def for_process_mp(for_func, worker, shape, thread_count=0, queue_limit=None, **extra):
''' Generator to process collection of arrays in parallel
:Parameters:
for_func : func
Generate a list of data
work : function
Function to preprocess the images
thread_count : int
Number of threads
shape : int
Shape of worker result array
extra : dict
Unused keyword arguments
:Returns:
index : int
Yields index of output array
out : array
Yields output array of worker
'''
if thread_count < 2:
for i, val in enumerate(for_func):
res = worker(val, i, **extra)
yield i, res
else:
if queue_limit is None: queue_limit = thread_count*8
else: queue_limit *= thread_count
qin, qout = process_queue.start_raw_enum_workers(process_worker2, thread_count, queue_limit, -1, worker, extra)
try:
total = 0
for i, val in enumerate(for_func):
if i >= thread_count:
pos = process_queue.safe_get(qout.get) #if i > thread_count else i
if pos is None or pos == -1: raise ValueError, "Error occured in process: %d"%pos
res, idx = pos
yield idx, res
else:
pos = i
total += 1
qin.put((val,i))
for i in xrange(total):
pos = process_queue.safe_get(qout.get)
if pos is None or pos == -1: raise ValueError, "Error occured in process: %d"%pos
res, idx = pos
yield idx, res
finally:
#_logger.error("Terminating %d workers"%(thread_count))
for i in xrange(thread_count):
qin.put((-1, -1))
pos = process_queue.safe_get(qout.get)
if pos != -1:
_logger.error("Wrong return value: %s"%str(pos))
assert(pos==-1)
raise StopIteration
def queue_iterator(qin, process_number):
try:
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
yield val
finally: pass
def queue_iterator(qin, process_number):
try:
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
yield val
finally: pass
def iterate_map(for_func, worker, thread_count, queue_limit=None, **extra):
''' Iterate over the input value and reduce after finished processing
'''
if thread_count < 2:
for val in worker(enumerate(for_func), process_number=0, **extra):
yield val
return
def queue_iterator(qin, process_number):
try:
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
yield val
finally:
pass
#_logger.error("queue-done")
def iterate_map_worker(qin, qout, process_number, process_limit, extra):
val = None
try:
val = worker(queue_iterator(qin, process_number),
process_number=process_number,
**extra)
except:
_logger.exception("Error in child process")
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
finally:
qout.put(val)
#process_queue.safe_get(qin.get)
if queue_limit is None: queue_limit = thread_count * 8
else: queue_limit *= thread_count
qin, qout = process_queue.start_raw_enum_workers(iterate_map_worker,
thread_count, queue_limit,
1, extra)
try:
for val in enumerate(for_func):
qin.put(val)
except:
_logger.error("for_func=%s" % str(for_func))
raise
for i in xrange(thread_count):
qin.put(None)
#qin.join()
for i in xrange(thread_count):
val = process_queue.safe_get(qout.get)
#qin.put(None)
if val is None: raise ValueError, "Exception in child process"
yield val
def iterate_map_worker(qin, qout, process_number, process_limit, extra):
val = None
try:
val = worker(queue_iterator(qin, process_number), process_number=process_number, **extra)
except:
_logger.exception("Error in child process")
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
finally:
qout.put(val)
def iterate_map_worker(qin, qout, process_number, process_limit, extra):
val = None
try:
val = worker(queue_iterator(qin, process_number), process_number=process_number, **extra)
except:
_logger.exception("Error in child process")
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
finally:
qout.put(val)
def iterate_map(for_func, worker, thread_count, queue_limit=None, **extra):
''' Iterate over the input value and reduce after finished processing
'''
if thread_count < 2:
for val in worker(enumerate(for_func), process_number=0, **extra):
yield val
return
