From where I'm sitting, it looks like Microsoft came up with this idea of "lost update anomaly", which would explain the need for locks, but that's not what the spec says.
Comments
Log in with your Bluesky account to leave a comment
It would be interesting to think what would happen if such a lock-free transaction would "cross paths" with a write-lock, but I think the answer is that they would have to wait on its monitor and then be applied at a (basically random) order.
Arguably, that the SQL-spec just lumps all "write-anomalies" under "write-skew" is the culprit here, but facts remain, this would be a compliant implementation of the specification.
Comments