Salome HOME
Updated copyright comment
[modules/kernel.git] / src / Container / SALOME_PyNode.py
index 892cca86bc9e7be91af3f1b8b2a98121ed90a3aa..453669085d858685a58a4677176cac180b9efeea 100644 (file)
@@ -1,5 +1,5 @@
 #  -*- coding: iso-8859-1 -*-
-# Copyright (C) 2007-2023  CEA/DEN, EDF R&D, OPEN CASCADE
+# Copyright (C) 2007-2024  CEA, EDF, OPEN CASCADE
 #
 # This library is free software; you can redistribute it and/or
 # modify it under the terms of the GNU Lesser General Public
@@ -29,6 +29,12 @@ import pickle
 import Engines__POA
 import SALOME__POA
 import SALOME
+import logging
+import os
+import sys
+from SALOME_ContainerHelper import ScriptExecInfo
+
+MY_CONTAINER_ENTRY_IN_GLBS = "my_container"
 
 class Generic(SALOME__POA.GenericObj):
   """A Python implementation of the GenericObj CORBA IDL"""
@@ -66,7 +72,7 @@ class PyNode_i (Engines__POA.PyNode,Generic):
     linecache.cache[nodeName]=0,None,code.split('\n'),nodeName
     ccode=compile(code,nodeName,'exec')
     self.context={}
-    self.context["my_container"] = self.my_container
+    self.context[MY_CONTAINER_ENTRY_IN_GLBS] = self.my_container
     exec(ccode, self.context)
 
   def getContainer(self):
@@ -113,8 +119,381 @@ class SenderByte_i(SALOME__POA.SenderByte,Generic):
 
   def sendPart(self,n1,n2):
     return self.bytesToSend[n1:n2]
+
+SALOME_FILE_BIG_OBJ_DIR = "SALOME_FILE_BIG_OBJ_DIR"
+    
+SALOME_BIG_OBJ_ON_DISK_THRES_VAR = "SALOME_BIG_OBJ_ON_DISK_THRES"
+
+# default is 50 MB
+SALOME_BIG_OBJ_ON_DISK_THRES_DFT = 50000000
+
+from ctypes import c_int
+TypeCounter = c_int
+
+def GetSizeOfTCnt():
+  return len( bytes(TypeCounter(0) ) )
+
+def GetSizeOfBufferedReader(f):
+  """
+  This method returns in bytes size of a file openned.
+
+  Args:
+  ----
+      f (io.IOBase): buffered reader returned by open
+      
+  Returns
+  -------
+      int: number of bytes
+  """
+  import io
+  pos = f.tell()
+  f.seek(0,io.SEEK_END)
+  pos2 = f.tell()
+  f.seek(pos,io.SEEK_SET)
+  return pos2-pos
+
+def GetObjectFromFile(fname, visitor = None):
+  with open(fname,"rb") as f:
+    cntb = f.read( GetSizeOfTCnt() )
+    cnt = TypeCounter.from_buffer_copy( cntb ).value
+    if visitor:
+      visitor.setHDDMem( GetSizeOfBufferedReader(f) )
+      visitor.setFileName( fname )
+    obj = pickle.load(f)
+  return obj,cnt
+
+def DumpInFile(obj,fname):
+  with open(fname,"wb") as f:
+    f.write( bytes( TypeCounter(1) ) )
+    f.write( obj )
+
+def IncrRefInFile(fname):
+  with open(fname,"rb") as f:
+    cntb = f.read( GetSizeOfTCnt() )
+  cnt = TypeCounter.from_buffer_copy( cntb ).value
+  with open(fname,"rb+") as f:
+    #import KernelServices ; KernelServices.EntryForDebuggerBreakPoint()
+    f.write( bytes( TypeCounter(cnt+1) ) )
+
+def DecrRefInFile(fname):
+  import os
+  with open(fname,"rb") as f:
+    cntb = f.read( GetSizeOfTCnt() )
+  cnt = TypeCounter.from_buffer_copy( cntb ).value
+  #
+  #import KernelServices ; KernelServices.EntryForDebuggerBreakPoint()
+  if cnt == 1:
+    os.unlink( fname )
+  else:
+    with open(fname,"rb+") as f:
+        f.write( bytes( TypeCounter(cnt-1) ) )
+
+def GetBigObjectOnDiskThreshold():
+  import os
+  if SALOME_BIG_OBJ_ON_DISK_THRES_VAR in os.environ:
+    return int( os.environ[SALOME_BIG_OBJ_ON_DISK_THRES_VAR] )
+  else:
+    return SALOME_BIG_OBJ_ON_DISK_THRES_DFT
+
+def ActivateProxyMecanismOrNot( sizeInByte ):
+  thres = GetBigObjectOnDiskThreshold()
+  if thres == -1:
+    return False
+  else:
+    return sizeInByte > thres
+
+def GetBigObjectDirectory():
+  import os
+  if SALOME_FILE_BIG_OBJ_DIR not in os.environ:
+    raise RuntimeError("An object of size higher than limit detected and no directory specified to dump it in file !")
