qemu-devel
[Top][All Lists]
Advanced

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: dirty bitmap migration refactor


From: Vladimir Sementsov-Ogievskiy
Subject: Re: dirty bitmap migration refactor
Date: Thu, 30 Apr 2020 07:53:09 +0300
User-agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101 Thunderbird/60.2.1

29.04.2020 16:29, John Snow wrote:
Hi all,

as you are probably aware I haven't been paying attention to dirty
bitmap work very much for the past month.

Around KVM Forum, we had a giant thread dedicated to discussing the
problems with dirty bitmap migration, which in a nutshell, are that it
migrates using the node name with no option for re-routing or re-naming
nodes.

IIRC, there was a patchset to fix this quickly sent by Virtuozzo, but
the series stalled because it was quite close to a release and was
deemed too risky.

What is the status of those patches, if any? Do we need to start from
scratch to implement the functionality that libvirt wants here?


Hi!

There are two series now in list:

"[PATCH v2 0/5] fix migration with bitmaps and mirror" - the series you are 
saying about

I made some changes to it downstream, to restrict migration by generated node 
names at all, as we had a bug. I can resend, if needed.

What the series does? It just tries to migrate by blk name even for filtered 
nodes. This fixes migration of bitmaps during mirror. But it doesn't apply to 
blockdev-era (doesn't hurt still). So can someone analyze, do we need this fix 
in current Qemu? Or is it for downstream only? Or should we take it just to 
make downstreaming cleaner?

===

The second is "[PATCH v2 00/22] Fix error handling during bitmap postcopy"

- it fixes, how postcopy bitmap migration behaves on errors, and we need it 
anyway.

===

What to do next? I have a plan to post series to implement new API, discussed 
on list, to make mapping from bitmaps on source to bitmaps on target by hand.

--
Best regards,
Vladimir



reply via email to

[Prev in Thread] Current Thread [Next in Thread]