I just getting familiar with Celery and have a question. My setup is Django-Redis-Celery
Lets take an example of a task sending email:
TASKS
#task
def send_email(message):
mailserver.sendOneMessage(message)
VIEWS
class newaccount(APIView):
def post(self, request, format=None):
send_email.delay(request.data.email)
This works perfectly, Django sends messages to Redis and those are picked up by Celery then to execute task. But I want to improve the system so that Celery picks up all messages from Redis at certain intervals and executes a single task with multiple messages. This because, connecting to the email server is slow and sending multiple messages as a single request will result in a faster process.
I want something like this to work:
TASKS
#task
def send_emails(messages):
mailserver.sendMultipleMessages(messages)
Thoughts?
Since i am using redis as a cache (django-redis) already i implemented the following workflow:
Step 1. Create a task that adds new emails to cache
#shared_task()
def add_email(user_id):
cache.set("email#{}".format(user_id), None, timeout=None)
Step 2. Create a periodic task that runs every second and looks up for new emails in the cache
class ProcessEmailsTask(PeriodicTask):
run_every = timedelta(seconds=1)
def run(self, **kwargs):
call_email()
def call_email():
item_exists = True
ids = []
while item_exists:
try:
key = next(cache.iter_keys("email#*"))
ids.append(key.split("email#")[1])
cache.delete_pattern(key)
except:
item_exists = False
if len(ids) > 0:
send_emails_to(ids)
Step 3. Run both celery workers and celery beat and profit!
Related
I have a task that I want to run every minute so the data is as fresh as possible. However depending on the size of the update it can take longer than one minute to finish. Django-Q creates new task and queues it every minute though so there is some overlap synchronizing the same data pretty much. Is it possible to not schedule the task that is already in progress?
I ended up creating decorator that locks the task execution and on new task run just returns immediately if the lock is not available. Timeout is 1 hour (enough in my case).
from functools import wraps
from django.core.cache import cache
from redis.exceptions import LockNotOwnedError
def django_q_task_lock(func):
"""
Decorator for django q tasks for preventing overlap in parallel task runs
"""
#wraps(func)
def wrapped_task(*args, **kwargs):
task_lock = cache.lock(f"django_q-{func.__name__}", timeout=60 * 60)
if task_lock.acquire(blocking=False):
try:
func(*args, **kwargs)
except Exception as e:
try:
task_lock.release()
except LockNotOwnedError:
pass
raise e
try:
task_lock.release()
except LockNotOwnedError:
pass
return wrapped_task
#django_q_task_lock
def potentialy_long_running_task():
...
# task logic
...
I am trying to build a REST API that will manage some machine learning classification tasks. I have written an API view, which when hit, will trigger the start of a classification task (such as: training an SVM classifier with the data the user provided previously). However, this is a long running task, so I would ideally not have the user wait once they have made a request to this view. Instead, I would like to start this task in the background and give them a response immediately. They can later view the results of the classification in a separate view (haven't implemented that yet.)
I am using ASGI_APPLICATION = 'mlxplorebackend.asgi.application' in settings.py.
Here's my API view in views.py
import asyncio
from concurrent.futures import ProcessPoolExecutor
from django import setup as SetupDjango
# ... other imports
loop = asyncio.get_event_loop()
def DummyClassification():
result = sum(i * i for i in range(10 ** 7))
print(result)
return result
# ... other API views
class TaskExecuteView(APIView):
"""
Once an API call is made to this view, the classification algorithm will start being processed.
Depends on:
1. Parser for the classification algorithm type and parameters
2. Classification algorithm implementation
"""
def get(self, request, taskId, *args, **kwargs):
try:
task = TaskModel.objects.get(taskId = taskId)
except TaskModel.DoesNotExist:
raise Http404
else:
# this is basically the classification task for now
# need to turn this to an async view
with ProcessPoolExecutor(initializer = SetupDjango) as pool:
loop.run_in_executor(pool, DummyClassification)
return Response({ "message": "The task with id: {} has been started".format(task.taskId) }, status = status.HTTP_200_OK)
The problem I am facing is the following:
When I do not use with ProcessPoolExecutor(initializer = SetupDjango) as pool: i.e. without the initializer, I get django.core.exceptions.AppRegistryNotReady: Apps aren't loaded yet. (full traceback at: https://paste.ubuntu.com/p/ctjmFNYMXW/)
When I do use the initializer, the view no longer remains async, it gets blocked. The response returns after the task is completed, which is about 5 seconds on my machine. I do realize I am not really making use of asyncio.sleep() inside my DummyClassification() function, but I can't figure out the way to do so.
I am guessing this is not the way to do it, therefore any suggestions would be appreciated. I would like to avoid celery if I can, since that seems a tad bit too complicated for me.
Edit:
If I get rid of ProcessPoolExecutor() and simply do loop.run_in_executor(None, DummyClassification), it works as expected, but then only one worker thread is working on the task, which doesn't seem remotely ideal for a classification task.
This was a ride. I at first went through the pain of setting up celery only to find out that the original problem of the classification task using one CPU core remains. Then I switched to django-rq with redis and it is currently working as expected.
from .tasks import Pipeline
class TaskExecuteView(APIView):
"""
Once an API call is made to this view, the classification algorithm will start being processed.
