def test():
NUMBER_OF_PROCESSES = 4
TASKS1 = [(mul, (i, 7)) for i in range(20)]
TASKS2 = [(plus, (i, 8)) for i in range(10)]
# Create queues
task_queue = Queue()
done_queue = Queue()
# Submit tasks
list(map(task_queue.put, TASKS1))
# Start worker processes
for i in range(NUMBER_OF_PROCESSES):
Process(target=worker, args=(task_queue, done_queue)).start()
# Get and print results
print('Unordered results:')
for i in range(len(TASKS1)):
print('\t', done_queue.get())
# Add more tasks using `put()` instead of `putMany()`
for task in TASKS2:
task_queue.put(task)
# Get and print some more results
for i in range(len(TASKS2)):
print('\t', done_queue.get())
# Tell child processes to stop
for i in range(NUMBER_OF_PROCESSES):
task_queue.put('STOP')
def _solve(self, X, Y, batch_size):
'''
Solve the multi-objective problem by multiple scalarized single-objective solvers.
'''
# generate scalarization weights
weights = np.random.random((batch_size, self.problem.n_obj))
weights /= np.expand_dims(np.sum(weights, axis=1), 1)
# initial solutions
X = np.vstack([X, lhs(X.shape[1], batch_size)])
F = self.problem.evaluate(X, return_values_of=['F'])
# optimization
xs, ys = [], []
queue = Queue()
n_active_process = 0
for i in range(batch_size):
x0 = X[np.argmin(augmented_tchebicheff(F, weights[i]))]
Process(target=optimization,
args=(self.problem, x0, weights[i], queue)).start()
n_active_process += 1
if n_active_process >= self.n_process:
x, y = queue.get()
xs.append(x)
ys.append(y)
n_active_process -= 1
# gather result
for _ in range(n_active_process):
x, y = queue.get()
xs.append(x)
ys.append(y)
return np.array(xs), np.array(ys)
def test():
NUMBER_OF_PROCESSES = 4
TASKS1 = [(mul, (i, 7)) for i in range(20)]
TASKS2 = [(plus, (i, 8)) for i in range(10)]
# Create queues
task_queue = Queue()
done_queue = Queue()
# Submit tasks
map(task_queue.put, TASKS1)
# Start worker processes
for i in range(NUMBER_OF_PROCESSES):
Process(target=worker, args=(task_queue, done_queue)).start()
# Get and print results
print "Unordered results:"
for i in range(len(TASKS1)):
print "\t", done_queue.get()
# Add more tasks using `put()` instead of `putMany()`
for task in TASKS2:
task_queue.put(task)
# Get and print some more results
for i in range(len(TASKS2)):
print "\t", done_queue.get()
# Tell child processes to stop
for i in range(NUMBER_OF_PROCESSES):
task_queue.put("STOP")
def solve(self, problem, X, Y):
'''
Solve the multi-objective problem by multiple scalarized single-objective solvers.
Parameters
----------
problem: mobo.surrogate_problem.SurrogateProblem
The surrogate problem to be solved.
X: np.array
Current design variables.
Y: np.array
Current performance values.
Returns
-------
solution: dict
A dictionary containing information of the solution.\n
- solution['x']: Proposed design samples.
- solution['y']: Performance of proposed design samples.
'''
# initialize population
sampling = self._get_sampling(X, Y)
if not isinstance(sampling, np.ndarray):
sampling = sampling.do(problem, self.pop_size)
# generate scalarization weights
weights = np.random.random((self.pop_size, Y.shape[1]))
weights /= np.expand_dims(np.sum(weights, axis=1), 1)
# optimization
xs, ys = [], []
queue = Queue()
n_active_process = 0
for i, x0 in enumerate(sampling):
Process(target=optimization,
args=(problem, x0, weights[i], queue)).start()
n_active_process += 1
if n_active_process >= self.n_process:
x, y = queue.get()
xs.append(x)
ys.append(y)
n_active_process -= 1
# gather result
for _ in range(n_active_process):
x, y = queue.get()
xs.append(x)
ys.append(y)
# construct solution
self.solution = {'x': np.array(xs), 'y': np.array(ys)}
return self.solution
def thread_samples(adj, mSamp):
if (mSamp <= 1):
lst = []
temp_que = Queue()
# Use thread to speed up processing (multithreading)
temp_thread = threading.Thread(
None,
target=make_sample,
name=("1: " + str(mSamp)),
args=[adj, mSamp, temp_que],
daemon=True,
)
# temp_thread = threading.Thread(None, target=make_sample, name=("2:" +
# str(
# max_Sample)),
# args=[adj, max_Sample, queue], )
temp_thread.start()
temp_thread.join()
time_samp_node = temp_que.get(
) # the resulting (samples,thread_time)
temp_que.task_done()
lst.append(time_samp_node)
return (lst, True)
else:
rec_val = thread_samples(adj, mSamp - 1)
temp_que = Queue()
# Use thread to speed up processing (multithreading)
temp_thread = threading.Thread(
None,
target=make_sample,
name=("1: " + str(mSamp)),
args=[adj, mSamp, temp_que],
daemon=True,
)
# temp_thread = threading.Thread(None, target=make_sample, name=("2:" +
# str(
# max_Sample)),
# args=[adj, max_Sample, queue], )
temp_thread.start()
temp_thread.join()
time_samp_node = temp_que.get(
) # the resulting (samples,thread_time)
temp_que.task_done()
# s = queue.get()
# Use thread to speed up processing (multithreading)
if (rec_val[1]):
rec_lst = rec_val[0]
rec_lst.append(time_samp_node)
ret = (rec_lst, True)
# Return the begining of a (time x sample) x boolean tuple list
return ret
def func(q: mp.Queue, pause: mp.Event
): # finish_table True will not read exit until all done
print(f'Process {os.getpid()}')
idle = True # A flag for whether player is actively playing
starttime = time.time()
while True: # global process loop
try: # non-blocking Queue get handler
val = q.get(block=False) # get from queue
# do when got from queue
if val == 'exit': # exit code for process
break
if type(val) == float:
print('time', val)
flag = False
idle = False
for item in val: # player loop
print(item, time.time() - starttime) # message event
# message delay
endtime = time.time() + 1
while pause.is_set():
continue
while time.time() < endtime:
if not q.empty(): # if new item in queue
flag = True # set exit flag
break
if flag: # play exit condition
idle = True
break
idle = True
except Exception: # handle Queue.Empty
pass # Do nothing if no input
def classic_gen_time_Sample2(max_Samp, sample_time_lst, queue):
if max_Samp <= 0:
queue.put(sample_time_lst)
return sample_time_lst
else:
temp_que = Queue()
# Use thread to speed up processing (multithreading)
temp_thread = threading.Thread(
None,
target=make_sample,
name=("2: " + str(max_Samp)),
args=[adj, max_Samp, temp_que],
daemon=True,
)
temp_thread.start()
s = temp_que.get() # the resulting (samples,thread_time)
temp_thread.join()
# s = queue.get()
total_n = s[0]
print("2", total_n, max_Samp)
sample_time_lst.append((total_n, max_Samp))
ret = classic_gen_time_Sample2(max_Samp - 1, sample_time_lst, queue)
queue.put(ret)
return ret
class BaseAggregator(object):
"""Base class for the data aggregator interface. This class is used
alongside the BaseListener class to spawn an aggregator process that
combines data from multiple crawl processes. The BaseAggregator class
manages the child listener process.
Parameters
----------
manager_params : dict
TaskManager configuration parameters
browser_params : list of dict
List of browser configuration dictionaries"""
__metaclass__ = abc.ABCMeta
def __init__(self, manager_params, browser_params):
self.manager_params = manager_params
self.browser_params = browser_params
self.logger = loggingclient(*manager_params['logger_address'])
self.listener_address = None
self.listener_process = None
@abc.abstractmethod
def save_configuration(self, openwpm_version, browser_version):
"""Save configuration details to the database"""
@abc.abstractmethod
def get_next_visit_id(self):
"""Return a unique visit ID to be used as a key for a single page visit"""
@abc.abstractmethod
def get_next_crawl_id(self):
"""Return a unique crawl ID to be used as a key for a browser instance"""
def launch(self, listener_process_runner):
"""Launch the aggregator listener process"""
self.status_queue = Queue()
self.listener_process = Process(
target=listener_process_runner,
args=(self.manager_params, self.status_queue))
self.listener_process.daemon = True
self.listener_process.start()
self.listener_address = self.status_queue.get()
def shutdown(self):
""" Terminate the aggregator listener process"""
self.logger.debug(
"Sending the shutdown signal to the %s listener process..." %
type(self).__name__
)
self.status_queue.put("SHUTDOWN")
start_time = time.time()
self.listener_process.join(300)
self.logger.debug(
"%s took %s seconds to close." % (
type(self).__name__,
str(time.time() - start_time)
)
)
self.listener_address = None
self.listener_process = None
def fit(self, X, Y):
assert not self.fit_done
assert len(X) == len(Y)
possible_labels = list(set(y_val for y in Y for y_val in y))
job_labels = np.array_split(possible_labels, self.n_jobs)
with Manager() as manager:
X_proxy = manager.list(X)
Y_proxy = manager.list(Y)
output_queue = Queue()
processes = [
Process(target=sequential_execute,
args=(output_queue, get_binary_clf_from_multilabel, [{
'X':
X_proxy,
'Y':
Y_proxy,
'label':
lbl,
'return_label':
True
} for lbl in job])) for job in job_labels
]
[p.start() for p in processes]
results = [output_queue.get()
for lbl in possible_labels] # needs to be flattened
[p.join() for p in processes]
self.classifiers = dict(results)
self.fit_done = True
def mp_process(self,nprocs,func,*args):
images=args[0]
# def worker(imgs,i,chunksize, out_q,func,*args):
# """ The worker function, invoked in a process. 'images' is a
# list of images to span the process upon. The results are placed in
# a dictionary that's pushed to a queue.
# """
# outdict = {}
# for imn in range(len(imgs)-1):
# print(i*chunksize+imn)
# outdict[i*chunksize+imn] = func(imgs[imn],imgs[imn+1],*args[1:],i*chunksize+imn)
# out_q.put(outdict)
# Each process will get 'chunksize' nums and a queue to put his out
# dict into
out_q = Queue()
chunksize = int(math.ceil((len(images)-1) / float(nprocs)))
procs = []
print("Chunks of size:",chunksize)
for i in range(nprocs):
if i == nprocs-1:
p = Process(
target=worker,
args=(images[chunksize * i:len(images)-1],i,chunksize,out_q,func,*args))
procs.append(p)
p.start()
self.loading.progress2['value']+=chunksize
self.update()
else:
p = Process(
target=worker,
args=(images[chunksize * i:chunksize * (i + 1)+1],i,chunksize,out_q,func,*args))
procs.append(p)
p.start()
self.loading.progress2['value']+=chunksize
self.update()
# Collect all results into a single result dict. We know how many dicts
# with results to expect.
resultdict = {}
for i in range(nprocs):
resultdict.update(out_q.get())
# Wait for all worker processes to finish
for p in procs:
p.join()
results=[]
for j in range(len(resultdict)):
results.append(resultdict[j])
return results
def test_worker(self):
tournament = axelrod.Tournament(
name=self.test_name,
players=self.players,
game=self.game,
turns=200,
repetitions=self.test_repetitions)
work_queue = Queue()
for repetition in range(self.test_repetitions):
work_queue.put(repetition)
work_queue.put('STOP')
done_queue = Queue()
tournament._worker(work_queue, done_queue)
for r in range(self.test_repetitions):
new_matches = done_queue.get()
self.assertEqual(len(new_matches), 15)
for index_pair, match in new_matches.items():
self.assertIsInstance(index_pair, tuple)
self.assertIsInstance(match, list)
queue_stop = done_queue.get()
self.assertEqual(queue_stop, 'STOP')
def mp_processTR(self, nprocs, func, *args):
images = args[0]
TrMat = args[1]
TrMatNull = np.array([[[1., 0., 0.], [0., 1., 0.]]])
