[Ocfs2-users] OCFS2 on CentOS 4.5 for CRS/RAC

[Mark Fasheh]

On Mon, Nov 26, 2007 at 07:37:06AM -0800, Anjan Chakraborty wrote:
> Hi,
> I sent an email to Mark Fisheh of Oracle Corp. & posted this issue at OTN under
> Linux thread this morning. I hope that someone among you might have experienced
> this and can help. On that basis, I am sending this to you too. I am stuck &
> will really appreciate if you can shed some light on this.

Probably a lot of folks in the US were on vacation this last week. You
should get better traction now since most of us are back :)


> Thanks.
> Anjan
> ***********************************************************************************************************
> I have a 2 node CentOS 4.5 86_64 system (kernel 2.6.9-55.EL). On this I
> installed Oracle OCFS2 1.2.7-1 (with exact kernel matching). After this I
> installed Oracle CRS 10.2.0.1 and that installation went fine. Then I tried to
> install Oracle RDBMS 10.2.0.1 and all the problems started from there. The /var
> /log/messages file got filled up with messages (giving some to avoid
> confusion):
> ocfs2_read_locked_inode: .. : ERROR: Invalid dinode #0 signature =
> ocfs2_lookup: .. : ERROR: Unable to create inode ....

Are there any other types of messages on either node? The "Invalid dinode"
message is very generic unfortunately, so typically we're looking for
something before that to indicate a root cause.


> Then OUI gave several error messages, e.g.
> .... Invalid stored block length on file ...../em/em.war followed by I/O error
> in file
> Errors in invoking to files ins_rdbms.mk and ins_ldap.mk
> 
> Then /var/log/messages gave:
> OCFS2: ERROR (device ....): ocfs2_extend_file: Dinode # ...... has bad
> signature O' # I ....
> And the installation failed & CRS died. And the machines reboot.
> I ran fsck.ocfs2 -n /dev/...., it came clean.
> I have tested this several timnes & always same thing happening.
> If I use RAW partitions, everything works fine. So, the problem may be in the
> OCFS2 & OS/Oracle -- but, not sure how to bypass this.
> I have to have OCFS2 -- can't use RAW for various reasons.
> Can somebody please help me to resolve this?

Can you describe your shared disk setup? Also, send me your cluster.conf
files from all nodes.

Considering it's a fresh file system and you've only just started putting
files on it, my initial reaction is to check the shared disk. It could be
that blocks are somehow being cached so the file system is getting stale
or invalid meta data.

I think Luis suggested trying an older version of Ocfs2. Feel free to do
that, it could only add a potentially useful data point. You really don't
have to jump far back though - just try 1.2.6 for starters.

Thanks,
	--Mark

--
Mark Fasheh
Senior Software Developer, Oracle
mark.fasheh at oracle.com