In message
<[EMAIL PROTECTED]>,
David A Barrett <[EMAIL PROTECTED]> writes
We have a monster file that hit the 2GB mark way back when, so we divided
it up in a distributed file.
It continues to grow, and the keys are largely sequential with a
partitioning algorithm that puts each block of one million records into a
part. Currently the file grows at a pace of about one million records
every 2 years, and it's a pain in the neck to add a new part because the
ten indexes on it need to be rebuilt each time you add a new part. With 8
million records, it takes a good long time to rebuild all of the indices.
The other thing about this is that I go through this operation only once
each two years, so that I forget the finer details about fiddling with
distributed files. So before I do it on our live system, I play around
with a test file to figure it out. This is where I found something
interesting.
Actually, I think all this was in the manuals as long ago as Prime
INFORMATION ... :-)
As Karl said, each partfile is a file in and of itself, with its own
indices. If you try and run an index-based command over a distributed
file, it will only display and/or use the indices that exist in EVERY
partfile. It certainly used to be (quite likely has changed now :-) that
if you tried to SELECT the distributed file and the index existed in
nine part-files but not the tenth, it would ignore the index
*completely*.
Bearing in mind that distributed files can be created easily with ED
(provided that the partfiles already exist as partfiles), I think UV
checks the partfiles when you access them (not looking up info from
&PARTFILES&). So again as Karl says, if you just create the new
partfile, index it, and add it, you won't upset the index status across
the distributed file.
Cheers,
Wol
--
Anthony W. Youngman <[EMAIL PROTECTED]>
'Yings, yow graley yin! Suz ae rikt dheu,' said the blue man, taking the
thimble. 'What *is* he?' said Magrat. 'They're gnomes,' said Nanny. The man
lowered the thimble. 'Pictsies!' Carpe Jugulum, Terry Pratchett 1998
Visit the MaVerick web-site - <http://www.maverick-dbms.org> Open Source Pick
-------
u2-users mailing list
[email protected]
To unsubscribe please visit http://listserver.u2ug.org/
