Hi Richard,
Schema evolution for data types that are used as keys is not allowed
because, potentially, if the schema of the key changes, hash codes of keys
may also change and can break partitioning for internal state managed by
Flink.
There are of course some evolution scenarios that would not re
Hi Richard,
I've pulled in Gordon who worked on this feature. He should be able to tell
you about the current limitations of Flink's schema evolution.
Cheers,
Till
On Wed, May 29, 2019 at 1:44 PM Richard Deurwaarder wrote:
> Hello,
>
> I am running into the problem where (avro) schema evolutio
Hello,
I am running into the problem where (avro) schema evolution works perfectly
for operator/keyed state but does not work when used with keyBy().
For example:
I have a job like so:
env.addSource(someSource())
.keyBy(object -> getMyAvroObject())
.process(doSomething()
