Kafka pours and Spark resolves! Alexey Zinovyev, Java/BigData Trainer in EPAM
Kafka pours and Spark
resolves!
Alexey Zinovyev, Java/BigData Trainer in EPAM
About
With IT since 2007
With Java since 2009
With Hadoop since 2012
With Spark since 2014
With EPAM since 2015
3Spark Streaming from Zinoviev Alexey
Contacts
E-mail : [email protected]
Twitter : @zaleslaw @BigDataRussia
Facebook: https://www.facebook.com/zaleslaw
vk.com/big_data_russia Big Data Russia
vk.com/java_jvm Java & JVM langs
4Spark Streaming from Zinoviev Alexey
Spark
Family
5Spark Streaming from Zinoviev Alexey
Spark
Family
6Spark Streaming from Zinoviev Alexey
Pre-summary
• Before RealTime
• Spark + Cassandra
• Sending messages with Kafka
• DStream Kafka Consumer
• Structured Streaming in Spark 2.1
• Kafka Writer in Spark 2.2
7Spark Streaming from Zinoviev Alexey
< REAL-TIME
8Spark Streaming from Zinoviev Alexey
Modern Java in 2016Big Data in 2014
9Spark Streaming from Zinoviev Alexey
Batch jobs produce reports. More and more..
10Spark Streaming from Zinoviev Alexey
But customer can wait forever (ok, 1h)
11Spark Streaming from Zinoviev Alexey
Big Data in 2017
12Spark Streaming from Zinoviev Alexey
Machine Learning EVERYWHERE
13Spark Streaming from Zinoviev Alexey
Data Lake in promotional brochure
14Spark Streaming from Zinoviev Alexey
Data Lake in production
15Spark Streaming from Zinoviev Alexey
Simple Flow in Reporting/BI systems
16Spark Streaming from Zinoviev Alexey
Let’s use Spark. It’s fast!
17Spark Streaming from Zinoviev Alexey
MapReduce vs Spark
18Spark Streaming from Zinoviev Alexey
MapReduce vs Spark
19Spark Streaming from Zinoviev Alexey
Simple Flow in Reporting/BI systems with Spark
20Spark Streaming from Zinoviev Alexey
Spark handles last year logs with ease
21Spark Streaming from Zinoviev Alexey
Where can we store events?
22Spark Streaming from Zinoviev Alexey
Let’s use Cassandra to store events!
23Spark Streaming from Zinoviev Alexey
Let’s use Cassandra to read events!
24Spark Streaming from Zinoviev Alexey
Cassandra
CREATE KEYSPACE mySpace WITH replication = {'class':
'SimpleStrategy', 'replication_factor': 1 };
USE test;
CREATE TABLE logs
( application TEXT,
time TIMESTAMP,
message TEXT,
PRIMARY KEY (application, time));
25Spark Streaming from Zinoviev Alexey
Cassandra
to Spark
val dataSet = sqlContext
.read
.format("org.apache.spark.sql.cassandra")
.options(Map( "table" -> "logs", "keyspace" -> "mySpace"
))
.load()
dataSet
.filter("message = 'Log message'")
.show()
26Spark Streaming from Zinoviev Alexey
Simple Flow in Pre-Real-Time systems
27Spark Streaming from Zinoviev Alexey
Spark cluster over Cassandra Cluster
28Spark Streaming from Zinoviev Alexey
More events every second!
29Spark Streaming from Zinoviev Alexey
SENDING MESSAGES
30Spark Streaming from Zinoviev Alexey
Your Father’s Messaging System
31Spark Streaming from Zinoviev Alexey
Your Father’s Messaging System
32Spark Streaming from Zinoviev Alexey
Your Father’s Messaging System
InitialContext ctx = new InitialContext();
QueueConnectionFactory f =
(QueueConnectionFactory)ctx.lookup(“qCFactory"); QueueConnection con =
f.createQueueConnection();
con.start();
33Spark Streaming from Zinoviev Alexey
KAFKA
34Spark Streaming from Zinoviev Alexey
Kafka
• messaging system
35Spark Streaming from Zinoviev Alexey
Kafka
• messaging system
• distributed
36Spark Streaming from Zinoviev Alexey
Kafka
• messaging system
• distributed
• supports Publish-Subscribe model
37Spark Streaming from Zinoviev Alexey
Kafka
• messaging system
• distributed
• supports Publish-Subscribe model
• persists messages on disk
38Spark Streaming from Zinoviev Alexey
Kafka
• messaging system
• distributed
• supports Publish-Subscribe model
• persists messages on disk
• replicates within the cluster (integrated with Zookeeper)
39Spark Streaming from Zinoviev Alexey
The main benefits of Kafka
Scalability with zero down time
Zero data loss due to replication
40Spark Streaming from Zinoviev Alexey
Kafka Cluster consists of …
• brokers (leader or follower)
• topics ( >= 1 partition)
• partitions
• partition offsets
• replicas of partition
• producers/consumers
41Spark Streaming from Zinoviev Alexey
Kafka Components with topic “messages” #1
Producer
Thread #1
Producer
Thread #2
Producer
Thread #3
Topic: MessagesData
Data
Data
Zookeeper
42Spark Streaming from Zinoviev Alexey
Kafka Components with topic “messages” #2
Producer
Thread #1
Producer
Thread #2
Producer
Thread #3
Topic: Messages
Part #1
Part #2
Data
Data
Data
Broker #1
Zookeeper
Part #1
Part #2
Leader
Leader
43Spark Streaming from Zinoviev Alexey
Kafka Components with topic “messages” #3
Producer
Thread #1
Producer
Thread #2
Producer
Thread #3
Topic: Messages
Part #1
Part #2
Data
Data
Data
Broker #1
Broker #2
Zookeeper
Part #1
Part #1
Part #2
Part #2
44Spark Streaming from Zinoviev Alexey
Why do we need Zookeeper?
