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

Re: [Xen-devel] Hyper and Xen Project



> On 25 Jun 2015, at 02:46, Wang Xu <gnawux@xxxxxxxxx> wrote:
> 
> Agree, but I think the document is a bit confused
> 
> > It is important that channel names are globally unique.
> 
> https://github.com/mirage/xen/blob/master/docs/misc/channel.txt#L94

I agreeâ that wording is definitely confusing. Perhaps the docs should compare 
the channel names to TCP/UDP port numbers? We could say that

- the IANA port registry ~=~ the channel registry in the docs/ directory
- a single IP can only have one binding for a particular port at a time ~=~ a 
domain can only have one binding for a particular name at a time
- lots of IPs can bind the same port ~=~ lots of domains can bind the same name

What do you think?

Thanks,
Dave

> 
> On Thu, Jun 25, 2015 at 2:29 AM Dave Scott <Dave.Scott@xxxxxxxxxx> wrote:
> Hi Xu,
> 
> > On 24 Jun 2015, at 14:44, Wang Xu <gnawux@xxxxxxxxx> wrote:
> >
> > Thank you Dave, I think I can also get work around for that.
> >
> > By the way, the document says the name should be global unique, but I can 
> > start 2 domains have channels with a same name, is there some potential 
> > problems?
> 
> The name needs to be unique within a domain. Itâs ok to have
> 
> 1. domid 10, channel name âagentâ
> 2. domid 11, channel name âagentâ
> 
> â this will be common, as multiple domains will have the same âagentâ 
> software installed.
> 
> but it will cause problems if the name is used twice within a domain. Itâs a 
> bad idea to have
> 
> 1: domid 10, channel name âagentâ
> 2: domid 10, channel name âagentâ
> 
> â although this will create 2 distinct /dev/hvc devices, it will be difficult 
> to tell which is which.
> 
> If libxl allows the name to be duplicated within a domain, then this is my 
> fault. We should add validation code to check uniqueness.
> 
> Thanks,
> Dave
> 
> >
> > Cheers
> >
> > Xu
> >
> > On Wed, Jun 24, 2015 at 9:03 PM Dave Scott <Dave.Scott@xxxxxxxxxx> wrote:
> > I donât think the frontend driver in Linux knows about the name key. In my 
> > testing I wrote a udev script which looks up the ânameâ key directly in 
> > xenstore and created a named device node using that. For reference my 
> > script is here:
> >
> > https://github.com/mirage/mirage-console/blob/master/udev/xenconsole-setup-tty
> >
> > Cheers,
> > Dave
> >
> > >> and I directly test `/dev/hvc1`, and it could communicate with the 
> > >> outside socket. Is there some mistake in my channel
> > >> name configuration?
> > >>
> > >> | static void hyper_config_channel(libxl_device_channel* ch, const char* 
> > >> name, const char* sock, int devid) {
> > >> |     libxl_device_channel_init(ch);
> > >> |     ch->backend_domid = 0;
> > >> |     ch->name = strdup(name);
> > >> |     ch->devid = devid;
> > >> |     ch->connection = LIBXL_CHANNEL_CONNECTION_SOCKET;
> > >> |     ch->u.socket.path = strdup(sock);
> > >> | }
> > >>
> > >> I tried to look at the oVirt code as it is mentioned in the dock, but I 
> > >> did not find xen console in its guest agent code.
> > >
> > > So the issue is that the name you assign here to the channel, doesn't
> > > come up anywhere in the guest. Is that correct?
> >
> >
> > >
> > >
> > >> Thank you!
> > >>
> > >>
> > >> On Tue, Jun 23, 2015 at 7:30 PM, Stefano Stabellini 
> > >> <stefano.stabellini@xxxxxxxxxxxxx> wrote:
> > >>      On Tue, 23 Jun 2015, Wang Xu wrote:
> > >>> On Sat, Jun 20, 2015 at 1:10 AM Stefano Stabellini 
> > >>> <stefano.stabellini@xxxxxxxxxxxxx> wrote:
> > >>>        Integrating hyper with Xen using libxl was the right decision 
> > >>> and it
