How to save and restore the state
1. Timing production distributed snapshot of the program state backup
2. failure:
the entire job all the task are rolled back to the last successful checkpoint, and then started from that point;
3. Prerequisites: support data retransmission
4. conformance statement: exactly once, at least once
checkpoint enforcement mechanisms
1.checkpoint coordinate sent to all the Trigger checkpoint Source
2. All task after receiving the barrier, will take a snapshot, and pass their output to the new barrier, their status will persist.
3. After the backup when the task is completed, the data will address notification checkpoint coordinate.