[libvirt-users] Network reconfiguration glitch after migration

[Joaquim Barrera]

Hello everyone,

I have the following scenario:

* 2 host machines (H1 and H2) running VMs. I execute migrations (QEMU) 
from one host to the other, and so on. The migrations are done in a 
dedicated LAN (eth1 in both hosts) while regular department LAN runs on eth0
* 1 computer (my desktop computer) also connected to this common LAN on 
eth0.

The VMs have a Python script sending UDP packets to my desktop computer 
(which has the server part of the Python script), every 250 ms, to be 
able to mesure downtime and other stuff. Once I run a live migration 
from H1 to H2, I would expect the downtime to be in the order of ms, or 
maybe a few seconds, but it takes something like 10 to 20 seconds on 
average for the packets to resume arriving at my computer. One could 
think about the network reconfiguration delay, or other reasons, but 
here comes the tricky part:

If I have a ping running on the destination host to the internal IP that 
the VM is supposed to have once the migration is done, the UDP packets 
resume just a few ms after the migration os done.

Is like: if the VM only has outgoing connections, it takes a lot of time 
to resume its operation, but if it has some incoming connection, the 
network resumes very quick.

I know it may sound wierd, but maybe there is a clear reason for this 
issue. Any ideas?

Thank you!