TrM = np.vstack((TrMatNull, TrMat))
fnames = args[2]
print(len(images), len(TrM), len(fnames))
out_q = Queue()
chunksize = int(math.ceil((len(images) - 1) / float(nprocs)))
procs = []
print("Chunks of size:", chunksize)
for i in range(nprocs):
if i == nprocs - 1:
p = Process(target=workerTR,
args=(images[chunksize * i:len(images)],
TrM[chunksize * i:len(images)],
fnames[chunksize * i:len(images)], i,
chunksize, out_q, func, *args[3:]))
procs.append(p)
p.start()
self.loading.progress2['value'] += chunksize
self.update()
else:
p = Process(target=workerTR,
args=(images[chunksize * i:chunksize * (i + 1)],
TrM[chunksize * i:chunksize * (i + 1)],
fnames[chunksize * i:chunksize * (i + 1)], i,
chunksize, out_q, func, *args[3:]))
procs.append(p)
p.start()
self.loading.progress2['value'] += chunksize
self.update()
# Collect all results into a single result dict. We know how many dicts
# with results to expect.
resultdict = {}
for i in range(nprocs):
resultdict.update(out_q.get())
# Wait for all worker processes to finish
for p in procs:
p.join()
results = []
for j in range(len(resultdict)):
results.append(resultdict[j])
return results
def async_generator(generator, backlog_size=10):
from multiprocess import Process, Queue
pipe = Queue(maxsize=backlog_size)
done_flag = 'FO0B@R'
def gen_reader(pipe):
for thing in generator:
pipe.put(thing) # And smoke it.
pipe.put(done_flag)
reader_proc = Process(target=gen_reader, args=(pipe, ))
reader_proc.start()
while True:
next_thing = pipe.get()
if next_thing == done_flag:
break
yield next_thing
reader_proc.join()
def test_start_workers(self):
workers = 2
work_queue = Queue()
done_queue = Queue()
for repetition in range(self.test_repetitions):
work_queue.put(repetition)
tournament = axelrod.Tournament(
name=self.test_name,
players=self.players,
game=self.game,
turns=200,
repetitions=self.test_repetitions)
tournament._start_workers(workers, work_queue, done_queue)
stops = 0
while stops < workers:
payoffs = done_queue.get()
if payoffs == 'STOP':
stops += 1
self.assertEqual(stops, workers)
def test_s3_callbacks(self):
TEST_SITE = BASE_TEST_URL + "/test_pages/simple_a.html"
manager_params, browser_params = self.get_config()
dataset = LocalS3Dataset(manager_params.s3_bucket, manager_params.s3_directory)
manager = task_manager.TaskManager(manager_params, browser_params)
queue = Queue()
def ensure_site_in_s3(success: bool):
# Ensure http table is created
queue.put(
TEST_SITE in dataset.load_table("http_requests").top_level_url.unique()
)
sequence = CommandSequence(
TEST_SITE, reset=True, blocking=True, callback=ensure_site_in_s3
)
sequence.get()
manager.execute_command_sequence(sequence)
manager.close()
assert queue.get()
def run(self, func, arglist):
# TODO: Generalise this a bit maybe?
# We know len(arglist) == self.workers in pretty much all cases
pool = []
queue = Queue()
for idx, split in enumerate(arglist):
proc = Process(target=self._eval, args=(idx, func, split, queue))
pool.append(proc)
proc.start()
for proc in pool:
proc.join()
yss = [[]] * len(arglist)
while not queue.empty():
idx, ys = queue.get()
yss[idx] = ys
ret = [y for ys in yss for y in ys]
#from code import interact
#interact(local=locals())
return ret
def mp_process(self, nprocs, func, *args):
images = args[0]
out_q = Queue()
chunksize = int(math.ceil((len(images) - 1) / float(nprocs)))
procs = []
print("Chunks of size:", chunksize)
for i in range(nprocs):
if i == nprocs - 1:
p = Process(target=worker,
args=(images[chunksize * i:len(images)], i,
chunksize, out_q, func, *args))
procs.append(p)
p.start()
self.loading.progress2['value'] += chunksize
self.update()
else:
p = Process(target=worker,
args=(images[chunksize * i:chunksize * (i + 1) +
1], i, chunksize, out_q, func, *args))
procs.append(p)
p.start()
self.loading.progress2['value'] += chunksize
self.update()
# Collect all results into a single result dict. We know how many dicts
# with results to expect.
resultdict = {}
for i in range(nprocs):
resultdict.update(out_q.get())
# Wait for all worker processes to finish
for p in procs:
p.join()
results = []
for j in range(len(resultdict)):
results.append(resultdict[j])
return results
def run(self):
"""Starts the run of the tests"""
results = []
worker_list = []
to_worker = Queue()
from_worker = Queue()
verbose = self.cl_args.verbose
failfast = self.cl_args.failfast
workers = int(not self.cl_args.parallel) or self.cl_args.workers
for suite in self.suites:
to_worker.put(suite)
for _ in range(workers):
to_worker.put(None)
start = time.time()
# A second try catch is needed here because queues can cause locking
# when they go out of scope, especially when termination signals used
try:
for _ in range(workers):
proc = Consumer(to_worker, from_worker, verbose, failfast)
worker_list.append(proc)
proc.start()
for _ in self.suites:
results.append(self.log_result(from_worker.get()))
end = time.time()
tests_run, errors, failures = self.compile_results(
run_time=end - start,
datagen_time=start - self.datagen_start,
results=results)
except KeyboardInterrupt:
print_exception("Runner", "run", "Keyboard Interrupt, exiting...")
os.killpg(0, 9)
return bool(sum([errors, failures, not tests_run]))
def run(self):
"""Starts the run of the tests"""
results = []
worker_list = []
to_worker = Queue()
from_worker = Queue()
verbose = self.cl_args.verbose
failfast = self.cl_args.failfast
workers = int(not self.cl_args.parallel) or self.cl_args.workers
for suite in self.suites:
to_worker.put(suite)
for _ in range(workers):
to_worker.put(None)
start = time.time()
# A second try catch is needed here because queues can cause locking
# when they go out of scope, especially when termination signals used
try:
for _ in range(workers):
proc = Consumer(to_worker, from_worker, verbose, failfast)
worker_list.append(proc)
proc.start()
for _ in self.suites:
results.append(self.log_result(from_worker.get()))
end = time.time()
tests_run, errors, failures = self.compile_results(
run_time=end - start, datagen_time=start - self.datagen_start,
results=results)
except KeyboardInterrupt:
print_exception("Runner", "run", "Keyboard Interrupt, exiting...")
os.killpg(0, 9)
return bool(sum([errors, failures, not tests_run]))
def _run(self, idx: int, mapper_queue: mp.Queue,
collector_queue: typing.Optional[mp.Queue]):
"""
Process's activity. It handles queue IO and invokes user's mapper handler.
(subprocess, blocked, only two queues can be used to communicate with main process)
"""
with self.mapper(idx) as mapper:
while True:
data = mapper_queue.get()
if data[0] == ParallelProcessor.CMD_STOP:
# print(idx, 'stop')
self._update_progress(mapper, finish=True)
if self.collector and collector_queue is not None:
collector_queue.put((ParallelProcessor.CMD_STOP, ))
return
elif data[0] == ParallelProcessor.CMD_DATA:
batch_result = []
for d in data[1]:
args, kwargs = d[0], d[1]
# print(idx, 'data')
self._update_progress(mapper,
type_=ProgressThread.P_LOADED)
result = mapper.process(*args, **kwargs)
self._update_progress(mapper,
type_=ProgressThread.P_PROCESSED)
if collector_queue is not None:
if self.collector:
if not isinstance(
result, tuple
): # collector must represent as tuple
result = (result, )
batch_result.append(result)
if collector_queue is not None and len(batch_result) > 0:
collector_queue.put(
(ParallelProcessor.CMD_DATA, batch_result))
batch_result = [] # reset buffer
def jacobianos(Xint, Ns, XextList, params, hessianos=True):
'''
funcion que calcula los jacobianos y hessianos de las variables intrinsecas
y extrinsecas. hace un hilo para cada cuenta
'''
# donde guardar resultado de derivadas de params internos
jInt = Queue()
if hessianos:
hInt = Queue()
# creo e inicializo los threads
if hessianos:
# print('cuentas intrinsecas, 2 processos')
pHInt = Process(target=procHint,
args=(Xint, Ns, XextList, params, hInt))
pHInt.start()
#else:
# print('cuentas intrinsecas, 1 processo')
pJInt = Process(target=procJint, args=(Xint, Ns, XextList, params, jInt))
pJInt.start() # inicio procesos
# donde guardar resultados de jaco y hess externos
n = len(XextList)
jExt = np.zeros((n, 1, 6), dtype=float)
qJext = [Queue() for nn in range(n)]
if hessianos:
hExt = np.zeros((n, 6, 6), dtype=float)
qHext = [Queue() for nn in range(n)]
# lista de threads
proJ = list()
if hessianos:
proH = list()
# creo e inicializo los threads
for i in range(n):
# print('starting par de processos ', i + 3)
pJ = Process(target=procJext,
args=(XextList[i], Xint, Ns, params, i, qJext[i]))
proJ.append(pJ)
if hessianos:
pH = Process(target=procHext,
args=(XextList[i], Xint, Ns, params, i, qHext[i]))
proH.append(pH)
pJ.start() # inicio procesos
if hessianos:
pH.start()
jInt = jInt.get() # saco los resultados
if hessianos:
hInt = hInt.get()
for i in range(n):
jExt[i] = qJext[i].get() # guardo resultados
if hessianos:
hExt[i] = qHext[i].get()
pJInt.join() # espero a que todos terminen
if hessianos:
pHInt.join()
[p.join() for p in proJ]
if hessianos:
[p.join() for p in proH]
if hessianos:
return jInt, hInt, jExt, hExt
else:
return jInt, jExt
pool = mp.Pool(processes = 4)
def func1(i):
time.sleep(20)
return 1
result = []
for i in range(100):
result.append(pool.apply_async(func1, (1)))
print(len(result))
pool.close()
pool.join()
'''
freeze_support()
NUMBER_OF_PROCESSES = 4
TASK1 = [(func, i) for i in range(10)]
task_queue = Queue()
done_queue = Queue()
list(map(task_queue.put, TASK1))
for i in range(NUMBER_OF_PROCESSES):
Process(target=worker, args=(task_queue, done_queue)).start()
print("submit complete")
for i in range(len(TASK1)):
print(done_queue.get())
class TaskManager:
"""
User-facing Class for interfacing with OpenWPM
The TaskManager spawns several child processes to run the automation tasks.
- DataAggregator to aggregate data in a SQLite database
- MPLogger to aggregate logs across processes
- BrowserManager processes to isolate Browsers in a separate process
<manager_params> dict of TaskManager configuration parameters
<browser_params> is a list of (or a single) dictionaries that specify
preferences for browsers to instantiate
<process_watchdog> will monitor firefox and Xvfb processes, killing
any not indexed in TaskManager's browser list.
NOTE: Only run this in isolated environments. It kills processes
by name, indiscriminately.
