Big, Fast, Easy Data: Distributed Stream Processing for Everyone with KSQL, the Streaming SQL Engine for Apache Kafka (Berlin Buzzwords 2018)

Big, Fast, Easy Data:
Distributed stream processing
for everyone with KSQL
The Streaming SQL Engine for Apache Kafka
Michael G. Noll, Confluent
@miguno

Founded by the creators
of Apache Kafka
Technology Developed
while at LinkedIn
Largest Contributor and
tester of Apache Kafka
• Founded in 2014
• Raised $84M from Benchmark, Index, Sequoia
• Transacting in 20 countries
• Commercial entities in US, UK, Germany, Australia

Apache Kafka Databases
SQLStream Processing

Booked hotel, flight Ordered a taxi
Chatted with friends
Listened to musicPaid money
Played a video game
Read a newspaper <add your example>

Billing Information
Purchases
Geolocation Updates
And more such data
STREAMS of
customer data
(continuously flowing)
TABLE of
customer profiles
(continuously updated)
Motivating example

KSQLis the
Streaming
SQL Engine
for
Apache Kafka
5+5

KSQL is the Easiest Way to Process with Kafka
Kafka
(data)
KSQL
(processing)
read,
write
network
All you need is Kafka – no complex deployments of
bespoke systems for stream processing!
CREATE STREAM
CREATE TABLE
SELECT
…and more…

KSQL is the Easiest Way to Process with Kafka
Runs
Everywhere
Elastic, Scalable,
Fault-Tolerant,
Distributed, S/M/L/XL
Powerful Processing incl.
Filters, Transforms, Joins,
Aggregations, Windowing
Supports Streams
and Tables
Free and
Open Source
Kafka Security
Integration
Event-Time
Processing
Zero Programming
in Java, Scala
0
Exactly-Once
Processing

Stream processing with Kafka
Example: Using Kafka’s Streams API for writing
elastic, scalable, fault-tolerant Java and Scala applications
Main
Logic

Stream processing with Kafka
CREATE STREAM fraudulent_payments AS
SELECT * FROM payments
WHERE fraudProbability > 0.8;
Same example, now with KSQL.
Not a single line of Java or Scala code needed.

Easier, faster workflow
write code in
package app
run app
write (K)SQL
Java or Scala
ksql>
Kafka Streams API KSQL
…
(1 or many instances)

Interactive KSQL usage
ksql> POST /query
CLI REST API1 3UI2

KSQL REST API example
POST /query HTTP/1.1
{
"ksql": "SELECT * FROM users WHERE name LIKE ‘a%’;"
"streamsProperties": {
"your.custom.setting": "value"
}
}
Here: run a query and stream back the results

KSQL
are some
what
use cases?
10+5

KSQL for Data Exploration
SELECT page, user_id, status, bytes
FROM clickstream
WHERE user_agent LIKE 'Mozilla%';
An easy way to inspect data in Kafka
SHOW TOPICS;
PRINT 'my-topic' FROM BEGINNING;

KSQL for Data Enrichment
CREATE STREAM enriched_payments AS
SELECT payment_id, u.country, total
FROM payments_stream p
LEFT JOIN users_table u
ON p.user_id = u.user_id;
Join data from a variety of sources to see the full picture
1 Stream-table join

KSQL for Streaming ETL
CREATE STREAM clicks_from_vip_users AS
SELECT user_id, u.country, page, action
FROM clickstream c
LEFT JOIN users u ON c.user_id = u.user_id
WHERE u.level ='Platinum';
Filter, cleanse, process data while it is moving

KSQL for Anomaly Detection
CREATE TABLE possible_fraud AS
SELECT card_number, COUNT(*)
FROM authorization_attempts
WINDOW TUMBLING (SIZE 30 SECONDS)
GROUP BY card_number
HAVING COUNT(*) > 3;
Aggregate data to identify patterns or anomalies in real-time
2 … per 30sec windows
1 Aggregate data

KSQL for Real-Time Monitoring
CREATE TABLE failing_vehicles AS
SELECT vehicle, COUNT(*)
FROM vehicle_telemetry_stream
WINDOW TUMBLING (SIZE 1 MINUTE)
WHERE event_type = 'ERROR’
GROUP BY vehicle
HAVING COUNT(*) >= 3;
Derive insights from events (IoT, sensors, etc.) and turn them into actions

KSQL for Data Transformation
CREATE STREAM clicks_by_user_id
WITH (PARTITIONS=6,
TIMESTAMP='view_time’
VALUE_FORMAT='JSON') AS
SELECT * FROM clickstream
PARTITION BY user_id;
Quickly make derivations of existing data in Kafka
1 Re-partition the data
2 Convert data to JSON

Where is KSQL not such a great fit?
BI reports
• Because no indexes
• No JDBC (most BI tools are not good
with continuous results!)
Ad-hoc queries
• Because no indexes
to facilitate efficient
random lookups on
arbitrary record fields

Shoulders of Streaming Giants
Consumer,
Producer
KSQL
Kafka Streams
powers
powers
Flexibility
Ease of Use
CREATE STREAM, CREATE TABLE,
SELECT, JOIN, GROUP BY, SUM, …
KStream, KTable,
filter(), map(), flatMap(),
join(), aggregate(), …
subscribe(), poll(), send(),
flush(), beginTransaction(), …

Shoulders of Streaming Giants
CREATE STREAM fraudulent_payments AS
SELECT * FROM payments
WHERE fraudProbability > 0.8;
KSQL
Kafka
Streams

