http://flink-forward.org/kb_sessions/beaming-flink-to-the-cloud-netflix/ Netflix is a data driven company and we process over 700 billion streaming events per day with at-least once processing semantics in the cloud. To enable extracting intelligence from this unbounded stream easily we are building Stream Processing as a Service (SPaaS) infrastructure so that the user can focus on extracting value and not have to worry about boilerplate infrastructure and scale. We will share our experience in building a scalable SPaaS using Flink, Apache Beam and Kafka as the foundation layer to process over 1.3 PB of event data without service disruption.

1. Monal Daxini
Engineering Manager
Stream Processing, Real Time Data Infrastructure
@monaldax @Netflix #keystone
Beaming Flink to the Cloud @ Netflix
Sep 12 2016

2. World’s Leading Internet Streaming Service
(Global launch Jan 6, 2016)

3. ● 83+ Million Members, 190+ Countries
● 1000+ device types
● 35% of downstream Internet traffic
Netflix Service Scale

4. Netflix Service Scale - Daily viewing hours
125,000,000,000+
Whoa!

5. Events Processed / day
1,000,000,000,000+
1.4 PB
That’s a huge number!

6. Event Scale
Peak
● 1T unique events ingested/day
● 16M / sec
● 43GB / sec
● 10MB / message
Daily Averages
● 1T+ events processed
● 600B unique events ingested
● 1.4 PB / day
● 4K / event

7. 99.99% + Availability / Four 9s
Keystone Scale

8. Keystone Events Trend
1/2014 80B / day 1/2015 300B / day 1/2016 1T+ / day

9. Evolution
SPaaS

10. Goal - Migrate 1.3 PB of event data to a new Pipeline
in flight, while ensuring data diff < 0.1%

11. Few years ago (Season 1)
EMR
Event
Producers

12. A year ago.. (Season 2)
Event
Producer
Stream
Consumers
EMR
Consumer
Kafka
Suro Router
Event
Producer
Suro
Kafka
Suro
Proxy

13. Season 3

14. Keystone
Keystone
Stream Processing
(SPaaS)
Keystone
Management
Keystone
Messaging
Schema Support
100% in AWS

15. Create DuploⓇ
Blocks:
Let reusability drive new value
Our Philosophy

16. Q4 2015
Event
Producers
Sinks
SPaaS

17. Keystone Management

18. Per Stream Auto Dashboard

19. Keystone SPaaS
Stream
Consumers
Router
EMR
Fronting
Kafka
Consumer
Kafka
Event
Producer
KSProxy
Control Plane
Self Service UI
100% in AWS
24 x 7
Region failover

20. Event flow
Keystone Pipeline As a Service

21. Keystone
Stream
Consumers
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
KSProxy
KSLib
Control Plane
Self Service UI

22. Keystone
Stream
Consumers
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
Control Plane
Self Service UI

23. Keystone
Stream
Consumers
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
Control Plane
Self Service UI

24. Keystone
Stream
Consumers
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
Control Plane
Self Service UI

25. Keystone
Stream
Consumers
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
Control Plane
Self Service UI

26. Routing

27. Keystone
Stream
Consumers
Samza
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
Control Plane
Self Service UI
Checkpoint
Cluster

28. Details..
○ Massively parallel use-case
■ Per element processing - declarative filtering &
projection
○ Stateless except Kafka offset checkpointing state

29. Routing Infrastructure
+
Checkpointing
Cluster
+ 0.9.1Go
C language

30. EC2 Instances
Zookeeper
(Instance Id assignment)
Job
Job
Job
NodeAgent
Checkpointing Cluster
ASG
Immutable Job Config
Self-Service UI /
Control Plane

31. Custom Go
Executor
./runJob
Logs
Snapshots
Attach Volumes
./runJob
./runJob
Reconcile Loop - 1 min
Health Check
What’s running on the host?
Zookeeper
(Instance Id assignment)

32. Yes! You inferred right!
No Mesos & No Yarn

33. Keystone Salient Features

34. Keystone Scale
● 4000+ Kafka brokers
● 14000+ Samza Jobs
○ Running in docker containers
○ On 1600+ nodes

35. Keystone Salient Features
● At-least-once delivery semantics*
● Multi-Tenant
● Self Service
● Scalable
● Fault Tolerant
100% in AWS

36. Keystone Salient Features
● Event payload is immutable
● Inject event metadata
○ guid, timestamp, host, app
● Custom extensible wire protocol
● Kafka producer wrapper

37. Keystone Salient Features
● Kafka Cluster failover
○ Kafka Kong
● Routing regional / failover
● Scales based on historical traffic
○ Externally driven

38. Season 4
Plot & Pilot...

39. Stream Processing As a Service (SPaaS)
(more capable)

40. SPaaS Vision (plot)
● Multi-tenant support for stateful stream processing apps
● Autoscaling managed infrastructure
● Support for schemas
● Self Service Tooling

