def _function_wrapper(args_package):
"""Allow multiprocessing Pool to call generic functions/args/kwargs.
Data are saved here rather than handed back to avoid the scenario where
all of the output from a batch run is held in memory.
"""
(identifier, directory, funcname, args, kwargs) = args_package
readable = utils.readable_call(funcname, args, kwargs)
filename = "%s/%s.shelve" % (directory, identifier)
filename = re.sub('/+', '/', filename)
print "%s -> %s" % (readable, filename)
result = utils.func_exec(funcname, args, kwargs, printcall=False)
outfile = shelve.open(filename, 'n', protocol=-1)
outfile["identifier"] = identifier
outfile["filename"] = filename
outfile["funcname"] = funcname
outfile["args"] = args
outfile["kwargs"] = kwargs
outfile["result"] = result
outfile.close()
return identifier
def scatter(self, *args, **kwargs):
if "execute_key" not in kwargs:
print "need to identify an execution key for the task"
return
rehashed = [_make_serializable(item) for item in args]
argpkl = pickle.dumps(
(self.funcname, rehashed, tuple(sorted(kwargs.items()))), -1)
identifier = hashlib.sha224(argpkl).hexdigest()
readable = utils.readable_call(self.funcname, rehashed, kwargs)
# delete the kwarg for this function to associate an ID to the output
# so it does not interfere with the function call.
execute_key = kwargs["execute_key"]
del kwargs["execute_key"]
jobfile_name = "%s/%s.job" % (pd.job_directory, identifier)
donefile_name = "%s/%s.done" % (pd.job_directory, identifier)
# first remove any lurking completed jobs
try:
os.remove(donefile_name)
os.remove(jobfile_name)
except OSError:
if self.verbose:
print "all clear: ", donefile_name
job_shelve = shelve.open(jobfile_name, "n", protocol=-1)
job_shelve['funcname'] = self.funcname
job_shelve['args'] = args
job_shelve['kwargs'] = kwargs
job_shelve['tag'] = execute_key
job_shelve['identifier'] = identifier
job_shelve['call'] = readable
#print job_shelve
job_shelve.close()
self.call_stack.append(identifier)
def scatter(self, *args, **kwargs):
if "execute_key" not in kwargs:
print "need to identify an execution key for the task"
return
rehashed = [_make_serializable(item) for item in args]
argpkl = pickle.dumps((self.funcname, rehashed, tuple(sorted(kwargs.items()))), -1)
identifier = hashlib.sha224(argpkl).hexdigest()
readable = utils.readable_call(self.funcname, rehashed, kwargs)
# delete the kwarg for this function to associate an ID to the output
# so it does not interfere with the function call.
execute_key = kwargs["execute_key"]
del kwargs["execute_key"]
jobfile_name = "%s/%s.job" % (pd.job_directory, identifier)
donefile_name = "%s/%s.done" % (pd.job_directory, identifier)
# first remove any lurking completed jobs
try:
os.remove(donefile_name)
os.remove(jobfile_name)
except OSError:
if self.verbose:
print "all clear: ", donefile_name
job_shelve = shelve.open(jobfile_name, "n", protocol=-1)
job_shelve["funcname"] = self.funcname
job_shelve["args"] = args
job_shelve["kwargs"] = kwargs
job_shelve["tag"] = execute_key
job_shelve["identifier"] = identifier
job_shelve["call"] = readable
# print job_shelve
job_shelve.close()
self.call_stack.append(identifier)
def memoize(*args, **kwargs):
funcname = func.__name__
rehashed = [_make_serializable(item) for item in args]
argpkl = pickle.dumps(
(funcname, rehashed, tuple(sorted(kwargs.items()))), -1)
identifier = hashlib.sha224(argpkl).hexdigest()
readable = utils.readable_call(funcname, rehashed, kwargs)
filename = "%s/%s.shelve" % (memoize_directory, identifier)
filename = re.sub('/+', '/', filename)
print "%s -> %s" % (readable, filename)
done_filename = filename + ".done"
busy_filename = filename + ".busy"
# prevent the race condition where many threads want to
# start writing a new cache file at the same time
time.sleep(random.uniform(0, 0.5))
# if the result is cahced or being calculated elsewhere,
if os.access(done_filename, os.F_OK) or \
os.access(busy_filename, os.F_OK):
printed = False
# wait for the other calculation process to finish
while (not os.access(done_filename, os.F_OK)):
time.sleep(1.)
if not printed:
print "waiting for %s" % filename
printed = True
# if many threads were waiting to read, space their reads
time.sleep(random.uniform(0, 0.5))
print "ready to read %s" % filename
try:
input_shelve = shelve.open(filename, "r", protocol=-1)
retval = input_shelve['result']
input_shelve.close()
print "used cached value %s" % filename
except:
raise ValueError
else:
# first flag the cachefile as busy so other threads wait
busyfile = open(busy_filename, "w")
busyfile.write("working")
busyfile.close()
# recalculate the function
print "no cache, recalculating %s" % filename
start = time.time()
retval = func(*args, **kwargs)
outfile = shelve.open(filename, "n", protocol=-1)
outfile["signature"] = identifier
outfile["filename"] = filename
outfile["funcname"] = funcname
outfile["args"] = rehashed
outfile["kwargs"] = kwargs
outfile["result"] = retval
outfile.close()
time.sleep(0.2)
# indicate that the function is done being recalculated
donefile = open(done_filename, "w")
donefile.write("%10.15f\n" % (time.time() - start))
donefile.close()
# remove the busy flag
os.remove(busy_filename)
return retval
def memoize(*args, **kwargs):
funcname = func.__name__
rehashed = [_make_serializable(item) for item in args]
argpkl = pickle.dumps((funcname, rehashed,
tuple(sorted(kwargs.items()))), -1)
identifier = hashlib.sha224(argpkl).hexdigest()
readable = utils.readable_call(funcname, rehashed, kwargs)
filename = "%s/%s.shelve" % (memoize_directory, identifier)
filename = re.sub('/+', '/', filename)
print "%s -> %s" % (readable, filename)
done_filename = filename + ".done"
busy_filename = filename + ".busy"
# prevent the race condition where many threads want to
# start writing a new cache file at the same time
time.sleep(random.uniform(0, 0.5))
# if the result is cahced or being calculated elsewhere,
if os.access(done_filename, os.F_OK) or \
os.access(busy_filename, os.F_OK):
printed = False
# wait for the other calculation process to finish
while(not os.access(done_filename, os.F_OK)):
time.sleep(1.)
if not printed:
print "waiting for %s" % filename
printed = True
# if many threads were waiting to read, space their reads
time.sleep(random.uniform(0, 0.5))
print "ready to read %s" % filename
try:
input_shelve = shelve.open(filename, "r", protocol=-1)
retval = input_shelve['result']
input_shelve.close()
print "used cached value %s" % filename
except:
raise ValueError
else:
# first flag the cachefile as busy so other threads wait
busyfile = open(busy_filename, "w")
busyfile.write("working")
busyfile.close()
# recalculate the function
print "no cache, recalculating %s" % filename
start = time.time()
retval = func(*args, **kwargs)
outfile = shelve.open(filename, "n", protocol=-1)
outfile["signature"] = identifier
outfile["filename"] = filename
outfile["funcname"] = funcname
outfile["args"] = rehashed
outfile["kwargs"] = kwargs
outfile["result"] = retval
outfile.close()
time.sleep(0.2)
# indicate that the function is done being recalculated
donefile = open(done_filename, "w")
donefile.write("%10.15f\n" % (time.time() - start))
donefile.close()
# remove the busy flag
os.remove(busy_filename)
return retval