def queue_iterator(qin, process_number):
try:
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
yield val
finally: pass
#_logger.error("queue-done")
def iterate_map_worker(qin, qout, process_number, process_limit, extra):
val = None
try:
val = worker(queue_iterator(qin, process_number), process_number=process_number, **extra)
except:
_logger.exception("Error in child process")
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
finally:
qout.put(val)
#process_queue.safe_get(qin.get)
if queue_limit is None: queue_limit = thread_count*8
else: queue_limit *= thread_count
qin, qout = process_queue.start_raw_enum_workers(iterate_map_worker, thread_count, queue_limit, 1, extra)
try:
for val in enumerate(for_func):
qin.put(val)
except:
_logger.error("for_func=%s"%str(for_func))
raise
for i in xrange(thread_count): qin.put(None)
#qin.join()
for i in xrange(thread_count):
val = process_queue.safe_get(qout.get)
#qin.put(None)
if val is None: raise ValueError, "Exception in child process"
yield val
def process_worker2(qin, qout, process_number, process_limit, worker, extra):
''' Worker in each process that preprocesses the images
:Parameters:
qin : multiprocessing.Queue
Queue with index for input images in shared array
qout : multiprocessing.Queue
Queue with index and offset for the output images in shared array
process_number : int
Process number
process_limit : int
Number of processes
worker : function
Function to preprocess the images
shmem_img : multiprocessing.RawArray
Shared memory image array
shape : tuple
Dimensions of the shared memory array
extra : dict
Keyword arguments
'''
_logger.debug("Worker %d of %d - started" %
(process_number, process_limit))
try:
while True:
pos = process_queue.safe_get(qin.get)
if pos is None or not hasattr(pos[0], 'ndim'): break
res, idx = pos
val = worker(res, idx, **extra)
qout.put((val, idx))
_logger.debug("Worker %d of %d - ending ..." %
(process_number, process_limit))
qout.put(-1)
except:
_logger.exception("Finished with error")
qout.put(None)
else:
_logger.debug("Worker %d of %d - finished" %
(process_number, process_limit))
def iterate_reduce_worker(qin, qout, process_number, process_limit, extra, shmem_map_base=None):#=shmem_map):
val = None
try:
if shmem_map_base is not None:
ar = shmem_map_base[process_number]
ar_map={}
for key in ar.iterkeys():
ar_map[key] = numpy.ctypeslib.as_array(ar[key])
ar_map[key] = ar_map[key].view(shmem_map[process_number][key].dtype).reshape(shmem_map[process_number][key].shape)
extra.update(ar_map)
val = worker(queue_iterator(qin, process_number), process_number=process_number, **extra)
except:
_logger.exception("Error in child process")
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
finally:
if shmem_map_base is not None:
qout.put(process_number)
else:
qout.put(val)
def iterate_reduce_worker(qin, qout, process_number, process_limit, extra, shmem_map_base=None):#=shmem_map):
val = None
try:
if shmem_map_base is not None:
ar = shmem_map_base[process_number]
ar_map={}
for key in ar.iterkeys():
ar_map[key] = numpy.ctypeslib.as_array(ar[key])
ar_map[key] = ar_map[key].view(shmem_map[process_number][key].dtype).reshape(shmem_map[process_number][key].shape)
extra.update(ar_map)
val = worker(queue_iterator(qin, process_number), process_number=process_number, **extra)
except:
_logger.exception("Error in child process")
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
finally:
if shmem_map_base is not None:
qout.put(process_number)
else:
qout.put(val)
def process_worker2(qin, qout, process_number, process_limit, worker, extra):
''' Worker in each process that preprocesses the images
:Parameters:
qin : multiprocessing.Queue
Queue with index for input images in shared array
qout : multiprocessing.Queue
Queue with index and offset for the output images in shared array
process_number : int
Process number
process_limit : int
Number of processes
worker : function