+  return os.path.expanduser( os.path.expandvars( os.environ[SALOME_FILE_BIG_OBJ_DIR] ) )
+
+def GetBigObjectFileName():
+  """
+  Return a filename in the most secure manner (see tempfile documentation)
+  """
+  import tempfile
+  with tempfile.NamedTemporaryFile(dir=GetBigObjectDirectory(),prefix="mem_",suffix=".pckl") as f:
+    ret = f.name
+  return ret
+
+class BigObjectOnDiskBase:
+  def __init__(self, fileName, objSerialized):
+    """
+    :param fileName: the file used to dump into.
+    :param objSerialized: the object in pickeled form
+    :type objSerialized: bytes
+    """
+    self._filename = fileName
+    # attribute _destroy is here to tell client side or server side
+    # only client side can be with _destroy set to True. server side due to risk of concurrency
+    # so pickled form of self must be done with this attribute set to False.
+    self._destroy = False
+    self.__dumpIntoFile(objSerialized)
+
+  def getDestroyStatus(self):
+    return self._destroy
+
+  def incrRef(self):
+    if self._destroy:
+      IncrRefInFile( self._filename )
+    else:
+      # should never happen !
+      RuntimeError("Invalid call to incrRef !")
+
+  def decrRef(self):
+    if self._destroy:
+      DecrRefInFile( self._filename )
+    else:
+      # should never happen !
+      RuntimeError("Invalid call to decrRef !")
+
+  def unlinkOnDestructor(self):
+    self._destroy = True
+
+  def doNotTouchFile(self):
+    """
+    Method called slave side. The life cycle management of file is client side not slave side.
+    """
+    self._destroy = False
+
+  def __del__(self):
+    if self._destroy:
+      DecrRefInFile( self._filename )
+
+  def getFileName(self):
+    return self._filename
+  
+  def __dumpIntoFile(self, objSerialized):
+    DumpInFile( objSerialized, self._filename )
+
+  def get(self, visitor = None):
+    obj, _ = GetObjectFromFile( self._filename, visitor )
+    return obj
+
+  def __float__(self):
+    return float( self.get() )
+    
+  def __int__(self):
+    return int( self.get() )
+    
+  def __str__(self):
+    obj = self.get()
+    if isinstance(obj,str):
+        return obj
+    else:
+        raise RuntimeError("Not a string")
+      
+class BigObjectOnDisk(BigObjectOnDiskBase):
+  def __init__(self, fileName, objSerialized):
+    BigObjectOnDiskBase.__init__(self, fileName, objSerialized)
+    
+class BigObjectOnDiskListElement(BigObjectOnDiskBase):
+  def __init__(self, pos, length, fileName):
+    self._filename = fileName
+    self._destroy = False
+    self._pos = pos
+    self._length = length
+
+  def get(self, visitor = None):
+    fullObj = BigObjectOnDiskBase.get(self, visitor)
+    return fullObj[ self._pos ]
+    
+  def __getitem__(self, i):
+    return self.get()[i]
+
+  def __len__(self):
+    return len(self.get())
+    
+class BigObjectOnDiskSequence(BigObjectOnDiskBase):
+  def __init__(self, length, fileName, objSerialized):
+    BigObjectOnDiskBase.__init__(self, fileName, objSerialized)
+    self._length = length
+
+  def __getitem__(self, i):
+    return BigObjectOnDiskListElement(i, self._length, self.getFileName())
+
+  def __len__(self):
+    return self._length
+
+class BigObjectOnDiskList(BigObjectOnDiskSequence):
+  def __init__(self, length, fileName, objSerialized):
+    BigObjectOnDiskSequence.__init__(self, length, fileName, objSerialized)
     
+class BigObjectOnDiskTuple(BigObjectOnDiskSequence):
+  def __init__(self, length, fileName, objSerialized):
+    BigObjectOnDiskSequence.__init__(self, length, fileName, objSerialized)
+
+def ProxyfyPickeled( obj, pickleObjInit = None, visitor = None ):
+  """
+  This method return a proxy instance of pickled form of object given in input.
