What is Kafka?
What is RabbitMQ?
Need advice about which tool to choose?Ask the StackShare community!
Sign up to add, upvote and see more prosMake informed product decisions
Sign up to get full access to all the companiesMake informed product decisions
Sign up to get full access to all the tool integrationsMake informed product decisions
Around the time of their Series A, Pinterest’s stack included Python and Django, with Tornado and Node.js as web servers. Memcached / Membase and Redis handled caching, with RabbitMQ handling queueing. Nginx, HAproxy and Varnish managed static-delivery and load-balancing, with persistent data storage handled by MySQL.
As of 2017, Slack was handling a peak of 1.4 billion jobs a day, (33,000 jobs every second). Until recently, Slack had continued to depend on their initial job queue implementation system based on Redis. While it had allowed them to grow exponentially and diversify their services, they soon outgrew the existing system. Also, dequeuing jobs required memory that was unavailable. Allowing job workers to scale up further burdened Redis, slowing the entire system.
Slack decided to use Kafka to ease the process and allow them to scale up without getting rid of the existing architecture. To build on it, they added Kafka in front of Redis leaving the existing queuing interface in place. A stateless service called Kafkagate was developed in Go to enqueue jobs to Kafka. It exposes an HTTP POST interface with each request comprising a topic, partition, and content. Kafkagate's design reduces latency while writing jobs and allows greater flexibility in job queue design. JQRelay, a stateless service, is used to relay jobs from a Kafka topic to Redis. It ensures only one relay process is assigned to each topic, failures are self-healing, and job-specific errors are corrected by re-enqueuing the job to Kafka. The new system was rolled out by double writing all jobs to both Redis and Kafka, with JQRelay operating in 'shadow mode' - dropping all jobs after reading it from Kafka. Jobs were verified by being tracked at each part of the system through its lifetime. By using durable storage and JQRelay, the enqueuing rate could be paused or adjusted to give Redis the necessary breathing room and make Slack a much more resilient service.
As Sentry runs throughout the day, there are about 50 different offline tasks that we execute—anything from “process this event, pretty please” to “send all of these cool people some emails.” There are some that we execute once a day and some that execute thousands per second.
Managing this variety requires a reliably high-throughput message-passing technology. We use Celery's RabbitMQ implementation, and we stumbled upon a great feature called Federation that allows us to partition our task queue across any number of RabbitMQ servers and gives us the confidence that, if any single server gets backlogged, others will pitch in and distribute some of the backlogged tasks to their consumers.
Lumosity is home to the world's largest cognitive training database, a responsibility we take seriously. For most of the company's history, our analysis of user behavior and training data has been powered by an event stream--first a simple Node.js pub/sub app, then a heavyweight Ruby app with stronger durability. Both supported decent throughput and latency, but they lacked some major features supported by existing open-source alternatives: replaying existing messages (also lacking in most message queue-based solutions), scaling out many different readers for the same stream, the ability to leverage existing solutions for reading and writing, and possibly most importantly: the ability to hire someone externally who already had expertise.
We ultimately migrated to Kafka in early- to mid-2016, citing both industry trends in companies we'd talked to with similar durability and throughput needs, the extremely strong documentation and community. We pored over Kyle Kingsbury's Jepsen post (https://aphyr.com/posts/293-jepsen-Kafka), as well as Jay Kreps' follow-up (http://blog.empathybox.com/post/62279088548/a-few-notes-on-kafka-and-jepsen), talked at length with Confluent folks and community members, and still wound up running parallel systems for quite a long time, but ultimately, we've been very, very happy. Understanding the internals and proper levers takes some commitment, but it's taken very little maintenance once configured. Since then, the Confluent Platform community has grown and grown; we've gone from doing most development using custom Scala consumers and producers to being 60/40 Kafka Streams/Connects.
We originally looked into Storm / Heron , and we'd moved on from Redis pub/sub. Heron looks great, but we already had a programming model across services that was more akin to consuming a message consumers than required a topology of bolts, etc. Heron also had just come out while we were starting to migrate things, and the community momentum and direction of Kafka felt more substantial than the older Storm. If we were to start the process over again today, we might check out Pulsar , although the ecosystem is much younger.
To find out more, read our 2017 engineering blog post about the migration!
