Redisq. An easy to use, reliable Java library for asynchronous messaging in Redis.
UPDATE: Redisq is no longer used and maintained by Grakn Labs.
This blog post is about a solution we built here at Grakn Labs for executing tasks asynchronously. We thought it would be nice to release it as a generic Java library built on Redis.
The Task of Choosing a Task Queue
We will discuss the design choices behind the engine in another post, but — motivated by the need to simplify our distribution — we decided to use Redis as a task queue.
Redis is often described as a in-memory database, but it offers good persistence guarantees and excellent primitives to use as a base for a distributed message queue. By message queue we mean something that supports a push primitive, used to store a message by a publisher, and a subscribe command, that defines how the messages are processed. Something similar to SQS, Kafka, or RabbitMQ. We also required a way for the publisher to be notified when the message was consumed.
We looked if something was available off-the-shelf for Java and, surprisingly, we only found a library called Jesque. We did not like a few things about Jesque. Unfortunately, its design had to be compatible with Resque and that constrained a lot what it could offer. Also, it was unnecessarily complex to create consumers, and it was lacking some features we needed (publisher notification, metrics, processing dead tasks).
So we made a new library. Go and check it out, it’s called Redisq! Here is a few things it provides:
- It is instrumented using my favourite library Metrics
- It has timeouts on tasks being stuck in the inflight queue so that they can be reprocessed or logged and discarded after a timeout
- It registers the state of a task
- Consumers are implemented as Java consumers, can be defined as lambda expressions, and are executed asynchronously. No need to instantiate threads
- The publisher can subscribe to a certain message (we call them documents) or block on its termination
How does it work?
When Redisq is instantiated, it creates a Redis list, used as queue. Every push operates as follows: it puts the id in the queue and it sets a key with the content of the document. It also creates a document state, which keeps track of when the document was pushed and whether it is being processed or if it is done. Finally, it creates a channel for subscriptions to changes in the state.
When a subscriber is started, one thread blocks on new content in the queue.
When something new appears in the queue, the task is sent to a consumer and executed asynchronously on a given thread pool. The id is moved atomically to another in-flight queue.
Another thread processes an inflight queue, and it looks for documents that have been there for too long.
Using Redisq
It’s really simple to use.
First step, as shown in the following snippet of code, you need to create a serialisable object so that Redisq knows how to store it. We use Jackson for that. We also need a method that we can use to extract the id getIdAsString. Everything that goes in the queue needs an id.
public class MyTask implements Document {
@JsonProperty
private String id; @JsonProperty
private String content; // Needed by Jackson
public MyTask() {}
public MyTask(String id, String content) {
this.id = id;
this.content = content;
} public String getContent() {
return content;
} @Override
@JsonIgnore
public String getIdAsString() {
return id;
}
}
In order to create a consumer, we need to choose a name for the queue, define the class of the document and define the logic. In this example we just print out the content to standard out.
Pool<Jedis> jedisPool = new JedisPool();
Queue<MyTask> redisq = new RedisqBuilder<MyTask>()
.setJedisPool(jedisPool)
.setName("my_queue")
.setConsumer((d) ->
System.out.println("I'm consuming "
+ d.getContent()))
.setDocumentClass(MyTask.class)
.createRedisq();redisq.startConsumer();
Note that when we start the consumer, we also start a inflight processor thread that periodically checks if something is stuck in the queue.
We can set the time a certain task is locked for execution using RedisqBuilder::setLockTime(Duration lockTime) and the time after which the task is discarded and locked RedisqBuilder::setDiscardTime(Duration discardTime). In other words if a consumer dies and the document is not deleted from the inflight queue, the inflight processor picks it up after lockTime. If the document has been there for less than discardTime, it’s put back in the main queue, otherwise it’s discarded.
This is how you push something to the queue:
redisq.push(new MyTask("documentid", "content"));
Remember to close the queue:
redisq.close();
Closing the queue terminates the inflight thread, the consumer thread and it shuts down the thread pool that executes the consumers.
The writer can also wait that the task is consumed. This is achieved using the awesome Redis SUBSCRIBE command.
redisq.pushAndWait(new MyTask("id", "waitforme"), 5, TimeUnit.SECONDS);
You can also push and have a future to process using Queue::getFutureForDocumentStateWait! Remember to subscribe before pushing so you don’t miss the state update.
Conclusion
Using queues within a micro-services architecture to deliver a stream of operations to be performed is a very common solution. We created a library so we could make the most of a technology, Redis, already used in Grakn to store metadata.
Redisq has a very minimal, easy to use interface and it provides some extra features compared to other similar solutions.
Check the GitHub repository at github.com/pluraliseseverythings/redisq for instructions on how to include Redisq in your project and get started! Please get in touch if you’ve any questions or comments, either below, via our Community Slack channel, or via our discussion forum.
If you enjoyed this article, please hit the clap button below, so others can find it, too. Find out more from https://grakn.ai.
*Diagram created with https://sketchboard.me/
Feature Image credit: “Queue” by Newtown grafitti is licensed under CC 2.0