Class: WorkflowManager::FGCZCluster

Inherits:
Cluster
  • Object
show all
Defined in:
lib/workflow_manager/cluster.rb

Direct Known Subclasses

FGCZCourseCluster

Instance Attribute Summary

Attributes inherited from Cluster

#log_dir, #name, #options

Instance Method Summary collapse

Methods inherited from Cluster

#default_node, #generate_new_job_script, #initialize

Constructor Details

This class inherits a constructor from WorkflowManager::Cluster

Instance Method Details

#cluster_nodes ⇒ Object 



165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
 
# File 'lib/workflow_manager/cluster.rb', line 165

def cluster_nodes
  nodes = {
    'fgcz-c-043: cpu 24,mem  23 GB,scr  11T' => 'fgcz-c-043',
    'fgcz-c-044: cpu 16,mem 128 GB,scr 500G' => 'fgcz-c-044',
    'fgcz-c-045: cpu 64,mem 504 GB,scr  15T' => 'fgcz-c-045',
    'fgcz-c-046: cpu 64,mem 504 GB,scr  11T' => 'fgcz-c-046',
    'fgcz-c-047: cpu 32,mem   1 TB,scr  28T' => 'fgcz-c-047',
    'fgcz-c-048: cpu 48,mem 252 GB,scr 3.5T' => 'fgcz-c-048',
    'fgcz-c-049: cpu  8,mem  63 GB,scr 1.7T' => 'fgcz-c-049',
    'fgcz-c-051: cpu  8,mem  31 GB,scr 800G' => 'fgcz-c-051',
    'fgcz-c-052: cpu  8,mem  31 GB,scr 800G' => 'fgcz-c-052',
    'fgcz-c-053: cpu  8,mem  31 GB,scr 800G' => 'fgcz-c-053',
    'fgcz-c-054: cpu  8,mem  31 GB,scr 800G' => 'fgcz-c-054',
    'fgcz-c-055: cpu  8,mem  31 GB,scr 800G' => 'fgcz-c-055',
    'fgcz-c-057: cpu  8,mem  31 GB,scr 200G' => 'fgcz-c-057',
    'fgcz-c-058: cpu  8,mem  31 GB,scr 200G' => 'fgcz-c-058',
    'fgcz-c-059: cpu  8,mem  31 GB,scr 200G' => 'fgcz-c-059',
    'fgcz-c-061: cpu  8,mem  31 GB,scr 200G' => 'fgcz-c-061',
    'fgcz-c-063: cpu 12,mem  70 GB,scr 450G' => 'fgcz-c-063',
    'fgcz-c-065: cpu 24,mem  70 GB,scr 197G' => 'fgcz-c-065',
    'fgcz-h-004: cpu 8,mem  30 GB,scr 400G' => 'fgcz-h-004',
    'fgcz-h-009: cpu 8,mem  30 GB,scr 500G' => 'fgcz-h-009',
    'fgcz-h-010: cpu 8,mem  30 GB,scr 400G' => 'fgcz-h-010',
  }
end

#copy_commands(org_dir, dest_parent_dir, now = nil) ⇒ Object 



149
150
151
152
153
154
155
156
157
158
 
# File 'lib/workflow_manager/cluster.rb', line 149

def copy_commands(org_dir, dest_parent_dir, now=nil)
  commands = if now == "force"
               target_file = File.join(dest_parent_dir, File.basename(org_dir))
               ["g-req copynow -f #{org_dir} #{dest_parent_dir}"]
             elsif now
               ["g-req copynow #{org_dir} #{dest_parent_dir}"]
             else
               ["g-req -w copy #{org_dir} #{dest_parent_dir}"]
             end
end

#delete_command(target) ⇒ Object 



162
163
164
 
# File 'lib/workflow_manager/cluster.rb', line 162

def delete_command(target)
  command = "g-req remove #{target}"
end

#job_ends?(log_file) ⇒ Boolean

Returns:

  • (Boolean) 


124
125
126
127
128
129
130
131
132
133
134
135
 
# File 'lib/workflow_manager/cluster.rb', line 124

def job_ends?(log_file)
  log_flag = false
  IO.popen("tail -n 10 #{log_file} 2> /dev/null") do |io|
    while line=io.gets
      if line =~ /__SCRIPT END__/
        log_flag = true
        break
      end
    end
  end
  log_flag
end

#job_pending?(job_id) ⇒ Boolean

Returns:

  • (Boolean) 


136
137
138
139
140
141
142
143
144
145
146
147
148
 
# File 'lib/workflow_manager/cluster.rb', line 136

def job_pending?(job_id)
 qstat_flag = false
  IO.popen('qstat -u "*"') do |io|
    while line=io.gets
      jobid, prior, name, user, state, *others = line.chomp.split
      if jobid.strip == job_id and state =~ /qw/
        qstat_flag = true
        break
      end
    end
  end
  qstat_flag
end

#job_running?(job_id) ⇒ Boolean

Returns:

  • (Boolean) 


111
112
113
114
115
116
117
118
119
120
121
122
123
 
# File 'lib/workflow_manager/cluster.rb', line 111

def job_running?(job_id)
 qstat_flag = false
  IO.popen('qstat -u "*"') do |io|
    while line=io.gets
      jobid, prior, name, user, state, *others = line.chomp.split
      if jobid.strip == job_id and state == 'r'
        qstat_flag = true
        break
      end
    end
  end
  qstat_flag
end

#kill_command(job_id) ⇒ Object 



159
160
161
 
# File 'lib/workflow_manager/cluster.rb', line 159

def kill_command(job_id)
  command = "qdel #{job_id}"
end

#submit_job(script_file, script_content, option = '') ⇒ Object 



94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
 
# File 'lib/workflow_manager/cluster.rb', line 94

def submit_job(script_file, script_content, option='')
  if script_name = File.basename(script_file) and script_name =~ /\.sh/
    script_name = script_name.split(/\.sh/).first + ".sh"
    new_job_script = generate_new_job_script(script_name, script_content)
    new_job_script_base = File.basename(new_job_script)
    log_file = File.join(@log_dir, new_job_script_base + "_o.log")
    err_file = File.join(@log_dir, new_job_script_base + "_e.log")
    command = "g-sub -o #{log_file} -e #{err_file} #{option} #{new_job_script}"
    job_id = `#{command}`
    job_id = job_id.match(/Your job (\d+) \(/)[1]
    [job_id, log_file, command]
  else
    err_msg = "FGCZCluster#submit_job, ERROR: script_name is not *.sh: #{File.basename(script_file)}"
    warn err_msg
    raise err_msg
  end
end