Source Code for Module Ganga.GPIDev.Adapters.IApplication

  1  ################################################################################ 
  2  # Ganga Project. http://cern.ch/ganga 
  3  # 
  4  # $Id: IApplication.py,v 1.1 2008-07-17 16:40:52 moscicki Exp $ 
  5  ################################################################################ 
  6   
  7  from Ganga.GPIDev.Base import GangaObject 
  8  from Ganga.GPIDev.Schema import * 
  9   
 10  import Ganga.Utility.logging 
 11  logger = Ganga.Utility.logging.getLogger() 
 12   
 13   
 14 -class PostprocessStatusUpdate(Exception): 
 15      """ This exception may be raised by postprocess hooks to change the job status.""" 
 16 -    def __init__(self,status): 
 17          Exception.__init__(self) 
 18          self.status = status 
 19   
 20   
 21 -class IApplication(GangaObject): 
 22   
 23      """ 
 24       Base class for all application objects. Derived classes represent 
 25       logical  applications  in  the  GPI and  implement  configuration 
 26       handler functionality. The application configuration is the first 
 27       phase of job submission. 
 28   
 29       General rules for implementing the configure methods: 
 30        
 31       In  general the  configure() and  master_configure()  methods are 
 32       called always in  the context of job submission,  so in principle 
 33       you may navigate to  the associated job object (including backend 
 34       information). However it  is not advised to use  backend or extra 
 35       sandbox  information at this  point.  Code  which depends  on the 
 36       backend  should be  put in  application-specific  runtime handler 
 37       which is the next step of job submission. 
 38        
 39       The input/output dataset information may be used if neccessary. 
 40       Objects in the job object tree should not be modified. 
 41   
 42       Efficient implementation of splitting: 
 43   
 44       If you want to enable the  typical case of splitting based on the 
 45       dataset  (defined at  the  job  level) then  it  is very  simple: 
 46       configure()  should only process  the inputdata  part of  the job 
 47       configuration and master_configure() should  do the rest. In that 
 48       case the splitter should not mutate the application object in the 
 49       subjobs, because such changes will not be taken into account (and 
 50       framework will have inconsistent behaviour). 
 51   
 52       You  may  also take  an  extreme  approach  and move  the  entire 
 53       application    configuration     to    configure().     Arbitrary 
 54       modifications of the application  object in the subjobs which are 
 55       done by  the splitter will  take effect.  But if  the application 
 56       configuration  process is time  consuming it  will be  repeated a 
 57       number of times which is inneficient. 
 58   
 59       Otherwise  you   should  first   identify  which  parts   of  the 
 60       application  object may be  altered by  the splitter  and process 
 61       them  in  configure()   method.   The  master_configure()  should 
 62       perform only  the time-consuming part of  the configuration which 
 63       is shared among the subjobs.  This means that splitter should not 
 64       try to  modify the application  parameters which are used  in the 
 65       common master_configure().  """ 
 66   
 67      _schema =  Schema(Version(0,0), {}) 
 68      _category='applications' 
 69      _hidden = 1 
 70   
 71 -    def master_configure(self): 
 72   
 73          """ Configure the shared/master  aspect of the application. 
 74          Return a tuple (modified_flag,  appconfig). 
 75   
 76          This method is always called exactly once, also in the case of 
 77          splitting. 
 78   
 79          Return tuple: 
 80   
 81            - appconfig  (also  known   as  appextra)  is  an  arbitrary 
 82              structure which will  be processed by application-specific 
 83              runtime handler as the next step of job submission. 
 84   
 85            - modified_flag  is True  if application  object  (self) has 
 86              been modified during configure() 
 87   
 88          If this method is not implemented in the derived class then it 
 89          is ignored.  """ 
 90           
 91          return (0,None) 
 92   
 93   
 94 -    def  configure(self,master_appconfig): 
 95   
 96          """ Configure  the specific aspect of the  application .  This 
 97          method has a similar  meaning as master_configure() method and 
 98          it should return a tuple (modified_flag,appconfig). 
 99   
100          This method must be  implemented in a derived class. Otherwise 
101          the submission will fail. 
102   
103          Arguments: 
104            - master_appconfig is a result of the master job master_configure() 
105              method. 
106   
107          In  case of  splitting this  method  will be  called for  each 
108          subjob  object  exactly  once  which  means that  it  will  be 
109          executed as many times as there are subjobs. 
110               
111          If there  is no splitting  this method will be  called exactly 
112          once, after  master_configure(). 
113   
114          Transition from Ganga 4.0.x: 
115           - master_appconfig should be ignored (it is None anyway) 
116          """ 
117   
118          raise NotImplementedError 
119   
120 -    def postprocess(self): 
121          """ Postprocessing after the job was reported as completed. By default do nothing. 
122          This method may raise an exception PostprocessStatusUpdate('failed'). In this 
123          case the job status will be 'failed'. The postprocess_failed() hook will NOT be called.""" 
124          pass 
125   
126 -    def postprocess_failed(self): 
127          """ Postprocessing after the job was reported as failed. By default do nothing.""" 
128          pass 
129       
130 -    def transition_update(self,new_status): 
131          """ 
132          This method will be called just before the status of the parent Job changes to new_status. 
133          The default it to do nothing. 
134          """ 
135          pass 
136