I’ve put together a more detailed question here on Stack Overflow that relates to an issue I’ve recently been encountering for a job that’s failing after a deployment (appears to be related to transactions as the failing sinks in question all either use EXACTLY_ONCE or AT_LEAST_ONCE processing). It’s a big long to share directly in the channel, but hoping that someone familiar with transactions, graph changes, or something else that might be related may be able to chime in.