> > >>>        looks like you did a good job. I think that you can go ahead 
> > >>> with the
> > >>>        PR!
> > >>>
> > >>>
> > >>>        But I did have a few issues building hyper. I am getting:
> > >>>
> > >>>        hyperd.go:11:2: cannot find package "hyper/daemon" in any of:
> > >>>        [...]
> > >>>
> > >>> I tried with a clean 0.2-dev branch
> > >>>> ./autogen.sh
> > >>>> ./configure
> > >>>> make
> > >>>
> > >>> It looks ok, are you work on the 0.2-dev branch, I did not write the 
> > >>> branch name in the instruction of
> > >>      Readme, sorry for
> > >>> that.
> > >>
> > >>      No worries, the most important part at this stage is the code, and 
> > >> that
> > >>      looks OK :-)
> > >>      Yes, I was using 0.2-dev and followed those steps. As I usually 
> > >> don't
> > >>      program in go, it is likely that my go working environment is 
> > >> missing
> > >>      something, or my go paths are wrong. This is the full error message:
> > >>
> > >>      CGO_LDFLAGS="-Lhypervisor/xen -lxenlight -lxenctrl -lhyperxl" godep 
> > >> go build hyperd.go
> > >>      hyperd.go:11:2: cannot find package "hyper/daemon" in any of:
> > >>              /local/scratch/sstabellini/go/src/hyper/daemon (from 
> > >> $GOROOT)
> > >>              
> > >> /local/scratch/sstabellini/hyper/Godeps/_workspace/src/hyper/daemon 
> > >> (from $GOPATH)
> > >>      hyperd.go:10:2: cannot find package "hyper/engine" in any of:
> > >>              /local/scratch/sstabellini/go/src/hyper/engine (from 
> > >> $GOROOT)
> > >>              
> > >> /local/scratch/sstabellini/hyper/Godeps/_workspace/src/hyper/engine 
> > >> (from $GOPATH)
> > >>      hyperd.go:12:2: cannot find package "hyper/lib/glog" in any of:
> > >>              /local/scratch/sstabellini/go/src/hyper/lib/glog (from 
> > >> $GOROOT)
> > >>              
> > >> /local/scratch/sstabellini/hyper/Godeps/_workspace/src/hyper/lib/glog 
> > >> (from $GOPATH)
> > >>      hyperd.go:13:2: cannot find package "hyper/utils" in any of:
> > >>              /local/scratch/sstabellini/go/src/hyper/utils (from $GOROOT)
> > >>              
> > >> /local/scratch/sstabellini/hyper/Godeps/_workspace/src/hyper/utils (from 
> > >> $GOPATH)
> > >>      godep: go exit status 1
> > >>
> > >>
> > >>>        Looking through the code, it seems that you are adding a
> > >>>        virtio-serial-pci device, why do you need it?  It is not used 
> > >>> very much
> > >>>        on Xen; the regular Xen uart is specified by setting
> > >>>        b_info->u.hvm.serial to "pty", and it looks like you are already 
> > >>> doing
> > >>>        that. If you need more than one console, you can have a list 
> > >>> setting
> > >>>        b_info->u.hvm.serial_list.
> > >>>
> > >>> What the difference between u.hvm.serial_list and channels in 
> > >>> domain_config. The channel looks having more
> > >>      features.
> > >>
> > >>      Actually I think that you are right: channels are better tested and 
> > >> more
> > >>      flexible.
> > >>
> > >>
> > >>>        virtio-9p-pci is also not used very much with Xen, but as we 
> > >>> don't have
> > >>>        an alternative yet, I think it is good for now.
> > >>>
> > >>>
> > >>>        Thanks again,
> > >>>
> > >>>        Stefano
> > >>>
> > >>>
> > >>>
> > >>>        On Fri, 19 Jun 2015, Sarah Conway wrote:
> > >>>        > Hi Xu,
> > >>>        > I'd be happy to work with you on some PR to promote this work. 
> > >>> I'll be in touch with some next steps
> > >>      next
> > >>>        week and look
> > >>>        > forward to Stefano's feedback.
> > >>>        >
> > >>>        > Sarah
> > >>>        >
> > >>>        > On Fri, Jun 19, 2015 at 12:09 PM, Lars Kurth 
