pbs 1.1.3

data/lib/pbs/error.rb

@@ -0,0 +1,403 @@
+module PBS
+  # The root exception class that all PBS-specific exceptions inherit from
+  class Error < StandardError; end
+
+  # Unknown job ID error
+  class UnkjobidError < Error; end
+
+  # Undefined attribute
+  class NoattrError < Error; end
+
+  # Cannot set attribute, read only or insufficient permission
+  class AttrroError < Error; end
+
+  # Invalid request
+  class IvalreqError < Error; end
+
+  # Unknown request
+  class UnkreqError < Error; end
+
+  # Too many submit retries
+  class ToomanyError < Error; end
+
+  # Unauthorized Request
+  class PermError < Error; end
+
+  # trqauthd unable to authenticate
+  class IffNotFoundError < Error; end
+
+  # Munge executable not found, unable to authenticate
+  class MungeNotFoundError < Error; end
+
+  # Access from host not allowed, or unknown host
+  class BadhostError < Error; end
+
+  # Job with requested ID already exists
+  class JobexistError < Error; end
+
+  # System error
+  class SystemError < Error; end
+
+  # PBS server internal error
+  class InternalError < Error; end
+
+  # Dependent parent job currently in routing queue
+  class RegrouteError < Error; end
+
+  # Unknown/illegal signal name
+  class UnksigError < Error; end
+
+  # Illegal attribute or resource value for
+  class BadatvalError < Error; end
+
+  # Cannot modify attribute while job running
+  class ModatrrunError < Error; end
+
+  # Request invalid for state of job
+  class BadstateError < Error; end
+
+  # Unknown queue
+  class UnkqueError < Error; end
+
+  # Invalid credential
+  class BadcredError < Error; end
+
+  # Expired credential
+  class ExpiredError < Error; end
+
+  # Queue is not enabled
+  class QunoenbError < Error; end
+
+  # Access to queue is denied
+  class QacessError < Error; end
+
+  # Bad UID for job execution
+  class BaduserError < Error; end
+
+  # Job routing over too many hops
+  class HopcountError < Error; end
+
+  # Queue already exists
+  class QueexistError < Error; end
+
+  # Incompatible type
+  class AttrtypeError < Error; end
+
+  # Cannot delete busy queue
+  class QuebusyError < Error; end
+
+  # Queue name too long
+  class QuenbigError < Error; end
+
+  # No support for requested service
+  class NosupError < Error; end
+
+  # Cannot enable queue, incomplete definition
+  class QuenoenError < Error; end
+
+  # Batch protocol error
+  class ProtocolError < Error; end
+
+  # Bad attribute list structure
+  class BadatlstError < Error; end
+
+  # No free connections
+  class NoconnectsError < Error; end
+
+  # No server specified
+  class NoserverError < Error; end
+
+  # Unknown resource type
+  class UnkrescError < Error; end
+
+  # Job exceeds queue resource limits
+  class ExcqrescError < Error; end
+
+  # No default queue specified
+  class QuenodfltError < Error; end
+
+  # Job is not rerunnable
+  class NorerunError < Error; end
+
+  # Job rejected by all possible destinations (check syntax, queue resources, …)
+  class RouterejError < Error; end
+
+  # Time in Route Queue Expired
+  class RouteexpdError < Error; end
+
+  # Execution server rejected request
+  class MomrejectError < Error; end
+
+  # (qsub) cannot access script file
+  class BadscriptError < Error; end
+
+  # Stage-in of files failed
+  class StageinError < Error; end
+
+  # Resource temporarily unavailable
+  class RescunavError < Error; end
+
+  # Bad GID for job execution
+  class BadgrpError < Error; end
+
+  # Maximum number of jobs already in queue
+  class MaxquedError < Error; end
+
+  # Checkpoint busy, may retry
+  class CkpbsyError < Error; end
+
+  # Resource limit exceeds allowable
+  class ExlimitError < Error; end
+
+  # Invalid Account
+  class BadacctError < Error; end
+
+  # Job already in exit state
+  class AlrdyexitError < Error; end
+
+  # Job files not copied
+  class NocopyfileError < Error; end
+
+  # Unknown job id after clean init
+  class CleanedoutError < Error; end
+
+  # No master found for sync job set
+  class NosyncmstrError < Error; end
+
+  # Invalid Job Dependency
+  class BaddependError < Error; end
+
+  # Duplicate entry in list
+  class DuplistError < Error; end
+
+  # Bad DIS based Request Protocol
+  class DisprotoError < Error; end
+
+  # Cannot execute at specified host because of checkpoint or stagein files
+  class ExecthereError < Error; end
+
+  # Sister rejected
+  class SisrejectError < Error; end
+
+  # Sister could not communicate
+  class SiscommError < Error; end
+
+  # Request not allowed: Server shutting down
+  class SvrdownError < Error; end
+
+  # Not all tasks could checkpoint
