After upgrade VMs getting 1500 MTU although jumbo frame is setup
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
kolla |
Expired
|
Undecided
|
Unassigned |
Bug Description
Steps to recreate the issue
set the jumbo frame on Machine and neutron.conf, ml2.conf and nova.conf
in the ansible code
l3 agent.ini
agent_mode = dvr_snat
network_device_mtu = 9000
{% elif inventory_hostname in groups['compute'] %}
agent_mode = dvr
{% endif %}
{% else %}
agent_mode = legacy
ml2 conf.ini
mechanism_drivers = openvswitch
path_mtu = 9000
{% elif enable_
type_drivers = flat,vlan
tenant_
{% else %}
# Changing type_drivers after bootstrap can lead to database inconsistencies
neutron.conf
# because it is used by the l3 and dhcp agents. The reason the path has 'kolla'
# in it is because we are sharing this socket in a volume which is it's own dir
metadata_
global_physnet_mtu = 9000
{% if neutron_
interface_driver = openvswitch
nova.conf
metadata_listen = {{ api_interface_
metadata_
network_device_mtu = 9000
use_neutron = True
firewall_driver = nova.virt.
then deploy using kolla (Ocata almost 3 weeks back)
2. create a VM and verify the VM is having 8950 MTU
3. Upgrade the system to latest version (stable/Ocata)
Observation:
VM is not accessible after upgrade
Reboot the VM
The VM which was created earlier having 1500 MTU where as its tap device is having 8950 MTU and VM is not able to get the IP
Create a New VM
now the new VM is accessible and having MTU as 8950
So after upgrade we will be loosing all the VMs which were created before
summary: |
- After upgrade VMs getting 1500 MTU although jumbo frame issetup + After upgrade VMs getting 1500 MTU although jumbo frame is setup |
After Rebooting the Host able to get the VMs back
but i dont think kolla recommends a reboot of Hosts after upgrade