def _send_status_rabbitmq(self):
"""
Send the status event to RabbitMQ
"""
dmpd_response_status = json.dumps(self.response)
drs = sizeof_fmt(len(dmpd_response_status))
executor_id = self.response['executor_id']
job_id = self.response['job_id']
rabbit_amqp_url = self.config['rabbitmq'].get('amqp_url')
status_sent = False
output_query_count = 0
params = pika.URLParameters(rabbit_amqp_url)
exchange = 'cloudbutton-{}-{}'.format(executor_id, job_id)
while not status_sent and output_query_count < 5:
output_query_count = output_query_count + 1
try:
connection = pika.BlockingConnection(params)
channel = connection.channel()
channel.exchange_declare(exchange=exchange, exchange_type='fanout', auto_delete=True)
channel.basic_publish(exchange=exchange, routing_key='',
body=dmpd_response_status)
connection.close()
logger.info("Execution status sent to rabbitmq - Size: {}".format(drs))
status_sent = True
except Exception as e:
logger.error("Unable to send status to rabbitmq")
logger.error(str(e))
logger.info('Retrying to send status to rabbitmq...')
time.sleep(0.2)
def put_object(self, bucket_name, key, data):
"""
Put an object in COS. Override the object if the key already exists.
:param key: key of the object.
:param data: data of the object
:type data: str/bytes
:return: None
"""
retries = 0
status = None
while status is None:
try:
res = self.cos_client.put_object(Bucket=bucket_name,
Key=key,
Body=data)
status = 'OK' if res['ResponseMetadata'][
'HTTPStatusCode'] == 200 else 'Error'
try:
logger.debug('PUT Object {} - Size: {} - {}'.format(
key, sizeof_fmt(len(data)), status))
except Exception:
logger.debug('PUT Object {} {}'.format(key, status))
except ibm_botocore.exceptions.ClientError as e:
if e.response['Error']['Code'] == "NoSuchKey":
raise StorageNoSuchKeyError(bucket_name, key)
else:
raise e
except ibm_botocore.exceptions.ReadTimeoutError as e:
if retries == OBJ_REQ_RETRIES:
raise e
logger.debug('PUT Object timeout. Retrying request')
retries += 1
return True
def _send_status_os(self):
"""
Send the status event to the Object Storage
"""
executor_id = self.response['executor_id']
job_id = self.response['job_id']
call_id = self.response['call_id']
act_id = self.response['activation_id']
if self.response['type'] == '__init__':
init_key = create_init_key(JOBS_PREFIX, executor_id, job_id, call_id, act_id)
self.internal_storage.put_data(init_key, '')
elif self.response['type'] == '__end__':
status_key = create_status_key(JOBS_PREFIX, executor_id, job_id, call_id)
dmpd_response_status = json.dumps(self.response)
drs = sizeof_fmt(len(dmpd_response_status))
logger.info("Storing execution stats - Size: {}".format(drs))
self.internal_storage.put_data(status_key, dmpd_response_status)
def get_memory_usage(formatted=True):
"""
Gets the current memory usage of the runtime.
To be used only in the action code.
"""
if not is_unix_system():
return
split_args = False
pids_to_show = None
discriminate_by_pid = False
ps_mem.verify_environment(pids_to_show)
sorted_cmds, shareds, count, total, swaps, total_swap = \
ps_mem.get_memory_usage(pids_to_show, split_args, discriminate_by_pid,
include_self=True, only_self=False)
if formatted:
return sizeof_fmt(int(ps_mem.human(total, units=1)))
else:
return int(ps_mem.human(total, units=1))
def put_object(self, container_name, key, data):
"""
Put an object in Swift. Override the object if the key already exists.
:param key: key of the object.
:param data: data of the object
:type data: str/bytes
:return: None
"""
url = '/'.join([self.endpoint, container_name, key])
try:
res = self.session.put(url, data=data)
status = 'OK' if res.status_code == 201 else 'Error'
try:
logger.debug('PUT Object {} - Size: {} - {}'.format(
key, sizeof_fmt(len(data)), status))
except Exception:
logger.debug('PUT Object {} - {}'.format(key, status))
except Exception as e:
print(e)
def _create_job(config,
internal_storage,
executor_id,
job_id,
func,
data,
runtime_meta,
runtime_memory=None,
extra_env=None,
invoke_pool_threads=128,
include_modules=[],
exclude_modules=[],
execution_timeout=None,
job_created_tstamp=None):
"""
:param func: the function to map over the data
:param iterdata: An iterable of input data
:param extra_env: Additional environment variables for CF environment. Default None.