+  class CkpshortError < Error; end
+
+  # Unknown node
+  class UnknodeError < Error; end
+
+  # Unknown node-attribute
+  class UnknodeatrError < Error; end
+
+  # Server has no node list
+  class NonodesError < Error; end
+
+  # Node name is too big
+  class NodenbigError < Error; end
+
+  # Node name already exists
+  class NodeexistError < Error; end
+
+  # Illegal value for
+  class BadndatvalError < Error; end
+
+  # Mutually exclusive values for
+  class MutualexError < Error; end
+
+  # Modification failed for
+  class GmoderrError < Error; end
+
+  # Server could not connect to MOM
+  class NorelymomError < Error; end
+
+  # No time-share node available
+  class NotsnodeError < Error; end
+
+  # Wrong job type
+  class JobtypeError < Error; end
+
+  # Bad ACL entry in host list
+  class BadaclhostError < Error; end
+
+  # Maximum number of jobs already in queue for user
+  class MaxuserquedError < Error; end
+
+  # Bad type in disallowedTypes list
+  class BaddisallowtypeError < Error; end
+
+  # Queue does not allow interactive jobs
+  class NointeractiveError < Error; end
+
+  # Queue does not allow batch jobs
+  class NobatchError < Error; end
+
+  # Queue does not allow rerunable jobs
+  class NorerunableError < Error; end
+
+  # Queue does not allow nonrerunable jobs
+  class NononrerunableError < Error; end
+
+  # Unknown Array ID
+  class UnkarrayidError < Error; end
+
+  # Bad Job Array Request
+  class BadArrayReqError < Error; end
+
+  # Bad data reading job array from file
+  class BadArrayDataError < Error; end
+
+  # Time out
+  class TimeoutError < Error; end
+
+  # Job not found
+  class JobnotfoundError < Error; end
+
+  # Queue does not allow fault tolerant jobs
+  class NofaulttolerantError < Error; end
+
+  # Queue does not allow fault intolerant jobs
+  class NofaultintolerantError < Error; end
+
+  # Queue does not allow job arrays
+  class NojobarraysError < Error; end
+
+  # Request was relayed to a MOM
+  class RelayedToMomError < Error; end
+
+  # Error allocating memory - out of memory
+  class MemMallocError < Error; end
+
+  # Error allocating controling mutex (lock/unlock)
+  class MutexError < Error; end
+
+  # Error setting thread attributes
+  class ThreadattrError < Error; end
+
+  # Error creating thread
+  class ThreadError < Error; end
+
+  # Error in socket select
+  class SelectError < Error; end
+
+  # Unable to get connection to socket
+  class SocketFaultError < Error; end
+
+  # Error writing data to socket
+  class SocketWriteError < Error; end
+
+  # Error reading data from socket
+  class SocketReadError < Error; end
+
+  # Socket close detected
+  class SocketCloseError < Error; end
+
+  # Error listening on socket
+  class SocketListenError < Error; end
+
+  # Invalid auth type in request
+  class AuthInvalidError < Error; end
+
+  # This functionality is not yet implemented
+  class NotImplementedError < Error; end
+
+  # Queue is currently not available
+  class QuenotavailableError < Error; end
+
+  # tmpdir owned by another user
+  class TmpdiffownerError < Error; end
+
+  # tmpdir exists but is not a directory
+  class TmpnotdirError < Error; end
+
+  # tmpdir cannot be named for job
+  class TmpnonameError < Error; end
+
+  # Cannot open demux sockets
+  class CantopensocketError < Error; end
+
+  # Cannot send join job to all sisters
+  class CantcontactsistersError < Error; end
+
+  # Cannot create tmpdir for job
+  class CantcreatetmpdirError < Error; end
+
+  # Mom is down, cannot run job
+  class BadmomstateError < Error; end
+
+  # Socket information is not accessible
+  class SocketInformationError < Error; end
+
+  # Data on socket does not process correctly
+  class SocketDataError < Error; end
+
+  # Client is not allowed/trusted
+  class ClientInvalidError < Error; end
+
+  # Premature End of File
+  class PrematureEofError < Error; end
+
+  # Error saving file
+  class CanNotSaveFileError < Error; end
+
+  # Error opening file
+  class CanNotOpenFileError < Error; end
+
+  # Error writing file
+  class CanNotWriteFileError < Error; end
+
+  # Job file corrupt
+  class JobFileCorruptError < Error; end
+
+  # Job can not be rerun
+  class JobRerunError < Error; end
+
+  # Can not establish connection
+  class ConnectError < Error; end
+
+  # Job function must be temporarily delayed
+  class JobworkdelayError < Error; end
+
+  # Parameter of function was invalid
+  class BadParameterError < Error; end
+
+  # Continue processing on job. (Not an error)
+  class ContinueError < Error; end
+
+  # Current sub state does not allow trasaction.
