Re: [Xen-users] domU sync I/O

[Roger Pau Monnà <roger.pau@xxxxxxxxxx>]

Hello,

El 29/10/15 a les 16.36, Simone Ricci ha escrit:
> Good evening everyone,
> 
> while trying out xenserver there was a thing that caught my attention: 
> apparently every I/O done in the guest magically ends being async toward the 
> storage (amongst other things I tried dd with oflags=sync). The storage BOX 
> exports the LUNs via FC, with writeback cache enabled. The very same dd 
> command issued from dom0, targeted to the same storage (onto an LV manually 
> created on the SR VG) acts correctly, i.e. I see SCSI command 0x35 
> (SYNCHRONIZE CACHE) issued to the target after every single block (confirmed 
> via quick-n-dirty dtrace script, storage is a zfs-based appliance); that 
> doesn't happen for the I/O issued from the guest.
> 
> In my point of view, this is a big issue which may compromise data integrity 
> in the event of a power failure or a generic HW fault in the storage BOX, 
> which may lead to guest filesystems corruption due to the possibility of 
> journal becoming corrupt. As a workaround I can obviously disable writeback 
> cache on the target, but that leads to unwarranted performance loss (because 
> every single I/O becomes synchronous, even there's no need to). Am I missing 
> something ?
> 
> Some hypothesis I made:
> 
> - tapblk3 drops sync cache commands from the guest

I cannot speak about tapdisk3, that's not part of the Xen Project source
code. Have you tried to ask on the xs-devel list?

https://lists.xenserver.org/sympa/subscribe/xs-devel

> - xen_blkfront into the domU ignores sync I/O semantics

xen-blkfront in Linux has support for flush since quite some time, can
you tell us which OS and version are you using in the guest?

Roger.


_______________________________________________
Xen-users mailing list
Xen-users@xxxxxxxxxxxxx
http://lists.xen.org/xen-users