On Tue, Feb 26, 2002 at 10:30:35PM +0100, Branko �ibej wrote:
> Daniel Berlin wrote:
> >On Tue, 2002-02-26 at 15:23, Branko Cibej wrote:
> >>Daniel Berlin wrote:
> >>>Of course, I was
> >>>
> >>>>worried about DB object size limits, not transaction logs -- but this
> >>>>solutino fixes both problems.
> >>>>
> >>>The db object sizes are only limited by available memory (assuming you
> >>>are doing non-partial object fetches), but are otherwise unlimited
> >>>length.
> >>>
> >>Not so. both size and offset are 32-bit in the DBT.
> >
> >The manual claims differently.
> >See the section on database limits.
> >
> I did. It just occured to me that "available memory" is usually less
> than 4 Gig ...
"But that says FOO" "No, that says BAR"
Ease up, guys.
DBT.doff is a 32-bit unsigned integer. Period. Done. Doesn't matter what any
document says, or how much memory we have, or whatever. That means that we
cannot index into a value greater than 4 gig.
> >>>We lose transaction control on fulltext
> >>>
> >>I know that. I don't think we need it, though.
> >
> >This is a complete kludge, and won't fix the underlying problem.
>
> Of course it will. The FS doesn't have to create log files.
We aren't going to add the use of files at this time. That is too large of a
delta, and increases the complexity of the underlying storage system. As
Daniel said, then we have to start worrying about backups, cleaning the
files when an FS txn is aborted, ensuring that we don't blow reasonable
directory sizes, unique names, contention, inode limits, blah blah blah.
No... we can continue to operate within the DB, but we simply need to review
how we're using it.
Cheers,
-g
--
Greg Stein, http://www.lyra.org/
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Sat Oct 21 14:37:10 2006