Hi Jean-Marc, Could you please double check that your code is having the mentioned fix + give a simple repro steps? Please hardcode the batch size value in the code [1] to avoid any yaml to config issues. I would take a look if you can help a bit.
BR, G [1] https://github.com/apache/flink/pull/25764/files#diff-278bac11f68be56ee499b24afe5e1d53a7c61b4d636654fd96b4167e2a45cbacR125 On Thu, Sep 18, 2025 at 5:30 PM Jean-Marc Paulin <[email protected]> wrote: > Hi, > > Using Flink 1.20.1, we get this error when trying to read a savepoint: > > Caused by: java.lang.RuntimeException: Record size is too large for > CollectSinkFunction. Record size is 5411925 bytes, but max bytes per batch > is only 2097152 bytes. Please consider increasing max bytes per batch value > by setting collect-sink.batch-size.max > at > org.apache.flink.streaming.api.operators.collect.CollectSinkFunction.invoke(CollectSinkFunction.java:288) > ... > > I tried to set collect-sink.batch-size.max in the flink-conf.yaml, but I > still hit the same error. It's like it's not taken into account. I see > there is a fix in 1.20.1 (https://github.com/apache/flink/pull/25764) for > this, but I still face the same issue, > > JM > > >
