |
|
|
if ($ARGV[0] eq '-state' && scalar @ARGV > 1) {
|
| 43 |
|
|
shift (@ARGV); $args{DROPDIR}= shift(@ARGV);
|
| 44 |
|
|
} elsif ($ARGV[0] eq '-next' && scalar @ARGV > 1) {
|
| 45 |
|
|
shift (@ARGV); push (@{$args{NEXTDIR}}, shift(@ARGV));
|
| 46 |
|
|
} elsif ($ARGV[0] eq '-wait' && scalar @ARGV > 1) {
|
| 47 |
|
|
shift (@ARGV); $args{WAITTIME} = shift(@ARGV);
|
| 48 |
|
|
} elsif ($ARGV[0] eq '-secs_between_inits' && scalar @ARGV > 1) {
|
| 49 |
|
|
shift (@ARGV); $args{SECS_BETWEEN_INITS} = shift(@ARGV);
|
| 50 |
|
|
} elsif ($ARGV[0] eq '-siteconfig' && scalar @ARGV > 1) {
|
| 51 |
|
|
shift (@ARGV); $args{SITECONFIG} = shift(@ARGV);
|
| 52 |
gutsche |
1.25 |
} elsif ($ARGV[0] eq '-dbsdlspath' && scalar @ARGV > 1) {
|
| 53 |
|
|
shift (@ARGV); $args{DBSDLSPATH} = shift(@ARGV);
|
| 54 |
gutsche |
1.21 |
} elsif ($ARGV[0] eq '-max-site-queue' && scalar @ARGV > 1) {
|
| 55 |
|
|
shift (@ARGV); $args{MAX_SITE_QUEUE} = shift(@ARGV);
|
| 56 |
gutsche |
1.22 |
} elsif ($ARGV[0] eq '-max-jobs' && scalar @ARGV > 1) {
|
| 57 |
|
|
shift (@ARGV); $args{MAX_JOBS} = shift(@ARGV);
|
| 58 |
gutsche |
1.21 |
} elsif ($ARGV[0] eq '-scheduler' && scalar @ARGV > 1) {
|
| 59 |
|
|
shift (@ARGV); $args{SCHEDULER} = shift(@ARGV);
|
| 60 |
|
|
} elsif ($ARGV[0] eq '-events' && scalar @ARGV > 1) {
|
| 61 |
|
|
shift (@ARGV); $args{EVENTS} = shift(@ARGV);
|
| 62 |
|
|
} elsif ($ARGV[0] eq '-pattern' && scalar @ARGV > 1) {
|
| 63 |
|
|
shift (@ARGV); $args{PATTERN} = shift(@ARGV);
|
| 64 |
|
|
} elsif ($ARGV[0] eq '-validtiers' && scalar @ARGV > 1) {
|
| 65 |
|
|
shift (@ARGV); $args{VALIDTIERS} = shift(@ARGV);
|
| 66 |
|
|
} elsif ($ARGV[0] eq '-simcfg' && scalar @ARGV > 1) {
|
| 67 |
|
|
shift (@ARGV); $args{SIMCFG} = shift(@ARGV);
|
| 68 |
|
|
} elsif ($ARGV[0] eq '-simoutput' && scalar @ARGV > 1) {
|
| 69 |
|
|
shift (@ARGV); $args{SIMOUTPUT} = shift(@ARGV);
|
| 70 |
|
|
} elsif ($ARGV[0] eq '-digicfg' && scalar @ARGV > 1) {
|
| 71 |
|
|
shift (@ARGV); $args{DIGICFG} = shift(@ARGV);
|
| 72 |
|
|
} elsif ($ARGV[0] eq '-digioutput' && scalar @ARGV > 1) {
|
| 73 |
|
|
shift (@ARGV); $args{DIGIOUTPUT} = shift(@ARGV);
|
| 74 |
|
|
} elsif ($ARGV[0] eq '-recocfg' && scalar @ARGV > 1) {
|
| 75 |
|
|
shift (@ARGV); $args{RECOCFG} = shift(@ARGV);
|
| 76 |
|
|
} elsif ($ARGV[0] eq '-recooutput' && scalar @ARGV > 1) {
|
| 77 |
|
|
shift (@ARGV); $args{RECOOUTPUT} = shift(@ARGV);
|
| 78 |
|
|
} elsif ($ARGV[0] eq '-h') {
|
| 79 |
|
|
&usage();
|
| 80 |
|
|
} else {
|
| 81 |
|
|
last;
|
| 82 |
|
|
}
|
| 83 |
gutsche |
1.1 |
}
|
| 84 |
|
|
|
| 85 |
gutsche |
1.21 |
if (@ARGV || !$args{DROPDIR}) {
|
| 86 |
|
|
die "Insufficient parameters, use -h for help.\n";
|
| 87 |
gutsche |
1.1 |
}
|
| 88 |
|
|
|
| 89 |
|
|
(new TaskSource (%args))->process();