> > >>> <lars.kurth@xxxxxxxxxx> wrote:
> > >>>        >       Xu,
> > >>>        > Fantastic! If you wanted to do some PR, such as a joint Xen 
> > >>> Project / Linux Foundation and Hyper
> > >>      press
> > >>>        release or
> > >>>        > other PR, we should start working on that after Stefano had a 
> > >>> look
> > >>>        > Let me know
> > >>>        > Regards
> > >>>        > Lars
> > >>>        >
> > >>>        > From: Wang Xu <gnawux@xxxxxxxxx>
> > >>>        > Date: Friday, 19 June 2015 16:54
> > >>>        > To: "stefano.stabellini@xxxxxxxxxx" 
> > >>> <stefano.stabellini@xxxxxxxxxx>
> > >>>        > Cc: "xu@xxxxxxxx" <xu@xxxxxxxx>, Sarah Conway 
> > >>> <sconway@xxxxxxxxxxxxxxxxxxx>, Lars Kurth
> > >>>        <lars.kurth@xxxxxxxxxx>,
> > >>>        > "feng@xxxxxxxx" <feng@xxxxxxxx>, "carmark.dlut@xxxxxxxxx" 
> > >>> <carmark.dlut@xxxxxxxxx>
> > >>>        > Subject: Re: Hyper and Xen Project
> > >>>        >
> > >>>        > Hi Stefano and Lars, Sarah,
> > >>>        > I'd like to share some progress of Xen support of Hyper with 
> > >>> you.
> > >>>        >
> > >>>        > The source code of hyper with xen support has been pushed to 
> > >>> 0.2-dev branch (
> > >>>        > https://github.com/hyperhq/hyper/tree/0.2-dev ). Before 
> > >>> release the binary packages, there are still
> > >>      some
> > >>>        test and
> > >>>        > tuning work to do, but I think it's better to share the 
> > >>> information with you firstly.
> > >>>        >
> > >>>        > Hyper supports HVM/Xen 4.5 as the first step, in which we used 
> > >>> C and Go to talk with Xen through
> > >>      libxl.
> > >>>        libxl is
> > >>>        > well organized and help us much, though It took us some time 
> > >>> to understand/debug the signal
> > >>      processing and
> > >>>        nic
> > >>>        > hotplug.
> > >>>        >
> > >>>        > You can check out the code and try to build it from source as 
> > >>> description listed in README if you
> > >>      have
> > >>>        interests.
> > >>>        > And any suggestions on performance and reliability are 
> > >>> appreciated.
> > >>>        >
> > >>>        > Cheers!
> > >>>        > Xu
> > >>>        >
> > >>>        > On Tue, Jun 9, 2015 at 5:41 PM Stefano Stabellini 
> > >>> <stefano.stabellini@xxxxxxxxxxxxx> wrote:
> > >>>        >       You are right, we don't really have an equivalent to 9p 
> > >>> in the Xen
> > >>>        >       ecosystem at the moment.  We had a prototype called 
> > >>> XenFS, like you
> > >>>        >       wrote, but it had never reached production quality, so 
> > >>> it was removed.
> > >>>        >       But we now have an OPW student working on a Xen 
> > >>> paravirtualized 9p fs
> > >>>        >       interface.  Once she finishes the initial prototype, we 
> > >>> could prioritize
> > >>>        >       it within the community and work together to complete it.
> > >>>        >
> > >>>        >       How are you using virtio 9p fs?  What Docker storage 
> > >>> backend are you
> > >>>        >       using?  I am asking because when I looked into running 
> > >>> Docker container
> > >>>        >       images as Xen VMs, I was able to boot VMs without any 
> > >>> filesystem share,
> > >>>        >       by using the devices set up by Docker's device mapper 
> > >>> backend directly.
> > >>>        >       I think that using storage devices directly as backends, 
> > >>> when possible,
> > >>>        >       is preferable because of performance and security.
> > >>>        >
> > >>>        >
> > >>>        >       On Tue, 9 Jun 2015, Wang Xu wrote:
> > >>>        >       > Hi Stefano,
> > >>>        >       >