:param extra_meta: Additional metadata to pass to CF. Default None.
:param remote_invocation: Enable remote invocation. Default False.
:param invoke_pool_threads: Number of threads to use to invoke.
:param data_all_as_one: upload the data as a single object. Default True
:param overwrite_invoke_args: Overwrite other args. Mainly used for testing.
:param exclude_modules: Explicitly keep these modules from pickled dependencies.
:return: A list with size `len(iterdata)` of futures for each job
:rtype: list of futures.
"""
log_level = os.getenv('CLOUDBUTTON_LOGLEVEL')
runtime_name = config['cloudbutton']['runtime']
if runtime_memory is None:
runtime_memory = config['cloudbutton']['runtime_memory']
ext_env = {} if extra_env is None else extra_env.copy()
if ext_env:
ext_env = utils.convert_bools_to_string(ext_env)
logger.debug("Extra environment vars {}".format(ext_env))
if not data:
return []
if execution_timeout is None:
execution_timeout = config['cloudbutton']['runtime_timeout'] - 5
job_description = {}
job_description['runtime_name'] = runtime_name
job_description['runtime_memory'] = runtime_memory
job_description['execution_timeout'] = execution_timeout
job_description['function_name'] = func.__name__
job_description['extra_env'] = ext_env
job_description['total_calls'] = len(data)
job_description['invoke_pool_threads'] = invoke_pool_threads
job_description['executor_id'] = executor_id
job_description['job_id'] = job_id
exclude_modules_cfg = config['cloudbutton'].get('exclude_modules', [])
include_modules_cfg = config['cloudbutton'].get('include_modules', [])
exc_modules = set()
inc_modules = set()
if exclude_modules_cfg:
exc_modules.update(exclude_modules_cfg)
if exclude_modules:
exc_modules.update(exclude_modules)
if include_modules_cfg is not None:
inc_modules.update(include_modules_cfg)
if include_modules_cfg is None and not include_modules:
inc_modules = None
if include_modules is not None and include_modules:
inc_modules.update(include_modules)
if include_modules is None:
inc_modules = None
host_job_meta = {'job_created_tstamp': job_created_tstamp}
logger.debug(
'ExecutorID {} | JobID {} - Serializing function and data'.format(
executor_id, job_id))
serializer = SerializeIndependent(runtime_meta['preinstalls'])
func_and_data_ser, mod_paths = serializer([func] + data, inc_modules,
exc_modules)
data_strs = func_and_data_ser[1:]
data_size_bytes = sum(len(x) for x in data_strs)
module_data = create_module_data(mod_paths)
func_str = func_and_data_ser[0]
func_module_str = pickle.dumps(
{
'func': func_str,
'module_data': module_data
}, -1)
func_module_size_bytes = len(func_module_str)
total_size = utils.sizeof_fmt(data_size_bytes + func_module_size_bytes)
host_job_meta['data_size_bytes'] = data_size_bytes
host_job_meta['func_module_size_bytes'] = func_module_size_bytes
if 'data_limit' in config['cloudbutton']:
data_limit = config['cloudbutton']['data_limit']
else:
data_limit = MAX_AGG_DATA_SIZE
if data_limit and data_size_bytes > data_limit * 1024**2:
log_msg = (
'ExecutorID {} | JobID {} - Total data exceeded maximum size '
'of {}'.format(executor_id, job_id,
utils.sizeof_fmt(data_limit * 1024**2)))
raise Exception(log_msg)
log_msg = ('ExecutorID {} | JobID {} - Uploading function and data '
'- Total: {}'.format(executor_id, job_id, total_size))
print(log_msg) if not log_level else logger.info(log_msg)
# Upload data
data_key = create_agg_data_key(JOBS_PREFIX, executor_id, job_id)
job_description['data_key'] = data_key
data_bytes, data_ranges = utils.agg_data(data_strs)
job_description['data_ranges'] = data_ranges
data_upload_start = time.time()
internal_storage.put_data(data_key, data_bytes)
data_upload_end = time.time()
host_job_meta['data_upload_time'] = round(
data_upload_end - data_upload_start, 6)
# Upload function and modules
func_upload_start = time.time()
func_key = create_func_key(JOBS_PREFIX, executor_id, job_id)
job_description['func_key'] = func_key
internal_storage.put_func(func_key, func_module_str)
func_upload_end = time.time()
host_job_meta['func_upload_time'] = round(
func_upload_end - func_upload_start, 6)
job_description['metadata'] = host_job_meta
return job_description
def run(self):
"""
Runs the function
