> Let's say that when you deploy, you want to direct 1% of traffic to the new code, then 5%, then 25%, then 50%, and finally 100%. And if monitoring metrics go red at any point, automatically roll back.
> How do you do that with a sproc?
Wrap the real sproc (well, the set of real sprocs, since old and new will sometimes coexist) with a wrapper sproc that probabilistically directs to the appropriate working sproc based on a table value (or anything else the sproc can read) that is updated to reflect the current stage in the conversion process, including any monitoring-based reversals.
> How do you do that with a sproc?
Wrap the real sproc (well, the set of real sprocs, since old and new will sometimes coexist) with a wrapper sproc that probabilistically directs to the appropriate working sproc based on a table value (or anything else the sproc can read) that is updated to reflect the current stage in the conversion process, including any monitoring-based reversals.