[Ocfs2-users] OCFS2 and LVM2

[Alexei_Roudnev]

If file system is designed in accordance with Linux rules, then it is not FS
concern which devices are used - disk, partition, lvm volume, evms volume,
raid device (RAID is another story in case of cluster, of course) and so on.
If file system depends of this, it is broken file system (first of all).

Of course, bugs are always possible. But these will be a bugs, not a natural
limitations.

(For example, one bug prevented Async IO on disk without partitions in SLES9
SP1-SP2; another existed for Async IO on LVM - but both was BUGS).

So, you don't need to be a hacker to try (OCFS on LVM or OCFS on EVMS). Of
course, it is Linux - everything which is out of the main road MUST BE
TESTED (because Linux vendors never test their products enough for
production - no one including Oracle or Redhat do it well enough) and better
to vbe avoided - just because of high probability for the bugs.

As I know, OCFSv2 do work on LVM - few people reported it. But of course,
I'd better avoid it - OCFSv2 itsels is not production-grade yet, and adding
one more complexity here is not a good thing.

(Btw, some Oracle engineers are saying that they are going to drop OCFSv2
certification.)

If someone decide to experiment - here is a list _TO TEST_:
- test startup/shutdown on primary node
- the same on other nodes
- test node reboot and failure
- make a policy _how to change volumes_ and test it
- test a heavy load, async io, and temporary disk errors (such as FCP
switchover for example).



----- Original Message ----- 
From: "Eckenfels. Bernd" <B.Eckenfels at seeburger.de>
To: <Ocfs2-users at oss.oracle.com>
Sent: Monday, February 26, 2007 11:39 AM
Subject: RE: [Ocfs2-users] OCFS2 and LVM2


Hello,

Alexei_Roudnev:
> You must understand how LVM2 works. It is all STATIC system,
> so it ALWAYS can be used in cluster until you don't change metadata.

Well, thats quite an over simplification. You have a layered kernel API,
and the File System can introduce subtile asumptions which dont work on
virtual volumes. So especially the error handling, but also trivial
things as device discovery may not work out of the box with LVMs. And
then you get deeper stakcs and also all kind of possible races,
especially when kernel threads are involved. Do not even think about
trying it unless you are kernel hacker or it is oficially supported.

Gruss
Bernd


SEEBURGER AG
Headquarter:
Edisonstraße 1
D-75015 Bretten
Tel.: 0 72 52/96-0
Fax: 0 72 52/96-2222
Internet: http://www.seeburger.de
e-mail: info at seeburger.de

Vorstand:
Bernd Seeburger, Axel Haas, Michael Kleeberg

Vorsitzender des Aufsichtsrats:
Dr. Franz Scherer

Handelsregister:
HRB 240708 Mannheim

_______________________________________________
Ocfs2-users mailing list
Ocfs2-users at oss.oracle.com
http://oss.oracle.com/mailman/listinfo/ocfs2-users