I am building a tool that fetches data from a different database, transforms it, and stores it in my own database. I'm migrating from APScheduler to Celery, but I ran into the following problem:
I use a class I call JobRecords to store when a job ran, whether it was successful and which errors it encountered. I use this to know not too look too far back for updated entries, especially since some tables have multiple millions of rows.
Since the system is the same for all jobs, I created a subclass from the celery Task object. I make sure the job is executed within the Flask app context, and I fetch the latest time this Job finished successfully. I also make sure I register a value for now to avoid timing issues between querying the database and adding the job record.
class RecordedTask(Task):
"""
Task sublass that uses JobRecords to get the last run date
and add new JobRecords on completion
"""
now: datetime = None
ignore_result = True
_session: scoped_session = None
success: bool = True
info: dict = None
#property
def session(self) -> Session:
"""Making sure we have one global session instance"""
if self._session is None:
from app.extensions import db
self._session = db.session
return self._session
def __call__(self, *args, **kwargs):
from app.models import JobRecord
kwargs['last_run'] = (
self.session.query(func.max(JobRecord.run_at_))
.filter(JobRecord.job_id == self.name, JobRecord.success)
.first()
)[0] or datetime.min
self.now = kwargs['now'] = datetime.utcnow()
with app.app_context():
super(RecordedTask, self).__call__(*args, **kwargs)
def on_failure(self, exc, task_id, args: list, kwargs: dict, einfo):
self.session.rollback()
self.success = False
self.info = dict(
args=args,
kwargs=kwargs,
error=exc.args,
exc=format_exception(exc.__class__, exc, exc.__traceback__),
)
app.logger.error(f"Error executing job '{self.name}': {exc}")
def on_success(self, retval, task_id, args: list, kwargs: dict):
app.logger.info(f"Executed job '{self.name}' successfully, adding JobRecord")
for entry in self.to_trigger:
if len(entry) == 2:
job, kwargs = entry
else:
job, = entry
kwargs = {}
app.logger.info(f"Scheduling job '{job}'")
current_celery_app.signature(job, **kwargs).delay()
def after_return(self, *args, **kwargs):
from app.models import JobRecord
record = JobRecord(
job_id=self.name,
run_at_=self.now,
info=self.info,
success=self.success
)
self.session.add(record)
self.session.commit()
self.session.remove()
I added an example of a job to update a model called Location, but there are a lot of jobs just like this one.
#celery.task(bind=True, name="update_locations")
def update_locations(self, last_run: datetime = datetime.min, **_):
"""Get the locations from the external database and check for updates"""
locations: List[ExternalLocation] = ExternalLocation.query.filter(
ExternalLocation.updated_at_ >= last_run
).order_by(ExternalLocation.id).all()
app.logger.info(f"ExternalLocation: collected {len(locations)} updated locations")
for update_location in locations:
existing_location: Location = Location.query.filter(
Location.external_id == update_location.id
).first()
if existing_location is None:
self.session.add(Location.from_worker(update_location))
else:
existing_location.update_from_worker(update_location)
The problem is that when I run this job, the Location objects are not committed with the JobRecord, so only the latter is created. If I track it with the debugger, Location.query.count() returns the correct value inside the function, but as soon as it enters the on_success callback, it's back to 0, and self._session.new returns an empty dict.
I already tried adding the session as a property to make sure it's the same instance everywhere, but the problem still persists. Maybe it has something to do with it being a scoped_session because of Flask-SQLAlchemy?
Sorry about the large amount of code, I did try to strip as much away as possible. Any help is welcome!
I found out that the culprit was the combination of scoped_session and the Flask app context. Like any contextmanager, running the code with app.app_context() triggered the __exit__ function on leaving, which in turn caused the ScopedRegistry, where the scoped_session was stored, to be cleared. Then, a new session was created, the JobRecords were added to that, and that session was committed. Therefore, the locations would not be written to the database.
