Beispiel #1
0
    def __call__(self, *args, **kwargs):
        # Add logger
        self.logger = LogHelper(self)

        # Add DB Helper
        self.db = MySQLHelper(self.logger)

        # Add Cache Helper
        self.cache_db = RedisHelper(self.logger)

        # Run
        try:
            return super(BaseResultSavingTask, self).__call__(*args, **kwargs)

        except Exception as e:
            for line in traceback.format_exc().splitlines():
                self.logger.error(line)

            raise
Beispiel #2
0
class BaseResultSavingTask(app.Task):
    '''
    Base result saving task class
    '''
    ignore_result = True

    def __call__(self, *args, **kwargs):
        # Add logger
        self.logger = LogHelper(self)

        # Add DB Helper
        self.db = MySQLHelper(self.logger)

        # Add Cache Helper
        self.cache_db = RedisHelper(self.logger)

        # Run
        try:
            return super(BaseResultSavingTask, self).__call__(*args, **kwargs)

        except Exception as e:
            for line in traceback.format_exc().splitlines():
                self.logger.error(line)

            raise

    def on_success(self, retval, task_id, args, kwargs):
        if self.request.called_directly:
            return

        self.logger.debug('[SUCCESS]')

    def on_failure(self, exc, task_id, args, kwargs, einfo):
        if self.request.called_directly:
            return

        self.logger.error('[FAILURE]')
Beispiel #3
0
class BaseTask(app.Task):
    '''
    Base task class

    1. Send Result Saving Task automatically.
    Overwrited hooks `on_success`, `on_failure` for sending Result Saving Task
    with an ID: `<Main Task ID>-RESULT`.

    Note: Sub class should provides following 2 tasks (See `example.py`):
        - `_success_result_saving_task` Result Saving Task for SUCCESS
        - `_failure_result_saving_task` Result Saving Task for FAILURE

    2. More detailed task status
    Before `__call__()` is called. Set extra information to `self.request` and Redis.
    Moreover, when `on_success`, `on_failure` is called, status will be updated by
    setting information to Redis.

    Note: Task inherit from the `BaseTask` will always store task result and status
    to the Celery original task result key(Overwrite).
    '''
    ignore_result = True

    _success_result_saving_task = None
    _failure_result_saving_task = None

    def _set_task_status(self, status, **next_context):
        '''
        Set task result for WAT's monitor.
        '''
        # Fixed in Celery for saving/publishing task result.
        # See [https://github.com/celery/celery/blob/v4.1.0/celery/backends/base.py#L518]
        if self.request.called_directly:
            return

        self.request.update(**next_context)

        task_key_prefix = 'celery-task-meta-'
        key = task_key_prefix + self.request.id
        content = {
            'task': self.name,
            'id': self.request.id,
            'args': self.request.args,
            'kwargs': self.request.kwargs,
            'queue': self.request.delivery_info['routing_key'],
            'origin': self.request.origin,
            'status': status,
            'startTime': self.request.x_start_time,
            'endTime': self.request.x_end_time,
            'retval': self.request.x_retval,
            'einfoTEXT': self.request.x_einfo_text,
            'exceptionMessage': self.request.x_exception_message,
            'exceptionDump': self.request.x_exception_dump,
        }

        if hasattr(self.request, 'extra'):
            content['extra'] = self.request.extra

        content = toolkit.json_safe_dumps(content, indent=None)

        # self.backend.client.setex(key, CONFIG['_WORKER_RESULT_EXPIRES'], content)

        if status in (celery_status.SUCCESS, celery_status.FAILURE):
            self.backend.client.publish(key, content)

    def __call__(self, *args, **kwargs):
        # Add logger
        self.logger = LogHelper(self)

        # Add DB Helper
        self.db = MySQLHelper(self.logger)

        # Add Cache Helper
        self.cache_db = RedisHelper(self.logger)

        # Add File Storage Helper
        self.file_storage = FileSystemHelper(self.logger)

        if CONFIG['MODE'] == 'prod':
            self.db.skip_log = True
            self.cache_db.skip_log = True

        # Add extra information
        if not self.request.called_directly:
            self._set_task_status(celery_status.PENDING,
                                  x_start_time=int(time.time()),
                                  x_end_time=None,
                                  x_retval=None,
                                  x_einfo_text=None,
                                  x_exception_message=None,
                                  x_exception_dump=None)

        # Sleep delay
        if 'sleepDelay' in kwargs:
            sleep_delay = 0
            try:
                sleep_delay = float(kwargs['sleepDelay'])
                self.logger.debug(
                    '[SLEEP DELAY] {} seconds...'.format(sleep_delay))

            except Exception as e:
                for line in traceback.format_exc().splitlines():
                    self.logger.error(line)
            else:
                time.sleep(sleep_delay)

