On 17.05.2013 10:16, Li Wei wrote:
On 05/17/2013 04:07 PM, Michal Privoznik wrote:
On 17.05.2013 07:47, Li Wei wrote:
Hi Michal, Daniel
It seems there is something wrong with the 2/2 part of this patchset. When I do an "change-media" command in virsh, it doesn't do any better than before, but even worse(I must to wait 5 secs to see the error).
I'm not family with libvirt, just add some log things in the qemu_hotplug.c and found the tray_status never change to open, but with michal's original patch (which do active poll on tray_status), I can do "change-media" successfully every time.
Yes & no. Thing is - originally we ignored tray status in the guest. So in case guest has locked the tray and thus ignored our eject request, we have changed the media anyway, resulting in possible I/O errors within guest.
With my patch, we send eject request to the qemu, and actively poll for the tray to open. We check for up to 10 times every 200ms. So guest has 2 seconds to open the tray. If the guest doesn't eject media and open the tray we don't proceed with forcibly changing the media.
These patches are actually a bug fix for https://bugzilla.redhat.com/show_bug.cgi?id=892289
Hmm... I think you misunderstood my meaning(sorry for my pool english), simply to say, I can get expected behavior by apply patch V1 but not V2.
In V1, we do active poll on qemuMonitorGetBlockInfo(), and in V2 we just wait for origdisk->tray_status to become VIR_DOMAIN_DISK_TRAY_OPEN but this never happened.
Thanks, Wei
Aaah, now I understand. So you are saying that the event is not being delivered. What's the qemu version? If you turn the debug logs on, do you see DEVICE_TRAY_MOVED in them? http://wiki.libvirt.org/page/DebugLogs Or maybe the timeout is just short for your guest to eject the tray. So after a while, does 'virsh change-media' succeed? Michal
IMHO, there must be some bug in the processing of qemu tray change.
Thanks, Wei
On 01/25/2013 08:20 PM, Michal Privoznik wrote:
The first patch is bare bug fix (without any bug reported though). The second is again a bug fix, but not so easy to spot. Basically, when we change a media, we should issue 'eject' first, then wait until the tray gets open, after which we can finally issue 'change' command. Even for bare 'eject' we ought to check status, shouldn't we?
Michal Privoznik (2): qemu_monitor: Fix tray-open attribute in query-block qemu_hotplug: Rework media changing process
src/qemu/qemu_hotplug.c | 51 +++++++++++++++++++++++++++++++++++++++++--- src/qemu/qemu_monitor_json.c | 2 +- src/qemu/qemu_monitor_text.c | 6 +++--- 3 files changed, 52 insertions(+), 7 deletions(-)
-- libvir-list mailing list libvir-list@redhat.com https://www.redhat.com/mailman/listinfo/libvir-list