[Ocfs2-users] PBL with RMAN and ocfs2

[Alexei Roudnev]

You compare 'shutdown abort / nointr' with 'no shutdown'. Unfortunately, we
must compare

- shutdown abourt / intr mode
with
- shutdown abort - everything freeze - power reset

Do they think that 'interrupt oracle' is worst then 'power reset serve in
the middle of HW operatioin' ?

Here is a problem - alternative  to 'intr' is 'POWER RESET' and not 'DONT
TOUCH AND WAIT'. If 'intr' can cause a data corruption then 'power reset'
can do exactly the same but with the higher probability.

And again - signal can interrupt IO only if these IO is already in the
frozen state (retry timeout happen on NFS) so it dont make any
difference for the normal shutdown.

(I know about this article and I can't agre with it).


----- Original Message ----- 
From: "Sunil Mushran" <Sunil.Mushran at oracle.com>
To: "Alexei Roudnev" <Alexei_Roudnev at exigengroup.com>
Cc: "Dheerendar Srivastav" <Dheerendars at bajajcapital.com>;
<Ocfs2-users at oss.oracle.com>
Sent: Monday, May 14, 2007 4:18 PM
Subject: Re: [Ocfs2-users] PBL with RMAN and ocfs2


http://www.netapp.com/tech_library/ftp/3183.pdf
<<<<<<<<
When running applications such as databases that depend on end-to-end
data integrity, you should use
“hard,nointr.” Oracle has verified that using intr instead of nointr can
expose your database to the risk
of corruption when a database instance is signaled (for example, during
a “shutdown abort” sequence).
 >>>>>>>>

Alexei Roudnev wrote:
> Oracle requires it BY MISTAKE and in documentation only (but not in the
> system). It works better with 'intr' mode on NFS.
>
> nointr prevents SHUTDOWN to kill application. As a result, if NFS server
> died you can't kill application (such as rman for example).
>
> I personally advice AGAINST 'nointr' in any environment esp. for the
things
> other then redo logs. The best setting for NFS should be
>
>   tcp,hard,intr,noac (rsize and wsize optional, and not 'bg')
>
> PS. Oracle dont require 'nointr' in reality; it require (verifies) 'hard'
> mount but 'nointr' is optional. 'hard' prevents io interruptions because
of
> network glitches and so is required for consistency, while having 'intr'
> option allows shutdown to work correctly and do not have any danger (no
> difference IO is sticked AND someone shutdown appllication).  Allowing
> 'nointr' mounts in production 100% guarantee problems with shutdowns. And
> btw, any non-network mount behaves as 'soft' and not 'hard' so it is even
> more amazing who selected these options in oracle (soft could be even
better
> choice because it translate IO errors to the application
> so appolication can make a preventive actions). I already had a fun
because
> of this strange requirements for 'hard' (we dont want 'hard' mount for
> backup area),
> but I can live with 'hard' if 'intr' is turned on (but not without it).
>
> PPS. What's the use of preventing interrupted IO-s? if someone can't
> shutdown your application because 'nointr' then he
> will power reset the whole server (so making things much worst because
> application can't shutdown normally). With 'intr', yes application got io
> error in io and signal in high level and will die (or shutdown depending
of
> signal) and it is better then power reset the whole box.
>
> Do you want application (oracle archiver for example) to be sticked
forever
> if one of NFS servers (one of a few archive destinations) don't answer? I
> dont think so,
>
>
> ----- Original Message ----- 
> From: "Sunil Mushran" <Sunil.Mushran at oracle.com>
> To: "Alexei_Roudnev" <Alexei_Roudnev at exigengroup.com>
> Cc: "Dheerendar Srivastav" <Dheerendars at bajajcapital.com>;
> <Ocfs2-users at oss.oracle.com>
> Sent: Monday, May 14, 2007 11:29 AM
> Subject: Re: [Ocfs2-users] PBL with RMAN and ocfs2
>
>
>
>> nointr blocks signals during the io itself. Prevents short ios.
>> AFAIK, Oracle requires a similar mount option with NFS too.
>>
>> Alexei_Roudnev wrote:
>>
>>> Btw, why Oracle like nointr?  Our analyze shows that 'intr' can be used
>>> instead in all cases andf is much safer. (at least for the NFS)
>>>
>>>  For NFS., intr means that mount is hard (application never got IO
>>>
> failure
>
>>> but just try IO forever - it's discussional too but we can agree with
>>>
> this
>
>>> approach)
>>> BUT if can be killed or interrupted. So that if IO is frozen and other
>>> application sends shutdown signal (such as kill -9) then application
>>>
> will be
>
>>> killed.
>>> Without intr, if your NFS server died, you wil not be able to shutdown
>>>
> your
>
>>> server without power reset (or reboot -n -f in some cases).
>>>
>>>
>>> ----- Original Message ----- 
>>> From: "Sunil Mushran" <Sunil.Mushran at oracle.com>
>>> To: "Dheerendar Srivastav" <Dheerendars at bajajcapital.com>
>>> Cc: <Ocfs2-users at oss.oracle.com>
>>> Sent: Monday, May 14, 2007 9:43 AM
>>> Subject: Re: [Ocfs2-users] PBL with RMAN and ocfs2
>>>
>>>
>>>
>>>
>>>> Ensure you have mounted the volume storing the
>>>> ocr and the voting diskfile with the "nointr,datavolume"
>>>> mount options.
>>>>
>>>>
>>>>
> http://oss.oracle.com/projects/ocfs2/dist/documentation/ocfs2_faq.html#RAC
>
>>>> Dheerendar Srivastav wrote:
>>>>
>>>>
>>>>> Dear sir ,
>>>>>
>>>>> We have used RHEL 4.0 with kernel 2.6.9-42.0.2.ELsmp with ocfs2 , we
>>>>> have oracle 10g 10.1.0.2
>>>>>
>>>>> I am working on a RAC installation (10.1.0.2)on RHEL 4.0 with EMC
>>>>> caliriion shared storage . The EMC device have been linked to raw
>>>>> devices .
>>>>>
>>>>> We are able to configure ocfs2 and
>>>>> ASM . When we will installed CRS the error message show " oracle
>>>>> cluster Registery can exists only as a shared file system file or as a
>>>>> shared raw partion .
>>>>>
>>>>> I would like request how to install the OCRFile .
>>>>>
>>>>>
>>>>>
>>>> _______________________________________________
>>>> Ocfs2-users mailing list
>>>> Ocfs2-users at oss.oracle.com
>>>> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>>>>
>>>>
>>>>
>>>
>>
>
>
> _______________________________________________
> Ocfs2-users mailing list
> Ocfs2-users at oss.oracle.com
> http://oss.oracle.com/mailman/listinfo/ocfs2-users
>