        # Run
        try:
            if LOG_CALL_ARGS:
                args_dumps = toolkit.json_safe_dumps(args, indent=None)
                if len(args_dumps) > LIMIT_ARGS_DUMP:
                    args_dumps = args_dumps[0:LIMIT_ARGS_DUMP - 3] + '...'

                kwargs_dumps = toolkit.json_safe_dumps(kwargs, indent=None)
                if len(kwargs_dumps) > LIMIT_ARGS_DUMP:
                    kwargs_dumps = kwargs_dumps[0:LIMIT_ARGS_DUMP - 3] + '...'

                self.logger.debug('[CALL] args: `{}`; kwargs: `{}`'.format(
                    args_dumps, kwargs_dumps))

            return super(BaseTask, self).__call__(*args, **kwargs)

        except (SoftTimeLimitExceeded, TimeLimitExceeded) as e:
            raise

        except Exception as e:
            for line in traceback.format_exc().splitlines():
                self.logger.error(line)

            raise

    def on_retry(self, exc, task_id, args, kwargs, einfo):
        if self.request.called_directly:
            return

        self._set_task_status(celery_status.RETRY)

        self.logger.warning('[{}]'.format(celery_status.RETRY))

    def on_success(self, retval, task_id, args, kwargs):
        if self.request.called_directly:
            return

        self._set_task_status(celery_status.SUCCESS,
                              x_end_time=int(time.time()),
                              x_retval=retval)

        if self._success_result_saving_task:
            args = (task_id, self.name, self.request.origin,
                    self.request.x_start_time, self.request.x_end_time,
                    self.request.args, self.request.kwargs, retval,
                    celery_status.SUCCESS, None)
            result_task_id = '{}-RESULT'.format(task_id)
            self._success_result_saving_task.apply_async(
                task_id=result_task_id, args=args)

        self.logger.debug('[{}]'.format(celery_status.SUCCESS))

    def on_failure(self, exc, task_id, args, kwargs, einfo):
        if self.request.called_directly:
            return

        # Exception Object can not be convert to JSON
        einfo_text = str(einfo)
        exception_message = str(einfo.exception)
        exception_dump = repr(einfo.exception)

        self._set_task_status(celery_status.FAILURE,
                              x_end_time=int(time.time()),
                              x_einfo_text=einfo_text,
                              x_exception_message=exception_message,
                              x_exception_dump=exception_dump)

        if self._failure_result_saving_task:
            args = (task_id, self.name, self.request.origin,
                    self.request.x_start_time, self.request.x_end_time,
                    self.request.args, self.request.kwargs, None,
                    celery_status.FAILURE, einfo_text)
            result_task_id = '{}-RESULT'.format(task_id)
            self._failure_result_saving_task.apply_async(
                task_id=result_task_id, args=args)

        self.logger.error('[{}]'.format(celery_status.FAILURE))
Beispiel #4
0
    def __call__(self, *args, **kwargs):
        # Add logger
        self.logger = LogHelper(self)

        # Add DB Helper
        self.db = MySQLHelper(self.logger)

        # Add Cache Helper
        self.cache_db = RedisHelper(self.logger)

        # Add File Storage Helper
        self.file_storage = FileSystemHelper(self.logger)

        if CONFIG['MODE'] == 'prod':
            self.db.skip_log = True
            self.cache_db.skip_log = True

        # Add extra information
        if not self.request.called_directly:
            self._set_task_status(celery_status.PENDING,
                                  x_start_time=int(time.time()),
                                  x_end_time=None,
                                  x_retval=None,
                                  x_einfo_text=None,
                                  x_exception_message=None,
                                  x_exception_dump=None)

        # Sleep delay
        if 'sleepDelay' in kwargs:
            sleep_delay = 0
            try:
                sleep_delay = float(kwargs['sleepDelay'])
                self.logger.debug(
                    '[SLEEP DELAY] {} seconds...'.format(sleep_delay))

            except Exception as e:
                for line in traceback.format_exc().splitlines():
                    self.logger.error(line)
            else:
                time.sleep(sleep_delay)

        # Run
        try:
            if LOG_CALL_ARGS:
                args_dumps = toolkit.json_safe_dumps(args, indent=None)
                if len(args_dumps) > LIMIT_ARGS_DUMP:
                    args_dumps = args_dumps[0:LIMIT_ARGS_DUMP - 3] + '...'

                kwargs_dumps = toolkit.json_safe_dumps(kwargs, indent=None)
                if len(kwargs_dumps) > LIMIT_ARGS_DUMP:
                    kwargs_dumps = kwargs_dumps[0:LIMIT_ARGS_DUMP - 3] + '...'

