Hi everyone, I'm trying for a while to answer myself a question about what are actually guarantees for state stores in regards to consistency when connected to transformers.
I have an application where a single (persistent, rocksdb backed) state store is connected to multiple transformers. Each transformer might both read (get) and write (put) data into the state store. All transformers receive data from multiple input topics in the same way (the same key, same number of partitions) that before sending it to transformers merged together. All transformers are located in the same sub-topology. What I observed is that even with 0 standby replicas I might get inconsistent results when querying this state store connected to multiple transformers. I have 2 instances and metadata on both instances agree on the active host for this state store and partition, but instance A returns result X for a partition I and instance B returns result Y for the same partition I. Any suggestions if this is a bug or is my assumption incorrect that the same state store should give the same result for the same key (same partition) in 2 distinct transformers fed from the same input? Thanks, Jiri
