[PATCH 0/9] Introduce instant memory snapshots

[Nikolay Shirokovskiy]

пт, 12 нояб. 2021 г. в 10:39, Peter Krempa <pkrempa at redhat.com>:

> On Thu, Nov 11, 2021 at 11:55:45 +0300, Nikolay Shirokovskiy wrote:
> > Hi, everyone.
> >
> > The details for instant snapshot are given in main patch "qemu: support
> > instant snapshots". Here I only want to mention two moments.
> >
> > First as instant snapshot API is synchronous we need to add an event
> > that snapshot moment in time is passed so that client is free to make
> > changes to guest. This way client can make use of instant snapshot while
> > memory is still being written to disk.
>
> This is a nice replacement for --live. I had a dummy implementation for
> this but I put it aside for the following reasons.
>
> 1) It doesn't work with all VM configs.
>
> My original approach wanted to use it as a full replacement for the
> snapshot code. That wouldn't work.
>
> Yours uses a new flag which is okay as it sidesteps the issue. (One
> detail, docs don't mention the caveat, more on that inline).
>
> 2) The format change for instant load
>
> Below you mention that you want to change the image format for instant
> load. Introducing the feature now, without the image format change means
> that there will be a need for yet another flag switching to the new
> format.
>
> I'd strongly suggest you reconsider whether to push this now or just
> simply go with the full new format once it's ready in qemu.


Hi, Peter.

Thank you for your review. Ok, let's wait for new format in qemu.

Nikolay



>
>
> > Second there is work in progress on adding asynchronous revert to
> > snapshot [1]. The idea is to write memory during snapshot in such a way
> > that it is possible to do a revert to snapshot in background. Namely
> > start guest on revert with only minimal amount of memory restored from
> > disk and then restore memory in background in postcopy migration
> > fashion. In order to make it possible snapshot memory should be written
> to
> > disk not just as plain migration stream. Thus the [1] patch adds
> > a special tool to write the migration stream during snapshot.
> >
> > So if/when patch [1] will be merged I plan to add 'instant revert' mode
> > to snapshot too.
> >
> > [1] [RFC PATCH v2 0/7] migration/snapshot: External snapshot utility
> > https://lists.nongnu.org/archive/html/qemu-devel/2021-05/msg03619.html
>
>
> I'll add few review points inline.
>
> Also note that Pavel Hrdina might want to comment too as he is currently
> making good progress on making reversion of external snapshots work.
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://listman.redhat.com/archives/libvir-list/attachments/20211115/9e68a812/attachment-0001.htm>