[libvirt] 答复: Re: 答复: Re: [PATCH] qemu: change the name of tap device for a tapand bridge network

Thu May 4 01:02:35 UTC 2017

(Please don't top-post replies. ALso, I added libvir-list back to the To:)

On 05/03/2017 08:46 PM, lu.zhipeng at zte.com.cn wrote:
> 
> Some other scenarios such as migration. 
> 
> Can we guarantee that the network card driver has been
> closed before  killing QEMU ?

During a migration, before we ever get to the point of killing the
source qemu process, we have already stopped the guest CPUs and made
sure that the entire state of the guest has been transferred to the qemu
process on the destination host. It doesn't matter if we detach the tap
device from the bridge or not - any traffic received on the source host
during that time (after stopping the guest CPUs) will be dropped (and
re-transmitted + received by the guest after it starts running on the
destination).

Since the guest CPU is already stopped on the source, and the guest on
the destination already has the full state of the guest ready to start
up (and will not receive any new updates), keeping the tap device
connected to the bridge for a small fraction of a second is not going to
make any difference at all.

Is that convincing enough? :-)

> 
> if we are sure,that's ok
> 
> 
> 
> 
> 
> *为了让您的VPlat虚拟化故障得到高效的处理，请上报故障到: $VPlat技术支持。*
> 
> 芦志朋 luzhipeng
> 
> 
> IT开发工程师 IT Development Engineer
> 操作系统产品部/中心研究院/系统产品 OS Product Dept./Central R＆D
> Institute/System Product
> 
> 
> 	
> 深圳市南山区科技南路55号中兴通讯研发大楼33楼
> 33/F, R&D Building, ZTE Corporation Hi-tech Road South,
> Hi-tech Industrial Park Nanshan District, Shenzhen, P.R.China, 518057
> T: +86 755 xxxxxxxx F:+86 755 xxxxxxxx
> M: +86 xxxxxxxxxxx
> E: lu.zhipeng at zte.com.cn
> www.zte.com.cn <http://www.zte.com.cn/>
> 
> 原始邮件
> *发件人：*＜laine at laine.org＞;
> *收件人：*芦志朋10108272;
> *日 期 ：*2017年05月04日 08:31
> *主 题 ：**Re: 答复: Re: [libvirt] [PATCH] qemu: change the name of tap
> device for a tapand bridge network*
> 
> 
> 
> 
> On May 3, 2017 8:28 PM,  ＜lu.zhipeng at zte.com.cn
> <mailto:lu.zhipeng at zte.com..cn>＞ wrote:
> 
> 
>     I'm worried that the network packet loss rate will increase after
>      reversing the order.
> 
> 
> Why? The guest is already shut down (or being shut down). By this time
> there is no traffic that would make any difference. And anyway, we've
> already determined that the current order causes a race so it is wrong.
> 
> 
> 
> 
> 
>     *为了让您的VPlat虚拟化故障得到高效的处理，请上报故障到: $VPlat技术支
>     持。*
> 
>     芦志朋 luzhipeng
> 
> 
>     IT开发工程师 IT Development Engineer
>     操作系统产品部/中心研究院/系统产品 OS Product Dept./Central R＆D
>     Institute/System Product
> 
> 
>     	
>     深圳市南山区科技南路55号中兴通讯研发大楼33楼
>     33/F, R&D Building, ZTE Corporation Hi-tech Road South,
>     Hi-tech Industrial Park Nanshan District, Shenzhen, P.R.China, 518057
>     T: +86 755 xxxxxxxx F:+86 755 xxxxxxxx
>     M: +86 xxxxxxxxxxx
>     E: lu.zhipeng at zte.com.cn <mailto:lu.zhipeng at zte.com.cn>
>     www.zte.com.cn <http://www.zte.com.cn/>
> 
>     原始邮件
>     *发件人：*＜berrange at redhat.com <mailto:berrange at redhat.com>＞;
>     *收件人：*＜laine at laine.org <mailto:laine at laine.org>＞;
>     *抄送人：*＜libvir-list at redhat.com <mailto:libvir-
>     list at redhat.com>＞;芦志朋10108272;
>     *日 期 ：*2017年05月02日 16:30
>     *主 题 ：**Re: [libvirt] [PATCH] qemu: change the name of tap device
>     for a tapand bridge network*
> 
> 
>     On Fri, Apr 28, 2017 at 01:08:45PM -0400, Laine Stump wrote:
>     ＞ On 04/28/2017 05:33 AM, Daniel P. Berrange wrote:
>     ＞ ＞ On Fri, Apr 28, 2017 at 05:23:19PM +0800, ZhiPeng Lu wrote:
>     ＞ ＞＞ Creating tap device and adding the device to bridge are not atomic operation.
>     ＞ ＞＞ Similarly deleting tap device and removing it from bridge are not atomic operation.
>     ＞ ＞＞ The Problem occurs when two vms start and shutdown. When one vm with the nic
>     ＞ ＞＞ named "vnet0" stopping, it deleted tap device but not removing port from bridge.
>     ＞ ＞＞ At this time, another vm created the tap device named "vnet0" and added port to the
>     ＞ ＞＞ same bridge. Then, the first vm deleted the tap device from the same bridge.
>     ＞ ＞＞ Finally, the tap device of the second vm don't attached to the bridge.
>     ＞ ＞＞ So, we can add domid to vm's nic name. For example, the vm's domid is 1 and vnet0
>     ＞ ＞＞ is renamed to vnet1.0.
>     ＞ ＞ 
>     ＞ ＞ Surely deleting the NIC automatically removes it from the bridge so we
>     ＞ ＞ can just remove the code that delets the bridge port.
>     ＞ 
>     ＞ That is true when using a Linux host bridge, but I recall that for
>     ＞ openvswitch (which I think is what ZhiPeng is using, based on an earlier
>     ＞ patch), you must explicitly remove the port from the bridge - apparently
>     ＞ the port is still there in openvswitch's table as some sort of "zombie"
>     ＞ connection even after the tap device itself no longer exists.
>     ＞ 
>     ＞ 
>     ＞ But instead of changing the naming scheme, maybe we should just delete
>     ＞ the bridge port *before* deleting the tap device instead of after. (Am I
>     ＞ recalling correctly that the tap device is deleted automatically when
>     ＞ the qemu process is killed? If so, then what's needed is to move the
>     ＞ loop in qemuProcessStop() that cleans up network interfaces so that it
>     ＞ happens before qemuProcessKill() is called.
> 
>     Agreed, we should just reverse the order.
> 
>     Regards,
>     Daniel
>     -- 
>     |: https://berrange.com      -o-    https://www.flickr.com/photos/dberrange
>     <https://www.flickr.com/photos/dberrange> :|
>     |: https://libvirt.org         -o-            https://fstop138.berrange.com
>     <https://fstop138..berrange.com> :|
>     |: https://entangle-photo.org    -o-    https://www.instagram.com/dberrange
>     <https://www.instagram.com/dberrange> :|
> 
>     --
>     libvir-list mailing list
>     libvir-list at redhat.com <mailto:libvir-list at redhat.com>
>     https://www.redhat.com/mailman/listinfo/libvir-list
>     <https://www.redhat.com/mailman/listinfo/libvir-list>
> 
> 
> 
> 
> 
> 