There is no reasonable scenario in which a QA department could be expected to verify those and no way that Operations/Support could be expected to put thousands into production at the same time.
Thus a planned rollout involving small sets is required. And small sets can be upgraded manually.
Even if everything went smoothly with say the first 100 one would still want to have a grace period between production and the next batch to insure that some serious problem didn't develop. Attempting to rollback tens versus thousands of apps is obviously very different.