+
+  Args:
+  ----
+      obj (pickelable type) : object to be proxified
+      pickleObjInit (bytes) : Optionnal. Original pickeled form of object to be proxyfied if already computed. If not this method generate it
+
+  Returns
+  -------
+      BigObjectOnDiskBase: proxy instance
+  """
+  pickleObj = pickleObjInit
+  if pickleObj is None:
+    pickleObj = pickle.dumps( obj , pickle.HIGHEST_PROTOCOL )
+  fileName = GetBigObjectFileName()
+  if visitor:
+    visitor.setHDDMem( len(pickleObj) )
+    visitor.setFileName(fileName)
+  if isinstance( obj, list):
+    proxyObj = BigObjectOnDiskList( len(obj), fileName, pickleObj )
+  elif isinstance( obj, tuple):
+    proxyObj = BigObjectOnDiskTuple( len(obj), fileName , pickleObj )
+  else:
+    proxyObj = BigObjectOnDisk( fileName , pickleObj )
+  return proxyObj
+
+def SpoolPickleObject( obj, visitor = None ):
+  import pickle
+  with InOutputObjVisitorCM(visitor) as v:
+    pickleObjInit = pickle.dumps( obj , pickle.HIGHEST_PROTOCOL )
+    if not ActivateProxyMecanismOrNot( len(pickleObjInit) ):
+      return pickleObjInit
+    else:
+      proxyObj = ProxyfyPickeled( obj, pickleObjInit, v.visitor() )
+      pickleProxy = pickle.dumps( proxyObj , pickle.HIGHEST_PROTOCOL )
+      return pickleProxy
+
+from SALOME_ContainerHelper import InOutputObjVisitorCM, InOutputObjVisitor
+
+def UnProxyObjectSimple( obj, visitor = None ):
+  """
+  Method to be called in Remote mode. Alterate the obj _status attribute. 
+  Because the slave process does not participate in the reference counting
+  
+  Args:
+  ----
+      visitor (InOutputObjVisitor): A visitor to keep track of amount of memory on chip and those on HDD
+
+  """
+  with InOutputObjVisitorCM(visitor) as v:
+    logging.debug( "UnProxyObjectSimple {}".format(type(obj)) )
+    if isinstance(obj,BigObjectOnDiskBase):
+      obj.doNotTouchFile()
+      return obj.get( v )
+    elif isinstance( obj, list):
+      retObj = []
+      for elt in obj:
+        retObj.append( UnProxyObjectSimple(elt,v.visitor()) )
+      return retObj
+    else:
+      return obj
+
+def UnProxyObjectSimpleLocal( obj ):
+  """
+  Method to be called in Local mode. Do not alterate the PyObj counter
+  """
+  if isinstance(obj,BigObjectOnDiskBase):
+    return obj.get()
+  elif isinstance( obj, list):
+    retObj = []
+    for elt in obj:
+      retObj.append( UnProxyObjectSimpleLocal(elt) )
+    return retObj
+  else:
+    return obj
+  
+class FileDeleter:
+  def __init__(self, fileName):
+    self._filename = fileName
+  @property
+  def filename(self):
+    return self._filename
+  def __del__(self):
+    import os
+    if os.path.exists( self._filename ):
+      os.unlink( self._filename )
+
+class MonitoringInfo:
+  def __init__(self, pyFileName, outFileName, pid):
+    self._py_file_name = pyFileName
+    self._out_file_name = outFileName
+    self._pid = pid
+
+  @property
+  def pyFileName(self):
+    return self._py_file_name
+
+  @property
+  def pid(self):
+    return self._pid
+  
+  @property
+  def outFileName(self):
+    return self._out_file_name
+
+def LaunchTimeCPUMonitoring( intervalInMs ):
+  """
+  Launch a subprocess monitoring self process.