"""
def __init__(self, manager_params, browser_params, process_watchdog=False):
# Make paths absolute in manager_params
for path in ['data_directory', 'log_directory']:
if manager_params[path] is not None:
manager_params[path] = os.path.expanduser(manager_params[path])
manager_params['database_name'] = os.path.join(
manager_params['data_directory'], manager_params['database_name'])
manager_params['log_file'] = os.path.join(
manager_params['log_directory'], manager_params['log_file'])
manager_params['screenshot_path'] = os.path.join(
manager_params['data_directory'], 'screenshots')
manager_params['source_dump_path'] = os.path.join(
manager_params['data_directory'], 'sources')
self.manager_params = manager_params
self.browser_params = browser_params
# Create data directories if they do not exist
if not os.path.exists(manager_params['screenshot_path']):
os.makedirs(manager_params['screenshot_path'])
if not os.path.exists(manager_params['source_dump_path']):
os.makedirs(manager_params['source_dump_path'])
# check size of parameter dictionary
self.num_browsers = manager_params['num_browsers']
if len(browser_params) != self.num_browsers:
raise Exception("Number of <browser_params> dicts is not the same "
"as manager_params['num_browsers']")
# Flow control
self.closing = False
self.failure_status = None
self.threadlock = threading.Lock()
self.failurecount = 0
if manager_params['failure_limit'] is not None:
self.failure_limit = manager_params['failure_limit']
else:
self.failure_limit = self.num_browsers * 2 + 10
self.process_watchdog = process_watchdog
# sets up logging server + connect a client
self.logging_status_queue = None
self.loggingserver = self._launch_loggingserver()
# socket location: (address, port)
self.manager_params['logger_address'] = self.logging_status_queue.get()
self.logger = MPLogger.loggingclient(
*self.manager_params['logger_address'])
# Mark if LDBAggregator is needed
# (if content saving is enabled on any browser)
self.ldb_enabled = False
for params in browser_params:
if params['save_javascript'] or params['save_all_content']:
self.ldb_enabled = True
break
# Initialize the data aggregators
self._launch_aggregators()
# sets up the BrowserManager(s) + associated queues
self.browsers = self._initialize_browsers(browser_params)
self._launch_browsers()
# start the manager watchdog
thread = threading.Thread(target=self._manager_watchdog, args=())
thread.daemon = True
thread.start()
# Save crawl config information to database
openwpm_v, browser_v = get_version()
self.data_aggregator.save_configuration(openwpm_v, browser_v)
self.logger.info(
get_configuration_string(self.manager_params, browser_params,
(openwpm_v, browser_v)))
def _initialize_browsers(self, browser_params):
""" initialize the browser classes, each its unique set of params """
browsers = list()
for i in range(self.num_browsers):
browser_params[i][
'crawl_id'] = self.data_aggregator.get_next_crawl_id()
browsers.append(Browser(self.manager_params, browser_params[i]))
return browsers
def _launch_browsers(self):
""" launch each browser manager process / browser """
for browser in self.browsers:
try:
success = browser.launch_browser_manager()
except Exception:
self._cleanup_before_fail(during_init=True)
raise
if not success:
self.logger.critical("Browser spawn failure during "
"TaskManager initialization, exiting...")
self.close()
break
def _manager_watchdog(self):
"""
Periodically checks the following:
- memory consumption of all browsers every 10 seconds
- presence of processes that are no longer in use
TODO: process watchdog needs to be updated since `psutil` won't
kill browser processes started by Selenium 3 (with `subprocess`)
"""
if self.process_watchdog:
self.logger.error("BROWSER %i: Process watchdog is not currently "
"supported." % self.crawl_id)
while not self.closing:
time.sleep(10)
# Check browser memory usage
for browser in self.browsers:
try:
process = psutil.Process(browser.browser_pid)
mem = process.memory_info()[0] / float(2**20)
if mem > BROWSER_MEMORY_LIMIT:
self.logger.info(
"BROWSER %i: Memory usage: %iMB"
", exceeding limit of %iMB" %
(browser.crawl_id, int(mem), BROWSER_MEMORY_LIMIT))
browser.restart_required = True
except psutil.NoSuchProcess:
pass
# Check for browsers or displays that were not closed correctly
# 300 second buffer to avoid killing freshly launched browsers
# TODO This buffer should correspond to the maximum spawn timeout
if self.process_watchdog:
browser_pids = set()
display_pids = set()
check_time = time.time()
for browser in self.browsers:
if browser.browser_pid is not None:
browser_pids.add(browser.browser_pid)
if browser.display_pid is not None:
display_pids.add(browser.display_pid)
for process in psutil.process_iter():
if (process.create_time() + 300 < check_time
and ((process.name() == 'firefox'
and process.pid not in browser_pids) or
(process.name() == 'Xvfb'
and process.pid not in display_pids))):
self.logger.debug("Process: %s (pid: %i) with start "
"time %s found running but not in "
"browser process list. Killing." %
(process.name(), process.pid,
process.create_time()))
process.kill()
def _launch_aggregators(self):
"""Launch the necessary data aggregators"""
self.data_aggregator = SqliteAggregator.SqliteAggregator(
self.manager_params, self.browser_params)
self.data_aggregator.launch()
self.manager_params[
'aggregator_address'] = self.data_aggregator.listener_address
# open connection to aggregator for saving crawl details
self.sock = clientsocket(serialization='dill')
self.sock.connect(*self.manager_params['aggregator_address'])
# TODO refactor ldb aggregator to use new base classes
if self.ldb_enabled:
self.ldb_status_queue = Queue()
self.ldb_aggregator = Process(
target=LevelDBAggregator.LevelDBAggregator,
args=(self.manager_params, self.ldb_status_queue))
self.ldb_aggregator.daemon = True
self.ldb_aggregator.start()
# socket location: (address, port)
self.manager_params['ldb_address'] = self.ldb_status_queue.get()
def _kill_aggregators(self):
"""Shutdown any currently running data aggregators"""
self.data_aggregator.shutdown()
# TODO refactor ldb aggregator to use new base classes
if self.ldb_enabled:
self.logger.debug("Telling the LevelDBAggregator to shut down...")
self.ldb_status_queue.put("DIE")
start_time = time.time()
self.ldb_aggregator.join(300)
self.logger.debug("LevelDBAggregator took %s seconds to close." %
(str(time.time() - start_time)))
def _launch_loggingserver(self):
""" sets up logging server """
self.logging_status_queue = Queue()
loggingserver = Process(target=MPLogger.loggingserver,
args=(
self.manager_params['log_file'],
self.logging_status_queue,
))
loggingserver.daemon = True
loggingserver.start()
return loggingserver
def _kill_loggingserver(self):
""" terminates logging server gracefully """
self.logging_status_queue.put("DIE")
self.loggingserver.join(300)
def _shutdown_manager(self, during_init=False):
"""
Wait for current commands to finish, close all child processes and
threads
<during_init> flag to indicator if this shutdown is occuring during
the TaskManager initialization
"""
self.closing = True
for browser in self.browsers:
browser.shutdown_browser(during_init)
self.sock.close() # close socket to data aggregator
self._kill_aggregators()
self._kill_loggingserver()
def _cleanup_before_fail(self, during_init=False):
"""
Execute shutdown commands before throwing an exception
This should keep us from having a bunch of hanging processes
and incomplete data.
<during_init> flag to indicator if this shutdown is occuring during
the TaskManager initialization
"""
self._shutdown_manager(during_init=during_init)
def _check_failure_status(self):
""" Check the status of command failures. Raise exceptions as necessary
The failure status property is used by the various asynchronous
command execution threads which interface with the
remote browser manager processes. If a failure status is found, the
appropriate steps are taken to gracefully close the infrastructure
"""
self.logger.debug("Checking command failure status indicator...")
if self.failure_status:
self.logger.debug(
"TaskManager failure status set, halting command execution.")
self._cleanup_before_fail()
if self.failure_status['ErrorType'] == 'ExceedCommandFailureLimit':
raise CommandExecutionError(
"TaskManager exceeded maximum consecutive command "
"execution failures.",
self.failure_status['CommandSequence'])
elif (self.failure_status['ErrorType'] == ("ExceedLaunch"
"FailureLimit")):
raise CommandExecutionError(
"TaskManager failed to launch browser within allowable "
"failure limit.", self.failure_status['CommandSequence'])
if self.failure_status['ErrorType'] == 'CriticalChildException':
reraise(*pickle.loads(self.failure_status['Exception']))
# CRAWLER COMMAND CODE
def _distribute_command(self, command_seq, index=None):
"""
parses command type and issues command(s) to the proper browser
<index> specifies the type of command this is:
= None -> first come, first serve
= # -> index of browser to send command to
= * -> sends command to all browsers
= ** -> sends command to all browsers (synchronized)
"""
if index is None:
# send to first browser available
command_executed = False
while True:
for browser in self.browsers:
if browser.ready():
browser.current_timeout = command_seq.total_timeout
thread = self._start_thread(browser, command_seq)
command_executed = True
break
if command_executed:
break
time.sleep(SLEEP_CONS)
elif index == '*':
# send the command to all browsers
command_executed = [False] * len(self.browsers)
while False in command_executed:
for i in range(len(self.browsers)):
if self.browsers[i].ready() and not command_executed[i]:
self.browsers[
i].current_timeout = command_seq.total_timeout
thread = self._start_thread(self.browsers[i],
command_seq)
command_executed[i] = True
time.sleep(SLEEP_CONS)
elif index == '**':
# send the command to all browsers and sync it
condition = threading.Condition() # block threads until ready
command_executed = [False] * len(self.browsers)
while False in command_executed:
for i in range(len(self.browsers)):
if self.browsers[i].ready() and not command_executed[i]:
self.browsers[
i].current_timeout = command_seq.total_timeout
thread = self._start_thread(self.browsers[i],
command_seq, condition)
command_executed[i] = True
time.sleep(SLEEP_CONS)
with condition:
condition.notifyAll() # All browsers loaded, start
elif 0 <= index < len(self.browsers):
# send the command to this specific browser
while True:
if self.browsers[index].ready():
self.browsers[
index].current_timeout = command_seq.total_timeout
thread = self._start_thread(self.browsers[index],
command_seq)
break
time.sleep(SLEEP_CONS)
else:
self.logger.info(
"Command index type is not supported or out of range")
return
if command_seq.blocking:
thread.join()
self._check_failure_status()
def _start_thread(self, browser, command_sequence, condition=None):
""" starts the command execution thread """
# Check status flags before starting thread
if self.closing:
self.logger.error(
"Attempted to execute command on a closed TaskManager")
return
self._check_failure_status()
browser.set_visit_id(self.data_aggregator.get_next_visit_id())
self.sock.send(("site_visits", {
"visit_id": browser.curr_visit_id,
"crawl_id": browser.crawl_id,
"site_url": command_sequence.url
}))
# Start command execution thread
args = (browser, command_sequence, condition)
thread = threading.Thread(target=self._issue_command, args=args)
browser.command_thread = thread
thread.daemon = True
thread.start()
return thread
def _issue_command(self, browser, command_sequence, condition=None):
"""
sends command tuple to the BrowserManager
"""
browser.is_fresh = False
# if this is a synced call, block on condition
if condition is not None:
with condition:
condition.wait()
reset = command_sequence.reset
start_time = None
for command_and_timeout in command_sequence.commands_with_timeout:
command, timeout = command_and_timeout
if command[0] in [
'GET', 'BROWSE', 'SAVE_SCREENSHOT', 'SCREENSHOT_FULL_PAGE',
'DUMP_PAGE_SOURCE', 'RECURSIVE_DUMP_PAGE_SOURCE'
]:
start_time = time.time()
command += (browser.curr_visit_id, )
elif command[0] in ['DUMP_FLASH_COOKIES', 'DUMP_PROFILE_COOKIES']:
command += (
start_time,
browser.curr_visit_id,
)
browser.current_timeout = timeout
# passes off command and waits for a success (or failure signal)
browser.command_queue.put(command)
command_succeeded = 0 # 1 success, 0 error, -1 timeout
command_arguments = command[1] if len(command) > 1 else None
# received reply from BrowserManager, either success or failure
try:
status = browser.status_queue.get(True,
browser.current_timeout)
if status == "OK":
command_succeeded = 1
elif status[0] == "CRITICAL":
self.logger.critical(
"BROWSER %i: Received critical error from browser "
"process while executing command %s. Setting failure "
"status." % (browser.crawl_id, str(command)))
self.failure_status = {
'ErrorType': 'CriticalChildException',
'CommandSequence': command_sequence,
'Exception': status[1]
}
return
else:
command_succeeded = 0
self.logger.info(
"BROWSER %i: Received failure status while executing "
"command: %s" % (browser.crawl_id, command[0]))
except EmptyQueue:
command_succeeded = -1
self.logger.info(
"BROWSER %i: Timeout while executing command, %s, killing "
"browser manager" % (browser.crawl_id, command[0]))
self.sock.send(("crawl_history", {
"crawl_id": browser.crawl_id,
"command": command[0],
"arguments": command_arguments,
"bool_success": command_succeeded
}))
if command_succeeded != 1:
with self.threadlock:
self.failurecount += 1
if self.failurecount > self.failure_limit:
self.logger.critical(
"BROWSER %i: Command execution failure pushes failure "
"count above the allowable limit. Setting "
"failure_status." % browser.crawl_id)
self.failure_status = {
'ErrorType': 'ExceedCommandFailureLimit',
'CommandSequence': command_sequence
}
return
browser.restart_required = True
self.logger.debug("BROWSER %i: Browser restart required" %
(browser.crawl_id))
else:
with self.threadlock:
self.failurecount = 0
if browser.restart_required:
break
# Sleep after executing CommandSequence to provide extra time for
# internal buffers to drain. Stopgap in support of #135
time.sleep(2)
if self.closing:
return
if browser.restart_required or reset:
success = browser.restart_browser_manager(clear_profile=reset)
if not success:
self.logger.critical(
"BROWSER %i: Exceeded the maximum allowable consecutive "
"browser launch failures. Setting failure_status." %
(browser.crawl_id))
self.failure_status = {
'ErrorType': 'ExceedLaunchFailureLimit',
'CommandSequence': command_sequence
}
return
browser.restart_required = False
def execute_command_sequence(self, command_sequence, index=None):
self._distribute_command(command_sequence, index)
# DEFINITIONS OF HIGH LEVEL COMMANDS
# NOTE: These wrappers are provided for convenience. To issue sequential
# commands to the same browser in a single 'visit', use the CommandSequence
# class directly.
def get(self, url, index=None, timeout=60, sleep=0, reset=False):
""" goes to a url """
command_sequence = CommandSequence.CommandSequence(url)
command_sequence.get(timeout=timeout, sleep=sleep)
command_sequence.reset = reset
self.execute_command_sequence(command_sequence, index=index)
def browse(self,
url,
num_links=2,
sleep=0,
index=None,
timeout=60,
reset=False):
""" browse a website and visit <num_links> links on the page """
command_sequence = CommandSequence.CommandSequence(url)
command_sequence.browse(num_links=num_links,
sleep=sleep,
timeout=timeout)
command_sequence.reset = reset
self.execute_command_sequence(command_sequence, index=index)
def close(self):
"""
Execute shutdown procedure for TaskManager
"""
if self.closing:
self.logger.error("TaskManager already closed")
return
self._shutdown_manager()
class BaseAggregator(object):
"""Base class for the data aggregator interface. This class is used
alongside the BaseListener class to spawn an aggregator process that
combines data from multiple crawl processes. The BaseAggregator class
manages the child listener process.