Depends on:
1. Parser for the classification algorithm type
2. Classification algorithm implementation
"""
def get(self, request, taskId, *args, **kwargs):
try:
task = TaskModel.objects.get(taskId = taskId)
except TaskModel.DoesNotExist:
raise Http404
else:
Pipeline.delay(taskId) # this is async now ✔
# mark this as an in-progress task
TaskModel.objects.filter(taskId = taskId).update(inProgress = True)
return Response({ "message": "The task with id: {}, title: {} has been started".format(task.taskId, task.taskTitle) }, status = status.HTTP_200_OK)
tasks.py
from django_rq import job
#job('default', timeout=3600)
def Pipeline(taskId):
# classification task
Our workflow is currently built around an old version of celery, so bear in mind things are already not optimal. We need to run a task and save a record of that task run in the database. If that task fails or hangs (it happens often), we want to re run, exactly as it was run the first time. This shouldn't happen automatically though. It needs to be triggered manually depending on the nature of the failure and the result needs to be logged in the DB to make that decision (via a front end).
How can we save a complete record of a task in the DB so that a subsequent process can grab the record and run a new identical task? The current implementation saves the path of the #task decorated function in the DB as part of a TaskInfo model. When the task needs to be rerun, we have a get_task() method on the TaskInfo model that gets the path from the DB, imports it using getattr, and another rerun() method that runs the task again with *args, **kwargs (also saved in the DB).
Like so (these are methods on the TaskInfo model instance):
def get_task(self):
"""Returns the task's decorated function, which can be delayed."""
module_name, object_name = self.path.rsplit('.', 1)
module = import_module(module_name)
task = getattr(module, object_name)
if inspect.isclass(task):
task = task()
# task = current_app.tasks[self.path]
return task
def rerun(self):
"""Re-run the task, and replace this one.
- A new task is scheduled to run.
- The new task's TaskInfo has the same parent as this TaskInfo.
- This TaskInfo is deleted.
"""
args, kwargs = self.get_arguments()
celery_task = self.get_task()
celery_task.delay(*args, **kwargs)
defaults = {
'path': self.path,
'status': Status.PENDING,
'timestamp': timezone.now(),
'args': args,
'kwargs': kwargs,
'parent': self.parent,
}
TaskInfo.objects.update_or_create(task_id=celery_task.id, defaults=defaults)
self.delete()
There must be a cleaner solution for saving a task in the DB to rerun later, right?
The latest version of Celery (4.4.0) included a param extended_result. You can set it to True, then the table (it is named celery_taskmeta by default) in the Result Backend Database will store the args and kwargs of the task.
Here is a demo:
app = Celery('test_result_backend')
app.conf.update(
broker_url='redis://localhost:6379/10',
result_backend='db+mysql://root:passwd#localhost/celery_toys',
result_extended=True
)
#app.task(bind=True, name='add')
def add(self, x, y):
self.request.task_name = 'add' # For saving the task name.
time.sleep(5)
return x + y
With the task info recorded in MySQL, you are able to re-run your task easily.
The idea is to run a background task on the worker.connect worker. While executing the task, I would like to send its progress to a connected client via the notifications Group.
The problem: the messages sent to the notifications Group are delayed until the task on the worker is finished. So: both messages 'Start' and 'Stop' appear simultaneously on the client, after a delay of 5 seconds (sleep(5)). I would expect the message 'Start', followed by a 5sec delay, followed by the message 'Stop'. Any idea why this is not the case?
I have the following three processes running:
daphne tests.asgi:channel_layer
python manage.py runworker --exclude-channel=worker.connect
python manage.py runworker --only-channel=worker.connect
In views.py:
def run(request, pk):
Channel('worker.connect').send({'pk': pk})
return HttpResponse(status=200)
In consumers.py:
def ws_connect(message):
Group('notifications').add(message.reply_channel)
message.reply_channel.send({"accept": True})
def worker_connect(message):
run_channel(message)
In views.py:
def run_channel(message):
Group('notifications').send({'text': 'Start'})
sleep(5)
Group('notifications').send({'text': 'Stop'})
routing.py
channel_routing = {
'websocket.connect': consumers.ws_connect,
'worker.connect': consumers.worker_connect,
}
You can add immediately=True as argument to the send function. According to the source:
Sends are delayed until consumer completion. To override this, you may pass immediately=True.
https://github.com/django/channels/blob/master/channels/channel.py#L32
We have a Django app running Gunicorn with sync workers that's deployed on Heroku. Our request response time shows several requests that hit 30s (and die), which is the default Gunicorn timeout.
What is the best way to log these requests and analyze the timeout? Gunicorn doesn't seem to provide a hook for catching these timeouts, at least not something that's obvious.
One rather rough way to do it is have a "watchdog" timer that interrupts the process after, say, 25 seconds. Once you have an idea of which procs are slow, you can refine the data to figure out what's going on.
Example:
import signal
def timeout(_signum, _frame):
print 'TIMEOUT'
signal.signal(signal.SIGALRM, timeout)
signal.alarm(1) # send SIGALRM in 1 second
print 'waiting'
signal.pause()
print 'done'
Another approach is to fire off a Thread which pokes the main code after a certain amount of elapsed time. It has several caveats -- be sure to read the ActiveState link.
Here's one implementation by Aaron Swartz from ActiveState.com
import threading
class TimeoutError(Exception): pass
def timelimit(timeout):
def internal(function):
def internal2(*args, **kw):
class Calculator(threading.Thread):
def __init__(self):
threading.Thread.__init__(self)
self.result = None
self.error = None
def run(self):
try:
self.result = function(*args, **kw)
except:
self.error = sys.exc_info()[0]
c = Calculator()
c.start()
c.join(timeout)
if c.isAlive():
raise TimeoutError
if c.error:
raise c.error
return c.result
return internal2
return internal
https://github.com/benoitc/gunicorn/pull/768/files added a worker_abort signal which is what I'm using in this case.