There are two possible solutions. If you don't use sessions in other files than in your task, you can add a session property to the task. This way, you avoid the scoped_session alltogether, and can clean up in your after_return function.
#property
def session(self):
if self._session is None:
from dashboard.extensions import db
self._session = db.create_session(options={})()
return self._session
However, I was accessing the session in my model definition files as well, through from extensions import db. Therefore, I was using two different sessions. I ended up using app.app_context().push() instead of the contextmanager, thus avoiding the __exit__ function
app.app_context().push()
super(RecordedTask, self).__call__(*args, **kwargs)
Related
I would like to use Django signals to trigger a celery task like so:
def delete_content(sender, instance, **kwargs):
task_id = uuid()
task = delete_libera_contents.apply_async(kwargs={"instance": instance}, task_id=task_id)
task.wait(timeout=300, interval=2)
But I'm always running into kombu.exceptions.EncodeError: Object of type MusicTracks is not JSON serializable
Now I'm not sure how to tread MusicTracks instance as it's a model class instance. How can I properly pass such instances to my task?
At my tasks.py I have the following:
#app.task(name="Delete Libera Contents", queue='high_priority_tasks')
def delete_libera_contents(instance, **kwargs):
libera_backend = instance.file.libera_backend
...
Never send instance in celery task, you only should send variables for example instanse primary key and then inside of the celery task via this pk find this instance and then do your logic
your code should be like this:
views.py
def delete_content(sender, **kwargs):
task_id = uuid()
task = delete_libera_contents.apply_async(kwargs={"instance_pk": sender.pk}, task_id=task_id)
task.wait(timeout=300, interval=2)
task.py
#app.task(name="Delete Libera Contents", queue='high_priority_tasks')
def delete_libera_contents(instance_pk, **kwargs):
instance = Instance.ojbects.get(pk = instance_pk)
libera_backend = instance.file.libera_backend
...
you can find this rule in celery documentation (can't find link), one of
reasons imagine situation:
you send your instance to celery tasks (it is delayed for any reason for 5 min)
then your project makes logic with this instance, before your task finished
then celery's task time come and it uses this instance old version, and this instance become corrupted
(this is the reason as I think it is, not from the documentation)
First off, sorry for making the question a bit confusing, especially for the people that have already written an answer.
In my case, the delete_content signal can be trigger from three different models, so it actually looks like this:
#receiver(pre_delete, sender=MusicTracks)
#receiver(pre_delete, sender=Movies)
#receiver(pre_delete, sender=TvShowEpisodes)
def delete_content(sender, instance, **kwargs):
delete_libera_contents.delay(instance_pk=instance.pk)
So every time one of these models triggers a delete action, this signal will also trigger a celery task to actually delete the stuff in the background (all stored on S3).
As I cannot and should not pass instances around directly as pointed out by #oruchkin, I pass the instance.pk to the celery task which I then have to find in the celery task as I don't know in the celery task what model has triggered the delete action:
#app.task(name="Delete Libera Contents", queue='high_priority_tasks')
def delete_libera_contents(instance_pk, **kwargs):
if Movies.objects.filter(pk=instance_pk).exists():
instance = Movies.objects.get(pk=instance_pk)
elif MusicTracks.objects.filter(pk=instance_pk).exists():
instance = MusicTracks.objects.get(pk=instance_pk)
elif TvShowEpisodes.objects.filter(pk=instance_pk).exists():
instance = TvShowEpisodes.objects.get(pk=instance_pk)
else:
raise logger.exception("Task: 'Delete Libera Contents', reports: No instance found (code: JFN4LK) - Warning")
libera_backend = instance.file.libera_backend
You might ask why do you not simply pass the sender from the signal to the celery task. I also tried this and again, as already pointed out, I cannot pass instances and I fail with:
kombu.exceptions.EncodeError: Object of type ModelBase is not JSON serializable
So it really seems I have to hard obtain the instance using the if-elif-else clauses at the celery task.
