I'm trying to bootstrap my operator broadcast state and keyed state using Flink's State API, in order to create a savepoint to initialize my job from.
As far as I can tell, I can create a transformation using a bootstrap function that is either bootstrapping the broadcast state or the keyed state.
The issue is that my operator is a KeyedBroadcastProcessFunction, which handles both regular events and broadcast events, and has a single UID.
When trying to create a savepoint using both transformations on the same UID, it throws an error saying I can't use the same UID twice.
I tried searching the documentation and the code itself for a solution such as bootstrap function that supports both broadcast and keyed state, without luck so far.
It seems that this is currently not supported, I found an open issue on their Jira.