+  This monitoring subprocess is a python process lauching every intervalInMs ms evaluation of
+  CPU usage and RSS memory.
+  Communication between subprocess and self is done by file.
+  """
+  import KernelBasis
+  def BuildPythonFileForCPUPercent( intervalInMs ):
+    import os
+    import tempfile
+    with tempfile.NamedTemporaryFile(prefix="htop_",suffix=".py") as f:
+      tempPyFile = f.name
+    tempOutFile = "{}.txt".format( os.path.splitext( tempPyFile )[0] )
+    pid = os.getpid()
+    with open(tempPyFile,"w") as f:
+      f.write("""import psutil
+pid = {}
+process = psutil.Process( pid )
+import time
+with open("{}","a") as f:
+  while True:
+    f.write( "{{}}\\n".format( str( process.cpu_percent() ) ) )
+    f.write( "{{}}\\n".format( str( process.memory_info().rss  ) ) )
+    f.flush()
+    time.sleep( {} / 1000.0 )
+""".format(pid, tempOutFile, intervalInMs))
+    return FileDeleter(tempPyFile), FileDeleter(tempOutFile)
+  pyFileName, outFileName = BuildPythonFileForCPUPercent( intervalInMs )
+  pid = KernelBasis.LaunchMonitoring(pyFileName.filename)
+  return MonitoringInfo(pyFileName, outFileName, pid)
+
+def StopMonitoring( monitoringInfo ):
+  """
+  Kill monitoring subprocess.
+
+  Args:
+  ----
+      monitoringInfo (MonitoringInfo): info returned by LaunchMonitoring
+  """
+  import KernelBasis
+  KernelBasis.StopMonitoring(monitoringInfo.pid)
+
+def ReadCPUMemInfo( monitoringInfo ):
+  """
+  Retrieve data of monitoring.
+
+  Args:
+  ----
+      monitoringInfo (MonitoringInfo): info returned by LaunchMonitoring
+  
+  Returns
+  -------
+    list<float,str> : list of pairs. First param of pair is CPU usage. Second param of pair is rss memory usage
+  """
+  import KernelBasis
+  ret = KernelBasis.ReadFloatsInFile( monitoringInfo.outFileName.filename )
+  cpu = ret[::2]
+  mem_rss = [ int(elt) for elt in ret[1::2]]
+  return [(a,b) for a,b in zip(cpu,mem_rss)]
+
 class SeqByteReceiver:
+  # 2GB limit to trigger split into chunks
   CHUNK_SIZE = 2000000000
   def __init__(self,sender):
     self._obj = sender
@@ -130,6 +509,9 @@ class SeqByteReceiver:
   def fetchOneShot(self,size):
     return self._obj.sendPart(0,size)
   def fetchByChunks(self,size):
+      """
+      To avoid memory peak parts over 2GB are sent using EFF_CHUNK_SIZE size.
