@tomp
We have reinstalled lxd and ovn and yesterday’s problem has gone.
All containers at node1 has internet by default when they are created on ovn0 network and on other nodes the created containers doesnt have internet and can not ping each other in ovn0 network subnet 
It is so strange that ovs-dbserver socket is created on ceph interface 192.168.1.X.
Is this the problem?
Uplink is on 10.0.0.0/24 and ovn and lxd cluster
And sfp port that is on 192.168.1.0/24 is for ceph cluster.
The problem solved by setting external_ids:ovn-encap-ip=192.168.1.50
No
Because of multiple interfaces that the servers have, ovs making genve tunnel on our 192.168 range and we didn’t know why. because of this we changed ovs command like I said.
Yes you do have to set external_ids:ovn-encap-ip= on each member an IP that all the other members can reach so that the geneve tunnels operate correctly. Glad you sorted it out.
BTW this is why doing tcpdump tests with an instance running on the active chassis is sometimes valuable in diagnosing OVN issues. If you can get it working with an instance running on the active gateway chassis then that bypasses any issues with the geneve tunnels. Then if you know it only doesn’t work on instances running on non-activate-chassis then you can be more confident its an issue with the geneve tunnels (either addressing, firewalling or MTU normally).