nova-compute is not setting MTU at all when following the plug_ovs_bridge() call path
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
OpenStack Compute (nova) |
Expired
|
Undecided
|
Unassigned |
Bug Description
Here is the specific code path I was talking about:
https:/
As you can see in the code,
1. self.plug_
2. however self.plug_
We believe this is an overlook from this MTU feature patch:
https:/
Let us know if you need more info then.
Particularly, this issue causes the connectivity problem between two VMs on two different hosts when they are based on a VxLAN private network. In this case, the default MTU size of 1500 bytes is not working due to the overhead of VxLAN header size.
Here is an example of the problem.
VM A runs on host B and VM C runs on host D. Each VM has a VxLAN-based private network interface. VM A has its own floating IP as well.
From a laptop, a user logins to VM A via its floating IP. And then, when s/he logins from VM A to VM B using the private IP of VM B, it fails due to the VxLAN overhead when MTU is 1500 bytes.
Only after manually increasing the MTU size of the tap interfaces of VM A and VM B to a jumbo frame, it is possible to login from VM A to VM B via their private IPs.
I think this is a serious bug.