Unified Data Processing with Apache Flink and Apache Pulsar_Seth Wiesman
•
1 like•893 views
Come learn how the combination of Apache Pulsar and Apache Flink is making stateful stream processing even more expressive and flexible to support applications in streaming that were previously not considered streamable. The new world of applications and fast data architectures has broken up the database: Raw data persistence comes in the form of event logs, and the state of the world is computed by a stream processor. Apache Pulsar provides a strong solution for the event log, while Apache Flink forms a powerful foundation for the computation over the event streams. We will discuss the key concepts behind Apache Flink's approach to stream processing and how it is a powerful abstraction for stateful event-driven applications. We will then see how to use Flink in conjunction with Apache Pulsar to creates a unified data processing platform.
Report
Share
![© 2019 Ververica29
DataStream API
Source
Transformation
Windowed Transformation
Sink
val lines: DataStream[String] = env.addSource(new FlinkKafkaConsumer011(…))
val events: DataStream[Event] = lines.map((line) => parse(line))
val stats: DataStream[Statistic] = stream
.keyBy("sensor")
.timeWindow(Time.seconds(5))
.sum(new MyAggregationFunction())
stats.addSink(new RollingSink(path))
Streaming
Dataflow
Source Transform Window
(state read/write)
Sink](https://cdn.statically.io/img/image.slidesharecdn.com/sethweismanday2keynote-200624090725/85/Unified-Data-Processing-with-Apache-Flink-and-Apache-Pulsar_Seth-Wiesman-29-320.jpg)
Unified Data Processing with Apache Flink and Apache Pulsar_Seth Wiesman
- 1. © 2019 Ververica Seth Wiesman Senior Solutions Architect @ Ververica Committer Apache Flink Unified Data Processing with Apache Flink and Apache Pulsar
- 2. © 2019 Ververica2 About Ververica (the company formerly known as “data Artisans”) Original Creators of Apache Flink® Enterprise Stream Processing With Ververica Platform Subsidiary of Alibaba Group
- 3. © 2019 Ververica3 Apache Flink
- 4. © 2019 Ververica4 Apache Flink
- 5. © 2019 Ververica5 2.5 B2M 985 PB Sub- Second 100TB containers data size throughput latency state size events / sec Apache Flink at The "Singles Day" (11/11/2019)
- 6. © 2019 Ververica6 Why Stream Processing?
- 7. © 2019 Ververica7 Stream Processing is real-time data processing and real-time data-driven actions
- 8. © 2019 Ververica8 Stream Processing is the unification of real-time and offline analytics
- 9. © 2019 Ververica9 Stream Processing is the intersection of data analytics and applications
- 10. © 2019 Ververica10 Stream Processing is to event-driven applications what the database is to request/response apps
- 11. © 2019 Ververica11 Stream Processing is a flexible and extensible architecture for data-driven applications
- 12. © 2019 Ververica12 Application / Business Logic Stream Processor (Datalake, Database) Application / Business Logic Batch Proc. or Req/resp. Stream Processing Stream Processing changes how Applications and Data interact request/trigger result/response event stream event stream events are the data events act as triggers application logic triggered by events/changes
- 13. © 2019 Ververica13 What is Stream Processing for? data changes slowly Ad-hoc queries, data exploration, ML model training Batch Proc. or Req/resp. Most business logic query/logic changes fast data changes fast query/logic changes slowly Continuous Streaming
- 14. © 2019 Ververica14 more lag time data warehousing OLAP / BI / reporting continuous monitoring (position, risk, …) real-time ML model training/evaluation distributed OLTP-style apps more real time continuous ETL real-time behavior modeling (recommenders, pricing, ..) The Spectrum of Streaming Data Use Cases machine learning model training unified offline/ real-time analytics real-time alerts (fraud, security, …)
