I've read the doc about Flink's checkpointing mechanism from Data Streaming Fault Tolerance, and I wonder how does it ensure the exactly once semantics. Consider there is a Flink job with checkpoint enabled and the interval of checkpointing is 500ms. If the last completed checkpoint is n and after that, a few records have been computed and then the job failed before the next checkpoint.So are these records will be recomputed after the job recovery.
Asked
Active
Viewed 52 times
2
-
Yes, some records will be re-processed during recovery. See this answer: https://stackoverflow.com/a/43442562/2000823 – David Anderson Jul 20 '17 at 15:36
-
I got it,thanks a lot – jack huang Jul 25 '17 at 08:47