                self.logger.debug('[CALL] args: `{}`; kwargs: `{}`'.format(
                    args_dumps, kwargs_dumps))

            return super(BaseTask, self).__call__(*args, **kwargs)

        except (SoftTimeLimitExceeded, TimeLimitExceeded) as e:
            raise

        except Exception as e:
            for line in traceback.format_exc().splitlines():
                self.logger.error(line)

            raise
Beispiel #5
0
# Redis config
redis_password_part = ''
if CONFIG['REDIS_PASSWORD']:
    redis_password_part = ':{}@'.format(CONFIG['REDIS_PASSWORD'])

redis_url = 'redis://{}{}:{}/{}'.format(redis_password_part,
                                        CONFIG['REDIS_HOST'],
                                        CONFIG['REDIS_PORT'],
                                        CONFIG['REDIS_DATABASE'])

app.conf.update(broker_url=redis_url, result_backend=redis_url)

# Redis helper
from worker.utils.log_helper import LogHelper
from worker.utils.extra_helpers import RedisHelper
WORKER_LOGGER = LogHelper()
REDIS_HELPER = RedisHelper(logger=WORKER_LOGGER)
REDIS_HELPER.skip_log = True


@signals.worker_ready.connect
def on_worker_ready(*args, **kwargs):
    after_app_created(app)

    print('Celery logging disabled')
    print('Celery is running (Press CTRL+C to quit)')
    print('Have fun!')


@signals.heartbeat_sent.connect
def on_heartbeat_sent(*args, **kwargs):
Beispiel #6
0
    def __call__(self, *args, **kwargs):
        # Add Queue Info
        self.worker_queue = self.request.delivery_info['routing_key']
        self.queue = self.worker_queue.split('@').pop()

        # Add logger
        self.logger = LogHelper(self)

        # Add DB Helper
        self.db = MySQLHelper(self.logger)

        # Add Cache Helper
        self.cache_db = RedisHelper(self.logger)

        # Add File Storage Helper
        self.file_storage = FileSystemHelper(self.logger)

        if CONFIG['MODE'] == 'prod':
            self.db.skip_log = True
            self.cache_db.skip_log = True

        # Add extra information
        if not self.request.called_directly:
            self._set_task_status(celery_status.PENDING,
                                  x_start_time=int(time.time()),
                                  x_end_time=None,
                                  x_retval=None,
                                  x_einfo_text=None,
                                  x_exception_message=None,
                                  x_exception_dump=None)

        # Sleep delay
        if 'sleepDelay' in kwargs:
            sleep_delay = 0
            try:
                sleep_delay = float(kwargs['sleepDelay'])
                self.logger.debug(
                    '[SLEEP DELAY] {} seconds...'.format(sleep_delay))

            except Exception as e:
                for line in traceback.format_exc().splitlines():
                    self.logger.error(line)
            else:
                time.sleep(sleep_delay)

        # Run
        try:
            task_info = '`{}`'.format(self.name)
            func_id = kwargs.get('funcId')
            if func_id:
                task_info = '`{}`'.format(func_id) + '@' + task_info

            self.logger.debug('[CALL] {}'.format(task_info))

            return super(BaseTask, self).__call__(*args, **kwargs)

        except (SoftTimeLimitExceeded, TimeLimitExceeded) as e:
            raise

        except Exception as e:
            for line in traceback.format_exc().splitlines():
                self.logger.error(line)

            raise
Beispiel #7
0
class BaseTask(app.Task):
    '''
    Base task class

    1. Send Result Saving Task automatically.
    Overwrited hooks `on_success`, `on_failure` for sending Result Saving Task
    with an ID: `<Main Task ID>-RESULT`.

    Note: Sub class should provides following 2 tasks (See `example.py`):
        - `_success_result_saving_task` Result Saving Task for SUCCESS
        - `_failure_result_saving_task` Result Saving Task for FAILURE

    2. More detailed task status
    Before `__call__()` is called. Set extra information to `self.request` and Redis.
    Moreover, when `on_success`, `on_failure` is called, status will be updated by
    setting information to Redis.

    Note: Task inherit from the `BaseTask` will always store task result and status
    to the Celery original task result key(Overwrite).
    '''
    ignore_result = True

    _success_result_saving_task = None
    _failure_result_saving_task = None

    def _set_task_status(self, status, **next_context):
        '''
        Set task result for WAT's monitor.
        '''
        # Fixed in Celery for saving/publishing task result.
        # See [https://github.com/celery/celery/blob/v4.1.0/celery/backends/base.py#L518]
        if self.request.called_directly:
            return

        self.request.update(**next_context)

        if status not in (celery_status.SUCCESS, celery_status.FAILURE):
            return

