vxlan mtu size not respected
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
StarlingX |
Won't Fix
|
Low
|
Unassigned |
Bug Description
StarlingX 3.0 release
AIO duplex with 2 datanetworks: physnet0 is based on vlan and physnet1 on vxlan. controller-0 and controller-1 are identically speced: eno1 to oam, eno2 to data0 (physnet0), enp1s0f0 to mgmt0, and enp1s0f1 to data1 (physnet1). eno* are 1gbe and enp* are 10gbe, not that it matters.
interface for data1 set to MTU=9216, accelerated=True
datanetwork physnet0 vlan mtu 1500; physnet1 vxlan mtu 9050
```
controller-1:~$ OS_CLOUD=
| ID | Name | Default | Shared | Project ID | Network Type | Physical Network | Min ID | Max ID
| ... | physnet0-vlan-r1 | False | False | ... | vlan | physnet0 | 120 | 130
| .. | physnet1-vxlan-r1 | False | True | None | vxlan | None | 100 | 99999
controller-1:~$ OS_CLOUD=
controller-1:~$ OS_CLOUD=
| mtu | 1450
```
The guest network testnet backed by vxlan, which is great. However, the mtu is messed up. I can modify the mtu manually or specify it when creating the network. Sadly, that is not possible in horizon and thus end users self servicing will end up with 1450.
Update: using openstack cli as non-admin: complains about max mtu=4500 if i specify 9000 mtu when creating my network, but i can set the mtu to 9000 in a subsequent call!
Update2: managed to modify the default vxlan tenant network mtu! i would be grateful if somebody can explain the mtu implications between cali*, tunl0@NONE, tapc*, and all the br-*. Is there any issue or optimization concerns if they are shown as these values (esp. tunl0@NONE which is of lower mtu than tapc*):
- cali* 9051
- tunl0@NONE 8991
- tapc* 9001
- br-phy0 1500
- br-int 9001
- br-tun 1500
- enp5s0f{0|1} (mgmt0 and data1 on vxlan respectively) 9216
description: | updated |
description: | updated |
tags: | added: stx.networking |
description: | updated |
description: | updated |
Added the domain & release tag. This will require further investigation by the networking team.