Hi all,
I have a working setup of oVirt (currently 4.3.7) as self-hosted engine consisting of
three virtualization hosts. Each of them has several VMs with public IP addresses which
work fine.
The problem is with virtual networks created on "ovirt-provider-ovn". I can
create a new network with the corresponding vNIC and attach two VMs to this vNIC.
If both VMs are on the same host, virtual network works, VMs can exchange packets.
However, if such VMs are on different hosts, then no packets are received on both VMs.
With tcpdump I see traffic going from sender VM over "ovirtmgmnt" interface. The
traffic successfully leaves one host and arrives to another, so I conclude that firewall
is configured properly to allow UDP 6081 traffic.
What I noted are the following messages in /var/log/openvswitch/ovs-vswitchd.log (on one
of the hosts):
2019-12-10T12:10:11.177Z|00009|tunnel(handler4)|WARN|receive tunnel port not found
(arp,tun_id=0x2,tun_src=172.18.53.254,tun_dst=172.18.53.202,tun_ipv6_src=::,tun_ipv6_dst=::,tun_gbp_id=0,tun_gbp_flags=0,tun_tos=0,tun_ttl=64,tun_erspan_ver=0,tun_flags=csum|key,in_port=2,vlan_tci=0x0000,dl_src=56:6f:03:4b:00:1f,dl_dst=ff:ff:ff:ff:ff:ff,arp_spa=10.1.1.102,arp_tpa=10.1.1.1,arp_op=1,arp_sha=56:6f:03:4b:00:1f,arp_tha=00:00:00:00:00:00)
Here 172.18.53.254 is IP of host1, 172.18.53.202 -- host2 (having this log), 10.1.1.102 --
VM IP on host1 and 10.1.1.1 -- VM IP on host2.
Could you please suggest me a way to properly debug and fix this problem? I would like to
be able to distribute several VMs connected to the single virtual network across hosts.
Thanks in advance!
--
WBR, Pavel
+32478910884
Show replies by date
Hi Pavel Nakonechnyi,
However, if such VMs are on different hosts, then no packets are received on both VMs.
Could you please suggest me a way to properly debug and fix this problem? I would like to
be able to distribute several VMs connected to the single virtual network across hosts.
It appears that the problem was (as usual) with my setup which is a bit messy as it was
based on experimental environment. The issues was triggered by changing the IP address of
one of the hosts. I actually just added an alias to ovrtmgmnt interface and then changed
the IP via oVirt engine...
Anyway, command `ovs-vsctl list Interface` showed that one of the switch peers was still
referenced by the old, unused IP address. The easiest way to fix it for me was to re-add
the host to cluster. After that cross-host communication between VMs started to work
properly.