perl/etc/Jobcluster/Jobguard.pl

   1 #!/usr/bin/perl -w
   2
   3 use strict;
   4 use FindBin qw/$Script $RealBin/;
   5 use lib "$RealBin/..";
   6 use Jobcluster::Jobguard qw/:DEFAULT $MAX_R_JOBS $DEL_BAD_LOG $CHECK_USER/;
   7 use Getopt::Long;
   8
   9 my ($shell,$memory,$sub_line,$submit_queue,$subsh_prefix,$check_job_time,$check_job_usage,$jump_user_determine);
  10 my ($log_directory,$interval,$reqsub,$max_eqw_time,$qhost_timespan,$queue_limit,@df_array,$save_bad_logs);
  11 my ($mem_check_sign,$mem_cycle,$mem_exceed,$mem_record);
  12 my $Help;
  13
  14 GetOptions(
  15         "s=s"  => \$shell,
  16         "m=s"  => \$memory,
  17         "l=i"  => \$sub_line,
  18         "q=s"  => \$submit_queue,
  19         "sp=s" => \$subsh_prefix,
  20         "ct"   => \$check_job_time,
  21         "cu"   => \$check_job_usage,
  22         "ju"   => \$jump_user_determine,
  23
  24         "d=s"  => \$log_directory,
  25         "t=i"  => \$interval,
  26         "r"    => \$reqsub,
  27         "eq=i" => \$max_eqw_time,
  28         "qs=i" => \$qhost_timespan,
  29         "ql"   => \$queue_limit,
  30         "df=s" => \@df_array,
  31         "sl"   => \$save_bad_logs,
  32
  33         "ms"   => \$mem_check_sign,
  34         "mc=i" => \$mem_cycle,
  35         "me=s" => \$mem_exceed,
  36         "mr"   => \$mem_record,
  37
  38     "mj:i" => \$MAX_R_JOBS,
  39
  40         "h"    => \$Help
  41 );
  42
  43 my $usage_info = "
  44 Name
  45
  46     Jobguard.pl
  47
  48 Description
  49
  50     works for Jobguard.pm
  51
  52 Version & author
  53
  54     Version: 1.0, 2011-08-15
  55     Author:  JiaWenlong, jiawenlong\@genomics.cn
  56
  57     Welcome any question, bug-report or suggestion, TIA.
  58
  59 Options
  60
  61     NOTE: for detail, please perldoc $RealBin/Jobguard.pm
  62
  63     -s    [s]   shell script. <required>
  64     -m    [s]   memory the shell script will use. <required>
  65     -l    [i]   line-number of the sub shells script splited from your -s. [1]
  66     -q    [s]   queue where you want your jobs to run. (strict format: *.q) [NA]
  67     -sp   [s]   prefix of each sub shells splited from the original shell ['work']
  68     -ct         sign of using system command 'time' to check the time each subshell used. [disabled]
  69     -cu         sign of using system commands 'qstat -j' and 'grep' to get the 'usage' info of each subshell. [disabled]
  70     -ju         sign of jumping userID determination. [disabled]
  71                 If you have used this script before, and be sure that userID got by it will be definitely your RealOne, set '-ju' for sving your 15s time.
  72     -d    [s]   directory where the guard_job_log will creat. [./]
  73                 The directory doesnot need to exist as it will be created automatically.
  74     -t    [i]   cycle time of guarding steps, and it is in second(s). [120,[300]]
  75     -r          sign of reqsub jobs run to error. [disabled]
  76                 It is suggestted that you should set -r if you prepare to run a multi-step-flow, for the existence of error-jobs.
  77     -sl         sign of saving logs of bad-jobs for user to check the errors. [disabled]
  78     -mj   [i]   maximum number of run/qw jobs, while others are hqw. [50]
  79     -eq   [i]   the maximum time of dealing Eqw status. [[60],100].
  80     -qs   [i]   time-span of qhost check. [1]
  81     -ql         the sign of job-submit-queue-limitation in the reqsubing error-jobs. (effective with '-q:*.q') [disabled]
  82     -df   [s]   info about the disk of which free space you want to check. [NA]
  83                 format: disk,min_space,rls_space
  84                 This parameters can be set in several times, so as to you may have several disks to check, but you should set all disks' info validly.