> > >>>        >       > Yes, we are working on HVM and using 9p and 
> > >>> virt-serial over virtio, and looks 9p do not
> > >>      have a
> > >>>        >       similar
> > >>>        >       > component in Xen ecosystem. What's the status of XenFS 
> > >>> and how do you think about filesystem
> > >>      share
> > >>>        in
> > >>>        >       Xen?
> > >>>        >       >
> > >>>        >       > On Mon, Jun 8, 2015 at 6:18 PM Stefano Stabellini 
> > >>> <stefano.stabellini@xxxxxxxxxxxxx> wrote:
> > >>>        >       >       Hi Xu,
> > >>>        >       >
> > >>>        >       >       Thank you for the very quick reply!  I am very 
> > >>> happy to hear that Xen
> > >>>        >       >       support is coming so soon.
> > >>>        >       >
> > >>>        >       >       >From the technical point of view, I think that 
> > >>> HVM guests with the
> > >>>        >       >       explicit kernel and initrd options is a very 
> > >>> good choice for the initial
> > >>>        >       >       implementation.
> > >>>        >       >
> > >>>        >       >       PVH will probably be a better alternative going 
> > >>> forward because startup
> > >>>        >       >       times are far lower, but at this stage is still 
> > >>> not as stable as regular
> > >>>        >       >       HVM guests, so I would wait for another Xen 
> > >>> release or two before using
> > >>>        >       >       them.  Old style PV guests might still be a 
> > >>> decent alternative for
> > >>>        >       >       people that really care about startup times, 
> > >>> because they boot really
> > >>>        >       >       quickly. They also have good performance in a 
> > >>> nested virtualization
> > >>>        >       >       environment, such as people deploying containers 
> > >>> on Amazon AWS. However
> > >>>        >       >       on native they offer lower performance than PVH 
> > >>> or HVM guests.
> > >>>        >       >
> > >>>        >       >       You mentioned virtio:  are you using virtio 9p 
> > >>> fs or one of the other
> > >>>        >       >       protocols (block, net, etc)?
> > >>>        >       >
> > >>>        >       >       Cheers,
> > >>>        >       >
> > >>>        >       >       Stefano
> > >>>        >       >
> > >>>        >       >
> > >>>        >       >       On Fri, 5 Jun 2015, Wang Xu wrote:
> > >>>        >       >       > Hello Stefano,
> > >>>        >       >       >
> > >>>        >       >       > It is very glad to get message from Xen.
> > >>>        >       >       >
> > >>>        >       >       > Xen is a great innovation and one of the most 
> > >>> widely adopted hypervisor, we believe
> > >>      we
> > >>>        should
> > >>>        >       >       support it as
> > >>>        >       >       > soon as possible.
> > >>>        >       >       >
> > >>>        >       >       > We are currently busy working on Xen support 
> > >>> and will announce the initial support
> > >>      in the
> > >>>        next
> > >>>        >       >       week.
> > >>>        >       >       >
> > >>>        >       >       > The initial support is working on Xen 4.5 hvm 
> > >>> mode, because it is easy to specify
> > >>      kernel and
> > >>>        >       >       initrd during
> > >>>        >       >       > boot and support virtio, which we have already 
> > >>> used in the kvm version.
> > >>>        >       >       >
> > >>>        >       >       > We also considered PVH on Xen 4.4, but met 
> > >>> some trouble with that.
> > >>>        >       >       >
> > >>>        >       >       > We are very interesting in cooperating with 
> > >>> Xen upstream to have a better support in
