Description of problem: When NeutronGlobalPhysnetMtu: 9000 is specified, compute nodes using standard OVS gets VXLAN tenant network interfaces with wrong MTU. MTU size has been also specified in the NIC template files. OVS-DPDK compute nodes do NOT have this issue. Version-Release number of selected component (if applicable): OSP13z1 How reproducible: - Deploy an environment with Neutron's global_physnet_mtu set at 9000. - Create a tenant VXLAN network without specifying the mtu param. As output Neutron will provide MTU size 50 Byte lower than 9000. # openstack network show mgmt --column mtu --format value 8950 - Deploy a VM using the just created tenant network and make sure it gets deployed on a compute node using plain OVS (without DPDK enabled) The virtio net tap device should have 8950byte MTU, but instead, it has the ethernet default. # ip link show tap1a858403-79 68: tap1a858403-79: <BROADCAST,MULTICAST,UP,LOWER_UP> mtu 1500 qdisc pfifo_fast master ovs-system state UNKNOWN mode DEFAULT group default qlen 1000 link/ether fe:16:3e:a5:bf:ca brd ff:ff:ff:ff:ff:ff Actual results: wrong mtu size following ethernet default of 1500byte (just like if it's not specified) Expected results: mtu size of 8950.
Created attachment 1477929 [details] sosreport Sosreport attached. Both Nova and Neutron put in debug mode.
This appears to be a duplicate of https://bugzilla.redhat.com/show_bug.cgi?id=1590771 we fix a bug upstream in rocky for the non dpdk ovs deployment case when hybrid plug = false and we have back-ported it to osp 13 as part of the recent z2 release. i am closing this as not a bug as i believe this will be resolved via a z stream upgrade. if this bug is reproducible on 13z2 please reopen. regards sean. *** This bug has been marked as a duplicate of bug 1590771 ***