梅源-Flink 新一代流计算和容错 - 阶段总结和展望.pdf

1、Flink 新一代流计算和容错-阶段总结和展望梅源（Yuan Mei）Flink引擎架构师，Flink Committer阿里云Flink存储引擎团队负责人Data Use Cases 图谱Continuous ProcessingBatch ProcessingData PipelinesStreaming AnalyticsEvent-drivenApplications Realtime AlertsFraud/Securitymore real timemore real timeless real timeless real timeData Use Cases 图谱Continuo

2、us ProcessingBatch ProcessingData PipelinesStreaming AnalyticsEvent-drivenApplications Realtime AlertsFraud/Securitymore real timemore real timeless real timeless real timeStream ModeBatch ModeData Use Cases 图谱Continuous ProcessingBatch ProcessingData PipelinesStreaming AnalyticsEvent-drivenApplicat

3、ions Realtime AlertsFraud/Securitymore real timemore real timeless real timeless real timeBatch ModeStream ModeData Use Cases 图谱Continuous ProcessingBatch ProcessingData PipelinesStreaming AnalyticsEvent-drivenApplications Realtime AlertsFraud/Securitymore real timemore real timeless real timeless r

4、eal timeBatch ModeStream ModeData Use Cases 图谱Continuous ProcessingBatch ProcessingData PipelinesStreaming AnalyticsEvent-drivenApplications Realtime AlertsFraud/Securitymore real timemore real timeless real timeless real timeq 正常处理：高效，低延迟（100ms）q 容错处理：代价相对较大（Sec Min）Flink 流式计算(Stream Mode)Batch Mod

5、eStream ModeData Use Cases 图谱Continuous ProcessingBatch ProcessingData PipelinesStreaming AnalyticsEvent-drivenApplications Realtime AlertsFraud/Securitymore real timemore real timeless real timeless real timeq 正常处理：高效，低延迟（100ms）q 容错处理：代价相对较大（Sec Min）Flink 流式计算(Stream Mode)Batch ModeStream ModeData

6、Use Cases 图谱Continuous ProcessingBatch ProcessingData PipelinesStreaming AnalyticsEvent-drivenApplications Realtime AlertsFraud/Securitymore real timemore real timeless real timeless real timeq 正常处理：高效，低延迟（100ms）q 容错处理：代价相对较大（Sec Min）Flink 流式计算(Stream Mode)Batch ModeStream ModeFault Tolerance 2.0Fau