Wednesday 16 January 2013

Signals - Part 2: Return to sender

This week we return with a follow up to one of the most popular topics on this blog - signals. In my first post on the topic (Signals: Letting People Know When You've Got Something To Say) I covered the basics of how to register a function to listen to signals as well as how to create your own signals and use them to send messages. Since that post there has been a new decorator added to help register your listeners which I will talk about more shortly. The other topics I will be covering are some of the other options that can be passed in when registering a signal and the mechanism provided for obtaining responses from listeners when a signal is sent.

The Receiver Decorator

This decorator was added in django 1.3 and provides a shortcut to having to call connect for your listener functions. It's very simple to use and takes the same parameters that connect does (behind the scenes it passes these straight through to the connect function for you) as shown in my first example (modified from example 1 of the first signals post.

# File: test_signal/models.py from django.db import models from django.contrib.auth.models import User from django.db.models.signals import post_save from django.dispatch import receiver class UserProfile(models.Model): user = models.ForeignKey(User, unique=True) allow_our_emails = models.BooleanField(default=False) allow_other_emails = models.BooleanField(default=False) @receiver(post_save, sender=User) def ensure_profile_exists(sender, **kwargs): if kwargs.get('created', False): UserProfile.objects.create(user=kwargs.get('instance')) post_save.connect(ensure_profile_exists, sender=User)

The second example shows a change from Django 1.5 to the receiver decorator which allows you to assign a function to listen to multiple signals in one go.

# File: test_signal/models.py from django.db import models from django.contrib.auth.models import User from django.db.models.signals import post_save, pre_delete from django.dispatch import receiver class UserProfile(models.Model): user = models.ForeignKey(User, unique=True) allow_our_emails = models.BooleanField(default=False) allow_other_emails = models.BooleanField(default=False) @receiver([post_save,pre_delete,], sender=User) def ensure_profile_exists(sender, **kwargs): if kwargs.get('created', False): UserProfile.objects.create(user=kwargs.get('instance')) signal = kwargs.get('signal', None) if signal == post_save: # After saving ensure that a UserProfile exists. if kwargs.get('created', False): UserProfile.objects.create(user=kwargs.get('instance')) elif signal == pre_delete: # Before deleting remove the UserProfile UserProfile.objects.filter(user=kwargs.get('instance')).delete()

Other Connection Arguments

So far I've only mentioned the sender argument that can be provided to connect or receiver there are a couple of other arguments that may be useful from time to time.

dispatch_uid
This allows you to pass in a unique identifier for your listener. If the signal you're trying to connect to already has a listener with that id then the new one won't be added. You can use this for situations where the code connecting receiver may end up running more than once, or for situations where different parts of the code may try to add a listener for a specific purpose but you only need one of them to actually receive the signals.
weak
This is True by default and refers to the way in which the signal stores the listener. If the value is True then a weak reference to the listening function is stored so if the function drops out of scope then it will be forgotten about and won't receive signals any more. If the value is False however you can pass in local functions without worrying that they'll be garbage collected.

Example 3 shows a situation where these parameters may be useful

# File: test_signal/models.py from django.db import models from django.db.models.signals import post_save from django.dispatch import receiver class LoggableModel1(models.Model): value = models.TextField() class LoggableModel2(models.Model): val = models.TextField() def add_logging_signals(models): for model_inst in models: uid = "%s-logger" % str(model_inst) if hasattr(model_inst, 'val') @receiver(post_save, sender=model_inst, weak=False, dispatch_uid=uid) def _listener(sender, kwargs**): print kwargs['instance'].val elif hasattr(model_inst, 'value') @receiver(post_save, sender=model_inst, weak=False, dispatch_uid=uid) def _listener(sender, kwargs**): print kwargs['instance'].value add_logging_signals([LoggableModel1, LoggableModel2, LoggableModel1,])

In this example because the listener functions are being created on the fly when add_logging_signals is called we specify weak=False so that the signal keeps hold of the function even when it drops out of scope.

We also make use of dispatch_uid so that even if we accidentally pass a model into the signal creator twice (which we then did) it will only add the one listener and so only output the logging code once.

Responding to a signal

The final topic for this week is on responding to signals. Signals don't have to be just a one-way mechanism!