        # Publish result by Redis
        key = 'celery-task-meta-' + self.request.id

        content = {
            'task': self.name,
            'id': self.request.id,
            'args': self.request.args,
            'kwargs': self.request.kwargs,
            'origin': self.request.
            origin,  # 注意:Celery中Task.requset.origin 和Func 业务中的Origin *不是*一回事
            'queue': self.worker_queue,
            'status': status,
            'startTime': self.request.x_start_time,
            'endTime': self.request.x_end_time,
            'retval': self.request.x_retval,
            'einfoTEXT': self.request.x_einfo_text,
            'exceptionMessage': self.request.x_exception_message,
            'exceptionDump': self.request.x_exception_dump,
        }

        if hasattr(self.request, 'extra'):
            content['extra'] = self.request.extra

        content = toolkit.json_dumps(content, indent=None)

        self.backend.client.publish(key, content)

    def __call__(self, *args, **kwargs):
        # Add Queue Info
        self.worker_queue = self.request.delivery_info['routing_key']
        self.queue = self.worker_queue.split('@').pop()

        # Add logger
        self.logger = LogHelper(self)

        # Add DB Helper
        self.db = MySQLHelper(self.logger)

        # Add Cache Helper
        self.cache_db = RedisHelper(self.logger)

        # Add File Storage Helper
        self.file_storage = FileSystemHelper(self.logger)

        if CONFIG['MODE'] == 'prod':
            self.db.skip_log = True
            self.cache_db.skip_log = True

        # Add extra information
        if not self.request.called_directly:
            self._set_task_status(celery_status.PENDING,
                                  x_start_time=int(time.time()),
                                  x_end_time=None,
                                  x_retval=None,
                                  x_einfo_text=None,
                                  x_exception_message=None,
                                  x_exception_dump=None)

        # Sleep delay
        if 'sleepDelay' in kwargs:
            sleep_delay = 0
            try:
                sleep_delay = float(kwargs['sleepDelay'])
                self.logger.debug(
                    '[SLEEP DELAY] {} seconds...'.format(sleep_delay))

            except Exception as e:
                for line in traceback.format_exc().splitlines():
                    self.logger.error(line)
            else:
                time.sleep(sleep_delay)

        # Run
        try:
            task_info = '`{}`'.format(self.name)
            func_id = kwargs.get('funcId')
            if func_id:
                task_info = '`{}`'.format(func_id) + '@' + task_info

            self.logger.debug('[CALL] {}'.format(task_info))

            return super(BaseTask, self).__call__(*args, **kwargs)

        except (SoftTimeLimitExceeded, TimeLimitExceeded) as e:
            raise

        except Exception as e:
            for line in traceback.format_exc().splitlines():
                self.logger.error(line)

            raise

    def on_retry(self, exc, task_id, args, kwargs, einfo):
        if self.request.called_directly:
            return

        self._set_task_status(celery_status.RETRY)

        self.logger.warning('[{}]'.format(celery_status.RETRY))

    def on_success(self, retval, task_id, args, kwargs):
        if self.request.called_directly:
            return

        self._set_task_status(celery_status.SUCCESS,
                              x_end_time=int(time.time()),
                              x_retval=retval)

        if self._success_result_saving_task:
            args = (task_id, self.name, self.request.origin,
                    self.request.x_start_time, self.request.x_end_time,
                    self.request.args, self.request.kwargs, retval,
                    celery_status.SUCCESS, None)
            result_task_id = '{}-RESULT'.format(task_id)
            self._success_result_saving_task.apply_async(
                task_id=result_task_id, args=args)

        self.logger.debug('[{}]'.format(celery_status.SUCCESS))

    def on_failure(self, exc, task_id, args, kwargs, einfo):
        if self.request.called_directly:
            return

        # Exception Object can not be convert to JSON
        einfo_text = str(einfo)
        exception_message = str(einfo.exception)
        exception_dump = repr(einfo.exception)

        self._set_task_status(celery_status.FAILURE,
                              x_end_time=int(time.time()),
                              x_einfo_text=einfo_text,
                              x_exception_message=exception_message,
                              x_exception_dump=exception_dump)

        if self._failure_result_saving_task:
            args = (task_id, self.name, self.request.origin,
                    self.request.x_start_time, self.request.x_end_time,
                    self.request.args, self.request.kwargs, None,
                    celery_status.FAILURE, einfo_text)
            result_task_id = '{}-RESULT'.format(task_id)
            self._failure_result_saving_task.apply_async(
                task_id=result_task_id, args=args)

        self.logger.error('[{}]'.format(celery_status.FAILURE))

    def launch_log(self):
        self.logger.info(f"`{self.name}` Task launched.")

    def lock(self, max_age=60):
        lock_key = toolkit.get_cache_key('lock', self.name)
        lock_value = toolkit.gen_uuid()
        if not self.cache_db.lock(lock_key, lock_value, max_age):
            self.logger.warning(f"`{self.name}` Task already launched.")
            return

        self.launch_log()

        return lock_key, lock_value

    def unlock(self, lock_key, lock_value):
        self.cache_db.unlock(lock_key, lock_value)