+      """
       data_for_split_case = bytes(0)
       EFF_CHUNK_SIZE = SeqByteReceiver.CHUNK_SIZE // 8
       iStart = 0 ; iEnd = EFF_CHUNK_SIZE
@@ -139,18 +521,38 @@ class SeqByteReceiver:
         iStart = iEnd; iEnd = min(iStart + EFF_CHUNK_SIZE,size)
       return data_for_split_case
 
+class LogOfCurrentExecutionSession:
+  def __init__(self, handleToCentralizedInst):
+    self._remote_handle = handleToCentralizedInst
+    self._current_instance = ScriptExecInfo()
+
+  def addInfoOnLevel2(self, key, value):
+    setattr(self._current_instance,key,value)
+
+  def finalizeAndPushToMaster(self):
+    self._remote_handle.assign( pickle.dumps( self._current_instance ) )
+
 class PyScriptNode_i (Engines__POA.PyScriptNode,Generic):
   """The implementation of the PyScriptNode CORBA IDL that executes a script"""
-  def __init__(self, nodeName,code,poa,my_container):
+  def __init__(self, nodeName,code,poa,my_container,logscript):
     """Initialize the node : compilation in the local context"""
     Generic.__init__(self,poa)
     self.nodeName=nodeName
     self.code=code
+    self.my_container_py = my_container
     self.my_container=my_container._container
     linecache.cache[nodeName]=0,None,code.split('\n'),nodeName
     self.ccode=compile(code,nodeName,'exec')
     self.context={}
-    self.context["my_container"] = self.my_container
+    self.context[MY_CONTAINER_ENTRY_IN_GLBS] = self.my_container
+    self._log_script = logscript
+    self._current_execution_session = None
+    sys.stdout.flush() ; sys.stderr.flush() # flush to correctly capture log per execution session
+      
+  def __del__(self):
+    # force removal of self.context. Don t know why it s not done by default
+    self.removeAllVarsInContext()
+    pass
 
   def getContainer(self):
     return self.my_container
@@ -180,6 +582,19 @@ class PyScriptNode_i (Engines__POA.PyScriptNode,Generic):
     except Exception:
       raise SALOME.SALOME_Exception(SALOME.ExceptionStruct(SALOME.BAD_PARAM,"","PyScriptNode.assignNewCompiledCode (%s) : code to be executed \"%s\"" %(self.nodeName,codeStr),0))
 
+  def executeSimple(self, key, val):
+    """
+    Same as execute method except that no pickelization mecanism is implied here. No output is expected
+    """
+    try:
+      self.context.update({ "env" : [(k,v) for k,v in zip(key,val)]})
+      exec(self.ccode,self.context)
+    except Exception:
+      exc_typ,exc_val,exc_fr=sys.exc_info()
+      l=traceback.format_exception(exc_typ,exc_val,exc_fr)
+      print("".join(l)) ; sys.stdout.flush() # print error also in logs of remote container
+      raise SALOME.SALOME_Exception(SALOME.ExceptionStruct(SALOME.BAD_PARAM,"".join(l),"PyScriptNode: %s" % (self.nodeName),0))
+    
   def execute(self,outargsname,argsin):
     """Execute the script stored in attribute ccode with pickled args (argsin)"""
     try:
@@ -196,42 +611,85 @@ class PyScriptNode_i (Engines__POA.PyScriptNode,Generic):
     except Exception:
       exc_typ,exc_val,exc_fr=sys.exc_info()
       l=traceback.format_exception(exc_typ,exc_val,exc_fr)
+      print("".join(l)) ; sys.stdout.flush() # print error also in logs of remote container
       raise SALOME.SALOME_Exception(SALOME.ExceptionStruct(SALOME.BAD_PARAM,"".join(l),"PyScriptNode: %s, outargsname: %s" % (self.nodeName,outargsname),0))
 
   def executeFirst(self,argsin):
     """ Same than first part of self.execute to reduce memory peak."""
-    import time
     try:
+      self.beginOfCurrentExecutionSession()
       data = None
+      self.addTimeInfoOnLevel2("startInputTime")
       if True: # to force call of SeqByteReceiver's destructor
         argsInPy = SeqByteReceiver( argsin )
         data = argsInPy.data()
+        self.addInfoOnLevel2("inputMem",len(data))
       _,kws=pickle.loads(data)
+      vis = InOutputObjVisitor()
+      for elt in kws:
+        # fetch real data if necessary
+        kws[elt] = UnProxyObjectSimple( kws[elt],vis)
+      self.addInfoOnLevel2("inputHDDMem",vis)
       self.context.update(kws)
+      self.addTimeInfoOnLevel2("endInputTime")
     except Exception:
       exc_typ,exc_val,exc_fr=sys.exc_info()
       l=traceback.format_exception(exc_typ,exc_val,exc_fr)
+      print("".join(l)) ; sys.stdout.flush() # print error also in logs of remote container
       raise SALOME.SALOME_Exception(SALOME.ExceptionStruct(SALOME.BAD_PARAM,"".join(l),"PyScriptNode:First %s" % (self.nodeName),0))
 
   def executeSecond(self,outargsname):
     """ Same than second part of self.execute to reduce memory peak."""