The next example shows this in action

# File: signals/signals.py from django.dispatch import Signal pizza_deliverer = Signal(providing_args=["topping","size",])
# File: signals/models.py from django.db import models from django.dispatch import receiver from django.db.models.signals import post_save, class_prepared from signals import pizza_deliverer # List of toppings that our pizzas can have: TOPPING_PLAIN = 0 TOPPING_PEPPERONI = 1 TOPPING_VEGETARIAN = 2 TOPPINGS = ((TOPPING_PLAIN, 'Plain',), (TOPPING_PEPPERONI, 'Pepperoni',), (TOPPING_VEGETARIAN, 'Vegetarian',), ) # List of sizes that the pizzas could be: SIZE_SMALL = 8 SIZE_MEDIUM = 10 SIZE_LARGE = 12 SIZES = ((SIZE_SMALL, 'Small',), (SIZE_MEDIUM, 'Medium',), (SIZE_LARGE, 'Large',), ) class Pizza(models.Model): topping = models.IntegerField(choices=TOPPINGS) size = models.IntegerField(choices=SIZES) customer = models.ForeignKey('Customer', null=True, blank=True, default=None) @property def uid(self): return "pizza-%s" % self.id def serve_pizza(self, customer): # We have a customer so stop listening for new ones. self.customer = customer self.save() customer.receive_pizza() pizza_deliverer.disconnect(dispatch_uid=self.uid, sender=Customer, weak=False) class Customer(models.Model): topping_wanted = models.IntegerField(choices=TOPPINGS) size_wanted = models.IntegerField(choices=SIZES) @property def uid(self): return "customer-%s" % self.id def receive_pizza(self): # We have our pizza so stop listening for new ones. pizza_deliverer.disconnect(dispatch_uid=self.uid, sender=Pizza, weak=False) def _process_pizza(pizza): hungry_customers = pizza_deliverer.send(sender=Pizza, topping=pizza.topping, size=pizza.size) for _receiver, customer in hungry_customers: if customer is not None: # Pizza goes to the first customer that wants it. pizza.serve_pizza(customer) break else: # Nobody wants this pizza yet so listen for someone that does. @receiver(pizza_deliverer, sender=Customer, weak=False, dispatch_uid=pizza.uid) def _wait_for_customer(sender, **kwargs): topping = kwargs.get('topping') size = kwargs.get('size') if topping == pizza.topping and size == pizza.size: return pizza @receiver(post_save, sender=Pizza) def new_pizza(sender, **kwargs): if kwargs.get('created', False): pizza = kwargs.get('instance') _process_pizza(pizza) def _process_customer(customer): prepared_pizzas = pizza_deliverer.send(sender=Customer, topping=customer.topping_wanted, size=customer.size_wanted) for _receiver, pizza in prepared_pizzas: if pizza is not None: # Customer takes the first pizza offered that matches their criteria. pizza.serve_pizza(customer) break else: # The pizza this customer wants isn't ready yet so wait for it. @receiver(pizza_deliverer, sender=Pizza, weak=False, dispatch_uid=customer.uid) def _wait_for_pizza(sender, **kwargs): topping = kwargs.get('topping') size = kwargs.get('size') if customer.topping_wanted == topping and customer.size_wanted == size: return customer @receiver(post_save, sender=Customer) def new_customer(sender, **kwargs): if kwargs.get('created', False): customer = kwargs.get('instance') _process_customer(customer) # Ensure listeners are added for any unallocated pizzas or customers. for pizza in Pizza.objects.filter(customer=None): _process_pizza(pizza) for customer in Customer.objects.filter(pizza=None): _process_customer(customer)

Now this rather convoluted example simulates a potential (inneficient) pizzaria and makes use of all the features of signals discussed so far.

The signal is created as before and is set up to take two parameters a topping and a size. This signal represents the person behind the counter who will look for a waiting customer when a new pizza arrives or look for a waiting pizza when there's a new customer.

When a new pizza is added, a call is sent out to all the listening customers with the topping and size. Received responses are then iterated through until a positive reply is found, at which point the pizza is allocated to that customer.

If no suitable customers are listening then the pizza registers a listener function with our deliverer signal so it can wait for an appropriate customer to arrive.

A similar setup is then created for the customers - new ones send out a signal looking for matching pizzas and if there's none they sit and wait.

The listening functions _wait_for_customer and _wait_for_pizza both require a strong reference as they're created on the fly as well as both utilising the dispatch_uid to ensure that there's only one listener per pizza or customer.

This code also make use of the disconnect method on a signal to ensure that once a pizza and customer are matched up neither of them listens out for any further messages.

The final mention on signals is for the send_robust this would be used the same way that we used send in example 4 however it will catch any exceptions thrown by the listening functions and add that as the response from that listener.

Therefore to use it in our example we would also need to change the check when we iterate through the responses to ensure that the response was not None and that the response was also not of type Exception

This concludes the post on advanced signalling in Django. Until next time - keep up the communicating!

No comments:

Post a Comment