Function to preprocess the images
shmem_img : multiprocessing.RawArray
Shared memory image array
shape : tuple
Dimensions of the shared memory array
extra : dict
Keyword arguments
'''
_logger.debug("Worker %d of %d - started"%(process_number, process_limit))
try:
while True:
pos = process_queue.safe_get(qin.get)
if pos is None or not hasattr(pos[0], 'ndim'): break
res, idx = pos
val = worker(res, idx, **extra)
qout.put((val, idx))
_logger.debug("Worker %d of %d - ending ..."%(process_number, process_limit))
qout.put(-1)
except:
_logger.exception("Finished with error")
qout.put(None)
else:
_logger.debug("Worker %d of %d - finished"%(process_number, process_limit))
def process_mp(process, vals, worker_count, init_process=None, ignored_errors=None, **extra):
''' Generator that runs a process functor in parallel (or serial if worker_count
is less than 2) over a list of given data values and returns the result
:Parameters:
process : function
Functor to be run in parallel (or serial if worker_count is less than 2)
vals : list
List of items to process in parallel
worker_count : int
Number of processes to run in parallel
init_process : function
Initalize the parameters for the child process
ignored_errors : list
Single element list with counter for ignored errors
extra : dict
Unused keyword arguments
:Returns:
val : object
Return value of process functor
'''
#_logger.error("worker_count1=%d"%worker_count)
if len(vals) < worker_count: worker_count = len(vals)
#_logger.error("worker_count2=%d"%worker_count)
if worker_count > 1:
def process_helper(val, **extra):
try:
return process(val, **extra)
except:
if ignored_errors is not None and len(ignored_errors) > 0:ignored_errors[0]+=1
if _logger.getEffectiveLevel()==logging.DEBUG or 1 == 1:
_logger.exception("Unexpected error in process - report this problem to the developer")
else:
_logger.warn("nexpected error in process - report this problem to the developer")
return extra.get('process_number', 0), val
qout = process_queue.start_workers_with_output(vals, process_helper, worker_count, init_process, ignore_error=True, **extra)
index = 0
while index < len(vals):
val = process_queue.safe_get(qout.get)
if isinstance(val, process_queue.ProcessException):
index = 0
while index < worker_count:
if process_queue.safe_get(qout.get) is None:
index += 1;
raise val
if val is None: continue
index += 1
yield val
else:
#_logger.error("worker_count3=%d"%worker_count)
logging.debug("Running with single process: %d"%len(vals))
for i, val in enumerate(vals):
try:
f = process(val, **extra)
except:
if ignored_errors is not None and len(ignored_errors) > 0: ignored_errors[0]+=1
if _logger.getEffectiveLevel()==logging.DEBUG or 1 == 1:
_logger.exception("Unexpected error in process - report this problem to the developer")
else:
_logger.warn("nexpected error in process - report this problem to the developer")
yield i, val
continue
yield i, f
def iterate_reduce(for_func, worker, thread_count, queue_limit=None, shmem_array_info=None, **extra):
''' Iterate over the input value and reduce after finished processing
'''
if thread_count < 2:
yield worker(enumerate(for_func), process_number=0, **extra)
return
shmem_map=None
shmem_map_base=None
if shmem_array_info is not None:
shmem_map=[]
shmem_map_base=[]
for i in xrange(thread_count):
base = {}
arr = {}
for key in shmem_array_info.iterkeys():
ar = shmem_array_info[key]
if ar.dtype.str[1]=='c':
typestr = ar.dtype.str[0]+'f'+str(int(ar.dtype.str[2:])/2)
ar = ar.view(numpy.dtype(typestr))
if ar.dtype == numpy.dtype(numpy.float64):
typecode="d"
elif ar.dtype == numpy.dtype(numpy.float32):
typecode="f"
else: raise ValueError, "dtype not supported: %s"%str(ar.dtype)
base[key] = multiprocessing.sharedctypes.RawArray(typecode, ar.ravel().shape[0])
arr[key] = numpy.ctypeslib.as_array(base[key])
arr[key] = arr[key].view(shmem_array_info[key].dtype).reshape(shmem_array_info[key].shape)