Parameters
----------
manager_params : dict
TaskManager configuration parameters
browser_params : list of dict
List of browser configuration dictionaries"""
__metaclass__ = abc.ABCMeta
def __init__(self, manager_params, browser_params):
self.manager_params = manager_params
self.browser_params = browser_params
self.listener_address = None
self.listener_process = None
self.status_queue = Queue()
self.shutdown_queue = Queue()
self._last_status = None
self._last_status_received = None
self.logger = logging.getLogger('openwpm')
@abc.abstractmethod
def save_configuration(self, openwpm_version, browser_version):
"""Save configuration details to the database"""
@abc.abstractmethod
def get_next_visit_id(self):
"""Return a unique visit ID to be used as a key for a single visit"""
@abc.abstractmethod
def get_next_crawl_id(self):
"""Return a unique crawl ID used as a key for a browser instance"""
def get_most_recent_status(self):
"""Return the most recent queue size sent from the listener process"""
# Block until we receive the first status update
if self._last_status is None:
return self.get_status()
# Drain status queue until we receive most recent update
while not self.status_queue.empty():
self._last_status = self.status_queue.get()
self._last_status_received = time.time()
# Check last status signal
if (time.time() - self._last_status_received) > STATUS_TIMEOUT:
raise RuntimeError(
"No status update from DataAggregator listener process "
"for %d seconds." % (time.time() - self._last_status_received))
return self._last_status
def get_status(self):
"""Get listener process status. If the status queue is empty, block."""
try:
self._last_status = self.status_queue.get(block=True,
timeout=STATUS_TIMEOUT)
self._last_status_received = time.time()
except queue.Empty:
raise RuntimeError(
"No status update from DataAggregator listener process "
"for %d seconds." % (time.time() - self._last_status_received))
return self._last_status
def launch(self, listener_process_runner, *args):
"""Launch the aggregator listener process"""
args = (self.manager_params, self.status_queue,
self.shutdown_queue) + args
self.listener_process = Process(target=listener_process_runner,
args=args)
self.listener_process.daemon = True
self.listener_process.start()
self.listener_address = self.status_queue.get()
def shutdown(self):
""" Terminate the aggregator listener process"""
self.logger.debug(
"Sending the shutdown signal to the %s listener process..." %
type(self).__name__)
self.shutdown_queue.put(SHUTDOWN_SIGNAL)
start_time = time.time()
self.listener_process.join(300)
self.logger.debug("%s took %s seconds to close." %
(type(self).__name__, str(time.time() - start_time)))
self.listener_address = None
self.listener_process = None
def _next(self):
'''
Core algorithm part in each iteration, see algorithm 1.
--------------------------------------
xs = stochastic_sampling(B, F, X)
for x in xs:
D = select_direction(B, x)
x_opt = local_optimization(D, F, X)
M = first_order_approximation(x_opt, F, X)
update_buffer(B, F(M))
--------------------------------------
where F is problem evaluation, X is design constraints
'''
# update optimization progress
# sys.stdout.write('\b' * len(str(self.n_gen - 1)) + str(self.n_gen))
# sys.stdout.flush()
# stochastic sampling by adding local perturbance
xs = self._stochastic_sampling()
# parallelize core pareto discovery process by multiprocess, see _pareto_discover()
# including select_direction, local_optimization, first_order_approximation in above algorithm illustration
x_batch = np.array_split(xs, self.n_process)
queue = Queue()
process_count = 0
for x in x_batch:
if len(x) > 0:
p = Process(target=_pareto_discover,
args=(x, self.problem.evaluate,
[self.problem.xl,
self.problem.xu], self.constr_func,
self.delta_s, self.buffer.origin,
self.buffer.origin_constant,
self.n_grid_sample, queue))
p.start()
process_count += 1
# gather results (new samples, new patch ids, new origin of performance buffer) from parallel discovery
new_origin = self.buffer.origin
x_samples_all = []
patch_ids_all = []
for _ in range(process_count):
x_samples, patch_ids, sample_num, origin = queue.get()
if x_samples is not None:
x_samples_all.append(x_samples)
patch_ids_all.append(
np.array(patch_ids) + self.patch_id
) # assign corresponding global patch ids to samples
self.patch_id += sample_num
new_origin = np.minimum(new_origin, origin)
# evalaute all new samples and adjust the origin point of buffer
x_samples_all = np.vstack(x_samples_all)
y_samples_all = self.problem.evaluate(x_samples_all,
return_values_of=['F'])
new_origin = np.minimum(np.min(y_samples_all, axis=0), new_origin)
patch_ids_all = np.concatenate(patch_ids_all)
# update buffer
self.buffer.move_origin(new_origin)
self.buffer.insert(x_samples_all, y_samples_all, patch_ids_all)
# update population by the best samples in the buffer
self.pop = self.pop.new('X', self.buffer.sample(self.pop_size))
self.evaluator.eval(self.problem, self.pop, algorithm=self)
class TaskManager:
"""
User-facing Class for interfacing with OpenWPM
The TaskManager spawns several child processes to run the automation tasks.
- DataAggregator to aggregate data in a SQLite database
- MPLogger to aggregate logs across processes
- BrowserManager processes to isolate Browsers in a separate process
<manager_params> dict of TaskManager configuration parameters
<browser_params> is a list of (or a single) dictionaries that specify preferences for browsers to instantiate
<process_watchdog> will monitor firefox and Xvfb processes, killing any not indexed in TaskManager's browser list.
NOTE: Only run this in isolated environments. It kills processes by name, indiscriminately.
"""
def __init__(self, manager_params, browser_params, process_watchdog=False):
# Make paths absolute in manager_params
for path in ['data_directory', 'log_directory']:
if manager_params[path] is not None:
manager_params[path] = os.path.expanduser(manager_params[path])
manager_params['database_name'] = os.path.join(
manager_params['data_directory'], manager_params['database_name'])
manager_params['log_file'] = os.path.join(
manager_params['log_directory'], manager_params['log_file'])
manager_params['screenshot_path'] = os.path.join(
manager_params['data_directory'], 'screenshots')
manager_params['source_dump_path'] = os.path.join(
manager_params['data_directory'], 'sources')
self.manager_params = manager_params
# Create data directories if they do not exist
if not os.path.exists(manager_params['screenshot_path']):
os.makedirs(manager_params['screenshot_path'])
if not os.path.exists(manager_params['source_dump_path']):
os.makedirs(manager_params['source_dump_path'])
# check size of parameter dictionary
self.num_browsers = manager_params['num_browsers']
if len(browser_params) != self.num_browsers:
raise Exception(
"Number of <browser_params> dicts is not the same as manager_params['num_browsers']"
)
# Flow control
self.closing = False
self.failure_status = None
self.threadlock = threading.Lock()
self.failurecount = 0
if manager_params['failure_limit'] is not None:
self.failure_limit = manager_params['failure_limit']
else:
self.failure_limit = self.num_browsers * 2 + 10
self.process_watchdog = process_watchdog
# sets up the crawl data database
db_path = manager_params['database_name']
if not os.path.exists(manager_params['data_directory']):
os.mkdir(manager_params['data_directory'])
self.db = sqlite3.connect(db_path)
with open(os.path.join(os.path.dirname(__file__), 'schema.sql'),
'r') as f:
self.db.executescript(f.read())
self.db.commit()
# sets up logging server + connect a client
self.logging_status_queue = None
self.loggingserver = self._launch_loggingserver()
# socket location: (address, port)
self.manager_params['logger_address'] = self.logging_status_queue.get()
self.logger = MPLogger.loggingclient(
*self.manager_params['logger_address'])
# Mark if LDBAggregator is needed (if js is enabled on any browser)
self.ldb_enabled = False
for params in browser_params:
if params['save_javascript'] or params['save_javascript_proxy']:
self.ldb_enabled = True
break
# Initialize the data aggregators
self._launch_aggregators()
# open client socket
self.sock = clientsocket(serialization='dill')
self.sock.connect(*self.manager_params['aggregator_address'])
print 1
self._save_configuration(browser_params)
print 2
# read the last used site visit id
cur = self.db.cursor()
cur.execute("SELECT MAX(visit_id) from site_visits")
last_visit_id = cur.fetchone()[0]
if last_visit_id is None:
last_visit_id = 0
self.next_visit_id = last_visit_id + 1
print 3
# sets up the BrowserManager(s) + associated queues
self.browsers = self._initialize_browsers(
browser_params) # List of the Browser(s)
print 5
self._launch_browsers()
print 4
# start the manager watchdog
thread = threading.Thread(target=self._manager_watchdog, args=())
thread.daemon = True
thread.start()
def _save_configuration(self, browser_params):
""" Saves crawl configuration details to db and logfile"""
cur = self.db.cursor()
# Get git version and commit information
openwpm_v, browser_v = get_version()
# Record task details
cur.execute(("INSERT INTO task "
"(manager_params, openwpm_version, browser_version) "
"VALUES (?,?,?)"),
(json.dumps(self.manager_params), openwpm_v, browser_v))
self.db.commit()
self.task_id = cur.lastrowid
# Record browser details for each brower
for i in xrange(self.num_browsers):
cur.execute(
"INSERT INTO crawl (task_id, browser_params) VALUES (?,?)",
(self.task_id, json.dumps(browser_params[i])))
self.db.commit()
browser_params[i]['crawl_id'] = cur.lastrowid
# Print the configuration details
self.logger.info(
get_configuration_string(self.manager_params, browser_params,
(openwpm_v, browser_v)))
def _initialize_browsers(self, browser_params):
""" initialize the browser classes, each its unique set of parameters """
browsers = list()
for i in xrange(self.num_browsers):
browsers.append(Browser(self.manager_params, browser_params[i]))
return browsers
def _launch_browsers(self):
print 8
print self.browsers
""" launch each browser manager process / browser """
for browser in self.browsers:
try:
print 9
success = browser.launch_browser_manager()
print 6
except:
print 7
self._cleanup_before_fail(during_init=True)
raise
if not success:
self.logger.critical(
"Browser spawn failure during TaskManager initialization, exiting..."
)
self.close()
break
# Update our DB with the random browser settings
# These are found within the scope of each instance of Browser in the browsers list
screen_res = str(browser.browser_settings['screen_res'])
ua_string = str(browser.browser_settings['ua_string'])
self.sock.send(("UPDATE crawl SET screen_res = ?, ua_string = ? \
WHERE crawl_id = ?", (screen_res, ua_string,
browser.crawl_id)))
def _manager_watchdog(self):
"""
Periodically checks the following:
- memory consumption of all browsers every 10 seconds
- presence of processes that are no longer in use
"""
while not self.closing:
time.sleep(10)
# Check browser memory usage
for browser in self.browsers:
try:
process = psutil.Process(browser.browser_pid)
mem = process.memory_info()[0] / float(2**20)
if mem > BROWSER_MEMORY_LIMIT:
self.logger.info(
"BROWSER %i: Memory usage: %iMB, exceeding limit of %iMB"
%
(browser.crawl_id, int(mem), BROWSER_MEMORY_LIMIT))
browser.restart_required = True
except psutil.NoSuchProcess:
pass
# Check for browsers or displays that were not closed correctly
# Provide a 300 second buffer to avoid killing freshly launched browsers
# TODO This buffer should correspond to the maximum browser spawn timeout
if self.process_watchdog:
browser_pids = set()
display_pids = set()
check_time = time.time()
for browser in self.browsers:
if browser.browser_pid is not None:
browser_pids.add(browser.browser_pid)
if browser.display_pid is not None:
display_pids.add(browser.display_pid)
for process in psutil.process_iter():
if (process.create_time() + 300 < check_time
and ((process.name() == 'firefox'
and process.pid not in browser_pids) or
(process.name() == 'Xvfb'
and process.pid not in display_pids))):
self.logger.debug(
"Process: %s (pid: %i) with start time %s found running but not in browser process list. Killing."
