Jeremy Katz wrote:
On Mon, 2006-02-06 at 15:06 -0800, Patrick Mansfield wrote:
On Mon, Feb 06, 2006 at 04:39:00PM -0500, Jeremy Katz wrote:
Okay with all your comments (thanks) except:
On Fri, 2006-02-03 at 14:04 -0800, Patrick Mansfield wrote:
+ def shutdown(self):
[snip]
+ # Note that iscsid has/had code to ignore 2 (SIGINT), hence the
+ # 9 (SIGKILL).
If we run iscsid with -p and save the pid in /tmp, then we can have a
much more robust kill.
There are really two iscsid processes, I think one is the logger and the
other the true daemon, but only one pid in the pid file, so I can't use
that to kill both.
There was a shutdown patch submitted (add iscsid '-k' option), that would
work really nice here but it is not in iscsi svn (nor I assume FC devel).
That's really broken ;)
Yeah.
Mike -- any chance of getting this included so that shutdown of iscsid
can be much nicer?
It is on the TODO.
It would be good for using in the initscript as
well. Just killing processes like this is really bad for cases where
people are using chroots and, potentially in the future, other types of
containers like the openvz patches that were sent to lkml recently.
I don't think the "iscsid -k" patch is what we want. The -k argument
meant that iscsid will run and cleanup running iscsi sessions in the
kernel before stating normal operations.
Are we just looking for a way to shutdown the daemon or both daemon and
running kernel sessions. Either way, I think it would be best to add a
iscsiadm command. iscaidm would then tell the daemon to either remove
all the running sessions in the kernel then stop or to just kill the
daemon in such a way that we can restart it later and it can rebuild its
state.