- 15. © 2019 Ververica15 Stateful Single Record Processing
- 16. © 2019 Ververica16 Everything is a Stream Streams Of Records in a Log or MQ
- 17. © 2019 Ververica17 Everything is a Stream Stream of Requests/Responses to/from Services Service DB à event sourcing architecture GET /a/b POST /b/c PUT /e/f 200 404 200 200 403
- 18. © 2019 Ververica18 Everything is a Stream Stream of Rows in a Table or in Files 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 …
- 19. © 2019 Ververica19 Everything is a Stream Stream of Rows in a Table or in Files 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 … a batch
- 20. © 2019 Ververica20 Everything is a Stream Streams may span storage systems 2016-3-1 12:00 am 2016-3-1 1:00 am 2016-3-1 2:00 am 2016-3-11 11:00pm 2016-3-11 10:00pm … Parquet files Avro records more distant past (e.g., compressed files in DFS/Object Store) recent past (e.g., events in MQ/Log)
- 22. © 2019 Ververica22 Bounded and Unbounded Streams
- 23. © 2019 Ververica23 Components of a Streaming Data Architecture Event producers (applications, servers, databases, sensors) Log / Stream Storage (Pulsar) Stream Processing Stream Processing Stream Processing Results (Views) (K/V stores, databases) Triggered Applications (Apache Flink)
- 24. © 2019 Ververica24 Flink Runtime Stateful Computations over Data Streams Stateful Stream Processing Streams, State, Time Event-driven Applications Stateful Functions Streaming Analytics SQL and Tables Apache Flink: Analytics and Applications on Streaming Data
- 25. © 2019 Ververica25 Stateful Stream Processing
- 26. © 2019 Ververica26 Flink Runtime Stateful Computations over Data Streams Stateful Stream Processing Streams, State, Time Event-driven Applications Stateful Functions Streaming Analytics SQL and Tables Apache Flink: Analytics and Applications on Streaming Data
- 27. © 2019 Ververica27 Stateful Stream Processing Computation Computation Computation Computation Source (Stream) Source (Static) Sink Sink Transformation State State State
- 28. © 2019 Ververica28 Example Use Cases •Real time search and recommendation models (e.g., Alibaba) •Build a real-time session behavior profile of users (e.g., Netflix) •Real time trade settlement dashboard (e.g., UBS) •Real time revenue accounting (various AdTechs) •Machine Learning-based anomaly/fraud detection (e.g., ING, Microsoft) •Real-time data refinement and data pipelines (many)
- 29. © 2019 Ververica29 DataStream API Source Transformation Windowed Transformation Sink val lines: DataStream[String] = env.addSource(new FlinkKafkaConsumer011(…)) val events: DataStream[Event] = lines.map((line) => parse(line)) val stats: DataStream[Statistic] = stream .keyBy("sensor") .timeWindow(Time.seconds(5)) .sum(new MyAggregationFunction()) stats.addSink(new RollingSink(path)) Streaming Dataflow Source Transform Window (state read/write) Sink
- 30. © 2019 Ververica30 DataStream API Process Functions 30
- 31. © 2019 Ververica31 Streaming Analytics
- 32. © 2019 Ververica32 Flink Runtime Stateful Computations over Data Streams Stateful Stream Processing Streams, State, Time Event-driven Applications Stateful Functions Streaming Analytics SQL and Tables Apache Flink: Analytics and Applications on Streaming Data
- 33. © 2019 Ververica33 Example Use Cases •Realtime Analytics Platforms (e.g., Alibaba, Uber, Lyft, Yelp!, Tencent) •Materializing Views (dashboards, data marts) •ETL - batch and continuous •Machine Learning Training (Alibaba, new ML library)
- 34. © 2019 Ververica34 SQL / Table API – Batch Queries SQL Query Batch Query Execution SELECT room, TUMBLE_END(rowtime, INTERVAL '1' HOUR), AVG(temperature) FROM sensors GROUP BY TUMBLE(rowtime, INTERVAL '1' HOUR), room Full TPC-DS support in Flink 1.10
