Tom Lane wrote:
> "Kevin Grittner" writes:
>> Tom Lane wrote:
>>> (Besides which the lock acquired by UPDATE isn't exclusive and
>>> wouldn't block anyway...)
>> It blocks other UPDATEs.
> Not at the table level.
The question was about whether we could change the timing of when
we get the current locks, not about getting different locks than we
currently do. Clearly, adding a table lock would make no sense.
> If you could lock only at the tuple level maybe you'd have
> something, but it seems like you can't find the target tuples
> without having acquired a snapshot.
That would be the fatal difference then. InnoDB doesn't actually
keep old versions of a row; it generates them from "undo" records in
the log when needed, which might be why it was feasible to get the
lock before the snapshot there. If we need the snapshot before we
can get the lock, it's not an optimization which is available to us.
Thanks,
-Kevin