Login to bookmark this video
Buy Access to Course
38.

Delaying in AMQP: Dead Letter Exchange

|

Share this awesome video!

|

Keep on Learning!

With a Subscription, click any sentence in the script to jump to that part of the video!

Login Subscribe

When we started working with AMQP, I told you to go into ImagePostController and remove the DelayStamp. This stamp is a way to tell the transport system to wait at least 500 milliseconds before allowing a worker to receive the message. Let's change this to 10 seconds - so 10000 milliseconds.

100 lines | src/Controller/ImagePostController.php
// ... lines 1 - 23
class ImagePostController extends AbstractController
{
// ... lines 26 - 40
public function create(Request $request, ValidatorInterface $validator, PhotoFileManager $photoManager, EntityManagerInterface $entityManager, MessageBusInterface $messageBus)
{
// ... lines 43 - 63
$envelope = new Envelope($message, [
new DelayStamp(10000)
]);
// ... lines 67 - 69
}
// ... lines 71 - 98
}

Now, move over to your terminal and make sure that your worker is not running.

Ok, let's see what happens! Right now both queues are empty. I'll upload 3 photos... then... quick, quick, quick! Go look at the queues. Suddenly, poof! A new queue appeared... with a strange name: delay_messages_high_priority__10000. And it has - dun, dun, dun! - three messages in it.

Let's look inside. Interesting, the messages were delivered here, instead of the normal queue. But then... they disappeared? The graph shows how the messages sitting in this queue went from 3 to 0. But... how? Our worker isn't even running!

Woh! This page just 404'ed! The queue is gone! Something is attacking our queues!

Head back to the queue list. Yea, that weird "delay" queue is gone... oh, but now the three messages are somehow in messages_high. What the heck just happened?

Well first, to prove that the whole system still works... regardless of what craziness just occurred... let's run our worker and consume from both the async_priority_high and async transports:

php bin/console messenger:consume -vv async_priority_high async

It consumes them and... when we move over, go to the homepage and refresh, yep! Ponka was added to those images.

The Delay Exchange

Ok, let's figure out how this worked. I mean, on the one hand, it's not important: if we had been running our worker the entire time, you would have seen that those messages were in fact delayed by 10 seconds. How you delay messages in RabbitMQ is kinda crazy... but if you don't care about the details, Messenger just takes care of it for you.

But I do want to see how this works... in part because it'll be a great chance to see how some of the more advanced features of AMQP work.

Click on "Exchanges". Surprise! There's a new exchange called delays. And instead of being a fanout type like our other two exchanges, this is a direct exchange. We'll talk about what that that means soon.

But the first thing to know is that when Messenger sees that a message should be delayed, it sends it to this exchange instead of sending it to the normal, "correct" exchange. At this moment, the delays exchange has no bindings... but that will change when we send a delayed message.

To be able to really see what's happening, let's increase the delay to 60 seconds.

100 lines | src/Controller/ImagePostController.php
// ... lines 1 - 23
class ImagePostController extends AbstractController
{
// ... lines 26 - 40
public function create(Request $request, ValidatorInterface $validator, PhotoFileManager $photoManager, EntityManagerInterface $entityManager, MessageBusInterface $messageBus)
{
// ... lines 43 - 63
$envelope = new Envelope($message, [
new DelayStamp(60000)
]);
// ... lines 67 - 69
}
// ... lines 71 - 98
}

Ok, upload 3 more photos: we now know that these were just sent to the delays exchange. And... if you refresh that exchange... it has a new binding! This says:

If a message sent here has a "routing key" set to delay_messages_high_priority__60000, then I will send that message to a queue called delay_messages_high_priority__60000

A "routing key" is an extra property that you can set on a message that's sent to AMQP. Normally Messenger doesn't set any routing key, but when a message has a delay, it does. And thanks to this binding - those three messages are sent to the delay_messages_high_priority__60000 queue. This is how a direct exchange works: instead of sending each message to all queues bound to it, it uses the "binding key" rules to figure out which queue - or queues - a message should go to.

Delay Queues: x-message-ttl and x-deal-letter-exchange

Click into the queue because it's super interesting. It has a few important properties. The first is an x-message-ttl set to 60 seconds. What does that means? When you set this on a queue, it means that, after a message has been sitting in this queue for 60 seconds, RabbitMQ should remove it... which seems crazy, right? Why would we want messages to only live for 60 seconds... and then be deleted? Well... it's by design... and works together with this second important property: x-dead-letter-exchange.

If a queue has this property, it tells Rabbit that when a message hits its 60 second TTL and needs to be removed, it should not be deleted. Instead, it should be sent to the messages_high_priority exchange.

So, Messenger delivers messages to the delays exchange with a routing key that makes it get sent here. Then, after sitting around for 60 seconds, the message is removed from this queue and sent to the messages_high_priority exchange. Yep, it's delivered to the correct place after 60 seconds!

And then... 404! Even the queue itself is marked as "temporary": once it doesn't have any messages left, it deletes itself.

When you click back to see the Queues, the messages were delivered to the messages_high queue... but that's already empty because our worker consumed them.

So... yea... wow! Whenever we publish a message with a delay, Messenger sets all of this up: it creates the temporary delay queue with the TTL and dead letter exchange settings, adds a binding to the delays exchange to route to this queue, and adds the correct routing key to the message to make sure it ends up in that queue.

You can really start to see how rich the features are in AMQP... even if you won't need them. The most important feature we just saw was the direct exchange type: an exchange that relies on routing keys to figure out where each message should go.

Next, could we use direct exchanges for our non-delayed messages? Instead of two exchanges that each "fan out" to a separate queue, could we create just one exchange that, by using routing keys, delivers the correct messages to the correct queues? Totally.