45Spark Streaming from Zinoviev Alexey
Kafka
Demo
46Spark Streaming from Zinoviev Alexey
REAL TIME WITH
DSTREAMS
47Spark Streaming from Zinoviev Alexey
RDD Factory
48Spark Streaming from Zinoviev Alexey
From socket to console with DStreams
49Spark Streaming from Zinoviev Alexey
DStream
val conf = new SparkConf().setMaster("local[2]")
.setAppName("NetworkWordCount")
val ssc = new StreamingContext(conf, Seconds(1))
val lines = ssc.socketTextStream("localhost", 9999)
val words = lines.flatMap(_.split(" "))
val pairs = words.map(word => (word, 1))
val wordCounts = pairs.reduceByKey(_ + _)
wordCounts.print()
ssc.start()
ssc.awaitTermination()
50Spark Streaming from Zinoviev Alexey
DStream
val conf = new SparkConf().setMaster("local[2]")
.setAppName("NetworkWordCount")
val ssc = new StreamingContext(conf, Seconds(1))
val lines = ssc.socketTextStream("localhost", 9999)
val words = lines.flatMap(_.split(" "))
val pairs = words.map(word => (word, 1))
val wordCounts = pairs.reduceByKey(_ + _)
wordCounts.print()
ssc.start()
ssc.awaitTermination()
51Spark Streaming from Zinoviev Alexey
DStream
val conf = new SparkConf().setMaster("local[2]")
.setAppName("NetworkWordCount")
val ssc = new StreamingContext(conf, Seconds(1))
val lines = ssc.socketTextStream("localhost", 9999)
val words = lines.flatMap(_.split(" "))
val pairs = words.map(word => (word, 1))
val wordCounts = pairs.reduceByKey(_ + _)
wordCounts.print()
ssc.start()
ssc.awaitTermination()
52Spark Streaming from Zinoviev Alexey
DStream
val conf = new SparkConf().setMaster("local[2]")
.setAppName("NetworkWordCount")
val ssc = new StreamingContext(conf, Seconds(1))
val lines = ssc.socketTextStream("localhost", 9999)
val words = lines.flatMap(_.split(" "))
val pairs = words.map(word => (word, 1))
val wordCounts = pairs.reduceByKey(_ + _)
wordCounts.print()
ssc.start()
ssc.awaitTermination()
53Spark Streaming from Zinoviev Alexey
DStream
val conf = new SparkConf().setMaster("local[2]")
.setAppName("NetworkWordCount")
val ssc = new StreamingContext(conf, Seconds(1))
val lines = ssc.socketTextStream("localhost", 9999)
val words = lines.flatMap(_.split(" "))
val pairs = words.map(word => (word, 1))
val wordCounts = pairs.reduceByKey(_ + _)
wordCounts.print()
ssc.start()
ssc.awaitTermination()
54Spark Streaming from Zinoviev Alexey
Kafka as a main entry point for Spark
55Spark Streaming from Zinoviev Alexey
DStreams
Demo
56Spark Streaming from Zinoviev Alexey
How to avoid DStreams with RDD-like API?
57Spark Streaming from Zinoviev Alexey
SPARK 2.2 DISCUSSION
58Spark Streaming from Zinoviev Alexey
Continuous Applications
59Spark Streaming from Zinoviev Alexey
Continuous Applications cases
• Updating data that will be served in real time
• Extract, transform and load (ETL)
• Creating a real-time version of an existing batch job
• Online machine learning
60Spark Streaming from Zinoviev Alexey
The main concept of Structured Streaming
You can express your streaming computation the
same way you would express a batch computation
on static data.
