Re: Connection Recovery

["Mallikarjun C." <cbm@rose.hp.com>]

Rahul Bhagwat wrote:
> 
> Hi,
> 
> The only possible state that CLEANUP_WAIT eventually ends up is
> FREE (on timeout or successful implicit/explicit logout).
> 
> If task reassignment happens unrelated to connection state, 

As I already said -
" A successful connection logout (implicit or explicit) must precede the
 task reassignments during a connection recovery operation."

Task reassignment *is* related to connection state FREE.

>is there any
> other resource associated with connection that can be still used while it is
> in CLEANUP_WAIT state.
> 
> Why do we have to wait for initiator to send logout to clean up things.

I am not sure I quite understood what you're saying here.  

There's an architected timeout (M1 transition) to reclaim the
resources, I only pointed out Logout as "highly desirable"
if all an implementation wants is to reclaim connection and
task resources.  

OTOH, Logout is mandatory if task reassignment is sought. 
A CLEANUP_WAIT state from initiator's perspective (let's say
due to an initiator NIC failure) does not imply a CLEANUP_WAIT
on the target (for the same connection).  Logout is a mechanism
that forces both sides to sync up on the FREE state, and is 
the architected mechanism to formally decouple the connection
allegiance of the task.  It is to avoid cases where target 
may receive stale PDUs on one connection, while receiving the
PDUs also on a (reassigned) different connection.  
-- 
Mallikarjun 


Mallikarjun Chadalapaka
Networked Storage Architecture
Network Storage Solutions Organization
MS 5668	Hewlett-Packard, Roseville.
cbm@rose.hp.com

 
> Regards,
> Rahul
> >
> > A successful connection logout (implicit or explicit) must precede the
> > task reassignments during a connection recovery operation.
> >
> > But please note that the notion of "connection cleanup" (graceful closing
> of
> > a
> > previously operational iSCSI connection) in the state diagrams goes beyond
> > the connection recovery (in fact, that is the reason I renamed from its
> > previous
> > name, please refer to my email to ips on 11/2/01 with the slide posting
> > announcement).
> > A connection cleanup is highly desirable even in the absence of task
> > reassignment,
> > to quickly reclaim the tags and buffers on either end (or, both sides
> would
> > have
> > to wait for a connection timeout to happen, symbolized by the M1
> > transition).
> >
> > >Once a CSM-E or a CSM-I
> > >drives the connection to free state, all the pending tasks need to be
> freed
> > up.
> >
> > Not correct.  The decision to free up the pending tasks is depedent on the
> > operational ErrorRecoveryLevel in the CSM-I case (please look at the
> > discussion in section 3.12.2), or is dependent on the Logout reason code
> > (recovery Vs close) in the CSM-E case.  All the FREE state symbolizes
> > really is that the iSCSI connection is gracefully closed with a successful
> > explicit/implit iSCSI Logout.  The pending tasks at this point have no
> > connection allegiance, and are loosely "owned" by the session.  It is
> > legitimate
> > for the pending tasks to be existent (waiting for reassignment) even when
> > all the connections reported FREE (please look at the discussion under
> > 3.15.2, Time2Retain).
> >
> > Regards.
> > --
> > Mallikarjun
> >
> > Mallikarjun Chadalapaka
> > Networked Storage Architecture
> > Network Storage Solutions Organization
> > Hewlett-Packard MS 5668
> > Roseville CA 95747
> >
> >
> > ----- Original Message -----
> > From: Rahul Bhagwat
> > To: ips@ece.cmu.edu
> > Sent: Monday, November 05, 2001 5:31 AM
> > Subject: Connection Recovery
> >
> >
> > Hi,
> >
> > Is there any order in task reassignments and connection logout (implicit
> or
> > explicit)
> > during a connection recovery.
> >
> > If these two are not related, what is the use of moving the connection to
> > CLEANUP_WAIT
> > state? CLEANUP_WAIT state typically means that there are pending tasks for
> > this
> > connection due to which it cannot be moved to FREE state. That is only
> > difference
> > betweeen FREE state and CLEANUP_WAIT state.
> >
> > Which probably means that it is mandatory that Task reassigment happens
> > before
> > logging out a failed connection (in CLEANUP_WAIT state). Once a CSM-E or a
> > CSM-I
> > drives the connection to free state, all the pending tasks need to be
> freed
> > up.
> >
> > Am I correct here?
> >
> > Regards,
> > Rahul
> >