[plug] kernel 2.6.3 production-ready?

[Craig Ringer]

On Thu, 2004-03-04 at 10:59, Denis Brown wrote:

> Ahhh... maybe I misunderstood.   So LVM is okay but LVM2 is the problem 
> child?

Pretty much, yes. That may be misconfiguration or site-specific issues,
though - LVM2 is considered stable and has had a fair bit of testing.
Perhaps it's something to do with our LVM1 to LVM2 migration that's
problematic.

> Given your environment I suspect that you've one of the 
> harder-working servers represented on this list so if anything is going to 
> break you should know about it.

It almost certainly has the most varied load, and is probably right up
there in overall workload, too. Still, we only have 2GB of RAM and a
dual Xeon - it's not a super-monster machine. (Hmm... quad Opteron...
*drools*).

> My plan of attack then will likely be (a) 
> ACL-patch and use a 2.4 series kernel; (b) wait for 2.6 to get a few more 
> notches in its belt; (c) implement 2.6 and explore LVM benefits.   Dim-dark 
> future: explore LVM2, unless I win Lotto in the meantime :-)

I don't think that'll work. LVM1 is, AFAIK, only available under 2.4.
2.6 includes the device-mapper based LVM2 in vanilla. As such, you're
_much_ better off using the wonderful LVM1 under 2.4, and later
investigating a move to LVM2 as part of a move to 2.6. I found LVM1
under 2.4 to be an absolute dream - easy to understand, easy to work
with, utterly reliable, etc. With LVM1, adding another RAID array and
expanding an existing mounted partition onto it becomes trivial.

I'd suggest testing LVM2 under 2.6 very carefully before you use it in
production, though. What I /should/ have done, had I not had externally
imposed priorities, would have been to move to LVM2 on 2.4 (via the
device-mapper patch) and only /then/ move to 2.6.x . If you start using
LVM, you may wish to consider the same course of action.

Craig Ringer