41. SPaaS Architecture (plot)
SPaaS Manager
Titus Container
Runtime
Framework Specific API
or
Common API (Beam)
[ Dockerized Job ]
1. Create
2. Submit 3. Launch
Runner
Flink / Spark /
Mantis
Running Job
1. Submit Job DSL (SQL)
Tooling/ Dashboard

42. Why Apache Beam?
○ Portable API layer to build sophisticated data processing apps
■ Support multiple execution engines
○ Unified model API over bounded and unbounded data sources
○ Millwheel, FlumeJava, Dataflow model lineage
SPaaS - “Beam Me Up, Scotty ! "

43. Iterative build out: then
● First - Flink on Titus in VPC, AWS
○ Titus is a cloud runtime platform for container based jobs
● Next - Apache Beam and Flink runner
SPaaS - Pilot

44. 2.
Keystone SPaaS-Flink Pilot Use Cases
Stream
Consumers
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
Demux
MergeControl Plane
Self Service UI

45. 2.
1.
Keystone SPaaS-Flink Pilot Use Cases
Stream
Consumers
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
3.
Demux
MergeControl Plane
Self Service UI

46. Broker
Router - Massively parallel use case
Router

47. Titus Job
Task Manager
IP
Titus Host 4 Titus Host 5
Flink (1.2) Program Deployment (prod shadow)
Zookeeper
Job Manager
(standby)
Job Manager
(master)
Task Manager
Titus Host 1
IP
Titus Host 2
….
Task Manager
Titus Host 3
IP
Titus Job
IPIP
AWS
VPC
ENI

48. Titus High Level Architecture
Titus UITitus UI
Docker
Registry
Docker
Registry
Rhea
container
container
container
docker
Titus Agent
metrics agent
container
container
SPaaS-Flink
Titus executor
logging agent
zfsmesos agent
docker
RheaTitus API
Cassandra
Titus Master
Job Management
& Scheduler
S3
Zookeeper
Docker
Registry
EC2 Autoscaling
API
Mesos Master
Titus UI
(CI/CD)
Fenzo

49. Titus UI

50. CD

51. Dashboard

56. Keystone SPaaS-Flink Pilot Use Case - 2

57. Keystone SPaaS-Flink Pilot Use Case - 2

58. Flink Router perf test (YMMV)
○ Note
■ The tests were performed on a specific use case,
■ running in a specific environment, and with
■ with one specific event stream, and setup.

59. 1.
Keystone
Stream
Consumers
Samza
Router
EMR
Fronting
Kafka
Event
Producer
Consumer
Kafka
Control Plane
Self Service UI

60. Details..
○ Different runtimes for Flink & Samza routers
○ Massively parallel use-case
■ Per element processing
○ Focused on net outcomes

61. Titus Job
Task Manager
IP
Titus Host 4 Titus Host 5
Flink (1.2) Router
Zookeeper
Job Manager
(standby)
Job Manager
(master)
Task Manager
Titus Host 1
IP
Titus Host 2
….
Task Manager
Titus Host 3
IP
Titus Job
IPIP
AWS
VPC
ENI
Backed state

62. Flink Router outcome (YMMV)
○ Cost ≅17% savings
○ Memory utilization ≅16% better
○ Cpu utilization ≅ 40% better
○ Network utilization ≅ 10% better
○ Msg. throughput ≅ 1% (avg) - 4% (peak) better

63. Awaiting Flink Features

64. Titus Job
Task Manager
IP
Titus Host 4 Titus Host 5
Fine Grained Recovery FLIP-1
Zookeeper
Job Manager
(standby)
Job Manager
(master)
Task Manager
Titus Host 1
IP
Titus Host 2
….
Task Manager
Titus Host 3
IP
Titus Job
IPIP
X
Flink Job
Restarts

65. Awaiting Flink Features (now)
● Checkpoints and savepoints
○ FLINK-4484 - Unify, Persistent checkpoints, periodic savepoints
○ Compatible across Flink version upgrades
○ Inspection tool for debugging

66. Awaiting Flink Features (now)
● Atomic savepoint and stop (pause) the job
● Dynamic Scaling
○ Resume from savepoint with different parallelism
○ Job elasticity
● FLINK-4545 Adjust TaskManager network buffer when scaling
DONE

67. Awaiting Flink Features (now)
● Cluster Runtime and Elasticity (FLIP - 6)
● Extending Window Function Metadata (FLIP-2)
● Large State support
○ Incremental checkpointing
○ hot standby

68. Awaiting Flink Features
● Metric tags as key-value pairs - FLINK-4245
● FLIP-9 Window Trigger DSL
● FLIP-11 Table API
● Side Inputs / Side Outputs (handle late data)
DONE

69. Ponder over
Could Stream Processing Engine enable
building non-analytics Applications?

70. More brain food...
● Netflix Keystone Pipeline Evolution
● Netflix Kafka in Keystone Pipeline
● Samza Meetup Presentation
● Titus talk
● Netflix OSS