torque_helper.rb in osc-machete-2.0.0

- old
+ new

@@ -10,26 +10,31 @@
   # FIXME: Use ood_cluster gem
   LIB = ENV['TORQUE_LIB'] || '/opt/torque/lib64'
   BIN = ENV['TORQUE_BIN'] || '/opt/torque/bin'
   HOSTS = {
     'oakley' => 'oak-batch.osc.edu',
-    'ruby'   => 'ruby-batch.osc.edu',
-    'quick'  => 'quick-batch.osc.edu',
-    'owens'  => 'owens-batch.ten.osc.edu'
+    'ruby'   => 'ruby-batch.ten.osc.edu',
+    'quick'  => 'quick-batch.ten.osc.edu',
+    'owens'  => 'owens-batch.ten.osc.edu',
+    :default => 'oak-batch.osc.edu'
   }
 
-  # Alias to initialize a new object.
-  def self.default
-    self::new()
+  class << self
+    #@!attribute default
+    #  @return [TorqueHelper] default TorqueHelper instance to use
+    attr_writer :default
+    def default
+      @default ||= self::new()
+    end
   end
 
   # Returns an OSC::Machete::Status ValueObject for a char
   #
   # @param [String] char The Torque status char
   #
-  # @example Completed
-  #   status_for_char("C") #=> OSC::Machete::Status.completed
+  # @example Passed
+  #   status_for_char("C") #=> OSC::Machete::Status.passed
   # @example Queued
   #   status_for_char("W") #=> OSC::Machete::Status.queued
   #
   # @return [OSC::Machete::Status] The status corresponding to the char
   def status_for_char(char)
@@ -58,20 +63,10 @@
   # Where depends_on is a hash with key being dependency type and array containing the
   # arguments. See documentation on dependency_list in qsub man pages for details.
   #
   # Bills against the project specified by the primary group of the user.
   def qsub(script, host: nil, depends_on: {}, account_string: nil)
-    # if the script is set to run on Oakley in PBS headers
-    # this is to obviate current torque filter defect in which
-    # a script with PBS header set to specify oak-batch ends
-    # isn't properly handled and the job gets limited to 4GB
-    pbs = PBS::Batch.new(
-      host: HOSTS.fetch( host || host_from_script_pbs_header(script) ),
-      lib: LIB,
-      bin: BIN
-    )
-
     headers = { depend: qsub_dependencies_header(depends_on) }
     headers.clear if headers[:depend].empty?
 
     # currently we set the billable project to the name of the primary group
     # this will probably be both SUPERCOMPUTER CENTER SPECIFIC and must change
@@ -81,11 +76,11 @@
       headers[PBS::ATTR[:A]] = account_string
     elsif account_string_valid_project?(default_account_string)
       headers[PBS::ATTR[:A]] = default_account_string
     end
 
-    pbs.submit_script(script, headers: headers, qsub: true)
+    pbs(host: host, script: script).submit_script(script, headers: headers, qsub: true)
   end
 
   # convert dependencies hash to a PBS header string
   def qsub_dependencies_header(depends_on = {})
     depends_on.map { |x|
@@ -115,17 +110,11 @@
   # @param [String] pbsid The pbsid of the job to inspect.
   #
   # @return [Status] The job state
   def qstat(pbsid, host: nil)
     id = pbsid.to_s
-    pbs = PBS::Batch.new(
-      host: HOSTS.fetch( host || host_from_pbsid(id) ),
-      lib: LIB,
-      bin: BIN
-    )
-
-    status = pbs.get_job(id, filters: [:job_state])
+    status = pbs(host: host, id: id).get_job(id, filters: [:job_state])
     status_for_char status[id][:job_state][0] # get status from status char value
   rescue PBS::UnkjobidError
     OSC::Machete::Status.passed
   end
 
@@ -134,19 +123,32 @@
   # @param [String] pbsid The pbsid of the job to be deleted.
   #
   # @return [nil]
   def qdel(pbsid, host: nil)
     id = pbsid.to_s
-    pbs = PBS::Batch.new(
-      host: HOSTS.fetch( host || host_from_pbsid(id) ),
+    pbs(host: host, id: id).delete_job(id)
+  rescue PBS::UnkjobidError
+    # Common use case where trying to delete a job that is no longer in the system.
+  end
+
+  def pbs(host: nil, id: nil, script: nil)
+    if host
+      # actually check if host is "oakley" i.e. a cluster key
+      host = HOSTS.fetch(host.to_s, host.to_s)
+    else
+      # try to determine host
+      key = host_from_pbsid(id) if id
+      key = host_from_script_pbs_header(script) if script && key.nil?
+
+      host = HOSTS.fetch(key, HOSTS.fetch(:default))
+    end
+
+    PBS::Batch.new(
+      host: host,
       lib: LIB,
       bin: BIN
     )
-
-    pbs.delete_job(id)
-  rescue PBS::UnkjobidError
-    # Common use case where trying to delete a job that is no longer in the system.
   end
 
   private
     # return the name of the host to use based on the pbs header
     # TODO: Think of a more efficient way to do this.
@@ -157,12 +159,10 @@
         "ruby"
       elsif (File.open(script) { |f| f.read =~ /#PBS -q @quick-batch/ })
         "quick"
       elsif (File.open(script) { |f| f.read =~ /#PBS -q @owens-batch/ })
         "owens"
-      else
-        "oakley"  # DEFAULT
       end
     end
 
     # Return the PBS host string based on a full pbsid string
     def host_from_pbsid(pbsid)
@@ -172,10 +172,8 @@
         "ruby"
       elsif (pbsid =~ /quick/ )
         "quick"
       elsif (pbsid =~ /owens/ )
         "owens"
-      else
-        "oakley"  # DEFAULT
       end
     end
 end