[stgt] [PATCH] release backing store when offline

[James R. Leu]

Hello,

We did our original proof of concept deleteing and adding LUNs as
part of the passive/active transition. We saw a few issues with that
approach:

- the iSCSI clients had alot of churn and multipath would get in a
  really bad state sometimes, the iSCSI clients handle the
  'offline' case much more gracefully
- their are many more failure cases that need to be handled 
  when deleting and adding LUNs as part of a resource migration
- writing trouble shooting scripts is easier when both sides of the
  redundant setup always list the same LUNs.

I can understand if this is not a feature that you would like
in the main distribution.  In that case could you please sanity
check my work to make sure I'm not missing some corner case?

On Sat, Jul 16, 2011 at 06:27:37PM +0900, FUJITA Tomonori wrote:
> On Thu, 14 Jul 2011 08:07:23 -0500
> "James R. Leu" <jleu at inoc.com> wrote:
> 
> > We are building a geographically redundant SAN that will allow 
> > each LUN to be active on one node and passive on the other node
> > independent of the other LUNs.  To accomplish this we we
> > need to be able to mark a LUN as offline and have it release the
> > reference to the underlying block device so we can make it passive.
> 
> Why can't the passive target simply add a LUN instead of keeping it as
> offline? Then when the active target fails, the passive target can add
> the LUN?
> 
> 
> > DRBD is doing the block level replication and asserts the notion
> > of active/passive.  It will not allow a block device to transition
> > to passive if anything is holding a reference to it.
> > 
> > The iSCSI clients of the SAN are geographically redundant as well.
> > The resources on those clients need to be able to migrate between the
> > datacenters independent of each other.  To accomplish this it is
> > highly desirable to keep the TGTD configuration nearly identical
> > on the nodes of the SAN so that the redundant iSCSI clients see
> > the same LUNs no matter if their local resources are active or passive.
> > 
> > If my patch does not appear to be anything you want in the
> > main distribution could you at least look it over to see if I am
> > overlooking anything?
> 
> I don't think that it's a good idea to close a LUN when making it
> offline. Making a LUN offline means that it reuses it later. There is
> no point to close (release) it.

-- 
James R. Leu
Software Architect
INOC
608.204.0203
608.663.4555 fax
jleu at inoc.com
www.inoc.com

*** DELIVERING UPTIME ***
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 198 bytes
Desc: not available
URL: <http://lists.wpkg.org/pipermail/stgt/attachments/20110716/498dd31b/attachment-0001.sig>