* shutting down logical volumes once an underlying device fails?
@ 2012-09-26 16:11 Christoph Hellwig
2012-09-26 20:11 ` Mike Snitzer
0 siblings, 1 reply; 3+ messages in thread
From: Christoph Hellwig @ 2012-09-26 16:11 UTC (permalink / raw)
To: dm-devel
A coworker has been testing failure scenarious using lvm, and noticed
that when a PV fail a filesystem on a logical volume that spans multiple
PVs only fails once it accesses the actually failed physical device.
At least for system using higher level error detection and replication
behaviour that fails a LV as soon as one of the underlying devices fails
would be much more helpful. Is the current LVM behaviour intention and
should we work around this using userspace monitoring, or would be
patches to introduce a shutdown state similar to what we have in XFS for
example be welcome?
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: shutting down logical volumes once an underlying device fails?
2012-09-26 16:11 shutting down logical volumes once an underlying device fails? Christoph Hellwig
@ 2012-09-26 20:11 ` Mike Snitzer
2012-09-28 13:05 ` Christoph Hellwig
0 siblings, 1 reply; 3+ messages in thread
From: Mike Snitzer @ 2012-09-26 20:11 UTC (permalink / raw)
To: Christoph Hellwig; +Cc: dm-devel
On Wed, Sep 26 2012 at 12:11pm -0400,
Christoph Hellwig <hch@infradead.org> wrote:
> A coworker has been testing failure scenarious using lvm, and noticed
> that when a PV fail a filesystem on a logical volume that spans multiple
> PVs only fails once it accesses the actually failed physical device.
>
> At least for system using higher level error detection and replication
> behaviour that fails a LV as soon as one of the underlying devices fails
> would be much more helpful. Is the current LVM behaviour intention and
> should we work around this using userspace monitoring, or would be
> patches to introduce a shutdown state similar to what we have in XFS for
> example be welcome?
Don't think any shutdown should be initiated by the kernel.
But events that userspace could respond to by deactivating (an
arbitrarily complex) lvm device would be the way forward IMHO.
Could this tie back to needing better UA handling/notification between
kernel<->userspace?
^ permalink raw reply [flat|nested] 3+ messages in thread
* Re: shutting down logical volumes once an underlying device fails?
2012-09-26 20:11 ` Mike Snitzer
@ 2012-09-28 13:05 ` Christoph Hellwig
0 siblings, 0 replies; 3+ messages in thread
From: Christoph Hellwig @ 2012-09-28 13:05 UTC (permalink / raw)
To: device-mapper development; +Cc: Christoph Hellwig
On Wed, Sep 26, 2012 at 04:11:17PM -0400, Mike Snitzer wrote:
>
> Don't think any shutdown should be initiated by the kernel.
>
> But events that userspace could respond to by deactivating (an
> arbitrarily complex) lvm device would be the way forward IMHO.
Deactivating is maybe a bit of a too strong word. But I'd expect we'll
get EIO from al sectors not just those on the failed device.
> Could this tie back to needing better UA handling/notification between
> kernel<->userspace?
That current proposed solution we have internally is to simply poll
the status output for the PVs and handle it all in userspace. To me
this seems a bit clumsy, but at least it's easily doable in short time.
^ permalink raw reply [flat|nested] 3+ messages in thread
end of thread, other threads:[~2012-09-28 13:05 UTC | newest]
Thread overview: 3+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2012-09-26 16:11 shutting down logical volumes once an underlying device fails? Christoph Hellwig
2012-09-26 20:11 ` Mike Snitzer
2012-09-28 13:05 ` Christoph Hellwig
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).