Postgres & large objects

[Matt Clark]

Hello all,

It seems I'm trying to solve the same problem as Richard Emberson had a
while ago (thread here:
http://archives.postgresql.org/pgsql-general/2002-03/msg01199.php).

Essentially I am storing a large number of large objects in the DB
(potentially tens or hundreds of gigs), and would like the pg_largeobject
table to be stored on a separate FS. But of course it's not just one file
to symlink and then forget about, it's a number of files that get created.

So, has anyone come up with a way to get the files for a table created in a
particular place? I know that tablespsaces aren't done yet, but a kludge
will do (or a patch come to that - we're runing redhat's 7.2.3 RPMs, but
could switch if necessary). I had thought that if the filenames were
predictable it might be possible to precreate a bunch of zero-length files
and symlink them in advance...

Cheers

Matt

[Bradley Kieser]

Matt,

Not really the answer that you are looking for and you may already do
this, but if it's a disk space or performance issue then I would suggest
moving the PGDATA dir (or the location if you are using locations) onto
a RAID5 disk array - means you can ramp up the space and you get the
performance gains of RAID5, not to mention the safety of a FS that
recovers from disk failure!

Brad

Post by Matt Clark
Hello all,
It seems I'm trying to solve the same problem as Richard Emberson had
http://archives.postgresql.org/pgsql-general/2002-03/msg01199.php).
Essentially I am storing a large number of large objects in the DB
(potentially tens or hundreds of gigs), and would like the
pg_largeobject table to be stored on a separate FS. But of course
it's not just one file to symlink and then forget about, it's a number
of files that get created.
So, has anyone come up with a way to get the files for a table created
in a particular place? I know that tablespsaces aren't done yet, but
a kludge will do (or a patch come to that - we're runing redhat's
7.2.3 RPMs, but could switch if necessary). I had thought that if the
filenames were predictable it might be possible to precreate a bunch
of zero-length files and symlink them in advance...
Cheers
Matt

---------------------------(end of broadcast)---------------------------
TIP 4: Don't 'kill -9' the postmaster

[Matt Clark]

Thanks, but it's already on a RAID array with a battery backed controller
and a journaled FS. The deal is that I don't really want to spend the money
on expanding that storage for data that isn't very critical at all. I want
to stick these blobs on a cheap bunch of ATA disks basically, as comparing
the price of a terabyte of ATA mirrored disks and the same TB on SCSI
hardware raid is enlightening.

M

-----Original Message-----
From: Bradley Kieser [mailto:***@kieser.net]
Sent: 06 May 2004 11:03
To: Matt Clark
Cc: pgsql-***@postgresql.org; ***@phc.net
Subject: Re: [ADMIN] Postgres & large objects


Matt,

Not really the answer that you are looking for and you may already do
this, but if it's a disk space or performance issue then I would suggest
moving the PGDATA dir (or the location if you are using locations) onto
a RAID5 disk array - means you can ramp up the space and you get the
performance gains of RAID5, not to mention the safety of a FS that
recovers from disk failure!

Brad

Post by Matt Clark
Hello all,
It seems I'm trying to solve the same problem as Richard Emberson had
http://archives.postgresql.org/pgsql-general/2002-03/msg01199.php).
Essentially I am storing a large number of large objects in the DB
(potentially tens or hundreds of gigs), and would like the
pg_largeobject table to be stored on a separate FS. But of course
it's not just one file to symlink and then forget about, it's a number
of files that get created.
So, has anyone come up with a way to get the files for a table created
in a particular place? I know that tablespsaces aren't done yet, but
a kludge will do (or a patch come to that - we're runing redhat's
7.2.3 RPMs, but could switch if necessary). I had thought that if the
filenames were predictable it might be possible to precreate a bunch
of zero-length files and symlink them in advance...
Cheers
Matt

---------------------------(end of broadcast)---------------------------
TIP 6: Have you searched our list archives?

http://archives.postgresql.org