shmem_map.append(arr)
shmem_map_base.append(base)
del shmem_array_info
def queue_iterator(qin, process_number):
try:
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
yield val
finally: pass
def iterate_reduce_worker(qin, qout, process_number, process_limit, extra, shmem_map_base=None):#=shmem_map):
val = None
try:
if shmem_map_base is not None:
ar = shmem_map_base[process_number]
ar_map={}
for key in ar.iterkeys():
ar_map[key] = numpy.ctypeslib.as_array(ar[key])
ar_map[key] = ar_map[key].view(shmem_map[process_number][key].dtype).reshape(shmem_map[process_number][key].shape)
extra.update(ar_map)
val = worker(queue_iterator(qin, process_number), process_number=process_number, **extra)
except:
_logger.exception("Error in child process")
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
finally:
if shmem_map_base is not None:
qout.put(process_number)
else:
qout.put(val)
if queue_limit is None: queue_limit = thread_count*8
else: queue_limit *= thread_count
qin, qout = process_queue.start_raw_enum_workers(iterate_reduce_worker, thread_count, queue_limit, 1, extra, shmem_map_base)
try:
for val in enumerate(for_func):
qin.put(val)
except:
_logger.error("for_func=%s"%str(for_func))
raise
for i in xrange(thread_count): qin.put(None)
#qin.join()
for i in xrange(thread_count):
val = process_queue.safe_get(qout.get)
if shmem_map is not None:
val = shmem_map[val]
#qin.put(None)
if val is None: raise ValueError, "Exception in child process"
yield val
def for_process_mp(for_func,
worker,
shape,
thread_count=0,
queue_limit=None,
**extra):
''' Generator to process collection of arrays in parallel
:Parameters:
for_func : func
Generate a list of data
work : function
Function to preprocess the images
thread_count : int
Number of threads
shape : int
Shape of worker result array
extra : dict
Unused keyword arguments
:Returns:
index : int
Yields index of output array
out : array
Yields output array of worker
'''
if thread_count < 2:
for i, val in enumerate(for_func):
res = worker(val, i, **extra)
yield i, res
else:
if queue_limit is None: queue_limit = thread_count * 8
else: queue_limit *= thread_count
qin, qout = process_queue.start_raw_enum_workers(
process_worker2, thread_count, queue_limit, -1, worker, extra)
try:
total = 0
for i, val in enumerate(for_func):
if i >= thread_count:
pos = process_queue.safe_get(
qout.get) #if i > thread_count else i
if pos is None or pos == -1:
raise ValueError, "Error occured in process: %d" % pos
res, idx = pos
yield idx, res
else:
pos = i
total += 1
qin.put((val, i))
for i in xrange(total):
pos = process_queue.safe_get(qout.get)
if pos is None or pos == -1:
raise ValueError, "Error occured in process: %d" % pos
res, idx = pos
yield idx, res
finally:
#_logger.error("Terminating %d workers"%(thread_count))
for i in xrange(thread_count):
qin.put((-1, -1))
pos = process_queue.safe_get(qout.get)
if pos != -1:
_logger.error("Wrong return value: %s" % str(pos))
assert (pos == -1)
raise StopIteration
def process_mp(process,
vals,
worker_count,
init_process=None,
ignored_errors=None,
**extra):
''' Generator that runs a process functor in parallel (or serial if worker_count
is less than 2) over a list of given data values and returns the result
:Parameters:
process : function
Functor to be run in parallel (or serial if worker_count is less than 2)
vals : list
List of items to process in parallel
worker_count : int
Number of processes to run in parallel
init_process : function
Initalize the parameters for the child process
ignored_errors : list
Single element list with counter for ignored errors
extra : dict
Unused keyword arguments
:Returns:
val : object
Return value of process functor
'''
#_logger.error("worker_count1=%d"%worker_count)
if len(vals) < worker_count: worker_count = len(vals)
#_logger.error("worker_count2=%d"%worker_count)
if worker_count > 1:
