Virtual Flink Forward 2020: Data driven matchmaking streaming at Hyperconnect - Jaehyeuk Oh, Gihoon Yeom
0 likes510 views
The document discusses the data-driven matchmaking process at Hyperconnect, highlighting the architecture and operational strategies involving stateful matchmaking and the use of Apache Flink for low-latency iteration. It outlines the steps in the matchmaking cycle, including data collection, feature engineering, pair generation, and state management, as well as the microservices structure to enhance scoring logic. Additionally, the document addresses performance tuning, fault tolerance, and the infrastructure required for effective staging and real-time operations.
Virtual Flink Forward 2020: Data driven matchmaking streaming at Hyperconnect - Jaehyeuk Oh, Gihoon Yeom
- 1. Data Driven Matchmaking Streaming at Hyperconnect Gihoon Yeom Jaehyeuk (Jacob) Oh 2020 Flink Forward Virtual Conference
- 2. Motivation HyperMatch Architecture Infra & Operation Stateful Matchmaking Contents
- 3. Video and AI Company Introduction
- 4. Scalability 5000 Match Requests Stability Stateful matchmaking Motivation
- 5. Why Flink? ● Capability for Low-latency iteration ● Loosely coupled architecture ● Intrinsic support for a state management
- 6. Motivation HyperMatch Architecture Infra & Operation Stateful Matchmaking Contents
- 7. Matchmaking Steps - Collect match requests from specific units of windows and select the best pairs among them
- 8. Matchmaking Steps Input Source Feature Engineering Pair Generator Match Maker Feedback System State Management Data Source Leftovers Scoring Ops Matchmaking Results Multiple Sink Score Microservices
- 9. Matchmaking Steps Input Source Feature Engineering Pair Generator Match Maker Feedback System State Management Data Source Leftovers Scoring Ops Matchmaking Results Multiple Sink Score Microservices
- 10. Matchmaking Steps Feature Engineering Pair Generator Match Maker Feedback System State Management Leftovers Scoring Ops Matchmaking Results Multiple Sink Input SourceData Source Score Microservices
- 11. Matchmaking Steps - Option 1. Built-in Kafka source - To simulate previous matchmaking steps - Reduce dependencies between applications - Option 2. Custom Netty Http Source - For real-time latency - Apache Bahir extension
- 12. Matchmaking Steps Feature Engineering Pair Generator Match Maker Feedback System State Management Leftovers Scoring Ops Matchmaking Results Multiple Sink Input SourceData Source Score Microservices
- 13. Matchmaking Steps - Preprocess data stream before entering match cycle - Conditional tagging for segmentations (ex) A/B test) - Vectorize user features for machine learning models - Save managed state on operator and connect it to feature engineering operator
- 14. Matchmaking Steps - Split match request attributes information 1) Match context infos - can be changed in matchmaking cycle - small size - direct transfer to next operator 2) Match properties - properties cannot be changed - large size - broadcast to only scoring operators Total Feature memory Small Mutable Context Large Immutable Properties
- 15. Matchmaking Steps Feature Engineering Pair Generator Match Maker Feedback System State Management Leftovers Scoring Ops Matchmaking Results Multiple Sink Input SourceData Source Score Microservices
- 16. Matchmaking Steps - Aggregate by window - Global tumbling window - Keyed tumbling window when segmentation enabled - Custom trigger & evictor depending on matchmaking function - Custom pair generating protocol with scoring operators
- 17. Matchmaking Steps - Efficient pair generation and bucketizing considering communication costs - Total communication costs = network consume time + serialization/deserialization time
- 18. Matchmaking Steps Feature Engineering Pair Generator Match Maker Feedback System State Management Leftovers Scoring Ops Matchmaking Results Multiple Sink Input SourceData Source Score Microservices
- 19. Matchmaking Steps Matchmaking logic is divided into several microservices - Many teams can be involved for scoring logic advancement - Scoring Operator with broadcasted request data or external in-memory cache - Prepare data to be provided to each microservices - Get responses from microservices - Scoring Microservices can be REST or Event driven - Use AsyncDataStream
- 20. Matchmaking Steps Feature Engineering Pair Generator Match Maker Feedback System State Management Leftovers Scoring Ops Matchmaking Results Multiple Sink Input SourceData Source Score Microservices
- 21. Matchmaking Steps Fault tolerance on scoring microservices - Custom trigger & evictor - Distributed sorting algorithm for finding best pairs - IterativeStream for leftovers
- 23. Matchmaking Steps - Trigger condition: Timeout or all candidates arrived - Timeout dummy token sent by generating operator for fault tolerance ScorersPair Generator Match Maker Leftovers Send Timeout Token for custom trigger
- 24. Matchmaking Steps - Create app level backpressure by buffering some requests when scorer fails consecutively ScorersPair Generator Match Maker Leftovers Too many leftovers! Backpressure: Buffering selected requests Some pair scorer failed => Increase requests wasn’t matched => Increase next match requests => Repeated.. => Service failed
- 25. Matchmaking Steps Feature Engineering Pair Generator Match Maker Feedback System State Management Leftovers Scoring Ops Matchmaking Results Multiple Sink Input SourceData Source Score Microservices
- 26. Matchmaking Steps - Matchmaking broker - Feedback controller: distribution of results - User state management - Logging handler - Metric reporter
- 27. Motivation HyperMatch Architecture Infra & Operation Stateful Matchmaking Contents
- 28. Infra & Operation - Easy to set up new staging infra - For performance testing - Metrics on prometheus/grafana Match App1 Job manager Task manager Match App2 Job manager Task manager Score App2 Score App1 HA Proxy ACTIVE STANDBY - For zero downtime: Blue green deployment - Score microservices can be rolling updated
- 29. Performance Tuning Network cost & General tips - Latency vs Throughput trade-off: setBufferTimeout - Forward vs Rebalance: setParallelism - For resource isolation: stream.slotSharingGroup
- 30. Motivation HyperMatch Architecture Infra & Operation Stateful Matchmaking Contents
- 31. Stateful MatchMaking Realtime State + Realtime Inference Stateful Matchmaking