% (process.name(), process.pid,
process.create_time()))
process.kill()
def _launch_aggregators(self):
"""
Launches the various data aggregators, which serialize data from all processes.
* DataAggregator - sqlite database for crawl data
* LevelDBAggregator - leveldb database for javascript files
"""
# DataAggregator
self.aggregator_status_queue = Queue()
self.data_aggregator = Process(target=DataAggregator.DataAggregator,
args=(self.manager_params,
self.aggregator_status_queue))
self.data_aggregator.daemon = True
self.data_aggregator.start()
self.manager_params[
'aggregator_address'] = self.aggregator_status_queue.get(
) # socket location: (address, port)
# LevelDB Aggregator
if self.ldb_enabled:
self.ldb_status_queue = Queue()
self.ldb_aggregator = Process(
target=LevelDBAggregator.LevelDBAggregator,
args=(self.manager_params, self.ldb_status_queue))
self.ldb_aggregator.daemon = True
self.ldb_aggregator.start()
self.manager_params['ldb_address'] = self.ldb_status_queue.get(
) # socket location: (address, port)
def _kill_aggregators(self):
""" Terminates the aggregators gracefully """
# DataAggregator
self.logger.debug("Telling the DataAggregator to shut down...")
self.aggregator_status_queue.put("DIE")
start_time = time.time()
self.data_aggregator.join(300)
self.logger.debug("DataAggregator took " +
str(time.time() - start_time) + " seconds to close")
# LevelDB Aggregator
if self.ldb_enabled:
self.logger.debug("Telling the LevelDBAggregator to shut down...")
self.ldb_status_queue.put("DIE")
start_time = time.time()
self.ldb_aggregator.join(300)
self.logger.debug("LevelDBAggregator took " +
str(time.time() - start_time) +
" seconds to close")
def _launch_loggingserver(self):
""" sets up logging server """
self.logging_status_queue = Queue()
loggingserver = Process(target=MPLogger.loggingserver,
args=(
self.manager_params['log_file'],
self.logging_status_queue,
))
loggingserver.daemon = True
loggingserver.start()
return loggingserver
def _kill_loggingserver(self):
""" terminates logging server gracefully """
self.logging_status_queue.put("DIE")
self.loggingserver.join(300)
def _shutdown_manager(self, failure=False, during_init=False):
"""
Wait for current commands to finish, close all child processes and
threads
<failure> flag to indicate manager failure (True) or end of crawl (False)
<during_init> flag to indicator if this shutdown is occuring during the TaskManager initialization
"""
self.closing = True
for browser in self.browsers:
browser.shutdown_browser(during_init)
if failure:
self.sock.send(
("UPDATE crawl SET finished = -1 WHERE crawl_id = ?",
(browser.crawl_id, )))
else:
self.sock.send(
("UPDATE crawl SET finished = 1 WHERE crawl_id = ?",
(browser.crawl_id, )))
self.db.close() # close db connection
self.sock.close() # close socket to data aggregator
self._kill_aggregators()
self._kill_loggingserver()
def _cleanup_before_fail(self, during_init=False):
"""
Execute shutdown commands before throwing an exception
This should keep us from having a bunch of hanging processes
and incomplete data.
<during_init> flag to indicator if this shutdown is occuring during
the TaskManager initialization
"""
self._shutdown_manager(failure=True, during_init=during_init)
def _check_failure_status(self):
""" Check the status of command failures. Raise exceptions as necessary
The failure status property is used by the various asynchronous
command execution threads which interface with the
remote browser manager processes. If a failure status is found, the
appropriate steps are taken to gracefully close the infrastructure
"""
self.logger.debug("Checking command failure status indicator...")
if self.failure_status:
self.logger.debug(
"TaskManager failure status set, halting command execution.")
self._cleanup_before_fail()
if self.failure_status['ErrorType'] == 'ExceedCommandFailureLimit':
raise CommandExecutionError(
"TaskManager exceeded maximum consecutive command "
"execution failures.",
self.failure_status['CommandSequence'])
elif self.failure_status[
'ErrorType'] == 'ExceedLaunchFailureLimit':
raise CommandExecutionError(
"TaskManager failed to launch browser within allowable "
"failure limit.", self.failure_status['CommandSequence'])
if self.failure_status['ErrorType'] == 'CriticalChildException':
reraise(*cPickle.loads(self.failure_status['Exception']))
# CRAWLER COMMAND CODE
def _distribute_command(self, command_sequence, index=None):
"""
parses command type and issues command(s) to the proper browser
<index> specifies the type of command this is:
= None -> first come, first serve
= # -> index of browser to send command to
= * -> sends command to all browsers
= ** -> sends command to all browsers (synchronized)
"""
if index is None:
#send to first browser available
command_executed = False
while True:
for browser in self.browsers:
if browser.ready():
browser.current_timeout = command_sequence.total_timeout
thread = self._start_thread(browser, command_sequence)
command_executed = True
break
if command_executed:
break
time.sleep(SLEEP_CONS)
elif 0 <= index < len(self.browsers):
#send the command to this specific browser
while True:
if self.browsers[index].ready():
self.browsers[
index].current_timeout = command_sequence.total_timeout
thread = self._start_thread(self.browsers[index],
command_sequence)
break
time.sleep(SLEEP_CONS)
elif index == '*':
#send the command to all browsers
command_executed = [False] * len(self.browsers)
while False in command_executed:
for i in xrange(len(self.browsers)):
if self.browsers[i].ready() and not command_executed[i]:
self.browsers[
i].current_timeout = command_sequence.total_timeout
thread = self._start_thread(self.browsers[i],
command_sequence)
command_executed[i] = True
time.sleep(SLEEP_CONS)
elif index == '**':
#send the command to all browsers and sync it
condition = threading.Condition(
) # Used to block threads until ready
command_executed = [False] * len(self.browsers)
while False in command_executed:
for i in xrange(len(self.browsers)):
if self.browsers[i].ready() and not command_executed[i]:
self.browsers[
i].current_timeout = command_sequence.total_timeout
thread = self._start_thread(self.browsers[i],
command_sequence,
condition)
command_executed[i] = True
time.sleep(SLEEP_CONS)
with condition:
condition.notifyAll(
) # All browsers loaded, tell them to start
else:
self.logger.info(
"Command index type is not supported or out of range")
return
if command_sequence.blocking:
thread.join()
self._check_failure_status()
def _start_thread(self, browser, command_sequence, condition=None):
""" starts the command execution thread """
# Check status flags before starting thread
if self.closing:
self.logger.error(
"Attempted to execute command on a closed TaskManager")
return
self._check_failure_status()
browser.set_visit_id(self.next_visit_id)
self.sock.send((
"INSERT INTO site_visits (visit_id, crawl_id, site_url) VALUES (?,?,?)",
(self.next_visit_id, browser.crawl_id, command_sequence.url)))
self.next_visit_id += 1
# Start command execution thread
args = (browser, command_sequence, condition)
thread = threading.Thread(target=self._issue_command, args=args)
browser.command_thread = thread
thread.daemon = True
thread.start()
return thread
def _issue_command(self, browser, command_sequence, condition=None):
"""
sends command tuple to the BrowserManager
"""
browser.is_fresh = False # since we are issuing a command, the BrowserManager is no longer a fresh instance
# if this is a synced call, block on condition
if condition is not None:
with condition:
condition.wait()
reset = command_sequence.reset
start_time = None # tracks when a site visit started, so that flash/profile
# cookies can be properly tracked.
for command_and_timeout in command_sequence.commands_with_timeout:
command, timeout = command_and_timeout
if command[0] in ['GET', 'BROWSE']:
start_time = time.time()
command += (browser.curr_visit_id, )
elif command[0] in ['DUMP_FLASH_COOKIES', 'DUMP_PROFILE_COOKIES']:
command += (
start_time,
browser.curr_visit_id,
)
browser.current_timeout = timeout
# passes off command and waits for a success (or failure signal)
browser.command_queue.put(command)
command_succeeded = 0 #1 success, 0 failure from error, -1 timeout
command_arguments = command[1] if len(command) > 1 else None
# received reply from BrowserManager, either success signal or failure notice
try:
status = browser.status_queue.get(True,
browser.current_timeout)
if status == "OK":
command_succeeded = 1
elif status[0] == "CRITICAL":
self.logger.critical(
"BROWSER %i: Received critical error "
"from browser process while executing "
"command %s. Setting failure status." %
(browser.crawl_id, str(command)))
self.failure_status = {
'ErrorType': 'CriticalChildException',
'CommandSequence': command_sequence,
'Exception': status[1]
}
return
else:
command_succeeded = 0
self.logger.info(
"BROWSER %i: Received failure status while"
" executing command: %s" %
(browser.crawl_id, command[0]))
except EmptyQueue:
command_succeeded = -1
self.logger.info(
"BROWSER %i: Timeout while executing command, "
"%s, killing browser manager" %
(browser.crawl_id, command[0]))
self.sock.send((
"INSERT INTO CrawlHistory (crawl_id, command, arguments, bool_success)"
" VALUES (?,?,?,?)", (browser.crawl_id, command[0],
command_arguments, command_succeeded)))
if command_succeeded != 1:
with self.threadlock:
self.failurecount += 1
if self.failurecount > self.failure_limit:
self.logger.critical(
"BROWSER %i: Command execution failure"
" pushes failure count above the allowable limit."
" Setting failure_status." % browser.crawl_id)
self.failure_status = {
'ErrorType': 'ExceedCommandFailureLimit',
'CommandSequence': command_sequence
}
return
browser.restart_required = True
else:
with self.threadlock:
self.failurecount = 0
if browser.restart_required:
break
if self.closing:
return
if browser.restart_required or reset:
success = browser.restart_browser_manager(clear_profile=reset)
if not success:
self.logger.critical(
"BROWSER %i: Exceeded the maximum allowable "
"consecutive browser launch failures. "
"Setting failure_status." % browser.crawl_id)
self.failure_status = {
'ErrorType': 'ExceedLaunchFailureLimit',
'CommandSequence': command_sequence
}
return
browser.restart_required = False
def execute_command_sequence(self, command_sequence, index=None):
self._distribute_command(command_sequence, index)
# DEFINITIONS OF HIGH LEVEL COMMANDS
# NOTE: These wrappers are provided for convenience. To issue sequential
# commands to the same browser in a single 'visit', use the CommandSequence
# class directly.
def get(self, url, index=None, timeout=60, sleep=0, reset=False):
""" goes to a url """
command_sequence = CommandSequence.CommandSequence(url)
command_sequence.get(timeout=timeout, sleep=sleep)
command_sequence.reset = reset
self.execute_command_sequence(command_sequence, index=index)
def browse(self,
url,
num_links=2,
sleep=0,
index=None,
timeout=60,
reset=False):
""" browse a website and visit <num_links> links on the page """
command_sequence = CommandSequence.CommandSequence(url)
command_sequence.get(sleep=sleep, timeout=timeout)
command_sequence.reset = reset
self.execute_command_sequence(command_sequence, index=index)
def close(self):
"""
Execute shutdown procedure for TaskManager
"""
if self.closing:
self.logger.error("TaskManager already closed")
return
self._shutdown_manager()
class BaseAggregator:
"""Base class for the data aggregator interface. This class is used
alongside the BaseListener class to spawn an aggregator process that
combines data from multiple crawl processes. The BaseAggregator class
manages the child listener process.
