Hi Jasvendra, In short 1.18 savepoint should be compatible from 1.20. We don't know such existing issue. Can you please come up with a bare minimal step-by-step or public repo where one can repro it easily?
BR, G On Tue, Apr 1, 2025 at 2:37 PM jasvendra kumar <javatech....@gmail.com> wrote: > Dear Flink Community, > > I am currently in the process of upgrading our Flink cluster from *version > 1.18.0 to 1.20.1*. The cluster itself is functioning correctly > post-upgrade, and I am able to deploy Flink jobs successfully. However, I > have encountered an issue when attempting to restore a job using a *savepoint > or state taken from Flink 1.18.0*. > *Issue Description* > > - > > When deploying the Flink job to the *Flink 1.20.1 cluster* using a > *savepoint > from Flink 1.18.0*, the job is assigned *only one Kafka partition > (partition 0)*. As a result, messages from the other partitions are > not being consumed. > - > > However, if I deploy the same job *without a savepoint*, the job > correctly assigns all three partitions (*0, 1, 2*) and consumes > messages as expected. > > I have researched this issue extensively but have not found a clear > explanation. I would appreciate any guidance on the following queries: > > 1. > > *Is this issue related to the compatibility of savepoint restoration > between Flink 1.18.0 and Flink 1.20.1?* > 2. > > *Is this behavior a known bug or an expected outcome?* > 3. > > *If this is a bug, what are the recommended steps to resolve it?* > - > > Are there any configuration changes required to properly restore > partitions? > - > > Would fixing this require modifications in the application code? > > Your insights and assistance on this matter would be highly appreciated. > > Thanks & Regards > Jasvendra Kumar >