Beamery runs a #microservices architecture in the backend on top of Google Cloud with Kubernetes There are a 100+ different microservice split between Node.js and Go . Data flows between the microservices over REST and gRPC and passes through Kafka RabbitMQ as a message bus. Beamery stores data in MongoDB with near-realtime replication to Elasticsearch . In addition, Beamery uses Redis for various memory-optimized tasks.
We've been using RabbitMQ as Zulip's queuing system since we needed a queuing system. What I like about it is that it scales really well and has good libraries for a wide range of platforms, including our own Python. So aside from getting it running, we've had to put basically 0 effort into making it scale for our needs.
However, there's several things that could be better about it:
* It's error messages are absolutely terrible; if ever one of our users ends up getting an error with RabbitMQ (even for simple things like a misconfigured hostname), they always end up needing to get help from the Zulip team, because the errors logs are just inscrutable. As an open source project, we've handled this issue by really carefully scripting the installation to be a failure-proof configuration (in this case, setting the RabbitMQ hostname to
127.0.0.1, so that no user-controlled configuration can break it). But it was a real pain to get there and the process of determining we needed to do that caused a significant amount of pain to folks installing Zulip.
pika library for Python takes a lot of time to startup a RabbitMQ connection; this means that Zulip server restarts are more disruptive than would be ideal.
* It's annoying that you need to run the
rabbitmqctl management commands as root.
But overall, I like that it has clean, clear semanstics and high scalability, and haven't been tempted to do the work to migrate to something like Redis (which has its own downsides).
Why we built Marmaray, an open source generic data ingestion and dispersal framework and library for Apache Hadoop :
Built and designed by our Hadoop Platform team, Marmaray is a plug-in-based framework built on top of the Hadoop ecosystem. Users can add support to ingest data from any source and disperse to any sink leveraging the use of Apache Spark . The name, Marmaray, comes from a tunnel in Turkey connecting Europe and Asia. Similarly, we envisioned Marmaray within Uber as a pipeline connecting data from any source to any sink depending on customer preference:
(Direct GitHub repo: https://github.com/uber/marmaray Kafka Kafka Manager )
I use Kafka because it has almost infinite scaleability in terms of processing events (could be scaled to process hundreds of thousands of events), great monitoring (all sorts of metrics are exposed via JMX).
Downsides of using Kafka are: - you have to deal with Zookeeper - you have to implement advanced routing yourself (compared to RabbitMQ it has no advanced routing)
The question for which Message Queue to use mentioned "availability, distributed, scalability, and monitoring". I don't think that this excludes many options already. I does not sound like you would take advantage of Kafka's strengths (replayability, based on an even sourcing architecture). You could pick one of the AMQP options.
I would recommend the RabbitMQ message broker, which not only implements the AMQP standard 0.9.1 (it can support 1.x or other protocols as well) but has also several very useful extensions built in. It ticks the boxes you mentioned and on top you will get a very flexible system, that allows you to build the architecture, pick the options and trade-offs that suite your case best.
For more information about RabbitMQ, please have a look at the linked markdown I assembled. The second half explains many configuration options. It also contains links to managed hosting and to libraries (though it is missing Python's - which should be Puka, I assume).
I used Kafka originally because it was mandated as part of the top-level IT requirements at a Fortune 500 client. What I found was that it was orders of magnitude more complex ...and powerful than my daily Beanstalkd , and far more flexible, resilient, and manageable than RabbitMQ.
So for any case where utmost flexibility and resilience are part of the deal, I would use Kafka again. But due to the complexities involved, for any time where this level of scalability is not required, I would probably just use Beanstalkd for its simplicity.
I tend to find RabbitMQ to be in an uncomfortable middle place between these two extremities.
As we've evolved or added additional infrastructure to our stack, we've biased towards managed services. Most new backing stores are Amazon RDS instances now. We do use self-managed PostgreSQL with TimescaleDB for time-series data—this is made HA with the use of Patroni and Consul.
We also use managed Amazon ElastiCache instances instead of spinning up Amazon EC2 instances to run Redis workloads, as well as shifting to Amazon Kinesis instead of Kafka.
Initially we had just 1 monolithic application with a PostgreSQL database (picked for performance, community & flexibility to work with GIS data), but as we've developed more features, it was clear that some stuff is relatively independent from the rest of the platform - it made sense to split the application into loosely coupled, asynchronously communicated services. As a communication broker we've used RabbitMQ (wrapped in our custom, ProtoBuff-based wrapper). To reduce some excessive inter-process (& inter-dyno) communication, we've applied Redis as a tool to keep short-lived, not-persistent information (but not as a cheap caching workaround for any kind of performance issues ;>).