"""
# self.stats.write('jobrunner_start', time.time())
logger.info("Started")
result = None
exception = False
try:
loaded_func_all = self._get_function_and_modules()
self._save_modules(loaded_func_all['module_data'])
function = self._unpickle_function(loaded_func_all['func'])
data = self._load_data()
if strtobool(os.environ.get('__PW_REDUCE_JOB', 'False')):
self._wait_futures(data)
elif is_object_processing_function(function):
self._load_object(data)
self._fill_optional_args(function, data)
logger.info("Going to execute '{}()'".format(str(
function.__name__)))
print('---------------------- FUNCTION LOG ----------------------',
flush=True)
function_start_tstamp = time.time()
result = function(**data)
function_end_tstamp = time.time()
print('----------------------------------------------------------',
flush=True)
logger.info("Success function execution")
self.stats.write('function_start_tstamp', function_start_tstamp)
self.stats.write('function_end_tstamp', function_end_tstamp)
self.stats.write(
'function_exec_time',
round(function_end_tstamp - function_start_tstamp, 8))
# Check for new futures
if result is not None:
self.stats.write("result", True)
if isinstance(result, ResponseFuture) or \
(type(result) == list and len(result) > 0 and isinstance(result[0], ResponseFuture)):
self.stats.write('new_futures', True)
logger.debug("Pickling result")
output_dict = {'result': result}
pickled_output = pickle.dumps(output_dict)
else:
logger.debug("No result to store")
self.stats.write("result", False)
except Exception:
exception = True
self.stats.write("exception", True)
exc_type, exc_value, exc_traceback = sys.exc_info()
print('----------------------- EXCEPTION !-----------------------',
flush=True)
traceback.print_exc(file=sys.stdout)
print('----------------------------------------------------------',
flush=True)
try:
logger.debug("Pickling exception")
pickled_exc = pickle.dumps(
(exc_type, exc_value, exc_traceback))
pickle.loads(
pickled_exc
) # this is just to make sure they can be unpickled
self.stats.write("exc_info", str(pickled_exc))
except Exception as pickle_exception:
# Shockingly often, modules like subprocess don't properly
# call the base Exception.__init__, which results in them
# being unpickleable. As a result, we actually wrap this in a try/catch block
# and more-carefully handle the exceptions if any part of this save / test-reload
# fails
self.stats.write("exc_pickle_fail", True)
pickled_exc = pickle.dumps({
'exc_type': str(exc_type),
'exc_value': str(exc_value),
'exc_traceback': exc_traceback,
'pickle_exception': pickle_exception
})
pickle.loads(pickled_exc
) # this is just to make sure it can be unpickled
self.stats.write("exc_info", str(pickled_exc))
finally:
store_result = strtobool(os.environ.get('STORE_RESULT', 'True'))
if result is not None and store_result and not exception:
output_upload_start_tstamp = time.time()
logger.info("Storing function result - Size: {}".format(
sizeof_fmt(len(pickled_output))))
self.internal_storage.put_data(self.output_key, pickled_output)
output_upload_end_tstamp = time.time()
self.stats.write(
"output_upload_time",
round(
output_upload_end_tstamp - output_upload_start_tstamp,
8))
self.jobrunner_conn.send("Finished")
logger.info("Finished")
def function_handler(event):
start_tstamp = time.time()
log_level = event['log_level']
cloud_logging_config(log_level)
logger.debug("Action handler started")
extra_env = event.get('extra_env', {})
os.environ.update(extra_env)
os.environ.update({'CLOUDBUTTON_FUNCTION': 'True',
'PYTHONUNBUFFERED': 'True'})
config = event['config']
call_id = event['call_id']
job_id = event['job_id']
executor_id = event['executor_id']
exec_id = "{}/{}/{}".format(executor_id, job_id, call_id)
logger.info("Execution-ID: {}".format(exec_id))
runtime_name = event['runtime_name']
runtime_memory = event['runtime_memory']
execution_timeout = event['execution_timeout']
logger.debug("Runtime name: {}".format(runtime_name))
logger.debug("Runtime memory: {}MB".format(runtime_memory))
logger.debug("Function timeout: {}s".format(execution_timeout))
func_key = event['func_key']
data_key = event['data_key']
data_byte_range = event['data_byte_range']
storage_config = extract_storage_config(config)