|
| 90 |
|
|
|
| 91 |
|
|
######################################################################
|
| 92 |
|
|
# Routines specific to this agent.
|
| 93 |
|
|
package TaskSource; use strict; use warnings; use base 'UtilsAgent';
|
| 94 |
|
|
use File::Path;
|
| 95 |
|
|
use UtilsCommand;
|
| 96 |
|
|
use UtilsLogging;
|
| 97 |
|
|
use UtilsTiming;
|
| 98 |
|
|
use UtilsNet;
|
| 99 |
|
|
use POSIX;
|
| 100 |
|
|
|
| 101 |
|
|
sub new
|
| 102 |
gutsche |
1.21 |
{
|
| 103 |
gutsche |
1.1 |
my $proto = shift;
|
| 104 |
|
|
my $class = ref($proto) || $proto;
|
| 105 |
|
|
my $self = $class->SUPER::new(@_);
|
| 106 |
gutsche |
1.21 |
my %params = (MAX_SITE_QUEUE => undef, # max number of jobs per site
|
| 107 |
gutsche |
1.22 |
MAX_JOBS => undef, # max number of jobs per project, -1 to deactivate
|
| 108 |
gutsche |
1.21 |
SECS_BETWEEN_INITS => 600, # path to siteconfig file
|
| 109 |
|
|
SITECONFIG => undef, # path to siteconfig file
|
| 110 |
gutsche |
1.25 |
DBSDLSPATH => "", # path to DBSDLS discovery ascii files
|
| 111 |
gutsche |
1.21 |
EVENTS => 1000, # number of events per job
|
| 112 |
|
|
PATTERN => "*", # DBS discovery pattern
|
| 113 |
|
|
VALIDTIERS => undef, # perl string pattern with valid tiers (DIGI|RECO)
|
| 114 |
|
|
SIMCFG => undef, # CMSSW parameter-set for SIM tier
|
| 115 |
|
|
DIGICFG => undef, # CMSSW parameter-set for DIGI tier
|
| 116 |
|
|
RECOCFG => undef, # CMSSW parameter-set for RECO tier
|
| 117 |
|
|
SIMOUTPUT => undef, # output (comma separated) for SIM tier
|
| 118 |
|
|
DIGIOUTPUT => undef, # output (comma separated) for DIGI tier
|
| 119 |
|
|
RECOOUTPUT => undef, # output (comma separated) for RECO tier
|
| 120 |
|
|
SCHEDULER => "edg"); # standard scheduler
|
| 121 |
gutsche |
1.1 |
my %args = (@_);
|
| 122 |
|
|
map { $self->{$_} = defined $args{$_} ? $args{$_} : $params{$_} } keys %params;
|
| 123 |
|
|
bless $self, $class;
|
| 124 |
|
|
return $self;
|
| 125 |
gutsche |
1.21 |
}
|
| 126 |
gutsche |
1.1 |
|
| 127 |
|
|
sub init
|
| 128 |
gutsche |
1.21 |
{
|
| 129 |
gutsche |
1.1 |
my ($self) = @_;
|
| 130 |
|
|
$self->{TASKREPO} = "$self->{DROPDIR}/tasks";
|
| 131 |
|
|
-d "$self->{TASKREPO}"
|
| 132 |
gutsche |
1.21 |
|| mkdir "$self->{TASKREPO}"
|
| 133 |
gutsche |
1.1 |
|| die "$self->{TASKREPO}: cannot create directory: $!\n";
