Re: "Transactions are bad, real bad" - discuss

"Marshall Spight" <mspight_at_dnai.com> wrote in message news:BCEua.261887$Si4.205632_at_rwcrnsc51.ops.asp.att.net...
> I've gotten a bit lost with all the various things discussed on
> this thread.
>
> Paul, I'm not sure if I understand you. Let me try saying
> it back. If I totally blow it, I apologize.

No worries.

> Is it that your objection to transactions is that they can
> stay open for a while?

That they are not atomic in time - yes that is one objection.

> You don't object to multiple operations
> being bundled together atomically, is that right?

Every atomic operation on a relational database is the replacement of the current database value with a new database value. The new value might be based on the old value after a multiple INSERTs, UPDATEs and/or DELETEs, but at the end of the day, those multiple IUD operations are just shorthands for a single database update statement.

> So if you had
> a transaction-like mechanism that made several operations
> atomic, but that took place all in an instant, would that
> be okay?

Yes, but I do much prefer to nest such operations into a single statement rather than to wrap a sequences of operations into an atomic whole. See my third reply in the first branch of this thread.

The point is that "transactions that make several operations atomic, and that take place in an logical instant" are nothing more than single database operations.

Transactions that are atomic in time are just not transactions at all in any meaningful sense

Regards
Paul Vernon
Business Intelligence, IBM Global Services Received on Fri May 09 2003 - 12:18:21 CEST