61Spark Streaming from Zinoviev Alexey
Batch
Spark 2.2
// Read JSON once from S3
logsDF = spark.read.json("s3://logs")
// Transform with DataFrame API and save
logsDF.select("user", "url", "date")
.write.parquet("s3://out")
62Spark Streaming from Zinoviev Alexey
Real
Time
Spark 2.2
// Read JSON continuously from S3
logsDF = spark.readStream.json("s3://logs")
// Transform with DataFrame API and save
logsDF.select("user", "url", "date")
.writeStream.parquet("s3://out")
.start()
63Spark Streaming from Zinoviev Alexey
WordCount
from
Socket
val lines = spark.readStream
.format("socket")
.option("host", "localhost")
.option("port", 9999)
.load()
val words = lines.as[String].flatMap(_.split(" "))
val wordCounts = words.groupBy("value").count()
64Spark Streaming from Zinoviev Alexey
WordCount
from
Socket
val lines = spark.readStream
.format("socket")
.option("host", "localhost")
.option("port", 9999)
.load()
val words = lines.as[String].flatMap(_.split(" "))
val wordCounts = words.groupBy("value").count()
65Spark Streaming from Zinoviev Alexey
WordCount
from
Socket
val lines = spark.readStream
.format("socket")
.option("host", "localhost")
.option("port", 9999)
.load()
val words = lines.as[String].flatMap(_.split(" "))
val wordCounts = words.groupBy("value").count()
66Spark Streaming from Zinoviev Alexey
WordCount
from
Socket
val lines = spark.readStream
.format("socket")
.option("host", "localhost")
.option("port", 9999)
.load()
val words = lines.as[String].flatMap(_.split(" "))
val wordCounts = words.groupBy("value").count()
Don’t forget
to start
Streaming
67Spark Streaming from Zinoviev Alexey
Unlimited Table
68Spark Streaming from Zinoviev Alexey
WordCount with Structured Streaming [Complete Mode]
69Spark Streaming from Zinoviev Alexey
Kafka -> Structured Streaming -> Console
70Spark Streaming from Zinoviev Alexey
Kafka To
Console
Demo
71Spark Streaming from Zinoviev Alexey
OPERATIONS
72Spark Streaming from Zinoviev Alexey
You can …
• filter
• sort
• aggregate
• join
• foreach
• explain
73Spark Streaming from Zinoviev Alexey
Operators
Demo
74Spark Streaming from Zinoviev Alexey
How it works?
75Spark Streaming from Zinoviev Alexey
Deep Diving in Spark Internals
Dataset
76Spark Streaming from Zinoviev Alexey
Deep Diving in Spark Internals
Dataset Logical Plan
77Spark Streaming from Zinoviev Alexey
Deep Diving in Spark Internals
Dataset Logical PlanOptimized
Logical Plan
78Spark Streaming from Zinoviev Alexey
Deep Diving in Spark Internals
Dataset Logical PlanOptimized
Logical PlanLogical Plan
Logical PlanLogical PlanPhysical
Plan
79Spark Streaming from Zinoviev Alexey
Deep Diving in Spark Internals
Dataset Logical PlanOptimized
Logical PlanLogical Plan
Logical PlanLogical PlanPhysical
Plan
Selected
Physical Plan
Planner
80Spark Streaming from Zinoviev Alexey
Deep Diving in Spark Internals
Dataset Logical PlanOptimized
Logical PlanLogical Plan
Logical PlanLogical PlanPhysical
Plan
Selected
Physical Plan
81Spark Streaming from Zinoviev Alexey
Deep Diving in Spark Internals
Dataset Logical PlanOptimized
Logical PlanLogical Plan
Logical PlanLogical PlanPhysical
Plan
Selected
Physical Plan
Incremental
#1
Incremental
#2
Incremental
#3
82Spark Streaming from Zinoviev Alexey
Incremental Execution: Planner polls
Planner
83Spark Streaming from Zinoviev Alexey
Incremental Execution: Planner runs
Planner
Offset: 1 - 5Incremental
#1
R
U
N
Count: 5
84Spark Streaming from Zinoviev Alexey
Incremental Execution: Planner runs #2
Planner
Offset: 1 - 5Incremental
#1
Incremental
#2Offset: 6 - 9
R
U
N
Count: 5
Count: 4
85Spark Streaming from Zinoviev Alexey
Aggregation with State
Planner
Offset: 1 - 5Incremental
#1
Incremental
#2Offset: 6 - 9
R
U
N
Count: 5
Count: 5+4=9
5
86Spark Streaming from Zinoviev Alexey
DataSet.explain()
== Physical Plan ==Project [avg(price)#43,carat#45]+- SortMergeJoin [color#21], [color#47]
:- Sort [color#21 ASC], false, 0: +- TungstenExchange hashpartitioning(color#21,200), None: +- Project [avg(price)#43,color#21]: +- TungstenAggregate(key=[cut#20,color#21], functions=[(avg(cast(price#25 as
bigint)),mode=Final,isDistinct=false)], output=[color#21,avg(price)#43]): +- TungstenExchange hashpartitioning(cut#20,color#21,200), None: +- TungstenAggregate(key=[cut#20,color#21],
functions=[(avg(cast(price#25 as bigint)),mode=Partial,isDistinct=false)], output=[cut#20,color#21,sum#58,count#59L])
: +- Scan CsvRelation(-----)+- Sort [color#47 ASC], false, 0
+- TungstenExchange hashpartitioning(color#47,200), None+- ConvertToUnsafe
+- Scan CsvRelation(----)
87Spark Streaming from Zinoviev Alexey
What’s the difference between
Complete and Append
output modes?
