Under the hood, how is a FIFO queue turned into a priority queue in a distributed fashion? Are they actually swapping the underlying datastructure, or is it a "hacked" fix
The underlying data structures are multiple queues, each assigned a priority. Each queue is an Erlang VM process. This is why having more than 10 or so priorities isn't recommended as performance suffers. If your load is light enough, this may be acceptable.
NOTE: the RabbitMQ team monitors the rabbitmq-users mailing list and only sometimes answers questions on StackOverflow.
When a message is published with a priority header, the message with the higher priority value gets placed on the head of the queue. This is done by actually swapping the messages in the queue. This is all done when the message is waiting to be consumed in the queue. In order to allow RabbitMQ to actually prioritise the messages, set the basic.qos of the consumer as low as possible. So if a consumer connects to an empty queue whose basic.qos is not set and to which messages are subsequently published, the messages may not spend any time at all waiting in the queue. In this case, the priority queue will not get any opportunity to prioritise them.
Reference: https://www.rabbitmq.com/priority.html
Related
I have a RabbitMQ setup in which jobs are sent to an exchange, which passes them to a queue. A consumer carries out the jobs from the queue correctly in turn. However, these jobs are long processes (several minutes at least). For scalability, I need to be able to have multiple consumers picking a job from the top of the queue and executing it.
The consumer is running on a Heroku dyno called 'queue'. When I scale the dyno, it appears to create additional consumers for each dyno (I can see these on the RabbitMQ dashboard). However, the number of tasks in the queue is unchanged - the extra consumers appear to be doing nothing. Please see the picture below to understand my setup.
Am I missing something here?
Why are the consumers showing as 'idle'? I know from my logs that at least one consumer is actively working through a task.
How can my consumer utilisation be 0% when at least one consumer is definitely working hard.
How can I make the other three consumers actually pull some jobs from the queue?
Thanks
EDIT: I've discovered that the round robin dispatching is actually working, but only if the additional consumers are already running when the messages are sent to the queue. This seems like counterintuitive behaviour to me. If I saw a large queue and wanted to add more consumers, the added consumers would do nothing until more items are added to the queue.
To pick out the key point from the other answer, the likely culprit here is pre-fetching, as described under "Consumer Acknowledgements and Publisher Confirms".
Rather than delivering one message at a time and waiting for it to be acknowledged, the server will send batches to the consumer. If the consumer acknowledges some but then crashes, the remaining messages will be sent to a different consumer; but if the consumer is still running, the unacknowledged messages won't be sent to any new consumer.
This explains the behaviour you're seeing:
You create the queue, and deliver some messages to it, with no consumer running.
You run a single consumer, and it pre-fetches all the messages on the queue.
You run a second consumer; although the queue isn't empty, all the messages are marked as sent to the first consumer, awaiting acknowledgement; so the second consumer sits idle.
A new message arrives in the queue; it is distributed in round-robin fashion to the second consumer.
The solution is to specify the basic.qos option in the consumer. If you set this to 1, RabbitMQ won't send a message to a consumer until it has acknowledged the previous message; multiple consumers with that setting will receive messages in strictly round-robin fashion.
I am not familiar to Heroku, so I don't know how Heroku worker build rabbitMQ consumer, I just have a quick view over Heroku document.
Why are the consumers showing as 'idle'?
I think your mean the queue is 'idle'? Because the queue's state is about the queue's traffic, it just means there is not on-doing job for the queue's job thread. And it will become 'running' when a message is published in the queue.
How can my consumer utilisation be 0% when at least one consumer is definitely working hard.
The same as queue state, from official explanation, consumer utilisation too low means:
There were more consumers
The consumers were faster
The consumers had a higher prefetch count
In your situation, prefetch_count = 0 means no limits on prefetch, so it's too large. And Messages.total = Messages.unacked = 78 means your consumer is too slow, there are two many messages have been processed by consumer.
So if your message rate is not large enough, the state and consumer utilisation field of the queue is useless.
If I saw a large queue and wanted to add more consumers, the added consumers would do nothing until more items are added to the queue.
Because these unacked messages have already been prefetched by exist consumers, they will not be consumed by new consumers unless you requeue the unacked messages.
I`ve been reading about the principles of AMQP messaging confirms. (https://www.rabbitmq.com/confirms.html). Really helpful and wel written article but one particular thing about consumer aknowledgments is really confusing, here is the quote:
Another things that's important to consider when using automatic acknowledgement mode is that of consumer overload.
Consumer overload? Message queue is processed and kept in RAM by broker (if I understand it correctly). What overload is it about? Does consumer have some kind of second queue?
Another part of that article is even more confusing:
Consumers therefore can be overwhelmed by the rate of deliveries, potentially accumulating a backlog in memory and running out of heap or getting their process terminated by the OS.
What backlog? How is this all works together? What part of job is done by consumer (besides consuming message and processing it of course)? I thought that broker is keeping queues alive and forwards the messages but now I am reading about some mysterious backlogs and consumer overloads. This is really confusing, can someone explain it a bit or at least point me to the good source?
I believe the documentation you're referring to deals with what, in my opinion, is sort of a design flaw in either AMQP 0-9-1 or RabbitMQ's implementation of it.
Consider the following scenario:
A queue has thousands of messages sitting in it
A single consumer subscribes to the queue with AutoAck=true and no pre-fetch count set
What is going to happen?
RabbitMQ's implementation is to deliver an arbitrary number of messages to a client who has not pre-fetch count. Further, with Auto-Ack, prefetch count is irrelevant, because messages are acknowledged upon delivery to the consumer.
