Sweta, Flink does not include watermarks in savepoints, nor are they included in aligned checkpoints. For what it's worth, I believe that with unaligned checkpoints in-flight watermarks are included in checkpoints, but I don't believe that would solve the problem, since the watermark strategy's state is still lost during a restart.
I can't think of any way to guarantee that all possibly late events will be deterministically identified as late. The commonly used bounded-out-of-orderness watermark strategy doesn't guarantee this either, even without a restart (because watermarks are delayed by the auto watermark interval, rather than being produced at every conceivable opportunity). If this is a strong requirement, you could decide not to rely on watermarks for dropping late events, and implement the logic yourself in a process function. Best, David On Wed, Jun 8, 2022 at 6:10 PM Sweta Kalakuntla <skalakun...@bandwidth.com> wrote: > Hi, > > I want to understand if flink saves a watermark during savepoint and if > not, how do we achieve this? > > We are seeing an issue where on recovery, the job processes some late > events which should have been discarded if the job were to be running > without any downtime. > > Thank you, > Sweta >