|
| 134 |
|
|
|
| 135 |
|
|
# Determine if links supports -dump-width option
|
| 136 |
|
|
$self->{LINKS_OPTS} = [];
|
| 137 |
|
|
open (LINKS_HELP, "links -help 2>/dev/null |");
|
| 138 |
gutsche |
1.4 |
if ( grep(/-no-numbering/, <LINKS_HELP>) ) {
|
| 139 |
|
|
push(@{$self->{LINKS_OPTS}}, qw(-dump-width 300 -no-numbering 1));
|
| 140 |
|
|
} elsif ( grep(/-dump-width/, <LINKS_HELP>) ) {
|
| 141 |
|
|
push(@{$self->{LINKS_OPTS}}, qw(-dump-width 300));
|
| 142 |
gutsche |
1.1 |
}
|
| 143 |
|
|
close (LINKS_HELP);
|
| 144 |
gutsche |
1.21 |
}
|
| 145 |
gutsche |
1.1 |
|
| 146 |
|
|
# Find out how many jobs are pending for each site. This is
|
| 147 |
|
|
# insensitive to the job type, and we only check once in the
|
| 148 |
|
|
# beginning to avoid favouring one dataset over another --
|
| 149 |
|
|
# once we decide to proceed for a site, we submit jobs for
|
| 150 |
|
|
# all datasets.
|
| 151 |
|
|
sub getSiteStatus
|
| 152 |
gutsche |
1.18 |
{
|
| 153 |
gutsche |
1.1 |
my ($self) = @_;
|
| 154 |
|
|
my %result = ();
|
| 155 |
|
|
my $taskrepo = $self->{TASKREPO};
|
| 156 |
gutsche |
1.21 |
foreach my $site (<$taskrepo/*/*>) {
|
| 157 |
|
|
my ($sitename) = ($site =~ m|.*/(.*)|);
|
| 158 |
|
|
foreach my $d (<$site/*/*>) {
|
| 159 |
|
|
if (! -f "$d/JOB_CREATE_LOG.txt") {
|
| 160 |
|
|
$result{$sitename}{C} ||= 0;
|
| 161 |
|
|
$result{$sitename}{C}++;
|
| 162 |
|
|
next;
|
| 163 |
|
|
}
|
| 164 |
gutsche |
1.18 |
|
| 165 |
gutsche |
1.21 |
my $f = (<$d/crab_*/share/db/jobs>)[0];
|
| 166 |
|
|
next if ! defined $f;
|
| 167 |
gutsche |
1.18 |
|
| 168 |
gutsche |
1.21 |
foreach my $status (split(/\n/, &input($f) || '')) {
|
| 169 |
|
|
my @statusarray = split('\|', $status);
|
| 170 |
|
|
$result{$sitename}{$statusarray[1]} ||= 0;
|
| 171 |
|
|
$result{$sitename}{$statusarray[1]}++;
|
| 172 |
|
|
}
|
| 173 |
gutsche |
1.18 |
}
|
| 174 |
gutsche |
1.21 |
}
|
| 175 |
gutsche |
1.18 |
|
| 176 |
gutsche |
1.1 |
return %result;
|
| 177 |
gutsche |
1.18 |
}
|
| 178 |
gutsche |
1.1 |
|
| 179 |
gutsche |
1.21 |
sub createHashFromString
|
| 180 |
|
|
{
|
| 181 |
|
|
my ($self, $string) = @_;
|
| 182 |
|
|
my $cmd = "mktemp /tmp/hash_XXXXXX";
|
| 183 |
|
|
open (HASHFILE, "$cmd 2>/dev/null |") or die "cannot run `$cmd': $!\n";
|
| 184 |
|
|
my $hash_file_name;
