Re: [Qemu-devel] [RFC PATCH] replication agent module

From: Paolo Bonzini <pbonzini@redhat.com>
To: Ori Mamluk <omamluk@zerto.com>
Cc: "Kevin Wolf" <kwolf@redhat.com>,
	"תומר בן אור" <tomer@zertodata.com>, "עודד קדם" <oded@zerto.com>,
	dlaor@redhat.com, qemu-devel@nongnu.org
Subject: Re: [Qemu-devel] [RFC PATCH] replication agent module
Date: Tue, 07 Feb 2012 15:40:27 +0100	[thread overview]
Message-ID: <4F3137DB.1040503@redhat.com> (raw)
In-Reply-To: <4F312FD3.5020206@zerto.com>

On 02/07/2012 03:06 PM, Ori Mamluk wrote:
> The main issue about it is that the Rephub also needs the other
> direction - to read the protected volume. I get the feeling that with
> live block copy and NBD there's probably something that might fit
> this need, no?

Yes, with two NBD sockets you could do it.  But would you use both at 
the same time?  I would have thought that either the rephub is streaming 
from the protected volume, or QEMU is streaming to the rephub.

The current streaming code in QEMU only deals with the former. 
Streaming to a remote server would not be supported.

> With a 'new' agent like I need this is relatively easily achieved by a
> bidirectional protocol, but I agree a more generic protocol would be
> more elegant, although it will probably require a socket per direction, no?
>
> I Some smaller questions:
> * Is there already a working iScsi initiator as a block driver (I hope
> I'm using the right terminology) in Qemu, or do I need to write one?

Yes, there is one using libiscsi.  But I think Anthony was not referring 
to iSCSI in particular, NBD would work just as well.

> * This driver would need to be added in run-time - to allow starting to
> protect a running VM. Maybe via a monitor command. I guess that's OK,
> right?

Yes, I think you can detach a block device from a drive and reattach the 
new mirroring device.

> * What can you say about NBD via iScsi - with respect to our
> requirements- who is more mature in Qemu?

Personally I prefer NBD because it is lighter-weight and there is a 
server inside QEMU (so you can use it easily with non-raw images).  It 
is more mature, but it is a bit less extensible.

> One more thing about the iScsi initiator - it will not be a standard
> backing for a drive, because the 'production' drive (i.e. the original
> image) is more important than the replicated one. This means that even
> though we use iScsi, this is still a replication agent - not a generic
> 'additional' iscsi backing.

Yes, understood.

Paolo