Re: Issue with QEMU Live Migration

[Arisetty, Chakri]

> * How is 'issue block-job-cancel' command issued exactly at the
> PRE_SWITCHOVER stage? virsh blockjob --abort?

We are currently not using libvirt/virsh to issue QMP command. Our software 
makes up QMP connection and sends the QMP command to cancel the job when RAM 
migration state is in pre-switchover.

qemu_blockdev_cancel($user, "node-ssci-disk-0");

sub qemu_blockdev_cancel {
    my $user   = shift;
    my $device = shift;

    return qemu_control(
        $user,
        {
            "execute"   => "block-job-cancel",
            "arguments" => {
                "device" => $device,
            }
        }
    );
}

>* Recently a postcopy issue, wherein the migrated guest on the
> destination machine hangs sometimes with migrate-postcopy but not with
> virsh ---postcopy-after-precopy. It seems virsh(1) handles the switch
> better. Wondering if it's similar with 'block-job-cancel'.

Thank you for the pointer, we are currently using pre-switchover. II will look 
more into libvirt as how it is implemented.

Thanks
Chakri


On 8/26/24, 5:05 AM, "Prasad Pandit" <ppandit@redhat.com 
<mailto:ppandit@redhat.com>> wrote:


!-------------------------------------------------------------------|
This Message Is From an External Sender
This message came from outside your organization.
|-------------------------------------------------------------------!


On Sun, 25 Aug 2024 at 22:40, Arisetty, Chakri <carisett@akamai.com 
<mailto:carisett@akamai.com>> wrote:
> > - start the mirror job
> > - qmp_migrate
> > - once PRE_SWITCHOVER is reached, issue block-job-cancel
> > - qmp_migrate_continue
>
> We use exact same steps to do live migration. I repeated the test now
>
> Sure, as you suggested to rule out any incorrect usage, I repeated the test 
> with above steps.
> once RAM migration state moved to pre-switchover, issued block-job-cancel. 
> There are no more dirty blocks.
> But all the disk writes from 'pre-switchover' state to 'complete' state are 
> lost.
> Thus, it is creating loss of customer data.
>


* How is 'issue block-job-cancel' command issued exactly at the
PRE_SWITCHOVER stage? virsh blockjob --abort?


* Recently a postcopy issue, wherein the migrated guest on the
destination machine hangs sometimes with migrate-postcopy but not with
virsh ---postcopy-after-precopy. It seems virsh(1) handles the switch
better. Wondering if it's similar with 'block-job-cancel'.


Thank you.
---
- Prasad