|
| 185 |
|
|
foreach my $hash_line (split(/\n/, <HASHFILE> || '')) {
|
| 186 |
|
|
chomp($hash_line);
|
| 187 |
|
|
$hash_file_name = $hash_line;
|
| 188 |
|
|
}
|
| 189 |
|
|
$cmd = "echo $string >> $hash_file_name";
|
| 190 |
|
|
system $cmd;
|
| 191 |
|
|
# open (FILLHASHFILE, "$cmd 2>/dev/null |") or die "cannot run `$cmd': $!\n";
|
| 192 |
|
|
$cmd = "cksum $hash_file_name";
|
| 193 |
|
|
open (CKSUMHASHFILE, "$cmd 2>/dev/null |") or die "cannot run `$cmd': $!\n";
|
| 194 |
|
|
my ($hash, $dummy_one, $dummy_two);
|
| 195 |
|
|
foreach my $hash_output (split(/\n/, <CKSUMHASHFILE> || '')) {
|
| 196 |
|
|
chomp($hash_output);
|
| 197 |
|
|
($hash, $dummy_one, $dummy_two) = split(/ /,$hash_output);
|
| 198 |
|
|
}
|
| 199 |
|
|
$cmd = "rm -f $hash_file_name";
|
| 200 |
|
|
system $cmd;
|
| 201 |
|
|
# open (RMHASHFILE, "$cmd 2>/dev/null |") or die "cannot run `$cmd': $!\n";
|
| 202 |
|
|
|
| 203 |
|
|
return $hash;
|
| 204 |
|
|
}
|
| 205 |
|
|
|
| 206 |
gutsche |
1.1 |
sub idle
|
| 207 |
|
|
{
|
| 208 |
gutsche |
1.21 |
&logmsg ("DBS/DLS discovery");
|
| 209 |
gutsche |
1.1 |
my ($self, @pending) = @_;
|
| 210 |
|
|
eval {
|
| 211 |
|
|
# Get status of how busy the sites are. We obtain this only once
|
| 212 |
|
|
# in order to not favour datasets "early on" in the list.
|
| 213 |
|
|
my %sitestats = $self->getSiteStatus ();
|
| 214 |
gutsche |
1.21 |
if (keys %sitestats) {
|
| 215 |
|
|
my @load;
|
| 216 |
|
|
foreach my $site (sort keys %sitestats) {
|
| 217 |
|
|
push (@load, "$site" . join("", map { " $_=$sitestats{$site}{$_}" }
|
| 218 |
|
|
sort keys %{$sitestats{$site}}));
|
| 219 |
gutsche |
1.1 |
}
|
| 220 |
gutsche |
1.21 |
&logmsg ("current load: ", join ("; ", @load));
|
| 221 |
|
|
}
|
| 222 |
gutsche |
1.1 |
|
| 223 |
gutsche |
1.25 |
# DBS/DLS discovery
|
| 224 |
|
|
# if ascii DBS/DLS discovery file exists in DBSDLSPATH with name: dlsdls_query_$cleaned_pattern.txt
|
| 225 |
|
|
# where cleaned pattern is the pattern with / replaced by _
|
| 226 |
|
|
# take ascii file, otherwise do discovery
|
| 227 |
|
|
my $cleaned_pattern = $self->{PATTERN};
|
| 228 |
|
|
$cleaned_pattern =~ s/\//_/g;
|
| 229 |
|
|
my $patternfile = $self->{DBSDLSPATH} . "/dbsdls_query_" . $cleaned_pattern . ".txt";
|
| 230 |
|
|
if ( -e "$patternfile" ) {
|
| 231 |
|
|
