[Stgt-devel] data corruption problems with stgt (aborted journal, remounting ro)?

[FUJITA Tomonori]

On Sat, 02 Feb 2008 10:39:02 +0100
Tomasz Chmielewski <mangoo at wpkg.org> wrote:

> FUJITA Tomonori schrieb:
> > On Fri, 01 Feb 2008 13:05:41 +0100
> > Tomasz Chmielewski <mangoo at wpkg.org> wrote:
> > 
> >> Tomasz Chmielewski schrieb:
> >>> Doesn't look my posts get to this list... or is it just lagged a lot? 
> >>> Resending.
> >>>
> >>>
> >>> Perhaps I'm doing something wrong - but with stgt I'm facing problems I 
> >>> didn't have with IET or SCST.
> >>>
> >>> Whenever I kill tgtd daemon and start it again (i.e., target server 
> >>> restart), the initiator detects an aborted journal and remount the 
> >>> device ro.
> >>>
> >>> Why is it so?
> >>>
> >>> What is the recommended way to kill the tgtd daemon? It doesn't seem to 
> >>> react on TERM signal.
> >> Hello, anyone there?
> >>
> >> Is there a way to restart tgtd daemon or a machine running tgtd, so that 
> >> iSCSI connections don't break?
> > 
> > What does your 'restart tgtd daemon' mean? For me, 'restart' involves
> > stopping tgtd daemon and it closes all the iSCSI connections.
> 
> Stop it, and start again?
> 
> Imagine you want to upgrade your tgtd daemon, a kernel running on that 
> machine, or you have to restart the target machine for some other reason 
> (i.e. your target machine died).
> 
> With IET or SCST there is no problem with that - stop the target, and 
> iSCSI initiator will try to reconnect.

How did you restart IET?

/etc/init.d/iscsi-target restart

?


> By default, open-iscsi tries to reconnect for 120 seconds without 
> returning an error to the SCSI layer, as defined /etc/iscsi/iscsid.conf:
> 
> node.session.timeo.replacement_timeout = 120
> 
> 
> But there is no problem to increase that value to even a couple of days 
> (imagine: admin stops the target machine erroneously on Friday, comes 
> back to work on Monday, starts the target, and initiators continue to 
> work as if nothing happened - processes were in an uninterruptible sleep 
> state, waiting for I/O operations to complete).
> 
> 
> With tgtd, it seems impossible to me - or am I wrong?
> 
> 
> -- 
> Tomasz Chmielewski
> http://wpkg.org