Parameters
----------
manager_params : ManagerParamsInternal
TaskManager configuration parameters
browser_params : list of BrowserParamsInternal
List of browser configuration class<BrowserParams>"""
__metaclass__ = abc.ABCMeta
def __init__(
self,
manager_params: ManagerParamsInternal,
browser_params: List[BrowserParamsInternal],
):
self.manager_params = manager_params
self.browser_params = browser_params
self.listener_address = None
self.listener_process = None
self.status_queue = Queue()
self.completion_queue = Queue()
self.shutdown_queue = Queue()
self._last_status = None
self._last_status_received = None
self.logger = logging.getLogger("openwpm")
@abc.abstractmethod
def save_configuration(self, openwpm_version, browser_version):
"""Save configuration details to the database"""
@abc.abstractmethod
def get_next_visit_id(self):
"""Return a unique visit ID to be used as a key for a single visit"""
@abc.abstractmethod
def get_next_browser_id(self):
"""Return a unique crawl ID used as a key for a browser instance"""
def get_most_recent_status(self):
"""Return the most recent queue size sent from the listener process"""
# Block until we receive the first status update
if self._last_status is None:
return self.get_status()
# Drain status queue until we receive most recent update
while not self.status_queue.empty():
self._last_status = self.status_queue.get()
self._last_status_received = time.time()
# Check last status signal
if (time.time() - self._last_status_received) > STATUS_TIMEOUT:
raise RuntimeError(
"No status update from DataAggregator listener process "
"for %d seconds." % (time.time() - self._last_status_received))
return self._last_status
def get_status(self):
"""Get listener process status. If the status queue is empty, block."""
try:
self._last_status = self.status_queue.get(block=True,
timeout=STATUS_TIMEOUT)
self._last_status_received = time.time()
except queue.Empty:
raise RuntimeError(
"No status update from DataAggregator listener process "
"for %d seconds." % (time.time() - self._last_status_received))
return self._last_status
def get_new_completed_visits(self) -> List[Tuple[int, bool]]:
"""
Returns a list of all visit ids that have been processed since
the last time the method was called and whether or not they
have been interrupted.
This method will return an empty list in case no visit ids have
been processed since the last time this method was called
"""
finished_visit_ids = list()
while not self.completion_queue.empty():
finished_visit_ids.append(self.completion_queue.get())
return finished_visit_ids
def launch(self, listener_process_runner, *args):
"""Launch the aggregator listener process"""
args = ((self.status_queue, self.completion_queue,
self.shutdown_queue), ) + args
self.listener_process = Process(target=listener_process_runner,
args=args)
self.listener_process.daemon = True
self.listener_process.start()
self.listener_address = self.status_queue.get()
def shutdown(self, relaxed: bool = True):
""" Terminate the aggregator listener process"""
self.logger.debug(
"Sending the shutdown signal to the %s listener process..." %
type(self).__name__)
self.shutdown_queue.put((SHUTDOWN_SIGNAL, relaxed))
start_time = time.time()
self.listener_process.join(300)
self.logger.debug("%s took %s seconds to close." %
(type(self).__name__, str(time.time() - start_time)))
self.listener_address = None
self.listener_process = None
class StorageControllerHandle:
"""This class contains all methods relevant for the TaskManager
to interact with the StorageController
"""
def __init__(
self,
structured_storage: StructuredStorageProvider,
unstructured_storage: Optional[UnstructuredStorageProvider],
) -> None:
self.listener_address: Optional[Tuple[str, int]] = None
self.listener_process: Optional[Process] = None
self.status_queue = Queue()
self.completion_queue = Queue()
self.shutdown_queue = Queue()
self._last_status = None
self._last_status_received: Optional[float] = None
self.logger = logging.getLogger("openwpm")
self.storage_controller = StorageController(
structured_storage,
unstructured_storage,
status_queue=self.status_queue,
completion_queue=self.completion_queue,
shutdown_queue=self.shutdown_queue,
)
def get_next_visit_id(self) -> VisitId:
"""Generate visit id as randomly generated positive integer less than 2^53.
Parquet can support integers up to 64 bits, but Javascript can only
represent integers up to 53 bits:
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Global_Objects/Number/MAX_SAFE_INTEGER
Thus, we cap these values at 53 bits.
"""
return VisitId(random.getrandbits(53))
def get_next_browser_id(self) -> BrowserId:
"""Generate crawl id as randomly generated positive 32bit integer
Note: Parquet's partitioned dataset reader only supports integer
partition columns up to 32 bits.
"""
return BrowserId(random.getrandbits(32))
def save_configuration(
self,
manager_params: ManagerParamsInternal,
browser_params: List[BrowserParamsInternal],
openwpm_version: str,
browser_version: str,
) -> None:
assert self.listener_address is not None
sock = DataSocket(self.listener_address)
task_id = random.getrandbits(32)
sock.store_record(
TableName("task"),
INVALID_VISIT_ID,
{
"task_id": task_id,
"manager_params": manager_params.to_json(),
"openwpm_version": openwpm_version,
"browser_version": browser_version,
},
)
# Record browser details for each browser
for browser_param in browser_params:
sock.store_record(
TableName("crawl"),
INVALID_VISIT_ID,
{
"browser_id": browser_param.browser_id,
"task_id": task_id,
"browser_params": browser_param.to_json(),
},
)
sock.finalize_visit_id(INVALID_VISIT_ID, success=True)
def launch(self) -> None:
"""Starts the storage controller"""
self.storage_controller = Process(
name="StorageController",
target=StorageController.run,
args=(self.storage_controller, ),
)
self.storage_controller.daemon = True
self.storage_controller.start()
self.listener_address = self.status_queue.get()
def get_new_completed_visits(self) -> List[Tuple[int, bool]]:
"""
Returns a list of all visit ids that have been processed since
the last time the method was called and whether or not they
ran successfully.
This method will return an empty list in case no visit ids have
been processed since the last time this method was called
"""
finished_visit_ids = list()
while not self.completion_queue.empty():
finished_visit_ids.append(self.completion_queue.get())
return finished_visit_ids
def shutdown(self, relaxed: bool = True) -> None:
"""Terminate the storage controller process"""
assert isinstance(self.storage_controller, Process)
self.logger.debug(
"Sending the shutdown signal to the Storage Controller...")
self.shutdown_queue.put((SHUTDOWN_SIGNAL, relaxed))
start_time = time.time()
self.storage_controller.join(300)
self.logger.debug("%s took %s seconds to close." %
(type(self).__name__, str(time.time() - start_time)))
def get_most_recent_status(self) -> int:
"""Return the most recent queue size sent from the Storage Controller process"""
# Block until we receive the first status update
if self._last_status is None:
return self.get_status()
# Drain status queue until we receive most recent update
while not self.status_queue.empty():
self._last_status = self.status_queue.get()
self._last_status_received = time.time()
# Check last status signal
if (time.time() - self._last_status_received) > STATUS_TIMEOUT:
raise RuntimeError(
"No status update from the storage controller process "
"for %d seconds." % (time.time() - self._last_status_received))
return self._last_status
def get_status(self) -> int:
"""Get listener process status. If the status queue is empty, block."""
try:
self._last_status = self.status_queue.get(block=True,
timeout=STATUS_TIMEOUT)
self._last_status_received = time.time()
except queue.Empty:
assert self._last_status_received is not None
raise RuntimeError(
"No status update from the storage controller process "
"for %d seconds." % (time.time() - self._last_status_received))
assert isinstance(self._last_status, int)
return self._last_status
class BaseAggregator(object):
"""Base class for the data aggregator interface. This class is used
alongside the BaseListener class to spawn an aggregator process that
combines data from multiple crawl processes. The BaseAggregator class
manages the child listener process.
Parameters
----------
manager_params : dict
TaskManager configuration parameters
browser_params : list of dict
List of browser configuration dictionaries"""
__metaclass__ = abc.ABCMeta
def __init__(self, manager_params, browser_params):
self.manager_params = manager_params
self.browser_params = browser_params
self.logger = loggingclient(*manager_params['logger_address'])
self.listener_address = None
self.listener_process = None
self.status_queue = Queue()
self.shutdown_queue = Queue()
self._last_status = None
self._last_status_received = None
@abc.abstractmethod
def save_configuration(self, openwpm_version, browser_version):
"""Save configuration details to the database"""
@abc.abstractmethod
def get_next_visit_id(self):
"""Return a unique visit ID to be used as a key for a single visit"""
@abc.abstractmethod
def get_next_crawl_id(self):
"""Return a unique crawl ID used as a key for a browser instance"""
def get_most_recent_status(self):
"""Return the most recent queue size sent from the listener process"""
# Block until we receive the first status update
if self._last_status is None:
return self.get_status()
# Drain status queue until we receive most recent update
while not self.status_queue.empty():
self._last_status = self.status_queue.get()
self._last_status_received = time.time()
# Check last status signal
if (time.time() - self._last_status_received) > STATUS_TIMEOUT:
raise RuntimeError(
"No status update from DataAggregator listener process "
"for %d seconds." % (time.time() - self._last_status_received)
)
return self._last_status
def get_status(self):
"""Get listener process status. If the status queue is empty, block."""
try:
self._last_status = self.status_queue.get(
block=True, timeout=STATUS_TIMEOUT)
self._last_status_received = time.time()
except queue.Empty:
raise RuntimeError(
"No status update from DataAggregator listener process "
"for %d seconds." % (time.time() - self._last_status_received)
)
return self._last_status
def launch(self, listener_process_runner, *args):
"""Launch the aggregator listener process"""
args = (self.manager_params, self.status_queue,
self.shutdown_queue) + args
self.listener_process = Process(
target=listener_process_runner,
args=args
)
self.listener_process.daemon = True
self.listener_process.start()
self.listener_address = self.status_queue.get()
def shutdown(self):
""" Terminate the aggregator listener process"""
self.logger.debug(
"Sending the shutdown signal to the %s listener process..." %
type(self).__name__
)
self.shutdown_queue.put(SHUTDOWN_SIGNAL)
start_time = time.time()
self.listener_process.join(300)
self.logger.debug(
"%s took %s seconds to close." % (
type(self).__name__,
str(time.time() - start_time)
)
)
self.listener_address = None
self.listener_process = None
def execute_parallel(farg_pairs, num_procs=None, verbose=False):
# see https://blog.ionelmc.ro/2014/12/21/compiling-python-extensions-on-windows/
from multiprocess import Process, Queue, cpu_count
if num_procs is None:
# leave 25%
num_procs = math.ceil(cpu_count() * .75)
print "using %d procs in execute parallel" % num_procs
processes = []
q = None
results = []
q = Queue()
num_jobs = len(farg_pairs)
if verbose:
print "execute_parallel num_procs=%d, num_jobs=%d" % (num_procs,
num_jobs)
i = -1
farg_pair = None
farg_pairs = copy.copy(farg_pairs)
while len(farg_pairs) > 0:
farg_pair = farg_pairs.pop(0)
i += 1
if verbose:
print "running job", i
def target_func(*args, **kwargs):
q.put((i, farg_pair[0](*args, **kwargs)))
if len(farg_pair) > 1:
p = Process(target=target_func, args=farg_pair[1])
else:
p = Process(target=target_func)
p.start()
processes.append(p)
# wait until we drop below num_procs
while len(processes) >= num_procs:
len1 = len(results)
results.append(q.get())
if len1 != len(results):
for j, p in enumerate(processes):
if p.exitcode is not None:
p.join()
break
processes = processes[:j] + processes[j + 1:]
else:
time.sleep(0.01)
while len(results) < num_jobs:
results.append(q.get())
time.sleep(0.01)
assert len(results) == num_jobs
# join remaining processes before exiting
for i, p in enumerate(processes):
p.join()
results = zip(*sorted(results, key=lambda x: x[0]))[1]
return results
class DataStreamRecorder(Process):
def __init__(self,
name,
data_sampler_method,
cache_path=None,
save_every=50):
""" Initializes a DataStreamRecorder
Parameters
----------
name : string
User-friendly identifier for this data stream
data_sampler_method : function
Method to call to retrieve data
"""
Process.__init__(self)
self._data_sampler_method = data_sampler_method
self._has_set_sampler_params = False
self._recording = False
self._name = name
self._cmds_q = Queue()
self._data_qs = [Queue()]
self._ok_q = None
self._tokens_q = None
self._save_every = save_every
self._cache_path = cache_path
self._saving_cache = cache_path is not None
if self._saving_cache:
self._save_path = os.path.join(cache_path, self.name)
if not os.path.exists(self._save_path):
os.makedirs(self._save_path)
self._start_data_segment = 0
self._cur_data_segment = 0
self._saving_ps = []
def run(self):
setproctitle('python.DataStreamRecorder.{0}'.format(self._name))
try:
logging.debug("Starting data recording on {0}".format(self.name))
self._tokens_q.put(("return", self.name))
while True:
if not self._cmds_q.empty():
cmd = self._cmds_q.get()
if cmd[0] == 'stop':
break
elif cmd[0] == 'pause':
self._recording = False
if self._saving_cache:
self._save_cache(self._cur_data_segment)
self._cur_data_segment += 1
self._data_qs.append(Queue())
elif cmd[0] == 'reset_data_segment':
self._start_data_segment = self._cur_data_segment
elif cmd[0] == 'resume':
self._recording = True
elif cmd[0] == 'save':
self._save_data(cmd[1], cmd[2], cmd[3])
elif cmd[0] == 'params':
self._args = cmd[1]
self._kwargs = cmd[2]
if self._recording and not self._ok_q.empty():
timestamp = self._ok_q.get()
self._tokens_q.put(("take", self.name))
data = self._data_sampler_method(*self._args,
**self._kwargs)
cur_data_q = self._data_qs[self._cur_data_segment]
if self._saving_cache and cur_data_q.qsize(
) == self._save_every:
self._save_cache(self._cur_data_segment)
cur_data_q = Queue()
self._data_qs.append(cur_data_q)
self._cur_data_segment += 1
cur_data_q.put((timestamp, data))
self._tokens_q.put(("return", self.name))
except KeyboardInterrupt:
logging.debug("Shutting down data streamer on {0}".format(
self.name))
sys.exit(0)
def _extract_q(self, i):
q = self._data_qs[i]
vals = []
while q.qsize() > 0:
vals.append(q.get())
self._data_qs[i] = None
del q
return vals
def _save_data(self, path, cb, concat):
if not os.path.exists(path):
os.makedirs(path)
target_filename = os.path.join(path, "{0}.jb".format(self.name))
if self._saving_cache:
while True in [p.is_alive() for p in self._saving_ps]:
sleep(1e-3)
p = Process(target=_caches_to_file,
args=(self._save_path, self._start_data_segment,
self._cur_data_segment, target_filename, cb,
concat))
p.start()
self._start_data_segment = self._cur_data_segment
else:
data = self._extract_q(0)
p = Process(target=_dump, args=(data, target_filename, cb))
p.start()
def _save_cache(self, i):
if not self._save_cache:
raise Exception(
"Cannot save cache if no cache path was specified.")
