What is use of kafka in Big Data cluster? - hadoop

I have recently deployed Big Data cluster. In that I've used Apache Kafka and zookeeper. But still I didn't understand about its usage in cluster. When both are required and for what purpose?

I am simplifying the concepts here. You can find detailed explanation at this article
Kafka is a fast, scalable, distributed in nature by its design, partitioned and replicated commit log service. It has a unique design.
A stream of Messages of a particular type is defined as a Topic.
A Producer can be anyone who can publish messages to a Topic.
The published messages are then stored at a set of servers called Brokers or Kafka Cluster.
A Consumer can subscribe to one or more Topics and consume the published Messages by pulling data from the Brokers.
ZooKeeper is a distributed, hierarchical file system that facilitates loose coupling between clients.
ZooKeeper achieves high availability by running multiple ZooKeeper servers, called an ensemble.
ZooKeeper is used for managing, coordinating Kafka broker.
Each Kafka broker is coordinating with other Kafka brokers using ZooKeeper.
Producer and consumer are notified by ZooKeeper service about the presence of new broker in Kafka system or failure of the broker in Kafka system.

Kafka is a distributed messaging system optimised for high throughput. It is has a persistent queue with messages being appended to to files with on disk structures and performs consistently, even with very modest hardware. In short you will use Kafka to load data into your big data clusters and you will be able to do this at a high speed even when using modest hardware because of the distributed nature of Kafka.
Regarding Zookeeper, its a centralized distributed configuration service and naming registry for large distributed systems. It is robust, since the persisted data is distributed between multiple nodes and one client connects to any of them , migrating if one node fails; as long as a strict majority of nodes are working. So in short, Zookeeper makes sure your big data cluster remains online even if some of its nodes are offline.

In regards to Kafka, I would add a couple things.
Kafka describes itself as being a log not a queue. A log is an append-only, totally-ordered sequence of records ordered by time.
In a strict data structures sense, a queue is FIFO collection that is designed to hold data, but then once it is taken out of the queue there's no way to get it back. Jaco does describe it has being a persistent queue, but using different terms (queue v. log) can help in understanding.
Kafka's log is saved to disk instead of being kept in memory. The designers of Kafka have chosen this because 1. They wanted to avoid a lot of the JVM overhead you get when storing things in data structures. 2. They wanted messages to persist even if the Java process dies for some reason.
Kafka is designed for multiple consumers (Kafka term) to read from the same logs. Each consumer tracks its own offset in the log, Consumer A might be at offset 2, Consumer B might be at offset 8, etc. Tracking Consumers by offset eliminates a lot of complexities from Kafka's side.
Reading that first link will explain a lot of the differences between Kafka and other messaging services.

Related

Fluentd vs Kafka

The use case is this:
I've several java applications running which all have to interact with different (each one has a specific target) elasticsearch indices. For instance an application A uses the indices A,B,C of ElasticSearch to query and update. Application B uses indices A,C,D(say).
Some common interface is required which can manage all these data streams. Currently I'm evaluating Kafka and fluentd for this purpose.
Can someone explain which will be better suited for this situation. I've looked at features of both Kafka and Fluentd and I don't really understand the difference it would make here.
Thanks a lot.
kafka provides publish/subscribe messaging as a distributed commit log. Usually you install kafka on each host where you need to produce some data to be forwarded somewhere else and all those hosts will together form a cluster. The good thing here is that if for some reason network connectivity becomes unstable or goes down, your application can continue to produce data/logs and they won't be lost. Whereas if your application directly sends logs to some remote centralized logging host, you might lose some logs during the time the network goes down.
fluentd is a centralized log collector which is commonly installed on one host (or more if you need horizontal scaling). It connects to remote data sources, applies filtering and sends unified log data to remote data sinks.
From the fluentd docs, you can see that fluentd can consume data from kafka and produce data towards kafka as well. This alone should hint that fluentd and kafka are on different layers since the former uses the latter.
It would be more logical to compare fluentd and logstash actually. As far as fluentd is concerned, kafka is just another data source and/or data sink, but they are different beasts altogether.
If you want the best of both worlds, use kafka as input/output data pipes from/to your apps and fluentd (or logstash) as your centralized logging system reading from those kafka topics.
If you want to read more on the topic, you can read how fluentd and kafka complement each other very well, read they are not competing against each other.
From: The Life Blood Of Your Data Pipeline
Kafka is primarily related to holding log data rather than moving log
data. Thus, Kafka producers need to write the code to put data in
Kafka, and Kafka consumers need to write the code to pull data out of
Kafka.
Fluentd has both input and output plugins for Kafka so that data
engineers can write less code to get data in and out of Kafka. We have
many users that use Fluentd as a Kafka producer and/or consumer.

