Help Center/ MapReduce Service/ Developer Guide (Normal_Earlier Than 3.x)/ Flink Development Guide/ Developing Flink Applications/ Asynchronous Checkpoint Mechanism Application/ Development Plan of Flink Asynchronous Checkpoint
Updated on 2024-08-16 GMT+08:00
View PDF
Share

Development Plan of Flink Asynchronous Checkpoint

Assume that you want to collect data volume in a 4-second time window every other second and the status of operators must be strictly consistent. That is, if an application recovers from a failure, the status of all operators must the same.

Data Planning

  1. Customized operators generate about 10,000 pieces of data per second.
  2. Generated data is of four tuples (Long, String, String, and Integer).
  3. Statistic results are printed on the devices.
  4. Printed data is of the Long type.

Development Guidelines

  1. A source operator sends 10,000 pieces of data and injects the data to a window operator every other second.
  2. The window operator collects the data volume statistics of the last 4 seconds every other second.
  3. The statistics is printed to the device every other second. For details, see Viewing the Running Result of a Flink Application.
  4. A checkpoint is triggered every other 6 seconds and the checkpoint result is stored in HDFS.