Flink at Big Data London 2017

1. 1
Kostas Tzoumas
@kostas_tzoumas
Big Data Ldn
November 4, 2016
Stream Processing with Apache
Flink®

2. 2
Kostas Tzoumas
@kostas_tzoumas
Big Data Ldn
November 4, 2016
Debunking Some Common Myths in
Stream Processing

3. 3
Original creators of Apache
Flink®
Providers of the
dA Platform, a supported
Flink distribution

4. Outline
 What is data streaming
 Myth 1: The throughput/latency tradeoff
 Myth 2: Exactly once not possible
 Myth 3: Streaming is for (near) real-time
 Myth 4: Streaming is hard
4

5. The streaming architecture
5

6. 6
Reconsideration of data architecture
 Better app isolation
 More real-time reaction to events
 Robust continuous applications
 Process both real-time and historical data

7. 7
app state
app state
app state
event log
Query
service

8. What is (distributed) streaming
 Computations on never-
ending “streams” of data
records (“events”)
 Stream processor
distributes the
computation in a cluster
8
Your
code
Your
code
Your
code
Your
code

9. What is stateful streaming
 Computation and state
• E.g., counters, windows of past
events, state machines, trained ML
models
 Result depends on history of
stream
 Stateful stream processor gives
the tools to manage state
• Recover, roll back, version,
upgrade, etc
9
Your
code
state

10. What is event-time streaming
 Data records associated with
timestamps (time series data)
 Processing depends on timestamps
 Event-time stream processor gives
you the tools to reason about time
• E.g., handle streams that are out of
order
• Core feature is watermarks – a clock
to measure event time
10
Your
code
state
t3 t1 t2t4 t1-t2 t3-t4

11. What is streaming
 Continuous processing on data that is
continuously generated
 I.e., pretty much all “big” data
 It’s all about state and time
11

12. Debunking some common stream
processing myths
12

13. Myth 1: Throughput/latency tradeoff
 Myth 1: you need to choose between high
throughput or low latency
 Physical limits
• In reality, network determines both the achievable
throughput and latency
• A well-engineered system achieves these limits
13

14. Flink performance
 10s of millions events per seconds in 10s of nodes
 scaled to 1000s of nodes
 with latency in single-digit milliseconds
14

15. Myth 2: Exactly once not possible
 Exactly once: under failures, system computes result
as if there was no failure
 In contrast to:
• At most once: no guarantees
• At least once: duplicates possible
 Exactly once state versus exactly once delivery
 Myth 2: Exactly once state not possible/too costly
15

16. Transactions
 “Exactly once” is transactions: either all
actions succeed or none succeed
 Transactions are possible
 Transactions are useful
 Let’s not start eventual consistency all over
again…
16

17. Flink checkpoints
 Periodic asynchronous consistent snapshots of
application state
 Provide exactly-once state guarantees under failures
17
9/2/2016 stream_barriers.svg
checkpoint
barrier n­1
data stream
stream record
(event)
checkpoint
barrier n
newer records
part of
checkpoint n­1
part of
checkpoint n
part of
checkpoint n+1
older records

18. End-to-end exactly once
 Checkpoints double as transaction coordination mechanism
 Source and sink operators can take part in checkpoints
 Exactly once internally, "effectively once" end to end: e.g.,
Flink + Cassandra with idempotent updates
18
transactional sinks

19. State management
 Checkpoints triple as state
versioning mechanism
(savepoints)
 Go back and forth in time while
maintaining state consistency
 Ease code upgrades (Flink or
app), maintenance, migration,
and debugging, what-if
simulations, A/B tests
19

20. Myth 3: Streaming and real time
 Myth 3: streaming and real-time are
synonymous
 Streaming is a new model
• Essentially, state and time
• Low latency/real time is the icing on the cake
20