> > >>      the
> > >>>        >       >       future.
> > >>>        >       >       >
> > >>>        >       >       > Thank you for your message!
> > >>>        >       >       >
> > >>>        >       >       > Cheers!
> > >>>        >       >       >
> > >>>        >       >       > Xu Wang
> > >>>        >       >       > Cofounder & CTO of Hyper
> > >>>        >       >       >
> > >>>        >       >       >
> > >>>        >       >       > Stefano Stabellini 
> > >>> <stefano.stabellini@xxxxxxxxxxxxx>ä2015å6æ6æåå 00:14åéï
> > >>>        >       >       >       Hello Xu, Feng, Lei,
> > >>>        >       >       >
> > >>>        >       >       >       My name is Stefano Stabellini and I lead 
> > >>> the Xen Project team within
> > >>>        >       >       >       Citrix.  I am also one of the Xen 
> > >>> maintainers.
> > >>>        >       >       >
> > >>>        >       >       >       Let me introduce you to Lars Kurth, 
> > >>> chair of the Xen Project advisory
> > >>>        >       >       >       board, and Sarah Conway, that represents 
> > >>> the Linux Foundation. As you
> > >>>        >       >       >       probably know, Xen Project is a Linux 
> > >>> Foundation collaborative project.
> > >>>        >       >       >
> > >>>        >       >       >
> > >>>        >       >       >       I wanted to get in touch with you 
> > >>> regarding your new project, Hyper:  I
> > >>>        >       >       >       am really glad to see that you are 
> > >>> stepping up to bring the security and
> > >>>        >       >       >       flexibility of hypervisor solutions to 
> > >>> the world of containers. I think is
> > >>>        >       >       >       a key project that will benefit the 
> > >>> containers and cloud industry as a
> > >>>        >       >       >       whole.  In fact I have been 
> > >>> investigating the best way to run containers
> > >>>        >       >       >       on Xen myself during the last few months.
> > >>>        >       >       >
> > >>>        >       >       >
> > >>>        >       >       >       I read that Hyper is hypervisor agnostic 
> > >>> and I am glad that Xen is
> > >>>        >       >       >       already on the roadmap.
> > >>>        >       >       >
> > >>>        >       >       >       I understand that Hyper is still young, 
> > >>> but do you have any timelines on
> > >>>        >       >       >       when you are planning to introduce Xen 
> > >>> support to Hyper?  Would you be
> > >>>        >       >       >       interested in discussing the best way to 
> > >>> do that?
> > >>>        >       >       >
> > >>>        >       >       >
> > >>>        >       >       >       Cheers,
> > >>>        >       >       >
> > >>>        >       >       >       Stefano
> > >>>        >       >       >
> > >>>        >       >       >
> > >>>        >       >       >
> > >>>        >       >
> > >>>        >       >
> > >>>        >       >
> > >>>        >
> > >>>        >
> > >>>        >
> > >>>        >
> > >>>        > --
> > >>>        > Sarah Conway
> > >>>        > PR Manager
> > >>>        > The Linux Foundation
> > >>>        > sconway@xxxxxxxxxxxxxxxxxxx
> > >>>        > (978) 578-5300  Cell
> > >>>        > Skype:  sarah.k.conway
> > >>>        >
> > >>>        >
> > >>>
> > >>>
> > >>>
> > >>
> > >>
> > >>
> > >>
> > >> --
> > >> Wang Xu
> > >>
> >
> 

_______________________________________________
Xen-devel mailing list
Xen-devel@xxxxxxxxxxxxx
http://lists.xen.org/xen-devel

 


Rackspace

Lists.xenproject.org is hosted with RackSpace, monitoring our
servers 24x7x365 and backed by RackSpace's Fanatical Support®.