Can Spark be run alongside Kafka on the same node to optimize by keeping the Spark Streaming ETL processes closer to real time data?

I have gotten the advice, and, read in a few places, that running Spark on the data nodes greatly improves the performance of batch processing. I have also gotten the advice to keep the Kafka service isolated on dedicated nodes.
If most of the consumers of Kafka data are Spark Streaming ETL processes that land a transformed version of the data either back on Kafka or some other storage mechanism, does it then not make since to run those process on the same nodes i.e. run the Spark Service alongside the Kafka service on the Kafka dedicated cluster?
Thanks

Apache Kafka vs Apache Storm

Apache Kafka: Distributed messaging system
Apache Storm: Real Time Message Processing
How we can use both technologies in a real-time data pipeline for processing event data?
In terms of real time data pipeline both seems to me do the job identical. How can we use both the technologies on a data pipeline?
You use Apache Kafka as a distributed and robust queue that can handle high volume data and enables you to pass messages from one end-point to another.
Storm is not a queue. It is a system that has distributed real time processing abilities, meaning you can execute all kind of manipulations on real time data in parallel.
The common flow of these tools (as I know it) goes as follows:
real-time-system --> Kafka --> Storm --> NoSql --> BI(optional)
So you have your real time app handling high volume data, sends it to Kafka queue. Storm pulls the data from kafka and applies some required manipulation. At this point you usually like to get some benefits from this data, so you either send it to some Nosql db for additional BI calculations, or you could simply query this NoSql from any other system.
I know that this is an older thread and the comparisons of Apache Kafka and Storm were valid and correct when they were written but it is worth noting that Apache Kafka has evolved a lot over the years and since version 0.10 (April 2016) Kafka has included a Kafka Streams API which provides stream processing capabilities without the need for any additional software such as Storm. Kafka also includes the Connect API for connecting into various sources and sinks (destinations) of data.
Announcement blog - https://www.confluent.io/blog/introducing-kafka-streams-stream-processing-made-simple/
Current Apache documentation - https://kafka.apache.org/documentation/streams/
In 0.11 Kafka the stream processing functionality was further expanded to provide Exactly Once Semantics and Transactions.
https://www.confluent.io/blog/exactly-once-semantics-are-possible-heres-how-apache-kafka-does-it/
Kafka and Storm have a slightly different purpose:
Kafka is a distributed message broker which can handle big amount of messages per second. It uses publish-subscribe paradigm and relies on topics and partitions. Kafka uses Zookeeper to share and save state between brokers. So Kafka is basically responsible for transferring messages from one machine to another.
Storm is a scalable, fault-tolerant, real-time analytic system (think like Hadoop in realtime). It consumes data from sources (Spouts) and passes it to pipeline (Bolts). You can combine them in the topology. So Storm is basically a computation unit (aggregation, machine learning).
But you can use them together: for example your application uses kafka to send data to other servers which uses storm to make some computation on it.
This is how it works
Kafka - To provide a realtime stream
Storm - To perform some operations on that stream
You might take a look at the GitHub project https://github.com/abhishekgoel137/kafka-nodejs-d3js.
(D3js is a graph-representation library)
Ideal case:
Realtime application -> Kafka -> Storm -> NoSQL -> d3js
This repository is based on:
Realtime application -> Kafka -> <plain Node.js> -> NoSQL -> d3js
As every one explain you that
Apache Kafka: is continuous messaging queue
Apache Storm: is continuous processing tool
here in this aspect Kafka will get the data from any website like FB,Twitter by using API's and that data is processed by using Apache Storm and you can store the processed data in either in any databases you like.
https://github.com/miguno/kafka-storm-starter
Just follow it you will get some idea
When I have a use case that requires me to visualize or alert on patterns (think of twitter trends), while continuing to process the events, I have a several patterns.
NiFi would allow me to process an event and update a persistent data store with low(er) batch aggregation with very, very little custom coding.
Storm (lots of custom coding) allows me nearly real time access to the trending events.
If I can wait for many seconds, then I can batch out of kafka, into hdfs (Parquet) and process.
If I need to know in seconds, I need NiFi, and probably even Storm. (Think of monitoring thousands of earth stations, where I need to see small region weather conditions for tornado warnings).
Simply Kafka send the messages from node to another , and Storm processing the messages . Check this example of how you can Integration Apache Kafka With Storm

