FlinkKafkaConsumer and FlinkKafkaProducer are now deprecated, KafkaSource and KafkaSink should be used instead.
Unfortunately as these operators do store a state some migration steps have to be respected:
- the job must be stopped with a savepoint and offsets committed to kafka
- the job must be started from this savepoint with most probably --allowNonRestoredState set to true
Given the risk of failures it might be wise to do this in two steps:
- add support for KafkaSink/KafkaSource in the codebase with an option to flip between the two implemations
- drop usage of FlinkKafkaConsumer and FlinkKafkaProducer once the job is proven to run properly with KafkaSink/KafkaSource
AC:
- FlinkKafkaConsumer and FlinkKafkaProducer usage are dropped from the codebase