Question about delaying dirty log stop to next vm_start

Question about delaying dirty log stop to next vm_start

[Yi Sun]

Hi, Jay,

We are testing some live migration cases recently. We found your below
patch delays the dirty log stop to the next vm_start.

commit 1931076077254a2886daa7c830c7838ebd1f81ef
Author: Jay Zhou <jianjay.zhou@huawei.com>
Date:   Fri Jul 28 18:28:53 2017 +0800

    migration: optimize the downtime

We understand this commit can optimize the downtime. But we observed
that the dirty log stop cannot be executed if user inputs "quit" in qemu
monitor after migration completes. Have you considered such case before?
Any suggestion? Thanks!

BRs,
Yi Sun

RE: Question about delaying dirty log stop to next vm_start

[Zhoujian (jay)]

Hi Yi,

> -----Original Message-----
> From: Yi Sun [mailto:yi.y.sun@linux.intel.com]
> Sent: Monday, August 2, 2021 3:54 PM
> To: Zhoujian (jay) <jianjay.zhou@huawei.com>
> Cc: qemu-devel@nongnu.org; pbonzini@redhat.com; sanjay.k.kumar@intel.com;
> yi.l.liu@intel.com; yi.y.sun@linux.intel.com
> Subject: Question about delaying dirty log stop to next vm_start
> 
> Hi, Jay,
> 
> We are testing some live migration cases recently. We found your below patch
> delays the dirty log stop to the next vm_start.
> 
> commit 1931076077254a2886daa7c830c7838ebd1f81ef
> Author: Jay Zhou <jianjay.zhou@huawei.com>
> Date:   Fri Jul 28 18:28:53 2017 +0800
> 
>     migration: optimize the downtime
> 
> We understand this commit can optimize the downtime. But we observed that
> the dirty log stop cannot be executed if user inputs "quit" in qemu monitor after
> migration completes. Have you considered such case before?

If the migration is successfully finished, the src VM should be destroyed. 

If user inputs "quit" in qemu monitor after migration completes, the VM is still
paused(vcpu is stopped), dirty log takes no effect; if the migration failed, the dirty log
stop can be executed after src VM started.

Thanks,
Jay Zhou