- 35. © 2019 Ververica35 Interpreting Streams as Tables
- 36. © 2019 Ververica36 SQL / Table API – Streaming Data Case SELECT room, TUMBLE_END(rowtime, INTERVAL '1' HOUR), AVG(temperature) FROM sensors GROUP BY TUMBLE(rowtime, INTERVAL '1' HOUR), room SQL Query Interpret Stream as Table Incremental Query Execution output result changes as stream update database with changes
- 37. © 2019 Ververica37 FLIP-72 Add Pulsar connectors and Catalog to Apache Flink > CREATE CATALOG my_pulsar ( ‘type’ = ‘pulsar’, ‘adminUrl’ = ‘localhost:9092’ ); > USE my_pulsar; > INSERT INTO aggregations SELECT room, TUMBLE_END(rowtime, INTERVAL '1' HOUR), AVG(temperature) FROM sensors GROUP BY TUMBLE(rowtime, INTERVAL '1' HOUR), room
- 38. © 2019 Ververica38 Materialized Views Example logCDC Continuous SQL Query Continuous SQL Query Continuous SQL Query Materialized View Materialized View Archive
- 39. © 2019 Ververica39 Materialized Views Example logCDC Continuous SQL Query Materialized Views View Materialization (streaming) Dashboard: Many short queries (batch)
- 40. © 2019 Ververica40 Many handy SQL features: Temporal Joins, Pattern Matching, … SELECT tf.time tf.price * rh.rate as conv_fare FROM taxiFare AS tf LATERAL TABLE (Rates(tf.time)) AS rh WHERE tf.currency = rh.currency;
- 42. © 2019 Ververica42 Flink Runtime Stateful Computations over Data Streams Stateful Stream Processing Streams, State, Time Event-driven Applications Stateful Functions Streaming Analytics SQL and Tables Apache Flink: Analytics and Applications on Streaming Data
- 43. © 2019 Ververica43 Classical Tiered Application Architecture App App App
- 44. © 2019 Ververica44 Consistency in Database Applications App App App
- 45. © 2019 Ververica45 Consistency in Database Applications App App App For any failure in any call, it becomes hard to reason about what effects did or did not already happen X
- 46. © 2019 Ververica46 Applying the Stream Processing Approach to Applications App App App X
- 47. © 2019 Ververica47 Stateful Functions
- 48. © 2019 Ververica48 Stream Processing F-a-a-S λ λ λ λ simplicity / generality state management composability lightweight resources performance event-driven Can we combine some of these properties ?
- 49. © 2019 Ververica49 Stateful Functions f(a,b) f(a,b) f(a,b) f(a,b) f(a,b) mass storage (S3, GCF, ECS, HDFS, …) event ingress event egress f(a,b) snapshot state
- 50. © 2019 Ververica50 Stateful Functions compared to Stream Proc. & Apache Flink Apache Flink DataStream/Table Stateful Functions f(a,b) f(a,b) f(a,b) Pool of Resources (Apache Flink Cluster) Arbitrary Function-to-Function messaging. Not restricted to a DAG. Functions are multiplexed and share resources. Makes it possible to run many very small jobs. Solves two major challenges f(a,b) f(a,b) f(a,b) f(a,b) f(a,b)
- 51. © 2019 Ververica51 Example: Ride Sharing App Driver status updates Passenger ride requests Ride status update Driver Ride Pass- enger Geo- index update create bill Inform / book bid lookup update cell seeking confirmed riding free bidding booked
- 52. © 2019 Ververica52 data preparation combining knowledge/information filtering, enriching, aggregating, joining events coordination, (interacting) state machines complex event/state interactions “occasional” actions or spiky loads compute-intensive or blocking Stream Processing Streaming SQL Stateful Functions F-a-a-S f(a,b) f(a,b) f(a,b) λ λ λ λ state-centricevent/stream-centric stateless / compute-centric
- 53. © 2019 Ververica53 Putting it all together f(a,b) f(a,b) f(a,b) λ λ λ λ FaaS render map/route image create a receipt PDF send email Stateful Functions ride life-cycle driver-to-ride matching Stream Processing traffic models demand forecast & pricing Billing Passenger updates Driver position updates Driver status updates
- 54. © 2019 Ververica54 ❤ Thank You!