+  class JobsubstateError < Error; end
+
+  # Error moving file
+  class CanNotMoveFileError < Error; end
+
+  # Job is being recycled
+  class JobRecycledError < Error; end
+
+  # Job is already in destination queue.
+  class JobAlreadyInQueueError < Error; end
+
+  # Mutex is NULL or otherwise invalid
+  class InvalidMutexError < Error; end
+
+  # The mutex is already locked by this object
+  class MutexAlreadyLockedError < Error; end
+
+  # The mutex has already been unlocked by this object
+  class MutexAlreadyUnlockedError < Error; end
+
+  # Command syntax invalid
+  class InvalidSyntaxError < Error; end
+
+  # A node is down. Check the MOM and host
+  class NodeDownError < Error; end
+
+  # Could not connect to batch server
+  class ServerNotFoundError < Error; end
+
+  # Server busy. Currently no available threads
+  class ServerBusyError < Error; end
+end

data/lib/pbs/job.rb

@@ -0,0 +1,189 @@
+require "socket"
+require "tempfile"
+require "open3"
+
+module PBS
+  class Job
+    HOSTNAME = Socket.gethostname
+
+    attr_accessor :id
+    attr_reader :conn
+
+    # Needs a connection object and headers
+    # Examples of headers found in 'headers.rb'
+    def initialize(args = {})
+      # Job specific args
+      @id = args[:id]
+      @conn = args[:conn] || Conn.new
+    end
+
+    # Put job on hold
+    def hold(args = {})
+      # hold_type::
+      # The parameter, hold_type, contains the type of hold to be applied. The possible values are (default is 'u'):
+      # "u" : Available to the owner of the job, the batch operator and the batch administrator.
+      # "o" : Available to the batch operator and the batch administrator.
+      # "s" : Available only to the batch administrator.
+      hold_type = args[:hold_type] || "u"
+
+      _pbs_hold(hold_type)
+      self
+    end
+
+    # Release job from hold
+    def release(args = {})
+      # hold_type::
+      # The parameter, hold_type, contains the type of hold to be applied. The possible values are (default is 'u'):
+      # "u" : Available to the owner of the job, the batch operator and the batch administrator.
+      # "o" : Available to the batch operator and the batch administrator.
+      # "s" : Available only to the batch administrator.
+      hold_type = args[:hold_type] || "u"
+
+      _pbs_release(hold_type)
+      self
+    end
+
+    # Delete job
+    def delete(args = {})
+      _pbs_delete()
+    end
+
+    # Get status of job by creating a Query object
+    def status(args = {})
+      q = Query.new(type: :job, conn: conn)
+      q.find(args.merge(id: id))[0]
+    end
+
+    # Can submit a script as a file or string
+    # @param args [Hash] The options when submitting a job.
+    # @option args [String] :string The batch script as a string.
+    # @option args [String] :file The batch script file if a string is not supplied.
+    # @option args [Boolean] :qsub (true) Whether the <tt>qsub</tt> command is used from command line.
+    # @option args [Hash] :headers ({}) PBS headers.
+    # @option args [Hash] :resources ({}) PBS resources.
+    # @option args [Hash] :envvars ({}) PBS environment variables.
+    # @raise [Error] if fail to submit batch job.
