[Ocfs-users] ORA-03113 EOF on Comm Channel when using OCFS + LVM + Striping

[John Smiley]

More info - CREATE DATABASE fails with ORA-03113.  An
examination of the trace files shows that the log
writer process (LGWR) terminated the instance due to
an interrupted system call.  The trace file is
attached.

P.S.  I have also upgraded my aic7xxx drivers from the
stock 6.2.36 to the latest version (6.3.3) on all
three nodes.  It didn't make any difference for this
problem (I just installed the 6.3.3 drivers today; all
prior testing was done with the 6.2.36 drivers).  

Here is the link for the drivers in case anyone's
interested.  These are maintained by Adaptec.
http://people.freebsd.org/~gibbs/linux/RPM/aic7xxx

--- John Smiley <pro_oracle at yahoo.com> wrote:
> While creating a new Oracle RAC database, I am
> getting
> "ORA-03113 - end-of-file on communication channel"
> during the create database or create tablespace
> phase.
>  This only happens if I use OCFS on top of LVM with
> striped logical volumes.  Everything works fine if
> the
> LV is not striped.
> 
> I have a three-node RAC cluster using a six disk
> JBOD
> connected to each node via a single SCSI bus (daisy
> chained).  I would like to use SAME (Stripe And
> Mirror
> Everything) and have been using the striping feature
> of LVM to do this.  Here are the configuration
> details:
> 
> 6 36GB SCSI disks as JBOD
> 1 36GB primary partition on each disk of type 8e
> (LVM)
> 1 Volume group consisting of all 6 PVs (total of
> ~200G) named /dev/oracle
> 1 LV striped across all 6 PVs with a stripe width of
> 128k of size 2G (lvcreate -C y -i 6 -I 128 -n demo
> -L
> 2g /dev/oracle
> OCFS file system built on /dev/oracle/demo with 128k
> block size (mkfs.ocfs -b 128 -F -g 502 -u 501 -L
> /u02
> -m /u02 /dev/oracle/demo
> 
> oracm and gsd start fine on all nodes, but when I
> try
> to create the database with dbca
> -datafileDestination
> /u02/oradata, it fails with ORA-03113 during
> database
> creation or tablespace creation.  The same error
> occurs if I try to create the database manually.
> 
> If I use the exact same configuration, but with a
> non-striped LV, everything works fine.
> 
> I have reproduced these results on RHEL 3 with the
> original and errata kernels (2.4.21-4.EL and
> 2.4.21-4.0.1.EL) and UL 1.0 with the SP2a kernel
> (304).  I have even tried upgrading the kernel LVM
> modules from the supplied 1.0.5 version to the
> latest
> 1.0.8 version from Sistina (patch the kernel with
> Sistina's code and re-compile the kernel, install
> the
> new modules and kernel, boot the custom kernel,
> etc.)
> 
> I've tried ocfs 1.0.9-9, 1.0.9-11, and 1.0.9-12.
> 
> I have also tried md to create a striped RAID 0
> device
> instead of using LVM, but got the same results.
> 
> Any thoughts on what I'm doing wrong or is there a
> problem with using ocfs on striped volumes?
> 
> John Smiley
> Sr. Database Architect
> Sprint Corporation
> _______________________________________________
> Ocfs-users mailing list
> Ocfs-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs-users
-------------- next part --------------
A non-text attachment was scrubbed...
Name: demo1_lgwr_7284.trc
Type: application/octet-stream
Size: 10071 bytes
Desc: demo1_lgwr_7284.trc
Url : http://oss.oracle.com/pipermail/ocfs-users/attachments/20031214/4b072086/demo1_lgwr_7284.obj