bad MTU with devstack and neutron

[Mike Spreitzer]

Following the instructions at http://docs.openstack.org/developer/devstack/guides/neutron.html#adding-additional-compute-nodes produces a system where every MTU is 1500. In this configuration, cross-host communication through Neutron networks fails silently, due to encapsulation overheads. There is not even any ICMP reporting the problem.

I tested this on Ubuntu 14.04.03, with exactly the "physical network setup" described in that document.

[Mike Spreitzer (mike-spreitzer)]

See also http://lists.openstack.org/pipermail/openstack-dev/2016-January/083298.html

[Mike Spreitzer (mike-spreitzer)]

I also tried again with the following local.conf, and I still got an MTU of 1500 on eth0 inside the VMs.

[[local|localrc]]
IP_VERSION=4
HOST_IP=172.18.161.6
SERVICE_HOST=172.18.161.6
MYSQL_HOST=172.18.161.6
RABBIT_HOST=172.18.161.6
GLANCE_HOSTPORT=172.18.161.6:9292
ADMIN_PASSWORD=secrete
DATABASE_PASSWORD=secrete
RABBIT_PASSWORD=secrete
SERVICE_PASSWORD=secrete
SERVICE_TOKEN=secrete

# Do not use Nova-Network
disable_service n-net
# Enable Neutron
ENABLED_SERVICES+=,q-svc,q-dhcp,q-meta,q-agt,q-l3

## Neutron options
Q_USE_SECGROUP=True
FLOATING_RANGE="172.18.161.0/24"
FIXED_RANGE="10.0.0.0/24"
Q_FLOATING_ALLOCATION_POOL=start=172.18.161.250,end=172.18.161.254
PUBLIC_NETWORK_GATEWAY="172.18.161.1"
Q_L3_ENABLED=True
PUBLIC_INTERFACE=eth0
Q_USE_PROVIDERNET_FOR_PUBLIC=True
OVS_PHYSICAL_BRIDGE=br-ex
PUBLIC_BRIDGE=br-ex
OVS_BRIDGE_MAPPINGS=public:br-ex

# Decent logging

LOG_COLOR=False
DEST=/opt/stack
LOGFILE=$DEST/logs/stack.sh.log
SCREEN_LOGDIR=$DEST/logs/screen

# Hack down the MTUs

[[post-config|$NEUTRON_CONF]]
[DEFAULT]
advertise_mtu=True
network_device_mtu=1450

[[post-config|$NOVA_CONF]]
[DEFAULT]
network_device_mtu=1450

[[post-config|/$Q_PLUGIN_CONF_FILE]]
[ml2]
physical_network_mtus = public:1450

[Sean M. Collins (scollins)]

This is the same as 1527675. The only difference is you did it manually - devstack-gate ends up doing the same thing, but in an automated fashion.

[Sean M. Collins (scollins)]

There may need to be a doc patch to the neutron guide in DevStack once we sort this thing out.