Then we look at the consistency checkpoint, and we can see,
Consistency checkpoint is the core of flink's fault recovery mechanism, which can record the status of all operators and record it only once.
Recover exactly once, right?
We can see, sum_even, 6 here 2+4 sum_odd here 9 1+3+5 we can see
Now, after the data of 5 is processed, all the states are recorded. After the data of 5 is processed,
All node status can be seen in the jobmanager record, 5, 6, 9 are all recorded, use when recovering
</