21. Low latency and high latency streams
21
2016-3-1
12:00 am
2016-3-1
1:00 am
2016-3-1
2:00 am
2016-3-11
11:00pm
2016-3-12
12:00am
2016-3-12
1:00am
2016-3-11
10:00pm
2016-3-12
2:00am
2016-3-12
3:00am…
partition
partition
Stream (low latency)
Batch
(bounded stream)
Stream (high latency)

22. Robust continuous applications
22

23. Accurate computation
 Batch processing is not an accurate
computation model for continuous data
• Misses the right concepts and primitives
• Time handling, state across batch boundaries
 Stateful stream processing a better model
• Real-time/low-latency is the icing on the cake
23

24. Myth 4: How hard is streaming?
 Myth 4: streaming is too hard to learn
 You are already doing streaming, just in an
ad hoc way
 Most data is unbounded and the code
changes slower than the data
• This is a streaming problem
24

25. It's about your data and code
 What's the form of your data?
• Unbounded (e.g., clicks, sensors, logs), or
• Bounded (e.g., ???*)
 What changes more often?
• My code changes faster than my data
• My data changes faster than my code
25
* Please help me find a great example of naturally bounded data

26. It's about your data and code
 If your data changes faster than your code
you have a streaming problem
• You may be solving it with hourly batch jobs
depending on someone else to create the
hourly batches
• You are probably living with inaccurate results
without knowing it
26

27. It's about your data and code
 If your code changes faster than your data
you have an exploration problem
• Using notebooks or other tools for quick data
exploration is a good idea
• Once your code stabilizes you will have a
streaming problem, so you might as well think
of it as such from the beginning
27

28. Flink in the real world
28

29. Flink community
 > 240 contributors, 95 contributors in Flink 1.1
 42 meetups around the world with > 15,000 members
 2x-3x growth in 2015, similar in 2016
29

30. Powered by Flink
30
Zalando, one of the largest ecommerce
companies in Europe, uses Flink for real-
time business process monitoring.
King, the creators of Candy Crush Saga,
uses Flink to provide data science teams
with real-time analytics.
Bouygues Telecom uses Flink for real-time
event processing over billions of Kafka
messages per day.
Alibaba, the world's largest retailer, built a
Flink-based system (Blink) to optimize
search rankings in real time.
See more at flink.apache.org/poweredby.html

31. 30 Flink applications in production for more than one
year. 10 billion events (2TB) processed daily
Complex jobs of > 30 operators running 24/7,
processing 30 billion events daily, maintaining state
of 100s of GB with exactly-once guarantees
Largest job has > 20 operators, runs on > 5000
vCores in 1000-node cluster, processes millions of
events per second
31

32. 32

33. Flink Forward 2016

34. Current work in Flink
34

35. Ongoing Flink development
35
Connectors
Session
Windows
(Stream) SQL
Library
enhancements
Metric
System
Operations
Ecosystem
Application
Features
Metrics &
Visualization
Dynamic Scaling
Savepoint
compatibility Checkpoints
to savepoints
More connectors Stream SQL
Windows
Large state
Maintenance
Fine grained
recovery
Side in-/outputs
Window DSL
Broader
Audience
Security
Mesos &
others
Dynamic Resource
Management
Authentication
Queryable State

36. A longer-term vision for Flink
36

37. Streaming use cases
Application
(Near) real-time apps
Continuous apps
Analytics on historical
data
Request/response apps
Technology
Low-latency streaming
High-latency streaming
Batch as special case of
streaming
Large queryable state
37

38. Request/response applications
 Queryable state: query Flink state directly instead
of pushing results in a database
 Large state support and query API coming in Flink
38
queries

39. In summary
 The need for streaming comes from a rethinking of
data infra architecture
• Stream processing then just becomes natural
 Debunking 4 common myths
• Myth 1: The throughput/latency tradeoff
• Myth 2: Exactly once not possible
• Myth 3: Streaming is for (near) real-time
• Myth 4: Streaming is hard
39

40. 4
Thank you!
@kostas_tzoumas
@ApacheFlink
@dataArtisans

41. 4
We are hiring!
data-artisans.com/careers