I'm trying to cache a large resource file among tasks using Celery 4.0.2.
Looking it in the documentation , I have reach with the task caching part.
http://docs.celeryproject.org/en/latest/userguide/tasks.html#instantiation
This can also be useful to cache resources, For example, a base Task class that caches a database connection:
from celery import Task
class DatabaseTask(Task):
_db = None
#property
def db(self):
if self._db is None:
self._db = Database.connect()
return self._db
In my case I have done some changes to cache my big file resource, and the object its shared among the tasks, but the memory used by big file resource are cached in the task forever.
from celery import Task
class BigResourceTask(Task):
_resource = None
#property
def resource(self):
if self._resource is None:
self._resource = load_big_resource()
return self._resource
How can I free that memory or make it expire after the execution of all the related tasks?
Since you are creating _resource on demand after checking if it exists, you can simply delete whenever you want it.
# complete all the tasks
del BigResourceTask._resource # free memory
# do something else
r = BigResourceTask.resource # create when needed
I'm trying to create some celery tasks as classes, but am having some difficulty. The classes are:
class BaseCeleryTask(app.Task):
def is_complete(self):
""" default method for checking if celery task has completed. """
# simply return result (since by default tasks return boolean indicating completion)
try:
return self.result
except AttributeError:
logger.error('Result not defined. Make sure task has run!')
return False
class MacroReportTask(BaseCeleryTask):
def run(self, params):
""" Override the default run method with signal factory run"""
# hold on to the factory
process = MacroCountryReport(params)
self.result = process.run()
return self.result
but when I initialize the app, and check app.tasks (or run worker), app doesn't seem to have these above tasks in its registry. Other function based tasks (using app.task() decorator) seem to be registered fine.
I run the above task as:
process = SignalFactoryTask()
process.delay(params)
Celery worker errors with the following message:
Received unregistered task of type None.
I think the issue I'm having is: how do I add custom classes to the task registry as I do with regular function based tasks?
Ran into the exact same issue, took hours to find the solution cause I'm 90% sure it's a bug. In your class tasks, try the following
class BaseCeleryTask(app.Task):
def __init__(self):
self.name = "[modulename].BaseCeleryTask"
class MacroReportTask(app.Task):
def __init__(self):
self.name = "[modulename].MacroReportTask"
It seems registering it with the app still has a bug where the name isn't automatically configured. Let me know if that works.
I want to run a function when instances of the Post model are committed. I want to run it any time they are committed, so I'd rather not explicitly call the function everywhere. How can I do this?
def notify_subscribers(post):
""" send email to subscribers """
...
post = Post("Hello World", "This is my first blog entry.")
session.commit() # How to run notify_subscribers with post as argument
# as soon as post is committed successfully?
post.title = "Hello World!!1"
session.commit() # Run notify_subscribers once again.
No matter which option you chose below, SQLAlchemy comes with a big warning about the after_commit event (which is when both ways send the signal).
The Session is not in an active transaction when the after_commit() event is invoked, and therefore can not emit SQL.
If your callback needs to query or commit to the database, it may have unexpected issues. In this case, you could use a task queue such as Celery to execute this in a background thread (with a separate session). This is probably the right way to go anyway, since sending emails takes a long time and you don't want your view to wait to return while it's happening.
Flask-SQLAlchemy provides a signal you can listen to that sends all the insert/update/delete ops. It needs to be enabled by setting app.config["SQLALCHEMY_TRACK_MODIFICATIONS"] = True because tracking modifications is expensive and not needed in most cases.