def process_helper(val, **extra):
try:
return process(val, **extra)
except:
if ignored_errors is not None and len(ignored_errors) > 0:
ignored_errors[0] += 1
if _logger.getEffectiveLevel() == logging.DEBUG or 1 == 1:
_logger.exception(
"Unexpected error in process - report this problem to the developer"
)
else:
_logger.warn(
"nexpected error in process - report this problem to the developer"
)
return extra.get('process_number', 0), val
qout = process_queue.start_workers_with_output(vals,
process_helper,
worker_count,
init_process,
ignore_error=True,
**extra)
index = 0
while index < len(vals):
val = process_queue.safe_get(qout.get)
if isinstance(val, process_queue.ProcessException):
index = 0
while index < worker_count:
if process_queue.safe_get(qout.get) is None:
index += 1
raise val
if val is None: continue
index += 1
yield val
else:
#_logger.error("worker_count3=%d"%worker_count)
logging.debug("Running with single process: %d" % len(vals))
for i, val in enumerate(vals):
try:
f = process(val, **extra)
except:
if ignored_errors is not None and len(ignored_errors) > 0:
ignored_errors[0] += 1
if _logger.getEffectiveLevel() == logging.DEBUG or 1 == 1:
_logger.exception(
"Unexpected error in process - report this problem to the developer"
)
else:
_logger.warn(
"nexpected error in process - report this problem to the developer"
)
yield i, val
continue
yield i, f
def iterate_reduce(for_func,
worker,
thread_count,
queue_limit=None,
shmem_array_info=None,
**extra):
''' Iterate over the input value and reduce after finished processing
'''
if thread_count < 2:
yield worker(enumerate(for_func), process_number=0, **extra)
return
shmem_map = None
shmem_map_base = None
if shmem_array_info is not None:
shmem_map = []
shmem_map_base = []
for i in xrange(thread_count):
base = {}
arr = {}
for key in shmem_array_info.iterkeys():
ar = shmem_array_info[key]
if ar.dtype.str[1] == 'c':
typestr = ar.dtype.str[0] + 'f' + str(
int(ar.dtype.str[2:]) / 2)
ar = ar.view(numpy.dtype(typestr))
if ar.dtype == numpy.dtype(numpy.float64):
typecode = "d"
elif ar.dtype == numpy.dtype(numpy.float32):
typecode = "f"
else:
raise ValueError, "dtype not supported: %s" % str(ar.dtype)
base[key] = multiprocessing.sharedctypes.RawArray(
typecode,
ar.ravel().shape[0])
arr[key] = numpy.ctypeslib.as_array(base[key])
arr[key] = arr[key].view(shmem_array_info[key].dtype).reshape(
shmem_array_info[key].shape)
shmem_map.append(arr)
shmem_map_base.append(base)
del shmem_array_info
def queue_iterator(qin, process_number):
try:
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
yield val
finally:
pass
def iterate_reduce_worker(qin,
qout,
process_number,
process_limit,
extra,
shmem_map_base=None): #=shmem_map):
val = None
try:
if shmem_map_base is not None:
ar = shmem_map_base[process_number]
ar_map = {}
for key in ar.iterkeys():
ar_map[key] = numpy.ctypeslib.as_array(ar[key])
ar_map[key] = ar_map[key].view(
shmem_map[process_number][key].dtype).reshape(
shmem_map[process_number][key].shape)
extra.update(ar_map)
val = worker(queue_iterator(qin, process_number),
process_number=process_number,
**extra)
except:
_logger.exception("Error in child process")
while True:
val = process_queue.safe_get(qin.get)
if val is None: break
finally:
if shmem_map_base is not None:
qout.put(process_number)
else:
qout.put(val)
if queue_limit is None: queue_limit = thread_count * 8
else: queue_limit *= thread_count
qin, qout = process_queue.start_raw_enum_workers(iterate_reduce_worker,
thread_count, queue_limit,
1, extra, shmem_map_base)
try:
for val in enumerate(for_func):
qin.put(val)
except:
_logger.error("for_func=%s" % str(for_func))
raise
for i in xrange(thread_count):
qin.put(None)
#qin.join()
for i in xrange(thread_count):
val = process_queue.safe_get(qout.get)
if shmem_map is not None:
val = shmem_map[val]
#qin.put(None)
if val is None: raise ValueError, "Exception in child process"
yield val
