Hi Robert, there's this one phrase I dont quite understand in the documentation, describing the RoyalFoodTaster, If a command fails, the taster is discarded and replaced with a clone of the real model, without requiring full restore. More importantly, the real model is readable while the taster is being rebuilt.
if the real model is still readable, (and valid) then why bother with rebuilding?
Ok got it; I had a performance test that appears superficially to contradict this; and my problem was that my test model was wastefully keeping a transaction history in memory that was never being queried. This resulted in a linear growth of model size in memory, resulting in a long clone times. (That's my theory) going to remove the transaction table from memory, ... that can always be recreated by replaying all the messages (wow.. that's incredible! ) ...and that should = a test that proves how it all works. If this works, I'd like to offer it for consideration as sample code for the docs?
so far, it looks like the sample code is thin on 'best practice' examples, or examples that show with tests, why (and when) you'd choose A over B.