[libvirt] mass create vm errors

[Michal Privoznik]

On 04.08.2015 14:31, Vasiliy Tolstov wrote:
> 2015-08-03 10:01 GMT+03:00 Michal Privoznik <mprivozn at redhat.com>:
>> There has not been much movement in that particular area since 1.2.16.
>> There's inherent race though: by the time that libvirt detects that a
>> certain port is free and qemu binds to it. During this window another
>> process may just allocate the port which will result in the error
>> message you're seeing.
> 
> This is very bad thing in case of mass start , for example when node
> dies i have 100-150 vm started at the same time, and sometimes they
> running on single host.
> Does it possible to create lock for this port when port passed to
> qemu, so another process when detect new port don't use it?

No. It would require change in other apps so that they use the lock file.

> 
>> One of the solutions may be to use static ports, or make sure that other
>> programs don't interfere with the range that's reserved for graphical
>> sessions (the range can be configured too in qemu.conf).
> But if i specify port ranges in qemu.conf, how this solve my issue? as
> i understand libvirt again randomly allocated port for vm...
> 

No, libvirt does not randomly allocate port. In fact, libvirt does not
allocate port at all. It merely checks on domain startup that the
graphics port is free. Or it searches for next free port within the
range set in qemu.conf (in case of autoport). Then, the port number is
put on qemu command line and qemu binds to it. However during this
window other apps may step in and allocate the port.
But if you choose a different range, unused one, you may be more
successful. BTW: to view which app has allocated the 5904 port you can
use: 'netstat -ltnp | grep 5904'.

Of course, the internal bitmap of free and taken ports is protected by a
mutex, so libvirt should not assign the same port to two domains.

Michal