logging.debug("Saving cache for {0} block {1}".format(
self.name, self._cur_data_segment))
data = self._extract_q(i)
p = Process(target=_dump_cache,
args=(data,
os.path.join(self._save_path, "{0}.jb".format(
self._cur_data_segment)), self.name,
self._cur_data_segment))
p.start()
self._saving_ps.append(p)
def _start_recording(self, *args, **kwargs):
""" Starts recording
Parameters
----------
*args : any
Ordinary args used for calling the specified data sampler method
**kwargs : any
Keyword args used for calling the specified data sampler method
"""
while not self._cmds_q.empty():
self._cmds_q.get_nowait()
while not self._data_qs[self._cur_data_segment].empty():
self._data_qs[self._cur_data_segment].get_nowait()
self._args = args
self._kwargs = kwargs
self._recording = True
self.start()
@property
def name(self):
return self._name
def _set_qs(self, ok_q, tokens_q):
self._ok_q = ok_q
self._tokens_q = tokens_q
def _flush(self):
""" Returns a list of all current data """
if self._recording:
raise Exception("Cannot flush data queue while recording!")
if self._saving_cache:
logging.warn(
"Flush when using cache means unsaved data will be lost and not returned!"
)
self._cmds_q.put(("reset_data_segment", ))
else:
data = self._extract_q(0)
return data
def save_data(self, path, cb=_NULL, concat=True):
if self._recording:
raise Exception("Cannot save data while recording!")
self._cmds_q.put(("save", path, cb, concat))
def _stop(self):
""" Stops recording. Returns all recorded data and their timestamps. Destroys recorder process."""
self._pause()
self._cmds_q.put(("stop", ))
try:
self._recorder.terminate()
except Exception:
pass
self._recording = False
def _pause(self):
""" Pauses recording """
self._cmds_q.put(("pause", ))
self._recording = False
def _resume(self):
""" Resumes recording """
self._cmds_q.put(("resume", ))
self._recording = True
def change_data_sampler_params(self, *args, **kwargs):
""" Chanes args and kwargs for data sampler method
Parameters
----------
*args : any
Ordinary args used for calling the specified data sampler method
**kwargs : any
Keyword args used for calling the specified data sampler method
"""
self._cmds_q.put(('params', args, kwargs))
pool.map(print_fibo, num_list) # pool에 일을 던져줍니다.
print("--- %s seconds ---" % (time.time() - start_time))
from multiprocess import Process, Queue
def f(q):
q.put('hello world')
if __name__ == '__main__':
q = Queue()
p = Process(target=f, args=[q])
p.start()
print(q.get())
p.join()
import time
start_time = time.time()
def fibo(n): # 회귀적 피보나치 함수
if n == 0:
return 0
elif n == 1:
return 1
else:
return fibo(n - 1) + fibo(n - 2)
class MultiprocessorFitnessCaller:
"""
Fitness caller used for multiprocessor parallelism.
Arguments
---------
num_workers : int
Number of worker nodes to create.
"""
def __init__(self, num_workers):
self.num_workers = num_workers
self.problem = None
self.total_tasks = 0
self.total_groups = 0
self.max_group_size = 0
self.tasks = []
self.num_calls = 0
self.max_nodes = 0
self.min_num_calls = 0
self.num_workers = self.num_workers
self.processes = []
self.task_queue = Queue()
self.result_queue = Queue()
def __enter__(self):
return self
def set_problem(self, problem):
"""
Sets the problem object to use to calculate the fitness.
Arguments
---------
problem
Problem object implementing the fitness method.
"""
for _ in range(self.num_workers):
p = Process(target=multiprocessor_process,
args=(problem, self.task_queue, self.result_queue))
p.start()
self.processes.append(p)
def add(self, location, userdata):
"""
Add a location to be evaluated.
Arguments
---------
location : numpy array
Location to be evaluated.
userdata
User data to be returned with the evaluation result.
"""
self.tasks.append([location, userdata])
def evaluate(self):
"""
Evaluates all the locations.
Returns
-------
list of (location, value, userdate) tuples
Tuples containing the location, value and corresponding user data
"""
num_tasks = len(self.tasks)
self.total_tasks += num_tasks
self.total_groups += 1
if num_tasks > self.max_group_size:
self.max_group_size = num_tasks
for i in range(num_tasks):
self.task_queue.put([i, self.tasks[i][0]]) # [index, loc]
y = 0
num_results = 0
results = []
while num_results < num_tasks:
result = self.result_queue.get() # [ index, y]
index = result[0]
y = result[1]
results.append((
self.tasks[index][0],
y,
self.tasks[index][1],
))
num_results += 1
self.tasks = []
return results
def __exit__(self, exc_type, exc_val, exc_tb):
self.finish()
def finish(self):
"""
Terminates the fitness caller.
"""
for p in self.processes:
self.task_queue.put([-1, -1])
for p in self.processes:
p.join()
class TaskManager:
"""
User-facing Class for interfacing with OpenWPM
The TaskManager spawns several child processes to run the automation tasks.
- DataAggregator to aggregate data in a SQLite database
- MPLogger to aggregate logs across processes
- BrowserManager processes to isolate Browsers in a separate process
<manager_params> dict of TaskManager configuration parameters
<browser_params> is a list of (or a single) dictionaries that specify preferences for browsers to instantiate
<process_watchdog> will monitor firefox and Xvfb processes, killing any not indexed in TaskManager's browser list.
NOTE: Only run this in isolated environments. It kills processes by name, indiscriminately.
"""
def __init__(self, manager_params, browser_params, process_watchdog=False):
# Make paths absolute in manager_params
for path in ['data_directory','log_directory']:
if manager_params[path] is not None:
manager_params[path] = os.path.expanduser(manager_params[path])
manager_params['database_name'] = os.path.join(manager_params['data_directory'],manager_params['database_name'])
manager_params['log_file'] = os.path.join(manager_params['log_directory'],manager_params['log_file'])
manager_params['screenshot_path'] = os.path.join(manager_params['data_directory'], 'screenshots')
manager_params['source_dump_path'] = os.path.join(manager_params['data_directory'], 'sources')
self.manager_params = manager_params
# Create data directories if they do not exist
if not os.path.exists(manager_params['screenshot_path']):
os.makedirs(manager_params['screenshot_path'])
if not os.path.exists(manager_params['source_dump_path']):
os.makedirs(manager_params['source_dump_path'])
# check size of parameter dictionary
self.num_browsers = manager_params['num_browsers']
if len(browser_params) != self.num_browsers:
raise Exception("Number of <browser_params> dicts is not the same as manager_params['num_browsers']")
# Flow control
self.closing = False
self.failure_status = None
self.threadlock = threading.Lock()
self.failurecount = 0
if manager_params['failure_limit'] is not None:
self.failure_limit = manager_params['failure_limit']
else:
self.failure_limit = self.num_browsers * 2 + 10
self.process_watchdog = process_watchdog
# sets up the crawl data database
db_path = manager_params['database_name']
if not os.path.exists(manager_params['data_directory']):
os.mkdir(manager_params['data_directory'])
self.db = sqlite3.connect(db_path)
with open(os.path.join(os.path.dirname(__file__), 'schema.sql'), 'r') as f:
self.db.executescript(f.read())
self.db.commit()
# sets up logging server + connect a client
self.logging_status_queue = None
self.loggingserver = self._launch_loggingserver()
# socket location: (address, port)
self.manager_params['logger_address'] = self.logging_status_queue.get()
self.logger = MPLogger.loggingclient(*self.manager_params['logger_address'])
# Mark if LDBAggregator is needed (if js is enabled on any browser)
self.ldb_enabled = False
for params in browser_params:
if params['save_javascript'] or params['save_javascript_proxy']:
self.ldb_enabled = True
break
# Initialize the data aggregators
self._launch_aggregators()
# open client socket
self.sock = clientsocket(serialization='dill')
self.sock.connect(*self.manager_params['aggregator_address'])
self._save_configuration(browser_params)
# read the last used site visit id
cur = self.db.cursor()
cur.execute("SELECT MAX(visit_id) from site_visits")
last_visit_id = cur.fetchone()[0]
if last_visit_id is None:
last_visit_id = 0
self.next_visit_id = last_visit_id + 1
# sets up the BrowserManager(s) + associated queues
self.browsers = self._initialize_browsers(browser_params) # List of the Browser(s)
self._launch_browsers()
# start the manager watchdog
thread = threading.Thread(target=self._manager_watchdog, args=())
thread.daemon = True
thread.start()
def _save_configuration(self, browser_params):
""" Saves crawl configuration details to db and logfile"""
cur = self.db.cursor()
# Get git version and commit information
openwpm_v, browser_v = get_version()
# Record task details
cur.execute(("INSERT INTO task "
"(manager_params, openwpm_version, browser_version) "
"VALUES (?,?,?)"),
(json.dumps(self.manager_params), openwpm_v, browser_v))
self.db.commit()
self.task_id = cur.lastrowid
# Record browser details for each brower
for i in xrange(self.num_browsers):
cur.execute("INSERT INTO crawl (task_id, browser_params) VALUES (?,?)",
(self.task_id, json.dumps(browser_params[i])))
self.db.commit()
browser_params[i]['crawl_id'] = cur.lastrowid
# Print the configuration details
self.logger.info(get_configuration_string(self.manager_params,
browser_params,
(openwpm_v, browser_v)))
def _initialize_browsers(self, browser_params):
""" initialize the browser classes, each its unique set of parameters """
browsers = list()
for i in xrange(self.num_browsers):
browsers.append(Browser(self.manager_params, browser_params[i]))
return browsers
def _launch_browsers(self):
""" launch each browser manager process / browser """
for browser in self.browsers:
try:
success = browser.launch_browser_manager()
except:
self._cleanup_before_fail(during_init=True)
raise
if not success:
self.logger.critical("Browser spawn failure during TaskManager initialization, exiting...")
self.close()
break
# Update our DB with the random browser settings
# These are found within the scope of each instance of Browser in the browsers list
screen_res = str(browser.browser_settings['screen_res'])
ua_string = str(browser.browser_settings['ua_string'])
self.sock.send(("UPDATE crawl SET screen_res = ?, ua_string = ? \
WHERE crawl_id = ?", (screen_res, ua_string, browser.crawl_id)))
def _manager_watchdog(self):
"""
Periodically checks the following:
- memory consumption of all browsers every 10 seconds
- presence of processes that are no longer in use
"""
while not self.closing:
time.sleep(10)
# Check browser memory usage
for browser in self.browsers:
try:
process = psutil.Process(browser.browser_pid)
mem = process.memory_info()[0] / float(2 ** 20)
if mem > BROWSER_MEMORY_LIMIT:
self.logger.info("BROWSER %i: Memory usage: %iMB, exceeding limit of %iMB"
% (browser.crawl_id, int(mem), BROWSER_MEMORY_LIMIT))
browser.restart_required = True
except psutil.NoSuchProcess:
pass
# Check for browsers or displays that were not closed correctly
# Provide a 300 second buffer to avoid killing freshly launched browsers
# TODO This buffer should correspond to the maximum browser spawn timeout
if self.process_watchdog:
browser_pids = set()
display_pids = set()
check_time = time.time()
for browser in self.browsers:
if browser.browser_pid is not None:
browser_pids.add(browser.browser_pid)
if browser.display_pid is not None:
display_pids.add(browser.display_pid)
for process in psutil.process_iter():
if (process.create_time() + 300 < check_time and
((process.name() == 'firefox' and process.pid not in browser_pids) or
(process.name() == 'Xvfb' and process.pid not in display_pids))):
self.logger.debug("Process: %s (pid: %i) with start time %s found running but not in browser process list. Killing."
