Module: CodeRunner::Slurm

Included in:
Helios
Defined in:
lib/coderunner/system_modules/slurm.rb

Overview

A module to let CodeRunner run using the SLURM queue system, used on certain HPC systems.

Instance Method Summary collapse

Instance Method Details

#batch_script ⇒ Object 



50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
 
# File 'lib/coderunner/system_modules/slurm.rb', line 50

def batch_script

  nodes, ppn = @nprocs.split(/x/)
  eputs "Warning: Underuse of nodes (#{ppn} cores per node instead of #{max_ppn})" if ppn.to_i < max_ppn 
  raise "Error: cores per node cannot excede #{max_ppn}" if ppn.to_i > max_ppn
#   raise "Error: project (i.e. budget) not specified" unless @project
  ppn ||= max_ppn
  raise "Please specify wall minutes" unless @wall_mins
  if @wall_mins
    ep @wall_mins
    hours = (@wall_mins / 60).floor
    mins = @wall_mins.to_i % 60
    secs = ((@wall_mins - @wall_mins.to_i) * 60).to_i
  end
  eputs "Allotted wall time is " + sprintf("%02d:%02d:%02d", hours, mins, secs)
  nprocstot = nodes.to_i * ppn.to_i
"#!/bin/bash\n#SBATCH -J \#{executable_name}.\#{job_identifier} # jobname\n#SBATCH -N \#{nodes.to_i}        # number of nodes\n#SBATCH -n \#{nprocstot}         # number of tasks\n#SBATCH -o \#{executable_name}.\#{job_identifier}.o%j              # strout filename (%j is jobid)\n#SBATCH -e \#{executable_name}.\#{job_identifier}.e%j               # stderr filename (%j is jobid)\n\#{@project ? \"#SBATCH -A \#@project # project to charge\" : \"\"}\n\#{@wall_mins ? \"#SBATCH -t \#{sprintf(\"%02d:%02d:%02d\", hours, mins, secs)} # walltime\" : \"\"}\n\n\#{code_run_environment}\necho \"Submitting \#{nodes}x\#{ppn} job on \#{CodeRunner::SYS} for project \#@project...\"\n\n\n\n"

end

#batch_script_file ⇒ Object 



42
43
44
 
# File 'lib/coderunner/system_modules/slurm.rb', line 42

def batch_script_file
  "#{executable_name}.#{job_identifier}.sh"
end

#cancel_job ⇒ Object 



85
86
87
88
89
90
91
92
 
# File 'lib/coderunner/system_modules/slurm.rb', line 85

def cancel_job
  if ((prefix = ENV['CODE_RUNNER_LAUNCHER']).size > 0 rescue false)
     fname = ENV['HOME'] + "/.coderunner_to_launch_#{prefix}/#{$$}.stop"
     File.open(fname, 'w'){|file| file.puts "\n"}
  else
    `scancel #{@job_no}`
  end
end

#error_file ⇒ Object 



94
95
96
 
# File 'lib/coderunner/system_modules/slurm.rb', line 94

def error_file
  return "#{executable_name}.#{job_identifier}.e#@job_no"
end

#execute ⇒ Object 



26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
 
# File 'lib/coderunner/system_modules/slurm.rb', line 26

def execute
  if ((prefix = ENV['CODE_RUNNER_LAUNCHER']).size > 0 rescue false)
    launch_id = "#{Time.now.to_i}#{$$}"
    fname = ENV['HOME'] + "/.coderunner_to_launch_#{prefix}/#{launch_id}"
    File.open(fname + '.start', 'w'){|file| file.puts "cd #{Dir.pwd};#{run_command}"}
    sleep 1 until FileTest.exist? fname + '.pid'
    pid = File.read(fname + '.pid').to_i
    FileUtils.rm fname + '.pid'
    return pid
  else
    File.open(batch_script_file, 'w'){|file| file.puts batch_script + run_command + "\n"}
    pid = %x[sbatch #{batch_script_file}].to_i
    return nil
  end
end

#get_run_status(job_no, current_status) ⇒ Object 



102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
 
# File 'lib/coderunner/system_modules/slurm.rb', line 102

def get_run_status(job_no, current_status)
  if ((prefix = ENV['CODE_RUNNER_LAUNCHER']).size > 0 rescue false)
    return :Unknown
  end
  line = current_status.split(/\n/).grep(Regexp.new(job_no.to_s))[0]
  unless line
    return :Unknown
  else 
    if line =~ /\sPD\s/
      return :Queueing
    elsif line =~ /\sR\s/
      return :Running
    elsif line =~ /\sC\s/
      return :Unknown
    else
      ep 'line', line
      raise 'Could not get run status'
    end
  end
end

#max_ppn ⇒ Object 



46
47
48
 
# File 'lib/coderunner/system_modules/slurm.rb', line 46

def max_ppn
  raise "Please define max_ppn for your system"
end

#output_file ⇒ Object 



98
99
100
 
# File 'lib/coderunner/system_modules/slurm.rb', line 98

def output_file
  return "#{executable_name}.#{job_identifier}.o#@job_no"
end

#queue_status ⇒ Object 



6
7
8
9
10
11
12
13
 
# File 'lib/coderunner/system_modules/slurm.rb', line 6

def queue_status
  if ((prefix = ENV['CODE_RUNNER_LAUNCHER']).size > 0 rescue false)
    %x[cat #{ENV['HOME']}/.coderunner_to_launch_#{prefix}/queue_status.txt]  +
    %x[cat #{ENV['HOME']}/.coderunner_to_launch_#{prefix}/queue_status2.txt] 
  else
    %x[squeue | grep #{ENV['USER'][0..7]}]
  end
end

#run_command ⇒ Object 



15
16
17
18
19
20
21
22
23
24
 
# File 'lib/coderunner/system_modules/slurm.rb', line 15

def run_command
#     "qsub #{batch_script_file}"
  if (ENV['CODE_RUNNER_LAUNCHER'].size > 0 rescue false)
    return %[mpiexec -np #{@nprocs} #{executable_location}/#{executable_name} #{parameter_string} > #{output_file} 2> #{error_file}]
  else
    nodes, ppn = @nprocs.split(/x/)
    nprocstot = nodes.to_i * ppn.to_i
    "mpirun -np #{nprocstot}  #{executable_location}/#{executable_name} #{parameter_string}"
  end
end