&logmsg("Use DBS/DLS ascii file: $patternfile");
|
| 232 |
|
|
open (PUBLISHED,"< $patternfile");
|
| 233 |
|
|
} else {
|
| 234 |
|
|
my $cmd = $ENV{PYTHONSCRIPT} . "/DBSDLSQuery.py --pattern \"$self->{PATTERN}\"";
|
| 235 |
|
|
open (PUBLISHED, "$cmd 2>/dev/null |") or die "cannot run `$cmd': $!\n";
|
| 236 |
|
|
}
|
| 237 |
gutsche |
1.21 |
while (<PUBLISHED>) {
|
| 238 |
|
|
chomp;
|
| 239 |
|
|
&timeStart($self->{STARTTIME});
|
| 240 |
|
|
my ($datapath, $site, $totalevents) = split(/ /, $_);
|
| 241 |
|
|
# read in siteconfig dynamically
|
| 242 |
|
|
if ( $self->{SITECONFIG} ) {
|
| 243 |
|
|
open(siteconfig_file,"$self->{SITECONFIG}");
|
| 244 |
|
|
my @line_array = <siteconfig_file>;
|
| 245 |
|
|
my $validsites = "(";
|
| 246 |
|
|
for (my $i = 0; $i < scalar @line_array; ++$i) {
|
| 247 |
|
|
if (defined $line_array[$i] ) {
|
| 248 |
|
|
if ( $line_array[$i] !~ /^#/ && $line_array[$i] !~ /^\n/ ) {
|
| 249 |
|
|
my ($tempsite, $tempshortsite, $templimit) = split(/ /,$line_array[$i]);
|
| 250 |
|
|
chomp($tempsite);
|
| 251 |
|
|
$validsites = $validsites . $tempsite . "|";
|
| 252 |
gutsche |
1.19 |
}
|
| 253 |
|
|
}
|
| 254 |
gutsche |
1.5 |
}
|
| 255 |
gutsche |
1.21 |
chop($validsites);
|
| 256 |
|
|
$validsites = $validsites . ")";
|
| 257 |
|
|
close siteconfig_file;
|
| 258 |
gutsche |
1.24 |
if ( $validsites eq ")" ) {
|
| 259 |
|
|
next;
|
| 260 |
|
|
} else {
|
| 261 |
|
|
next if ($validsites && $site !~ /$validsites/);
|
| 262 |
|
|
}
|
| 263 |
gutsche |
1.5 |
}
|
| 264 |
gutsche |
1.22 |
# check if totalevents is larger than $EVENTS * $MAX_JOBS
|
| 265 |
|
|
if ($self->{MAX_JOBS} != -1) {
|
| 266 |
|
|
if ( $self->{MAX_JOBS} * $self->{EVENTS} < $totalevents ) {
|
| 267 |
|
|
$totalevents = $self->{MAX_JOBS} * $self->{EVENTS};
|
| 268 |
|
|
}
|
| 269 |
|
|
}
|
| 270 |
gutsche |
1.21 |
$self->createTask($datapath, $site, $totalevents, %sitestats);
|
| 271 |
gutsche |
1.1 |
}
|
| 272 |
|
|
};
|
| 273 |
|
|
do { chomp ($@); &alert ($@); } if $@;
|
| 274 |
|
|
|
| 275 |
|
|
$self->nap ($self->{WAITTIME});
|
| 276 |
|
|
}
|
| 277 |
|
|
|
| 278 |
|
|
sub createTask()
|
| 279 |
|
|
{
|
| 280 |
corvo |
1.3 |
my ($self) = shift(@_);
|
| 281 |
gutsche |
1.5 |
my ($datasetpath, $site, $totalevents, %sitestats) = @_;
|
| 282 |
|
|
|
| 283 |
|
|