Websphere MQ and High Availability

When I read about HA in Websphere MQ I always come to the point, when the best practise is to create two Queue Managers handling the same queue and use out-of-the-box load balancing. Therefore, when one is down, the other takes over his job.
Well, this is great but what about the messages in the queue that belong to the Queue Manager that went down? I mean do these messages reside there (when queue is persistent of course) until QM is up and running again?
Furthermore, is it possible to create a common storage for this doubled Queue Managers? Then no message would wait for the QM to be up. Every message would be delivered in the proper order. Is this correct?
WebSphere MQ provides different capabilities for HA, depending on your requirements. WebSphere MQ clustering uses parallelism to distribute load across multiple instances of a queue. This provides availability of the service but not for in-flight messages.
Hardware clustering and Multi-Instance Queue Manager (MIQM) are both designs using multiple instances of a queue manager that see a single disk image of that queue manager's state. These provide availability of in-flight messages but the service is briefly unavailable while the cluster fails over.
Using these in combination it is possible to provide recovery of in-flight messages as well as availability of the service across multiple queue instances.
In hardware cluster model the disk is mounted to only one server and the cluster software monitors for failure and swaps the disk, IP address and possibly other resources to the secondary node. This requires a hardware cluster monitor such as PowerHA to manage the cluster.
The Multi-Instance QMgr is implemented entirely within WebSphere MQ and needs no other software. It works by having two running instances of the QMgr pointing to the same NFS 4 shared disk mount. Both instances compete for locks on the files. The first one to acquire a lock becomes the active QMgr. Because there is no hardware cluster monitor to perform IP address takeover this type of cluster will have multiple IP addresses. Any modern version of WMQ allows for this using multi-instance CONNAME where you can supply a comma-separated list of IP or DNS names. Client applications that previously used Client Channel Definition Tables (CCDT) to manage failover across multiple QMgrs will continue to work and CCDT continues to be supported in current versions of WMQ.
Please see the Infocenter topic Using WebSphere MQ with high availability configurations for details of hardware cluster and MIQM support.
Client Channel Definition Table files are discussed in the Infocenter topic Client Channel Definition Table file.

How To Load-Distribution in RabbitMQ cluster?

Hi I create three RabbitMQ servers running in cluster on EC2
I want to scale out RabbitMQ cluster base on CPU utilization but when I publish message only one server utilizes CPU and other RabbitMQ-server not utilize CPU
so how can i distribute the load across the RabbitMQ cluster
RabbitMQ clusters are designed to improve scalability, but the system is not completely automatic.
When you declare a queue on a node in a cluster, the queue is only created on that one node. So, if you have one queue, regardless to which node you publish, the message will end up on the node where the queue resides.
To properly use RabbitMQ clusters, you need to make sure you do the following things:
have multiple queues distributed across the nodes, such that work is distributed somewhat evenly,
connect your clients to different nodes (otherwise, you might end up funneling all messages through one node), and
if you can, try to have publishers/consumers connect to the node which holds the queue they're using (in order to minimize message transfers within the cluster).
Alternatively, have a look at High Availability Queues. They're like normal queues, but the queue contents are mirrored across several nodes. So, in your case, you would publish to one node, RabbitMQ will mirror the publishes to the other node, and consumers will be able to connect to either node without worrying about bogging down the cluster with internal transfers.
That is not really true. Check out the documentation on that subject.
Messages published to the queue are replicated to all mirrors. Consumers are connected to the master regardless of which node they connect to, with mirrors dropping messages that have been acknowledged at the master. Queue mirroring therefore enhances availability, but does not distribute load across nodes (all participating nodes each do all the work).

Resources