+    def submit(args)
+      string = args.fetch(:string) { File.open(args[:file]).read }
+      queue  = args.fetch(:queue, nil)
+      qsub   = args.fetch(:qsub, true)
+
+      headers   = args.fetch(:headers,   {})
+      resources = args.fetch(:resources, {})
+      envvars   = args.fetch(:envvars,   {})
+
+      # Create batch script in tmp file, submit, remove tmp file
+      script = Tempfile.new('qsub.')
+      begin
+        script.write string
+        script.close
+        if qsub
+          _qsub_submit(script.path, queue, headers, resources, envvars)
+        else
+          _pbs_submit(script.path, queue, headers, resources, envvars)
+        end
+      ensure
+        script.unlink # deletes the temp file
+      end
+
+      self
+    end
+
+    private
+      # Connect to batch server, put job on hold,
+      # disconnect, and finally check for errors
+      def _pbs_hold(hold_type)
+        conn.connect unless conn.connected?
+        Torque.pbs_holdjob(conn.conn_id, id, hold_type, nil)
+        conn.disconnect
+        Torque.check_for_error
+      end
+
+      # Connect to batch server, release job from hold,
+      # disconnect, and finally check for errors
+      def _pbs_release(hold_type)
+        conn.connect unless conn.connected?
+        Torque.pbs_rlsjob(conn.conn_id, id, hold_type, nil)
+        conn.disconnect
+        Torque.check_for_error
+      end
+
+      # Connect to batch server, delete job,
+      # disconnect, and finally check for errors
+      def _pbs_delete()
+        conn.connect unless conn.connected?
+        Torque.pbs_deljob(conn.conn_id, id, nil)
+        conn.disconnect
+        Torque.check_for_error
+      end
+
+      # Connect to server, submit job with headers,
+      # disconnect, and finally check for errors
+      def _pbs_submit(script, queue, headers, resources, envvars)
+        # Generate attribute hash for this job
+        attribs = _default_headers.merge(headers)
+        attribs[ATTR[:l]] = _default_resources.merge(resources)
+        attribs[ATTR[:v]] = _default_envvars.merge(envvars).map{|k,v| "#{k}=#{v}"}.join(",")
+
+        # Filter some of the attributes
+        attribs[ATTR[:o]].prepend("#{HOSTNAME}:")
+        attribs[ATTR[:e]].prepend("#{HOSTNAME}:")
+
+        # Submit job
+        conn.connect unless conn.connected?
+        attropl = Torque::Attropl.from_hash(attribs)
+        self.id = Torque.pbs_submit(conn.conn_id, attropl, script, queue, nil)
+        conn.disconnect
+        Torque.check_for_error
+      end
+
+      # Submit using system call `qsub`
+      # Note: Do not need to filter as OSC has personal torque filter
+      def _qsub_submit(script, queue, headers, resources, envvars)
+        params = "-q #{queue}@#{conn.server}"
+        params << resources.map{|k,v| " -l '#{k}=#{v}'"}.join("")
+        params << " -v '#{envvars.map{|k,v| "#{k}=#{v}"}.join(",")}'" unless envvars.empty?
+        params << headers.map do |k,v|
+          param = ATTR.key(k)
+          if param && param.length == 1
+            " -#{param} '#{v}'"
+          else
+            " -W '#{k}=#{v}'"
+          end
+        end.join("")
+        cmd = "#{conn.qsub} #{params} #{script}"
+        Open3.popen3(cmd) do |stdin, stdout, stderr, wait_thr|
+          exit_status = wait_thr.value
+          unless exit_status.success?
+            raise PBS::Error, "#{stderr.read}"
+          end
+
+          self.id = stdout.read.chomp   # newline char at end of job id
+        end
+      end
+
+      # Hash representing the job headers
+      def _default_headers
+        {
+          ATTR[:N] => "Jobname",
+          ATTR[:o] => "#{Dir.pwd}/",
+          ATTR[:e] => "#{Dir.pwd}/",
+          ATTR[:S] => "/bin/bash",
+        }
+      end
+
+      # Hash representing the resources used
+      def _default_resources
+        {
+          walltime: "01:00:00",
+        }
+      end
+
+      # Hash representing the PBS working directory
+      def _default_envvars
+        {
+          PBS_O_WORKDIR: "#{Dir.pwd}",
+        }
+      end
+  end
+end