I use Redis because, based on the case studies I have reviewed, it appears to be the most performant cache database for my Django projects. The ease of configuration and deployment is also a big plus.
Using both higher level view caching as well as low-level QuerySet caching with Redis has allowed me to improve HTTP request times by an order of magnitude.
Automations are what makes a CRM powerful. With Celery and RabbitMQ we've been able to make powerful automations that truly works for our clients. Such as for example, automatic daily reports, reminders for their activities, important notifications regarding their client activities and actions on the website and more.
We use Celery basically for everything that needs to be scheduled for the future, and using RabbitMQ as our Queue-broker is amazing since it fully integrates with Django and Celery storing on our database results of the tasks done so we can see if anything fails immediately.
I developed one of the largest queue based medical results delivery systems in the world, 18,000+ queues and still growing over a decade later all using MQSeries, later called Websphere MQ. When I left that company I started using RabbitMQ after doing some research on free offerings.. it works brilliantly and is incredibly flexible from small scale single instance use to large scale multi-server - multi-site architectures.
If you can think in queues then RabbitMQ should be a viable solution for integrating disparate systems.
Redis is a good caching tool for a cluster, but our application had performance issues while using Aws Elasticache Redis since some page had 3000 cache hits per a page load and Redis just couldn't quickly process them all in once + latency and object deseialization time - page load took 8-9 seconds. We create a custom hybrid caching based on Redis and EhCache which worked great for our goals. Check it out on github, it's called HybriCache - https://github.com/batir-akhmerov/hybricache.
Front-end messages are logged to Kafka by our API and application servers. We have batch processing (on the middle-left) and real-time processing (on the middle-right) pipelines to process the experiment data. For batch processing, after daily raw log get to s3, we start our nightly experiment workflow to figure out experiment users groups and experiment metrics. We use our in-house workflow management system Pinball to manage the dependencies of all these MapReduce jobs.
Redis is used for storing all ephemeral (that's data you don't necessarily want to store permanently) user data, such as mapping of session IDs (stored in cookies) to current session variables at Cloudcraft.co. The many datastructures supported by Redis also makes it an excellent caching and realtime statistics layer. It doesn't hurt that the author, Antirez, is the nicest guy ever! These days, I would be really hard pressed to find any situation where I would pick something like Memcached over Redis.
Trello uses Redis for ephemeral data that needs to be shared between server processes but not persisted to disk. Things like the activity level of a session or a temporary OpenID key are stored in Redis, and the application is built to recover gracefully if any of these (or all of them) are lost. We run with allkeys-lru enabled and about five times as much space as its actual working set needs, so Redis automatically discards data that hasn’t been accessed lately, and reconstructs it when necessary.
The UI has message inbox that is sent a message when you get a new badge, receive a message, significant event, etc. Done using WebSockets and is powered by redis. Redis has 2 slaves, SQL has 2 replicas, tag engine has 3 nodes, elastic has 3 nodes - any other service has high availability as well (and exists in both data centers).
Redis makes certain operations very easy. When I need a high-availability store, I typically look elsewhere, but for rapid development with the ability to land on your feet in prod, Redis is great. The available data types make it easy to build non-trivial indexes that would require complex queries in postgres.
I use Redis for cacheing, data storage, mining and augmentation, proprietary distributed event system for disparate apps and services to talk to each other, and more. Redis has some very useful native data types for tracking, slicing and dicing information.
The poster child for scalable messaging systems, RabbitMQ has been used in countless large scale systems as the messaging backbone of any large cluster, and has proven itself time and again in many production settings.
Rabbit acts as our coordinator for all actions that happen during game time. All worker containers connect to rabbit in order to receive game events and emit their own events when applicable.
Used as central Message Broker; off-loading tasks to be executed asynchronous, used as communication tool between different microservices, used as tool to handle peaks in incoming data, etc.
RabbitMQ is the enterprise message bus for our platform, providing infrastructure for managing our ETL queues, real-time event notifications for applications, and audit logging.
RabbitMQ is an all purpose queuing service for our stack. We use it for user facing jobs as well as keeping track of behind the scenes jobs.
Building out real-time streaming server to present data insights to Coolfront Mobile customers and internal sales and marketing teams.