Re: iSCSI draft 02: logout

[Mark Bakke <mbakke@cisco.com>]

Matt and Somesh-

I agree that a target should be allowed to do this.  I would
agree with either a separate command or an addition to the
async_event.  This would be extremely useful when rebooting a
target, or performing a manual shutdown, to ensure a smooth
failover.

I would add a second timer to the one specified by Somesh, and
have these determined by the target, and communicated as part of
the message.  How about these:

1) Open time - the amount of time the connection will remain
   available (e.g. the time until the reboot or shutdown occurs).

2) Wait time - the amount of time to wait after the reboot, before
   attempting a new connection to the same address.  This would
   normally be the expected amount of time for the target to reboot,
   or for the address to become available via a failover to another
   unit.

After receiving this message, the initiator would stop sending
new commands, try to let any currently-executing commands finish,
and perform a normal connection logout.  It would then wait for
(open time + wait time) before attempting a new connection.

I would expect these timers to normally be a few seconds each, but
perhaps milliseconds would be an appropriate unit measure.

--
Mark

somesh gupta wrote:
> 
> --- Matt Wakeley <matt_wakeley@agilent.com> wrote:
> > Some comments on the logout command:
> >
> > - In Section 2.15, it indicates that the logout command is only used
> > for error
> > recovery.  I thought there were those that wanted to always use the
> > logout to
> > close a connection cleanly.
> >
> > - One would think that a "logout" on the last TCP connection of a
> > session
> > would close the session.  However, the error recovery steps in
> > section 4.1
> > seem to imply otherwise.  If closing the last TCP connection of a
> > session does
> > not close the session, what does?
> >
> > - A target should be allowed to "logout" an initiator.  This is
> > allowed in
> > fibre channel.  It might be nice to have a reason code field
> > indicating why
> > the logout was performed.
> 
> I had proposed (in a coversation with Julain in Pittsburg) a
> logout_requested command which the target could send to the initiator,
> and the target -- after waiting for some time (the target determines
> it)
> could then terminate/close any still open connections.
> 
> This feature would avoid entering recovery states (which in my
> experience aren't implemented/tested thoroughly) in cases where e.g.
> a controller board on a multi-controller array was to be serviced.
> 
> Initiators could divert all their traffic to other ports without having
> to go through session and command recovery.
> 
> Somesh
> 
> __________________________________________________
> Do You Yahoo!?
> Yahoo! Shopping - Thousands of Stores. Millions of Products.
> http://shopping.yahoo.com/

-- 
Mark A. Bakke
Cisco Systems
mbakke@cisco.com
763.398.1054