internal_storage = InternalStorage(storage_config)
call_status = CallStatus(config, internal_storage)
call_status.response['host_submit_tstamp'] = event['host_submit_tstamp']
call_status.response['start_tstamp'] = start_tstamp
context_dict = {
'cloudbutton_version': os.environ.get("CLOUDBUTTON_VERSION"),
'call_id': call_id,
'job_id': job_id,
'executor_id': executor_id,
'activation_id': os.environ.get('__PW_ACTIVATION_ID')
}
call_status.response.update(context_dict)
show_memory_peak = strtobool(os.environ.get('SHOW_MEMORY_PEAK', 'False'))
try:
if version.__version__ != event['cloudbutton_version']:
msg = ("Cloudbutton version mismatch. Host version: {} - Runtime version: {}"
.format(event['cloudbutton_version'], version.__version__))
raise RuntimeError('HANDLER', msg)
# send init status event
call_status.send('__init__')
# call_status.response['free_disk_bytes'] = free_disk_space("/tmp")
custom_env = {'CLOUDBUTTON_CONFIG': json.dumps(config),
'CLOUDBUTTON_EXECUTION_ID': exec_id,
'PYTHONPATH': "{}:{}".format(os.getcwd(), LIBS_PATH)}
os.environ.update(custom_env)
jobrunner_stats_dir = os.path.join(STORAGE_FOLDER,
storage_config['bucket'],
JOBS_PREFIX, executor_id,
job_id, call_id)
os.makedirs(jobrunner_stats_dir, exist_ok=True)
jobrunner_stats_filename = os.path.join(jobrunner_stats_dir, 'jobrunner.stats.txt')
jobrunner_config = {'cloudbutton_config': config,
'call_id': call_id,
'job_id': job_id,
'executor_id': executor_id,
'func_key': func_key,
'data_key': data_key,
'log_level': log_level,
'data_byte_range': data_byte_range,
'output_key': create_output_key(JOBS_PREFIX, executor_id, job_id, call_id),
'stats_filename': jobrunner_stats_filename}
if show_memory_peak:
mm_handler_conn, mm_conn = Pipe()
memory_monitor = Thread(target=memory_monitor_worker, args=(mm_conn, ))
memory_monitor.start()
handler_conn, jobrunner_conn = Pipe()
jobrunner = JobRunner(jobrunner_config, jobrunner_conn, internal_storage)
logger.debug('Starting JobRunner process')
local_execution = strtobool(os.environ.get('__PW_LOCAL_EXECUTION', 'False'))
jrp = Thread(target=jobrunner.run) if local_execution else Process(target=jobrunner.run)
jrp.start()
jrp.join(execution_timeout)
logger.debug('JobRunner process finished')
if jrp.is_alive():
# If process is still alive after jr.join(job_max_runtime), kill it
try:
jrp.terminate()
except Exception:
# thread does not have terminate method
pass
msg = ('Function exceeded maximum time of {} seconds and was '
'killed'.format(execution_timeout))
raise TimeoutError('HANDLER', msg)
if show_memory_peak:
mm_handler_conn.send('STOP')
memory_monitor.join()
peak_memory_usage = int(mm_handler_conn.recv())
logger.info("Peak memory usage: {}".format(sizeof_fmt(peak_memory_usage)))
call_status.response['peak_memory_usage'] = peak_memory_usage
if not handler_conn.poll():
logger.error('No completion message received from JobRunner process')
logger.debug('Assuming memory overflow...')
# Only 1 message is returned by jobrunner when it finishes.
# If no message, this means that the jobrunner process was killed.
# 99% of times the jobrunner is killed due an OOM, so we assume here an OOM.
msg = 'Function exceeded maximum memory and was killed'
raise MemoryError('HANDLER', msg)
if os.path.exists(jobrunner_stats_filename):
with open(jobrunner_stats_filename, 'r') as fid:
for l in fid.readlines():
key, value = l.strip().split(" ", 1)
try:
call_status.response[key] = float(value)
except Exception:
call_status.response[key] = value
if key in ['exception', 'exc_pickle_fail', 'result', 'new_futures']:
call_status.response[key] = eval(value)
except Exception:
# internal runtime exceptions
print('----------------------- EXCEPTION !-----------------------', flush=True)
traceback.print_exc(file=sys.stdout)
print('----------------------------------------------------------', flush=True)
call_status.response['exception'] = True
pickled_exc = pickle.dumps(sys.exc_info())
pickle.loads(pickled_exc) # this is just to make sure they can be unpickled
call_status.response['exc_info'] = str(pickled_exc)
finally:
call_status.response['end_tstamp'] = time.time()
call_status.send('__end__')
for key in extra_env:
os.environ.pop(key)
logger.info("Finished")