  85     -ms         sign of checking memory when jobs are running. [disabled]
  86     -mc   [i]   time-span of checking memory. (effective with -ms) [5]
  87     -me   [s]   limited memory that the job can exceed its required memory(-m:vf_mem). (effective with -ms) [0.2]
  88                 it can be decimals(fraction{[0,1]} of vf_mem) or specific memory(in uint 'M/m/G/g').
  89     -mr         sign of recording memory guard log. (effective with -ms) [abled]
  90     -h          show this help.
  91
  92 Usage
  93
  94     perl Jobguard.pl -s <shell_for_qsub> -m <memory_for_run_shell> [Options: -h]
  95
  96 ";
  97
  98 die $usage_info if ($Help);
  99 die "perl $Script -s <shell_for_qsub> -m <memory_for_run_shell> [Options: -h]\n" unless($shell && $memory);
 100 my (@qsubmulti,@jobguard);
 101 &std_para();
 102 #------ qsub --------
 103 qsubmulti(@qsubmulti);
 104 #------ check ---------
 105 jobguard(@jobguard);
 106 #-------- sub-routines -----------
 107 sub std_para{
 108         #%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 109         #--- check paras of qsubmulti ---
 110         #%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 111         #------ check the shell --------
 112         die "Cannot find the shellpath: $shell\n" unless(-e $shell);
 113         push @qsubmulti,"-s:$shell";
 114         #------ check the memory -------
 115         if($memory =~ /^\d+$/){
 116                 print STDERR "WARNING: Memory has been reset as 1G\nBecause your input $memory has no unit!\n";
 117                 $memory = '1G';
 118         }
 119         push @qsubmulti,"-m:$memory";
 120         #------- check the subline -----
 121         if($sub_line){
 122                 if($sub_line !~ /^\d+$/){
 123                         print STDERR "WARNING: Subline has been reset as 1\nBecause your input $sub_line is not whole-digital!\n";
 124                         $sub_line = 1;
 125                 }
 126                 push @qsubmulti,"-l:$sub_line";
 127         }
 128         #------- check the submit queue -------
 129         if($submit_queue){
 130                 push @qsubmulti,"-q:$submit_queue";
 131         }
 132         #------- check the subsh prefix -------
 133         if($subsh_prefix){
 134                 push @qsubmulti,"-sp:$subsh_prefix";
 135         }
 136         #------- check the sign of check usage and time ---------
 137         if($check_job_time){
 138                 push @qsubmulti,"-ct:1";
 139         }
 140         if($check_job_usage){
 141                 push @qsubmulti,"-cu:1";
 142         }
 143         #------- check the sign of jump user determine ----------
 144         if($jump_user_determine){
 145                 $CHECK_USER = 0;
 146         }
 147         #%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 148         #--- check paras of jobguard ---
 149         #%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
 150         #------ check the log_dir --------
 151         $log_directory ||= './';
 152         `mkdir -p $log_directory` unless(-d $log_directory);
 153         push @jobguard,"-d:$log_directory";
 154         #------ check the interval -------
 155         if($interval){
 156                 push @jobguard,"-t:$interval";
 157         }
 158         #------- check the reqsub sign -----
 159         if($reqsub){
 160                 push @jobguard,"-r:1";
 161         }
 162         #------- check the max eqw time -----
 163         if($max_eqw_time){
 164                 push @jobguard,"-e:$max_eqw_time";
 165         }
 166         #------- check the qhost time span -----
 167         if($qhost_timespan){
 168                 push @jobguard,"-h:$qhost_timespan";
 169         }
 170         #------- check job queue limit sign ---------
 171         if($queue_limit){
 172                 push @jobguard,"-ql:1";
 173         }
 174         #------- disk free space -------
 175         if(@df_array){
 176                 push @jobguard,"-df:$_" for @df_array;
 177         }
 178         #------- check mem_check_sign --------
 179         if($mem_check_sign){
 180                 push @jobguard,"-ms:1";
 181         }
 182         #------- check mem_cycle -------
 183         if($mem_check_sign && $mem_cycle){
 184                 push @jobguard,"-mc:$mem_cycle";
 185         }
 186         #------- check mem_exceed -------
 187         if($mem_check_sign && $mem_exceed){
 188                 push @jobguard,"-me:$mem_exceed";
 189         }
 190         #------- check memory_record_sign -------
 191         if($mem_record){
 192                 push @jobguard,"-mr:1";
 193         }
 194         #------- check the save of bad-jobs' logs ------
 195         $DEL_BAD_LOG = !$save_bad_logs;
 196 }