In-memory buffers:
The default client API implementations of the consumer have an in-memory buffer (in .NET it is some type of blocking collection (if I remember correctly). So, before the message is processed, but after the message is received from the broker, it goes into this in-memory holding area. Now, the design flaw is this holding area. A consumer has no choice but to accept the message coming from the broker, as it is published to the client asynchronously. This is a flaw with the AMQP protocol specification (see page 53).
Thus, every message in the queue at that point will be delivered to the consumer immediately and the consumer will be inundated with messages. Assuming each message is small, but takes 5 minutes to process, it is entirely possible that this one consumer will be able to drain the entire queue before any other consumers can attach to it. And since AutoAck is turned on, the broker will forget about these messages immediately after delivery.
Obviously this is not a good scenario if you'd like to get those messages processed, because they've left the relative safety of the broker and are now sitting in RAM at the consuming endpoint. Let's say an exception is encountered that crashes the consuming endpoint - poof, all the messages are gone.
How to work around this?
You must turn Auto-Ack off, and generally it is also a good idea to set reasonable pre-fetch count (usually 2-3 is sufficient).
Being able to signal back pressure a basic problem in distributed systems. Without explicit acknowledgements, the consumer does not have any way to say "Slow down" to broker. With auto-ack on, as soon as the TCP acknowledgement is received by broker, it deletes the message from its memory/disk.
However, it does not mean that the consuming application has processed the message or ave enough memory to store incoming messages. The backlog in the article is simply a data structure used to store unprocessed messages (in the consumer application)
So I'm testing RabbitMQ in one node. Plain and simple,
One producer sends messages to the queue,
Multiple consumers take tasks from that queue.
Currently consumers execute thousands of messages per second, they are too fast so I need them to slow down. Managing consumer-side throttling is not possible due to network unreliable nature.
Collectively consumers must not take more than 10 messages per second altogether from that queue.
Is there a way to configure RabbitMQ so as the queue dispatches a maximum of 10 messages per second?
If I remember correctly, once Rabbit MQ has delivered a message to the queue, it's up to consumers to consume a message. There are various consumers in different languages, you haven't mentioned anything specific, so I'm giving a generic answer.
In my understanding, you shouldn't try to impose any restrictions on Rabbit MQ itself, instead, consider implementing connection pool of message consumers that will be able to handle not more than X messages simultaneously on the client side. Alternatively, you can provide some kind of semaphore at the handler itself, but not on the Rabbit MQ server itself.
We're seeing an issue where consumers of our message queues are picking up messages from queues at the top of the alphabetical range. We have two applications: a producer, and a subscriber. We're using RabbitMQ 3.6.1.
Let's say that the message queues are setup like so:
Our first application, the producer, puts say 100 messages/second onto each queue:
Our second application, the subscriber, has five unique consumer methods that can deal with messages on each respective queue. Each method binds to it's respective queue. A subscriber has a prefetch of 1 meaning it can only hold one message at a time, regardless of queue. We may run numerous instances of the subscriber like so:
So the situation is thus: each queue is receiving 100 msg/sec, and we have four instances of subscriber consuming these messages, so each queue has four consumers. Let's say that the consumer methods can deal with 25 msg/sec each.
What happens is that instead of all the queues being consumed equally, the alphabetically higher queues instead get priority. It's seems as though when the subscriber becomes ready, RabbitMQ looks down the list of queues that this particular ready channel is bound to, and picks the first queue with pending messages.
In our situation, A_QUEUE will have every message consumed. B_QUEUE may have some consumed in certain race conditions, but C_QUEUE/D_QUEUE and especially E_QUEUE will rarely get touched.
If we turn off the publisher, the queues will eventually drain, top to bottom.
Is it possible to configure either RabbitMQ itself or possibly even the channel to use some sort of round robin distribution policy or maybe even random policy so that when a channel has numerous bound queues, all with messages pending, the distribution is even?
to clarify: you have a single subscriber application with multiple consumers in it, right?
I'm guessing you're using a single RabbitMQ Connection within the subscriber app.
Are you also re-using a single RabbitMQ Channel for all of your consumers? If so, that would be a problem. Be sure to use a new Channel for each consumer you start.
Maybe the picture is wrong, but if it's not then your setup is wrong. You don't need 4 queues if you are going to have subscribers that listen to each and every queue. You'd just need one queue, that has multiple instances of the same subscriber consuming from it.
Now to answer, yes (but no need to configure, as long as prefetch is 1), actually rabbitmq does distribute messages evenly. You can find about about that here, and on the same place actually how your setup should look like. Here is a quote from the link.
RabbitMQ just dispatches a message when the message enters the queue.
It doesn't look at the number of unacknowledged messages for a
consumer. It just blindly dispatches every n-th message to the n-th
consumer.
I have been looking at message queues (currently between Kafka and RabbitMQ) for one of my projects where these are biggest must have features.
Must have features
Messages in queues should be persistent. (only until they are processed successfully by consumers.)
Messages in queues should be removed only when downstream consumers were able to process the message successfully. Basically, a consumer should ACK. that it processed a message successfully.
Good to have features
To increase throughput, consumers should be able to pull batch of messages from queue.
If you are going with Kafka it will only retains message for a configurable duration of time after which the messages will be discarded to free up spaces no matter consumed or not.
And it is simply the responsibilities of the Kafka consumers to keep a track of what has been consumed.
IMHO if you require to keep the messages persisted for ever than consider using a different storage medium (database may be).