[padb-devel] [padb] r123 committed - Get slurm support working again on the branch, basically query squeue ...

[codesite-noreply at google.com]

Revision: 123
Author: apittman
Date: Wed Aug 19 07:28:24 2009
Log: Get slurm support working again on the branch, basically query squeue  
for a host
list when setting up the job to know how many hosts to expect signons from.

http://code.google.com/p/padb/source/detail?r=123

Modified:
  /branches/full-duplex/src/padb

=======================================
--- /branches/full-duplex/src/padb	Tue Aug 18 14:13:05 2009
+++ /branches/full-duplex/src/padb	Wed Aug 19 07:28:24 2009
@@ -319,11 +319,12 @@
  };

  $rmgr{"slurm"} = {
-    'is_installed'    => \&slurm_is_installed,
-    'get_active_jobs' => \&slurm_get_jobs,
-    'job_is_running'  => \&slurm_job_is_running,
-    'setup_pcmd'      => \&slurm_setup_pcmd,
-    'find_pids'       => \&slurm_find_pids,
+    'is_installed'           => \&slurm_is_installed,
+    'get_active_jobs'        => \&slurm_get_jobs,
+    'job_is_running'         => \&slurm_job_is_running,
+    'setup_pcmd'             => \&slurm_setup_pcmd,
+    'find_pids'              => \&slurm_find_pids,
+    'require_inner_callback' => 1,
  };

  $rmgr{"local"} = {
@@ -397,6 +398,8 @@
  $conf{"prun-exittimeout"} = 120;
  $conf{"rmgr"}             = "auto";

+$conf{"slurm-job-step"} = 0;
+
  # These settings are passed onto inner only.
  $conf{"edbopt"} = "";

@@ -604,6 +607,7 @@
  $debugModes{"tree"}        = undef;
  $debugModes{"verbose"}     = undef;
  $debugModes{"signon"}      = undef;
+$debugModes{"rmgr"}        = undef;

  sub parse_args_outer {

@@ -2206,6 +2210,21 @@
      return $tasks if $have_tasks;
      return undef;
  }
+
+sub slurm_job_to_nodelist {
+    my $job   = shift;
+    my @steps = `squeue -s -o "%i %N" 2>/dev/null`;
+    return undef if ( $? != 0 );
+
+    my $hosts;
+    my $s = "$job." . $conf{"slurm-job-step"};
+    foreach my $step (@steps) {
+        my ( $left, $right ) = split( " ", $step );
+        $hosts = $right if ( $left eq $s );
+
+    }
+    return $hosts;
+}

  sub slurm_job_is_running {
      my $job    = shift;
@@ -2215,9 +2234,11 @@
  }

  sub slurm_setup_pcmd {
-    my $job  = shift;
-    my $cpus = slurm_job_to_ncpus($job);
-    return ( "srun --jobid=$job", $cpus );
+    my $job   = shift;
+    my $cpus  = slurm_job_to_ncpus($job);
+    my @nodes = slurm_job_to_nodelist($job);
+    my $nc    = $#nodes + 1;
+    return ( "srun --jobid=$job", $cpus, $nc );
  }

   
###############################################################################
@@ -2720,6 +2741,7 @@

  sub get_all_jobids {
      my $user = shift;
+    debug_log( "rmgr", undef, "Loading active jobs list", undef );
      return $rmgr{ $conf{"rmgr"} }{get_active_jobs}($user);
  }

@@ -3683,6 +3705,7 @@
      # Check that the file is mode 100600 (Octal)
      if ( $mode != 33152 ) {
          printf("Wrong permissions on secret file, should be 0600  
($file)\n");
+        exit(1);
      }

      open( SFD, $file ) or return;