Re: Linux Native Multipath, ASM and Instance Failures

Hi Guillermo,

On 26/07/2012 18:44, Guillermo Alan Bort wrote:
> We have the following set up:
>
>    1. RHEL 5.8 (standard RH kernel)
>    2. Oracle RAC 11.2.0.3 (Jan PSU)
>    3. Linux Native Multipath (/dev/mapper)
>    4. 3PAR storage (don't know much about the storage layer, though).
>    5. NO ASMLIB is used, the asm diskstring is /dev/mapper/*p1
>
>    We were running some redundancy tests (pulling cables and seeing what
> happens) and when the servers lost a path, the instances crashed. I'm still
> gathering logs, but OS errors looks like this:
> Jul 26 09:40:41 tvl-p-orep001 kernel: end_request: I/O error, dev sdbg,
> sector 4151
[...]
> and then
>
> Jul 26 09:40:43 tvl-p-orep001 kernel: device-mapper: multipath: Failing
> path 65:192.
[...]
> In the meantime ASM logs show this:
>
> WARNING: Read Failed. group:0 disk:22 AU:0 offset:0 size:4096
> Errors in file
> /u01/ORAUTL/grid/base/diag/asm/+asm/+ASM/trace/+ASM_ora_18784.trc:
> ORA-27061: waiting for async I/Os failed
> Linux-x86_64 Error: 5: Input/output error
> Additional information: -1
> Additional information: 4096
Just for your information:
we have no problems with RHEL 5.7 (RH kernel), RAC 11.2.0.3.2, 3PAR 
_and_ ASMLib.
We did the same tests and we had no problems (there were messages for 
the failing paths in the OS logs [as expected], but the Oracle stack
remained up and running (no error messages at all in the various alert 
logs).

If I recall correctly some MOS notes suggest to set ORACLEASM_SCANORDER 
to dm (/dev/dm-* as opposed to /dev/mapper/* ).
As far as I know the fact that the names dm-* are not persistent 
shouldn't be a problem when clusterware files (voting/ocr) are in ASM 
disk groups (11.2).
I would try to set asm_diskstring to /dev/dm-* and then I would repeat 
the tests.


Regards
Dimitre
--
http://www.freelists.org/webpage/oracle-l


Other related posts: