RE: active-pasive cluster in linux

One question is whether you want the diagnostic logs inside what it might be
logging about. So even if you do use Oracle clustered file systems it can be
of value to understand that non-clustered file systems on shareable volumes
can be mounted read only with no increase in the chances of corruption. And
I believe the OP of this thread (Juan Cruz Miranda Vigo
[jcmiranda@xxxxxxxxx]) was asking about alternatives to using Oracle
clusterware. That was David's point if I understood him. If you're going
active-passive, just can just mount the file systems on the active node. I
was adding that it can be useful and is no additional risk to mount the file
systems containing the diagnostic logs read only on the passive node.

 

  _____  

From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of LS Cheng
Sent: Tuesday, February 23, 2010 2:24 PM
To: Mark W. Farnham
Cc: ballester.david@xxxxxxxxx; jcmiranda@xxxxxxxxx; oracle-l@xxxxxxxxxxxxx
Subject: Re: active-pasive cluster in linux

 

That is my point even mounting and dismounting filesystems is doable do we
really want to do it working with a clustering software which does not know
about filesystems except his own "filesystem" which is ASM? That is why I
said ASM is a must if we want to use Oracle Clusterware to setup an
active-passive cluster.

Thanks

--
LSC





On Tue, Feb 23, 2010 at 7:08 PM, Mark W. Farnham <mwf@xxxxxxxx> wrote:

Most ports support a read only mount of a volume mounted read/write on one
node. The only wrinkle I've noticed is that the appearance of new objects on
the non-rw node may be slightly delayed. I'm uncertain of the exact
mechanism of this delay, and it may vary by port.

 

An interesting application of this is mounting all the log directories (both
archived redo and diagnostic) as read only without the need for clustering
software at the file system level. (They still have to be shareable volumes,
of course). IF you're not storing online redo in ASM this can also be used
for the online redo logs, but of course you need the dbfiles multi-rw-able
for any flavor of RAC. But this thread isn't really about RAC.

 

Purely for the active-passive idea, this of course means you can tail files
on read mounted volumes and see the dying diagnostics from a crashed node
without waiting for some other node to start or waiting to mount the
volumes. Then you have the last messages at hand if you want to look at them
before you activate the passive node. This can save a lot of time if the
logs are telling you about, for example, media failure or some other issue
that will just be repeated on the passive node.

 

Presumably the interest in this type of architecture is minimizing the time
until the applications are available on the passive node if the active node
is in some sort of failure, as well as for routine planned switchover for
preventive maintenance and just to prove it will work.

 

<I wrote too much. I just wanted to introduce the notion about
non-corrupting read only mountablility even without a clustered file system;
David is absolutely correct you don't want multiple concurrent rw mounts
without a clustered file system, even if you can trick the OS into doing it
for you.>

 

mwf

  _____  

From: oracle-l-bounce@xxxxxxxxxxxxx [mailto:oracle-l-bounce@xxxxxxxxxxxxx]
On Behalf Of David Ballester
Sent: Monday, February 22, 2010 6:26 PM
To: LS Cheng
Cc: jcmiranda@xxxxxxxxx; oracle-l@xxxxxxxxxxxxx
Subject: Re: active-pasive cluster in linux

 

 

2010/2/22 LS Cheng <exriscer@xxxxxxxxx>

Hi

You can use OCFS2 if you wish but basically you need some sort of Cluster
Filesystem or Volume Manager such as ASM.

I dont mention OCFS2 because introducing two clustering software (OCFS2 and
CRS) for a Single instance Active Passive solution is a bit too much in my
opinion that is why I said you are forced to use ASM (runs on top of CRS).

Thanks

 


Or not use ocfs2 or any cluster filesystem, the only node that needs to have
mounted the filesystems is the active one

If you try to mount a non cluster filesystem in another node while is
mounted in a first one, the other node should answer 'partition used' or
something similar ( prevents corruption )

when switching from nodes, as part of the automatic process, one node should
umount and the other mount ( may be fsck first )


Regards


D.

Other related posts: