As per the CE partitioning extension, the "partitionkey" attribute "can be used to determine which bucket each event goes into". We should set it to a value derived from the change event key. I can think of two ways:

• Stringified JSON-representation of the key, e.g. "{\"id\" : 1234, \"sub-id\" : 56}"
• semicolon-separated concatention of all key field(s), e.g. "id:1234;sub-id:56", similar to what we do for the "id" field

I'd vote for the latter. Any thoughts?