I am implementing a custom partitioner by extending the org.apache.flink.api.common.functions.Partitioner interface in Apache Flink. This requires me to override the partition method. My question is: do I need to ensure that the partition method is thread-safe?

I have searched online but have not found any definitive resources regarding the thread-safety requirements of the partition method when used in Flink’s distributed processing environment. Since Flink jobs can run in parallel with multiple tasks potentially using the same partitioner instance, I am concerned about potential race conditions or inconsistent partitioning if the method is not thread-safe.

Any insights or pointers to relevant documentation would be greatly appreciated.