That only covers errors that occur inside the sproc, not errors that are caused by the sproc. For example: What if the new version of the sproc succeeds, but the data it returns causes the application code to fail in some way? The sproc metrics would report 100% success and the application error rate wouldn't get above noise until a significant fraction of traffic was using the new buggy version of the sproc.
It's possible to add more epicycles to try and correct for these deficiencies, but the truth is that sprocs are dangerously double-edged. Yes, your code runs next to your data, but to get parity with application code (canary deploys, debugging, logging, etc) you must reinvent lots of tooling.
It's possible to add more epicycles to try and correct for these deficiencies, but the truth is that sprocs are dangerously double-edged. Yes, your code runs next to your data, but to get parity with application code (canary deploys, debugging, logging, etc) you must reinvent lots of tooling.