[FLINK-19999] State Processor API classes leaking into savepoint

• Type: Bug
• Status: Open
• Priority: Not a Priority
• Resolution: Unresolved
• Affects Version/s: 1.11.2, 1.12.0
• Component/s: API / State Processor
• Labels: auto-deprioritized-major auto-deprioritized-minor auto-unassigned
• Assignee: Unassigned
• Reporter: Nico Kruber
• Created: 05/Nov/20 10:15
• Updated: 12/Dec/21 22:38

Description

Currently, any configuration for serializers that you are using when writing a State Processor API job will be shared with the serializers that are used for writing a savepoint. However, your normal job shouldn't necessarily depend on (helper) classes that you only use in the StateProc API job.

By default, for example, ExecutionConfig#autoTypeRegistrationEnabled = true and thus classes like org.apache.flink.runtime.checkpoint.OperatorSubtaskState will be registered with Kryo and will thus also be needed when reading the created savepoint if you have Kryo serialization in your job.

This particular instance can be worked around by calling ExecutionConfig#disableAutoTypeRegistration() but the problem is probably bigger and extends to other type registrations, e.g. POJOs, as well.