$ ksql-server-start
KSQL Architecture
KSQL
Engine
REST
API
Processing happens here,
powered by Kafka Streams
ksql>
Programmatic access from
Go, Python, .NET, Java,
JavaScript, …
UI
CLI
KSQL Server (JVM process)
Physical
…

Runs Everywhere, Viable for S/M/L/XL Use Cases
Physical
…and many more…

KSQL Architecture
Kafka
(your data)
KSQL
read,
write
…
More KSQL
…
FraudTeam
…
MobileTeam
KSQLCluster
Servers form a
Kafka consumer group
to process data
collaboratively
network

KSQL Interactive Usage
Start 1+ KSQL servers
$ ksql-server-start
Interact with
KSQL CLI, UI, etc.
$ ksql http://ksql-server:8088
ksql>
REST API

KSQL Headless, Non-Interactive Usage
$ ksql-server-start --queries-file application.sql
ksql>
Typically version
controlled for auditing,
rollbacks, etc.
REST API
disabled
Start 1+ KSQL servers with .sql file containing pre-defined queries.

Example Journey from Idea to Production
Interactive KSQL
for development and testing
Headless KSQL
for Production
Desired KSQL queries
have been identified
and vetted
REST
“Hmm, let me try
out this idea...”

Stream-Table
The
Duality
22+15

Stream-Table Duality
CREATE STREAM enriched_payments AS
SELECT payment_id, u.country, total
FROM payments_stream p
LEFT JOIN users_table u
ON p.user_id = u.user_id;
CREATE TABLE failing_vehicles AS
SELECT vehicle, COUNT(*)
FROM vehicle_monitoring_stream
WINDOW TUMBLING (SIZE 1 MINUTE)
WHERE event_type = 'ERROR’
GROUP BY vehicle
HAVING COUNT(*) >= 3;
Stream Table
(from previous slides)

Do you think that’s a table you are querying ?

Stream Table
Alice 1
Alice 1
Charlie 5
Alice 3
Charlie 5
(Alice, 1)
(Charlie, 5)
(Alice, 3)
Alice 1
Alice 1
Charlie 5
Alice 3
Charlie 5
Table
https://www.confluent.io/blog/introducing-kafka-streams-stream-processing-made-simple/
https://www.michael-noll.com/blog/2018/04/05/of-stream-and-tables-in-kafka-and-stream-processing-part1/

CREATE TABLE current_location_per_user
WITH (KAFKA_TOPIC='input-topic’, ...);
This is actually an animation, but the
PDF format does not support this.

CREATE TABLE visited_locations_per_user AS
SELECT username, COUNT(*)
FROM location_updates
GROUP BY username;
This is actually an animation, but the
PDF format does not support this.

aggregation
changelog
“materialized view”
of the stream
(like SUM, COUNT)
Stream Table
(CDC)

Apache Kafka Databases

How you benefit from this as a KSQL user.

Example: CDC from DB via Kafka to Elastic
customers
Kafka Connect
streams data in
Kafka Connect
streams data out
KSQL processes
table changes
in real-time

Example: Real-time Data Enrichment
Kafka Connect
streams data in
<wherever>
Kafka Connect
streams data out
Devices write
directly via
Kafka API
KSQL joins the stream
and table in real-time
customers

How KSQL itself benefits from this – a closer technical look

Fault-Tolerance, powered by Kafka
Server A:
“I do stateful stream
processing, like tables,
joins, aggregations.”
“streaming
restore” of
A’s local state to BChangelog Topic
“streaming
backup” of
A’s local state
KSQL
Kafka
A key challenge of distributed stream processing is fault-tolerant state.
State is automatically migrated
in case of server failure
Server B:
“I restore the state and
continue processing where
server A stopped.”

Fault-Tolerance, powered by Kafka
Processing fails over automatically, without data loss or miscomputation.
1 Kafka consumer group
rebalance is triggered
2 Processing and state of #3
is migrated via Kafka to
remaining servers #1 + #2
#3 died so #1 and #2 take over
2 Part of processing incl.
state is migrated via Kafka
from #1 + #2 to server #3
#3 is back so the work is split again

Elasticity and Scalability, powered by Kafka
You can add, remove, restart servers in KSQL clusters during live operations.
2 Part of processing incl.
state is migrated via Kafka
to additional server processes
“We need more processing power!”
Kafka consumer group
1
2 Processing incl. state of
stopped servers is migrated
via Kafka to remaining servers
“Ok, we can scale down again.”

Want to take a deeper dive?
https://kafka.apache.org/documentation/streams/architecture
KSQL is built on top of Kafka Streams:
Read up on Kafka Streams’ architecture
including threading model, elasticity,
fault-tolerance, state stores for stateful
computation, etc. to learn more about how
all this works behind the scenes.

KSQLis the
Streaming
SQL Engine
for
Apache Kafka

Where to go from here
http://confluent.io/ksql
https://slackpass.io/confluentcommunity #ksql
https://github.com/confluentinc/ksql

Big, Fast, Easy Data: Distributed Stream Processing for Everyone with KSQL, the Streaming SQL Engine for Apache Kafka (Berlin Buzzwords 2018)

More Related Content

What's hot (20)

Similar to Big, Fast, Easy Data: Distributed Stream Processing for Everyone with KSQL, the Streaming SQL Engine for Apache Kafka (Berlin Buzzwords 2018) (20)

Recently uploaded (20)

Big, Fast, Easy Data: Distributed Stream Processing for Everyone with KSQL, the Streaming SQL Engine for Apache Kafka (Berlin Buzzwords 2018)