Flink Forward China 2018—为流处理世界重新设计的存储.pdf

主要章节：

• Market Drivers
• The Quest for Unified Batch and Stream Processing
• Querying the Past
• Querying the Future
• Querying the Past and the Future
• Today’s “Accidental Architecture”
• Our Goal: Refactor the “Accidental Storage Stack”
• Building a Unified Batch and Stream Stack
• The Stack
• Pravega Architecture Goals
• Pravega Fundamentals
• Segments
• Streams
• Beyond the Fundamentals
• Stream Elasticity
• Daily Cycles
• Pravega Auto Scaling
• Exactly Once – Ack on write
• Exactly Once – Transactions
• Pravega with Flink
• The Complete Pipeline
• Outlook: Autoscaling with Flink
• Connecting Flink and Pravega
• FlinkPravegaReader
• Details on the Transactional Flink Pravega Writer
• Transactional Writes
• The FlinkPravegaWriter
• Exactly-Once Writes
• Fails after local snapshot
• Transaction fails before commit…
• … commit on recovery
• A Unified Data Pipeline