my($n, $dataset, $tier, $owner) = split(/\//, $datasetpath);
|
| 284 |
|
|
|
| 285 |
gutsche |
1.21 |
# filter on tier
|
| 286 |
|
|
next if ($self->{VALIDTIERS} && $tier !~ /$self->{VALIDTIERS}/);
|
| 287 |
|
|
|
| 288 |
|
|
my ($cfg, $output);
|
| 289 |
|
|
if ($tier =~ /RECO/) {
|
| 290 |
|
|
$cfg = $self->{RECOCFG};
|
| 291 |
|
|
$output = $self->{RECOOUTPUT};
|
| 292 |
|
|
} elsif ($tier =~ /DIGI/) {
|
| 293 |
|
|
$cfg = $self->{DIGICFG};
|
| 294 |
|
|
$output = $self->{DIGIOUTPUT};
|
| 295 |
|
|
} elsif ($tier =~ /SIM/) {
|
| 296 |
|
|
$cfg = $self->{SIMCFG};
|
| 297 |
|
|
$output = $self->{SIMOUTPUT};
|
| 298 |
|
|
} else {
|
| 299 |
|
|
&logmsg ("No valid tier, defaults for cfg and output");
|
| 300 |
|
|
$cfg = "robot.cfg";
|
| 301 |
|
|
$output = "";
|
| 302 |
gutsche |
1.5 |
}
|
| 303 |
|
|
|
| 304 |
gutsche |
1.8 |
# take site specific or if not found default max site queue
|
| 305 |
gutsche |
1.20 |
# read in from siteconfig file
|
| 306 |
|
|
my %sitelimits = ();
|
| 307 |
gutsche |
1.21 |
my %siteNameTranslation = ();
|
| 308 |
gutsche |
1.20 |
if ( $self->{SITECONFIG} ) {
|
| 309 |
|
|
open(siteconfig_file,"$self->{SITECONFIG}");
|
| 310 |
|
|
my @line_array = <siteconfig_file>;
|
| 311 |
|
|
for (my $i = 0; $i < scalar @line_array; ++$i) {
|
| 312 |
|
|
if (defined $line_array[$i] ) {
|
| 313 |
|
|
if ( $line_array[$i] !~ /^#/ && $line_array[$i] !~ /^\n/ ) {
|
| 314 |
gutsche |
1.21 |
my ($tempsite, $tempshortsite, $templimit) = split(/ /,$line_array[$i]);
|
| 315 |
gutsche |
1.20 |
chomp($tempsite);
|
| 316 |
gutsche |
1.21 |
chomp($tempshortsite);
|
| 317 |
gutsche |
1.20 |
chomp($templimit);
|
| 318 |
|
|
$sitelimits{$tempsite} = $templimit;
|
| 319 |
gutsche |
1.21 |
$siteNameTranslation{$tempsite} = $tempshortsite;
|
| 320 |
gutsche |
1.20 |
}
|
| 321 |
|
|
}
|
| 322 |
|
|
}
|
| 323 |
|
|
close siteconfig_file;
|
| 324 |
|
|
}
|
| 325 |
gutsche |
1.21 |
|
| 326 |
|
|
# If the site isn't too busy already, ignore.
|
| 327 |
|
|
my $pending = ($sitestats{$siteNameTranslation{$site}}{S} || 0);
|
| 328 |
|
|
$pending += ($sitestats{$siteNameTranslation{$site}}{C} || 0);
|
| 329 |
gutsche |
1.20 |
my $max_queue = $sitelimits{$site} || $self->{MAX_SITE_QUEUE};
|
| 330 |
gutsche |
1.26 |
if ( $pending > $max_queue ) {
|
| 331 |
gutsche |
1.4 |
next;