88Spark Streaming from Zinoviev Alexey
COMPLETE, APPEND &
UPDATE
89Spark Streaming from Zinoviev Alexey
There are two main modes and one in future
• append (default)
90Spark Streaming from Zinoviev Alexey
There are two main modes and one in future
• append (default)
• complete
91Spark Streaming from Zinoviev Alexey
There are two main modes and one in future
• append (default)
• complete
• update [in dreams]
92Spark Streaming from Zinoviev Alexey
Aggregation with watermarks
93Spark Streaming from Zinoviev Alexey
SOURCES & SINKS
94Spark Streaming from Zinoviev Alexey
Spark Streaming is a brick in the Big Data Wall
95Spark Streaming from Zinoviev Alexey
Let's save to Parquet files
96Spark Streaming from Zinoviev Alexey
Let's save to Parquet files
97Spark Streaming from Zinoviev Alexey
Let's save to Parquet files
98Spark Streaming from Zinoviev Alexey
Can we write to Kafka?
99Spark Streaming from Zinoviev Alexey
Nightly Build
100Spark Streaming from Zinoviev Alexey
Kafka-to-Kafka
101Spark Streaming from Zinoviev Alexey
J-K-S-K-S-C
102Spark Streaming from Zinoviev Alexey
Pipeline
Demo
103Spark Streaming from Zinoviev Alexey
I didn’t find sink/source for XXX…
104Spark Streaming from Zinoviev Alexey
Console
Foreach
Sink
import org.apache.spark.sql.ForeachWriter
val customWriter = new ForeachWriter[String] {
override def open(partitionId: Long, version: Long) = true
override def process(value: String) = println(value)
override def close(errorOrNull: Throwable) = {}
}
stream.writeStream
.queryName(“ForeachOnConsole")
.foreach(customWriter)
.start
105Spark Streaming from Zinoviev Alexey
Pinch of wisdom
• check checkpointLocation
• don’t use MemoryStream
• think about GC pauses
• be careful about nighty builds
• use .groupBy.count() instead count()
• use console sink instead .show() function
106Spark Streaming from Zinoviev Alexey
We have no ability…
• join two streams
• work with update mode
• make full outer join
• take first N rows
• sort without pre-aggregation
107Spark Streaming from Zinoviev Alexey
Roadmap 2.2
• Support other data sources (not only S3 + HDFS)
• Transactional updates
• Dataset is one DSL for all operations
• GraphFrames + Structured MLLib
• KafkaWriter
• TensorFrames
108Spark Streaming from Zinoviev Alexey
IN CONCLUSION
109Spark Streaming from Zinoviev Alexey
Final point
Scalable Fault-Tolerant Real-Time Pipeline with
Spark & Kafka
is ready for usage
110Spark Streaming from Zinoviev Alexey
A few papers about Spark Streaming and Kafka
Introduction in Spark + Kafka
http://bit.ly/2mJjE4i
111Spark Streaming from Zinoviev Alexey
Source code from Demo
Spark Tutorial
https://github.com/zaleslaw/Spark-Tutorial
112Spark Streaming from Zinoviev Alexey
Contacts
E-mail : [email protected]
Twitter : @zaleslaw @BigDataRussia
Facebook: https://www.facebook.com/zaleslaw
vk.com/big_data_russia Big Data Russia
vk.com/java_jvm Java & JVM langs
113Spark Streaming from Zinoviev Alexey
Any questions?