[libvirt] problems with current libvirt and qemu-kvm

[Gene Czarcinski]

On 11/30/2012 08:19 AM, Gene Czarcinski wrote:
> On 11/30/2012 03:24 AM, Daniel P. Berrange wrote:
>> On Thu, Nov 29, 2012 at 03:48:52PM -0500, Gene Czarcinski wrote:
>>> When I rebased my DHCPv6, etc. patches, I did that not only on to of
>>> Laine Stump's dnsmasq-capabiliotes/bind-dynamic updates but I also
>>> rebased to "current" git master.
>>>
>>> Well, the problem with the large number/frequent RTR-ADVERT syslog
>>> messages from dnsmasq that I had previously seen were fixed ... they
>>> stopped and dnsmasq produced expected results.  But, starting
>>> virt-manager hung and there were new error messages in syslog about
>>> "internal error cannot find suitable emulator for X86_64.
>> Until I pushed the patches from Victor yesterday, libvirt GIT was
>> broken when used against QEMU GIT. The effect was that libvirt
>> could not detect any QEMU binaries. So if you're using latest QEMU
>> GIT that might be what you've hit.
>>
>>
> No, not the latest ... qemu-1.0.1-2.fc17.
>
> But, this problem is getting more than a bit strange.
>
> I have two systems which are pretty near identical (one id AMD-6-core 
> and the other is AMD-8-core).  They have the same kernel, the same 
> libvirt, qemu, virt-manager, kernel, dnsmasq.
>
> The "problem system" only has the RTR-ADVERT problem with one network 
> definition! [IPv4 with DHCP and IPv6 with SLAAC].  Destroy that 
> network and start another with different addresses but the same 
> specification and no problems.  The is one difference I need to check 
> ... the problem network is autostarted.
>
> I just installed the libvirt version that has a problem with qemu. I 
> did not bother trying virt-manage since I just noticed a bunch of 
> really strange syslog messages from libvirtd "unknown OS type hvm" so 
> I downgraded back to something that works.
>
> The last update I have listed in the git-log is 
> b7aba48bcaf315c27430a7d0edf8ff702d293527 and that one changed 
> src/conf/domain_conf.c.
>
> I am going to give two things a try:
>
> 1. Back off the top few commits that hist domain stuff and
>
> 2. Try do a pull to update my copy of libvirt's git.
>
> About the only thing I am sure of is I believe that the three patches 
> I submitted and Laine's three patches for bind-dynamic are not the 
> problem ... they are also applied to the version that works and only 
> deal with network stuff.
>
> Gene
That was fun.   Good news!  I did a pull to get my clone up to date, saw 
that Laine's updates had been pushed, rebased my updates, created a 
tarball and used it to build a set of libvirt rpms. Installed the 
updated rpms and ... everything seems to work again! I must have caught 
something in the middle when I did my last pull because it is fine now.

And, the RTR-ADVERT messages (and associated issuing of RA packets) is 
back to "normal" ... an initial flurry and then slowing to one every 
none/ten minutes.

However, I have noticed one thing.  Simon Kelley had me add a little 
"instrumentation" to dnsmasq trying to pursue this large number of 
RTR-ADVERT messages.  I am going to do a little more instrumentation.

What I have noticed is that there seems to be a broadcast of something 
whenever a network is started so that all dnsmasq instances (one per 
started network) go into "flurry" mode with issuing RA packets and 
syslog messages.

Does anyone know if something in libvirt is doing that?

BTW, I will shortly be sending in a small rebase of my patches based on 
today's git.

Gene