|
| 332 |
|
|
}
|
| 333 |
|
|
|
| 334 |
|
|
# Find out what is already pending for this task. First find all
|
| 335 |
|
|
# existing tasks in the repository, the latest generation.
|
| 336 |
|
|
my $datestamp = strftime ("%y%m%d", gmtime(time()));
|
| 337 |
gutsche |
1.21 |
my $taskdir = "$self->{TASKREPO}/$datestamp/$siteNameTranslation{$site}/$tier";
|
| 338 |
|
|
|
| 339 |
|
|
# create short unique taskbase from $datestamp.$site.$dataset.$tier.$owner
|
| 340 |
|
|
my $taskbase = $self->createHashFromString ("$datestamp.$site.$dataset.$tier.$owner");
|
| 341 |
|
|
chomp($taskbase);
|
| 342 |
|
|
|
| 343 |
gutsche |
1.4 |
my @existing = sort { $a <=> $b } map { /.*\.(\d+)$/ } <$taskdir/$taskbase.*>;
|
| 344 |
|
|
my $curgen = pop(@existing) || 0;
|
| 345 |
|
|
my $nextgen = $curgen + 1;
|
| 346 |
|
|
|
| 347 |
|
|
# OK to create the task if enough time has passed from previous
|
| 348 |
|
|
# task creation, or there is no previous task.
|
| 349 |
|
|
if (! -f "$taskdir/$taskbase.$curgen/crab.cfg"
|
| 350 |
gutsche |
1.21 |
|| (((stat("$taskdir/$taskbase.$curgen/crab.cfg"))[9]
|
| 351 |
|
|
< time() - $self->{SECS_BETWEEN_INITS})))
|
| 352 |
gutsche |
1.4 |
{
|
| 353 |
|
|
my $mydir = $0; $mydir =~ s|/[^/]+$||;
|
| 354 |
|
|
my $drop = sprintf("%s.%03d", $taskbase, $nextgen);
|
| 355 |
gutsche |
1.21 |
my $ret = &runcmd ("$mydir/CrabJobs",
|
| 356 |
|
|
"-cfg", "$mydir/$cfg",
|
| 357 |
|
|
"-datasetpath", $datasetpath,
|
| 358 |
gutsche |
1.4 |
"-output", $output,
|
| 359 |
gutsche |
1.21 |
"-totalevents", $totalevents,
|
| 360 |
|
|
"-whitelist", $site,
|
| 361 |
|
|
"-eventsperjob", $self->{EVENTS},
|
| 362 |
gutsche |
1.4 |
"-scheduler", $self->{SCHEDULER},
|
| 363 |
gutsche |
1.21 |
"-jobname", "$taskdir/$drop");
|
| 364 |
gutsche |
1.4 |
die "$drop: failed to create task: @{[&runerror($ret)]}\n" if $ret;
|
| 365 |
gutsche |
1.21 |
|
| 366 |
gutsche |
1.4 |
&output ("$taskdir/$drop/TASK_INIT.txt",
|
| 367 |
|
|
&mytimeofday () . "\n");
|
| 368 |
|
|
|
| 369 |
|
|
my $dropdir = "$self->{WORKDIR}/$drop";
|
| 370 |
|
|
mkdir "$dropdir" || die "$dropdir: cannot create: $!\n";
|
| 371 |
gutsche |
1.21 |
if (&output ("$dropdir/task", "$taskdir/$drop")) {
|
| 372 |
|
|
&touch ("$dropdir/done");
|
| 373 |
|
|
$self->relayDrop ($drop);
|
| 374 |
gutsche |
1.23 |
&logmsg("stats: $drop $datestamp $site $dataset $tier $owner @{[&formatElapsedTime($self->{STARTTIME})]} success");
|
| 375 |
gutsche |
1.21 |
} else {
|
| 376 |
gutsche |
1.23 |
&alert ("$drop $datestamp $site $dataset $tier $owner: failed to create drop");
|
| 377 |
gutsche |
1.21 |
&rmtree ([ "$self->{WORKDIR}/$drop" ]);
|
| 378 |
|
|
}
|
| 379 |
gutsche |
1.4 |
}
|
| 380 |
|
|
}
|