Then listen for the signal:
from flask_sqlalchemy import models_committed
def notify_subscribers(app, changes):
new_posts = [target for target, op in changes if isinstance(target, Post) and op in ('insert', 'update')]
# notify about the new and updated posts
models_committed.connect(notify_subscribers, app)
app.config["SQLALCHEMY_TRACK_MODIFICATIONS"] = True
You can also implement this yourself (mostly by copying the code from Flask-SQLAlchemy). It's slightly tricky, because model changes occur on flush, not on commit, so you need to record all changes as flushes occur, then use them after the commit.
from sqlalchemy import event
class ModelChangeEvent(object):
def __init__(self, session, *callbacks):
self.model_changes = {}
self.callbacks = callbacks
event.listen(session, 'before_flush', self.record_ops)
event.listen(session, 'before_commit', self.record_ops)
event.listen(session, 'after_commit', self.after_commit)
event.listen(session, 'after_rollback', self.after_rollback)
def record_ops(self, session, flush_context=None, instances=None):
for targets, operation in ((session.new, 'insert'), (session.dirty, 'update'), (session.deleted, 'delete')):
for target in targets:
state = inspect(target)
key = state.identity_key if state.has_identity else id(target)
self.model_changes[key] = (target, operation)
def after_commit(self, session):
if self._model_changes:
changes = list(self.model_changes.values())
for callback in self.callbacks:
callback(changes=changes)
self.model_changes.clear()
def after_rollback(self, session):
self.model_changes.clear()
def notify_subscribers(changes):
new_posts = [target for target, op in changes if isinstance(target, Post) and op in ('insert', 'update')]
# notify about new and updated posts
# pass all the callbacks (if you have more than notify_subscribers)
mce = ModelChangeEvent(db.session, notify_subscribers)
# or you can append more callbacks
mce.callbacks.append(my_other_callback)
I'm creating a task (by subclassing celery.task.Task) that creates a connection to Twitter's streaming API. For the Twitter API calls, I am using tweepy. As I've read from the celery-documentation, 'a task is not instantiated for every request, but is registered in the task registry as a global instance.' I was expecting that whenever I call apply_async (or delay) for the task, I will be accessing the task that was originally instantiated but that doesn't happen. Instead, a new instance of the custom task class is created. I need to be able to access the original custom task since this is the only way I can terminate the original connection created by the tweepy API call.
Here's some piece of code if this would help:
from celery import registry
from celery.task import Task
class FollowAllTwitterIDs(Task):
def __init__(self):
# requirements for creation of the customstream
# goes here. The CustomStream class is a subclass
# of tweepy.streaming.Stream class
self._customstream = CustomStream(*args, **kwargs)
#property
def customstream(self):
if self._customstream:
# terminate existing connection to Twitter
self._customstream.running = False
self._customstream = CustomStream(*args, **kwargs)
def run(self):
self._to_follow_ids = function_that_gets_list_of_ids_to_be_followed()
self.customstream.filter(follow=self._to_follow_ids, async=False)
follow_all_twitterids = registry.tasks[FollowAllTwitterIDs.name]
And for the Django view
def connect_to_twitter(request):
if request.method == 'POST':
do_stuff_here()
.
.
.
follow_all_twitterids.apply_async(args=[], kwargs={})
return
Any help would be appreciated. :D
EDIT:
For additional context for the question, the CustomStream object creates an httplib.HTTPSConnection instance whenever the filter() method is called. This connection needs to be closed whenever there is another attempt to create one. The connection is closed by setting customstream.running to False.
The task should only be instantiated once, if you think it is not for some reason,
I suggest you add a
print("INSTANTIATE")
import traceback
traceback.print_stack()
to the Task.__init__ method, so you could tell where this would be happening.
I think your task could be better expressed like this:
from celery.task import Task, task
class TwitterTask(Task):
_stream = None
abstract = True
def __call__(self, *args, **kwargs):
try:
return super(TwitterTask, self).__call__(stream, *args, **kwargs)
finally:
if self._stream:
self._stream.running = False
#property
def stream(self):
if self._stream is None:
self._stream = CustomStream()
return self._stream
#task(base=TwitterTask)
def follow_all_ids():
ids = get_list_of_ids_to_follow()
follow_all_ids.stream.filter(follow=ids, async=false)