Re: imp commit=n and rbs

Norman Dunbar <Norman.Dunbar_at_lfs.co.uk> wrote in message news:<E2F6A70FE45242488C865C3BC1245DA703994AD3_at_lnewton.leeds.lfs.co.uk>...
> Morning Joel,
>
> commit=y means that a commit is done whenever a buffer's worth of rows
> are written to a table. Unless the table being imported has a LONG
> column, in which case you get a commit for each row in that table.
>
> commit=n means that a commit takes place at the end of each table. You
> will have the index builds as well I think, constraints etc - all of
> which would have to be rolledback in case of a failure.

Yeah, I know all that, I'm just wondering why the rollback is _so_ much larger than the table? This has bothered me for years. I've seen ocassional posts asking it, never seen any answer other than what the docs say. And they don't say _why_. It's obviously much more than just a before image of the table. Bigger even than the sum of before images of the table for each index, if that's what it does. I was hoping someone else may have been curious about this and figured it out before I start dumpster diving.

>
> I use commit=y and a buffer of 1024000 or 2048000 bytes when I import
> depending on the size of the dump file.

I usually take a tip from tkyte and use 20000000, but I was thinking too far ahead of my fingers and didn't type it or commit on the line. And then it blew up, surprising me. So many interesting things come from someone saying "hmmmm, that's odd..."

jg

--
@home.com
"So what did you say you were doing to it?"  Overheard comment from
Network Engineer as I was typing this.