+    import sys
     try:
+      self.addTimeInfoOnLevel2("startExecTime")
+      ##
+      self.addInfoOnLevel2("measureTimeResolution",self.my_container_py.monitoringtimeresms())
+      monitoringParams = LaunchTimeCPUMonitoring( self.my_container_py.monitoringtimeresms() )
       exec(self.ccode, self.context)
+      StopMonitoring( monitoringParams )
+      cpumeminfo = ReadCPUMemInfo( monitoringParams )
+      ##
+      self.addInfoOnLevel2("CPUMemDuringExec",cpumeminfo)
+      del monitoringParams
+      self.addTimeInfoOnLevel2("endExecTime")
+      self.addTimeInfoOnLevel2("startOutputTime")
       argsout=[]
       for arg in outargsname:
         if arg not in self.context:
           raise KeyError("There is no variable %s in context" % arg)
         argsout.append(self.context[arg])
-      argsout=pickle.dumps(tuple(argsout),-1)
-      ret = SenderByte_i( self.poa,argsout )
-      id_o = self.poa.activate_object(ret)
-      retObj = self.poa.id_to_reference(id_o)
-      return retObj._narrow( SALOME.SenderByte )
+      ret = [ ]
+      outputMem = 0
+      vis = InOutputObjVisitor()
+      for arg in argsout:
+        # the proxy mecanism is catched here
+        argPickle = SpoolPickleObject( arg, vis )
+        retArg = SenderByte_i( self.poa,argPickle )
+        id_o = self.poa.activate_object(retArg)
+        retObj = self.poa.id_to_reference(id_o)
+        ret.append( retObj._narrow( SALOME.SenderByte ) )
+        outputMem += len(argPickle)
+      self.addInfoOnLevel2("outputMem",outputMem)
+      self.addInfoOnLevel2("outputHDDMem",vis)
+      self.addTimeInfoOnLevel2("endOutputTime")
+      self.endOfCurrentExecutionSession()
+      return ret
     except Exception:
       exc_typ,exc_val,exc_fr=sys.exc_info()
       l=traceback.format_exception(exc_typ,exc_val,exc_fr)
+      print("".join(l)) ; sys.stdout.flush() # print error also in logs of remote container
       raise SALOME.SALOME_Exception(SALOME.ExceptionStruct(SALOME.BAD_PARAM,"".join(l),"PyScriptNode:Second %s, outargsname: %s" % (self.nodeName,outargsname),0))
 
+  def listAllVarsInContext(self):
+      import re
+      pat = re.compile("^__([a-z]+)__$")
+      return [elt for elt in self.context if not pat.match(elt) and elt != MY_CONTAINER_ENTRY_IN_GLBS]
+      
+  def removeAllVarsInContext(self):
+      for elt in self.listAllVarsInContext():
+        del self.context[elt]
+
   def getValueOfVarInContext(self,varName):
     try:
       return pickle.dumps(self.context[varName],-1)
@@ -258,3 +716,17 @@ class PyScriptNode_i (Engines__POA.PyScriptNode,Generic):
       l=traceback.format_exception(exc_typ,exc_val,exc_fr)
       raise SALOME.SALOME_Exception(SALOME.ExceptionStruct(SALOME.BAD_PARAM,"".join(l),"PyScriptNode: %s" %self.nodeName,0))
     pass
+
+  def beginOfCurrentExecutionSession(self):
+    self._current_execution_session = LogOfCurrentExecutionSession( self._log_script.addExecutionSession() )
+  
+  def endOfCurrentExecutionSession(self):
+    self._current_execution_session.finalizeAndPushToMaster()
+    self._current_execution_session = None
+
+  def addInfoOnLevel2(self, key, value):
+    self._current_execution_session.addInfoOnLevel2(key, value)
+      
+  def addTimeInfoOnLevel2(self, key):
+    from datetime import datetime
+    self._current_execution_session.addInfoOnLevel2(key,datetime.now())