Exactly-once fault-tolerance in Apache Flink - CS 591 K1: Data Stream Processing and Analytics Spring 2020interfere with processing • processing and messages do not stop • Each process cast locally record its own state • Any process can initiate the algorithm 19 The Chandy-Lamport Algorithm ??? Vasiliki Kalavri Kalavri | Boston University 2020 Initiating a snapshot The initiator process: 1. Records its own state. 2. Sends a marker out on each of its outgoing channels. a. The marker is a special message that 2020 On receiving a marker (I) A process receiving a marker for the first time: 1. Records its own state. 2. Marks the channel that the marker came in on as empty. a. Future messages arriving on0 码力 | 81 页 | 13.18 MB | 1 年前3
Graph streaming algorithms - CS 591 K1: Data Stream Processing and Analytics Spring 2020to vertex ID • Iterative step: For each vertex • choose the min of neighbors’ component IDs and own component ID as the new ID • if the component ID changed since the last iteration, notify neighbors0 码力 | 72 页 | 7.77 MB | 1 年前3
Stream processing fundamentals - CS 591 K1: Data Stream Processing and Analytics Spring 2020be the result of one or more base and/or derived streams • Each query (operator) maintains its own state • Queries process raw streams, not synopses => results are typically exact • Challenges:0 码力 | 45 页 | 1.22 MB | 1 年前3
PyFlink 1.15 Documentationprojects by creating a separate environment for each project. It is a directory tree which contains its own Python executable files and the installed Python packages. It is useful for local development to create0 码力 | 36 页 | 266.77 KB | 1 年前3
PyFlink 1.16 Documentationprojects by creating a separate environment for each project. It is a directory tree which contains its own Python executable files and the installed Python packages. It is useful for local development to create0 码力 | 36 页 | 266.80 KB | 1 年前3
共 5 条
- 1













