John R Pierce schrieb am 20.02.2016 um 12:05:
> near as I can tell, the OP has used some sort of SQL (unspecified) where multiple inserts
>within a transaction are individually inserted, regardless of one failing.
At least Oracle does it this way (and I think DB2 as well).
Oracle gets really slow if you do a row-by-row commit with large inserts. That's why
most people don't use auto-commit and just ignore any errors during inserts for batch loads.
> to me this seems to break the rules of transaction semantics
I agree, the expected behaviour from the OP does violate the A in the ACID principle,
but apparently it's popular enough that people think the correct behaviour is a bug:
http://dba.stackexchange.com/a/129822/1822