[dm-devel] Does multipath require disabled write cache?

Mike Snitzer snitzer at redhat.com
Mon Jul 21 17:49:39 UTC 2014


On Sun, Jul 20 2014 at  5:20pm -0400,
Spelic <spelic at shiftmail.org> wrote:

> Hello all,
> I am wondering if multipath requires a disabled write cache at the
> remote disks.
> We have some external SAS disks, for which the connection is flaky
> and sometimes is lost and then they are detected again. Multipath
> seems perfect for this situation.
> However, when the disks are lost, I suppose resets are issued, which
> might clear the write cache so that data located there might not be
> committed (true? well at least on power loss this is true).

You should disable both the controller cache and disk caches if they are
volatile.  Filesystems will issue flushes (and DM will handle those
flushes) but until those flushes return successfully the associated
outstanding data is not considered on permanent storage.

> It would be great if multipath would issue a flush every N  writes,
> and then remember the last N  writes and resubmit them when the path
> is reestablished, so that there is again chance to flush them to
> stable storage. (similar to DRBD epochs) However I don't know if
> this smart mechanism is implemented.

It isn't implemented.




More information about the dm-devel mailing list