Some CDC messages seem to be delivered with significant latency [O(minutes)] and/or out of order (with regards to commit timestamp but in correct order w.r.t. replay id).
This seems to happen for large commits with spread out over 50+ messages.
Is this expected? Any way we can handle this on our end?
We do not document an SLA here, as it is really based on the customer and what they are doing. The p99 is the experience that we have seen on a general day to day basis.
I can say that CDC is slightly slower than PE (since there is more logic in the capture and the flush to the event bus happens after the entire commit, where Platform Events is a direct publish). CDC is similar to OBMM in terms of having to wait until after the commit, though since we make outbound calls with OBMM, it is really on a regular cycle that batches up messages with OBMM, as opposed to per event