% (process.name(), process.pid, process.create_time()))
process.kill()
def _launch_aggregators(self):
"""
Launches the various data aggregators, which serialize data from all processes.
* DataAggregator - sqlite database for crawl data
* LevelDBAggregator - leveldb database for javascript files
"""
# DataAggregator
self.aggregator_status_queue = Queue()
self.data_aggregator = Process(target=DataAggregator.DataAggregator,
args=(self.manager_params, self.aggregator_status_queue))
self.data_aggregator.daemon = True
self.data_aggregator.start()
self.manager_params['aggregator_address'] = self.aggregator_status_queue.get() # socket location: (address, port)
# LevelDB Aggregator
if self.ldb_enabled:
self.ldb_status_queue = Queue()
self.ldb_aggregator = Process(target=LevelDBAggregator.LevelDBAggregator,
args=(self.manager_params, self.ldb_status_queue))
self.ldb_aggregator.daemon = True
self.ldb_aggregator.start()
self.manager_params['ldb_address'] = self.ldb_status_queue.get() # socket location: (address, port)
def _kill_aggregators(self):
""" Terminates the aggregators gracefully """
# DataAggregator
self.logger.debug("Telling the DataAggregator to shut down...")
self.aggregator_status_queue.put("DIE")
start_time = time.time()
self.data_aggregator.join(300)
self.logger.debug("DataAggregator took " + str(time.time() - start_time) + " seconds to close")
# LevelDB Aggregator
if self.ldb_enabled:
self.logger.debug("Telling the LevelDBAggregator to shut down...")
self.ldb_status_queue.put("DIE")
start_time = time.time()
self.ldb_aggregator.join(300)
self.logger.debug("LevelDBAggregator took " + str(time.time() - start_time) + " seconds to close")
def _launch_loggingserver(self):
""" sets up logging server """
self.logging_status_queue = Queue()
loggingserver = Process(target=MPLogger.loggingserver,
args=(self.manager_params['log_file'], self.logging_status_queue, ))
loggingserver.daemon = True
loggingserver.start()
return loggingserver
def _kill_loggingserver(self):
""" terminates logging server gracefully """
self.logging_status_queue.put("DIE")
self.loggingserver.join(300)
def _shutdown_manager(self, failure=False, during_init=False):
"""
Wait for current commands to finish, close all child processes and
threads
<failure> flag to indicate manager failure (True) or end of crawl (False)
<during_init> flag to indicator if this shutdown is occuring during the TaskManager initialization
"""
self.closing = True
for browser in self.browsers:
browser.shutdown_browser(during_init)
if failure:
self.sock.send(("UPDATE crawl SET finished = -1 WHERE crawl_id = ?",
(browser.crawl_id,)))
else:
self.sock.send(("UPDATE crawl SET finished = 1 WHERE crawl_id = ?",
(browser.crawl_id,)))
self.db.close() # close db connection
self.sock.close() # close socket to data aggregator
self._kill_aggregators()
self._kill_loggingserver()
def _cleanup_before_fail(self, during_init=False):
"""
Execute shutdown commands before throwing an exception
This should keep us from having a bunch of hanging processes
and incomplete data.
<during_init> flag to indicator if this shutdown is occuring during
the TaskManager initialization
"""
self._shutdown_manager(failure=True, during_init=during_init)
def _check_failure_status(self):
""" Check the status of command failures. Raise exceptions as necessary
The failure status property is used by the various asynchronous
command execution threads which interface with the
remote browser manager processes. If a failure status is found, the
appropriate steps are taken to gracefully close the infrastructure
"""
self.logger.debug("Checking command failure status indicator...")
if self.failure_status:
self.logger.debug("TaskManager failure status set, halting command execution.")
self._cleanup_before_fail()
if self.failure_status['ErrorType'] == 'ExceedCommandFailureLimit':
raise CommandExecutionError(
"TaskManager exceeded maximum consecutive command "
"execution failures.", self.failure_status['CommandSequence']
)
elif self.failure_status['ErrorType'] == 'ExceedLaunchFailureLimit':
raise CommandExecutionError(
"TaskManager failed to launch browser within allowable "
"failure limit.", self.failure_status['CommandSequence']
)
if self.failure_status['ErrorType'] == 'CriticalChildException':
reraise(*cPickle.loads(self.failure_status['Exception']))
# CRAWLER COMMAND CODE
def _distribute_command(self, command_sequence, index=None):
"""
parses command type and issues command(s) to the proper browser
<index> specifies the type of command this is:
= None -> first come, first serve
= # -> index of browser to send command to
= * -> sends command to all browsers
= ** -> sends command to all browsers (synchronized)
"""
if index is None:
#send to first browser available
command_executed = False
while True:
for browser in self.browsers:
if browser.ready():
browser.current_timeout = command_sequence.total_timeout
thread = self._start_thread(browser, command_sequence)
command_executed = True
break
if command_executed:
break
time.sleep(SLEEP_CONS)
elif 0 <= index < len(self.browsers):
#send the command to this specific browser
while True:
if self.browsers[index].ready():
self.browsers[index].current_timeout = command_sequence.total_timeout
thread = self._start_thread(self.browsers[index], command_sequence)
break
time.sleep(SLEEP_CONS)
elif index == '*':
#send the command to all browsers
command_executed = [False] * len(self.browsers)
while False in command_executed:
for i in xrange(len(self.browsers)):
if self.browsers[i].ready() and not command_executed[i]:
self.browsers[i].current_timeout = command_sequence.total_timeout
thread = self._start_thread(self.browsers[i], command_sequence)
command_executed[i] = True
time.sleep(SLEEP_CONS)
elif index == '**':
#send the command to all browsers and sync it
condition = threading.Condition() # Used to block threads until ready
command_executed = [False] * len(self.browsers)
while False in command_executed:
for i in xrange(len(self.browsers)):
if self.browsers[i].ready() and not command_executed[i]:
self.browsers[i].current_timeout = command_sequence.total_timeout
thread = self._start_thread(self.browsers[i], command_sequence, condition)
command_executed[i] = True
time.sleep(SLEEP_CONS)
with condition:
condition.notifyAll() # All browsers loaded, tell them to start
else:
self.logger.info("Command index type is not supported or out of range")
return
if command_sequence.blocking:
thread.join()
self._check_failure_status()
def _start_thread(self, browser, command_sequence, condition=None):
""" starts the command execution thread """
# Check status flags before starting thread
if self.closing:
self.logger.error("Attempted to execute command on a closed TaskManager")
return
self._check_failure_status()
browser.set_visit_id(self.next_visit_id)
self.sock.send(("INSERT INTO site_visits (visit_id, crawl_id, site_url) VALUES (?,?,?)",
(self.next_visit_id, browser.crawl_id, command_sequence.url)))
self.next_visit_id += 1
# Start command execution thread
args = (browser, command_sequence, condition)
thread = threading.Thread(target=self._issue_command, args=args)
browser.command_thread = thread
thread.daemon = True
thread.start()
return thread
def _issue_command(self, browser, command_sequence, condition=None):
"""
sends command tuple to the BrowserManager
"""
browser.is_fresh = False # since we are issuing a command, the BrowserManager is no longer a fresh instance
# if this is a synced call, block on condition
if condition is not None:
with condition:
condition.wait()
reset = command_sequence.reset
start_time = None # tracks when a site visit started, so that flash/profile
# cookies can be properly tracked.
for command_and_timeout in command_sequence.commands_with_timeout:
command, timeout = command_and_timeout
if command[0] in ['GET', 'BROWSE']:
start_time = time.time()
command += (browser.curr_visit_id,)
elif command[0] in ['DUMP_FLASH_COOKIES', 'DUMP_PROFILE_COOKIES']:
command += (start_time, browser.curr_visit_id,)
browser.current_timeout = timeout
# passes off command and waits for a success (or failure signal)
browser.command_queue.put(command)
command_succeeded = 0 #1 success, 0 failure from error, -1 timeout
command_arguments = command[1] if len(command) > 1 else None
# received reply from BrowserManager, either success signal or failure notice
try:
status = browser.status_queue.get(True, browser.current_timeout)
if status == "OK":
command_succeeded = 1
elif status[0] == "CRITICAL":
self.logger.critical("BROWSER %i: Received critical error "
"from browser process while executing "
"command %s. Setting failure status." % (
browser.crawl_id, str(command)))
self.failure_status = {
'ErrorType': 'CriticalChildException',
'CommandSequence': command_sequence,
'Exception': status[1]
}
return
else:
command_succeeded = 0
self.logger.info("BROWSER %i: Received failure status while"
" executing command: %s" % (browser.crawl_id, command[0]))
except EmptyQueue:
command_succeeded = -1
self.logger.info("BROWSER %i: Timeout while executing command, "
"%s, killing browser manager" % (browser.crawl_id, command[0]))
self.sock.send(("INSERT INTO CrawlHistory (crawl_id, command, arguments, bool_success)"
" VALUES (?,?,?,?)",
(browser.crawl_id, command[0], command_arguments, command_succeeded)))
if command_succeeded != 1:
with self.threadlock:
self.failurecount += 1
if self.failurecount > self.failure_limit:
self.logger.critical("BROWSER %i: Command execution failure"
" pushes failure count above the allowable limit."
" Setting failure_status." % browser.crawl_id)
self.failure_status = {
'ErrorType': 'ExceedCommandFailureLimit',
'CommandSequence': command_sequence
}
return
browser.restart_required = True
else:
with self.threadlock:
self.failurecount = 0
if browser.restart_required:
break
# Sleep after executing CommandSequence to provide extra time for
# internal buffers to drain. Stopgap in support of #135
time.sleep(2)
if self.closing:
return
if browser.restart_required or reset:
success = browser.restart_browser_manager(clear_profile = reset)
if not success:
self.logger.critical("BROWSER %i: Exceeded the maximum allowable "
"consecutive browser launch failures. "
"Setting failure_status." % browser.crawl_id)
self.failure_status = {
'ErrorType': 'ExceedLaunchFailureLimit',
'CommandSequence': command_sequence
}
return
browser.restart_required = False
def execute_command_sequence(self, command_sequence, index=None):
self._distribute_command(command_sequence, index)
# DEFINITIONS OF HIGH LEVEL COMMANDS
# NOTE: These wrappers are provided for convenience. To issue sequential
# commands to the same browser in a single 'visit', use the CommandSequence
# class directly.
def get(self, url, index=None, timeout=60, sleep=0, reset=False):
""" goes to a url """
command_sequence = CommandSequence.CommandSequence(url)
command_sequence.get(timeout=timeout, sleep=sleep)
command_sequence.reset = reset
self.execute_command_sequence(command_sequence, index=index)
def browse(self, url, num_links=2, sleep=0, index=None, timeout=60, reset=False):
""" browse a website and visit <num_links> links on the page """
command_sequence = CommandSequence.CommandSequence(url)
command_sequence.browse(num_links=num_links, sleep=sleep, timeout=timeout)
command_sequence.reset = reset
self.execute_command_sequence(command_sequence, index=index)
def close(self):
"""
Execute shutdown procedure for TaskManager
"""
if self.closing:
self.logger.error("TaskManager already closed")
return
self._shutdown_manager()
