hi Krzysztof, That's why my goal is to always set env.getConfig().disableGenericTypes(); in my streaming jobs. This way, you will receive an early crash if GenericTypes are used somewhere. (They are bad for the performance so I try to avoid them everywhere).
Sadly, if you build up streaming jobs based on kafka, you must have Flink 1.10.1 or flink 1.11 to be able to have this setting set. ( https://issues.apache.org/jira/browse/FLINK-15904 ). As my project is still on Flink 1.9, I currently have an e2e-like test with "disableGenericTypes", which differs from the real job only by not using kafka consumer and instead have another input, so I can make sure the remaining part of my pipeline has no generic types within. Best regards Theo ----- Ursprüngliche Mail ----- Von: "KristoffSC" <krzysiek.chmielew...@gmail.com> An: "user" <user@flink.apache.org> Gesendet: Donnerstag, 16. Juli 2020 20:57:09 Betreff: Re: Flink Pojo Serialization for Map Values Theo, thank you for clarification and code examples. I was actually suspectign that this is becase the Java type erasure.s The thing that bothers me though is fact that Flink was failing over to Kryo silently in my case. Without any information in the logs. And actually we found it just by luck.
