Unable to add additional Compute Node to DevStack Neutron environment

asked 2014-04-01 08:41:05 -0500

Gessler gravatar image

updated 2014-04-01 08:43:18 -0500

Hey folks,

I'm working on starting my own OpenStack cloud with a setup of three physical nodes - one controller (10.8.234.100), two computes compute1 (10.8.234.101) and compute2 (10.8.234.102). This takes place in a private network (10.8.234.0/24) with internet-access through a proxy (set system-wide) on Ubuntu 12.04 as the base system on all nodes.

So far I've managed to configure a controller node using DevStack and the following local.conf:

[[local|localrc]]

DATABASE_PASSWORD=password
RABBIT_PASSWORD=password
SERVICE_TOKEN=password
SERVICE_PASSWORD=password
ADMIN_PASSWORD=password

FLOATING_RANGE=10.8.234.128/27
FLAT_INTERFACE=eth0
LOGFILE=/home/stack/stack.sh.log
Q_FLOATING_ALLOCATION_POOL=start=10.8.234.110,end=10.8.234.120
PUBLIC_NETWORK_GATEWAY=10.8.234.100

disable_service n-net
enable_service q-svc
enable_service q-agt
enable_service q-dhcp
enable_service q-l3
enable_service q-meta
enable_service neutron
enable_service tempest

After completion of stack.sh I am able to launch instances on the controller which itself provides nova-compute service. However, I am not sure how to correctly add another compute node to the environment. To be more precise, I don't know which services are essential to provide computing and to be able to become part of the whole Neutron setup.

By now, I ran stack.sh on compute1 with this local.conf:

[[local|localrc]]
HOST_IP=10.8.234.101
SERVICE_HOST=10.8.234.100
PUBLIC_NETWORK_GATEWAY=10.8.234.100
FLAT_INTERFACE=eth0
FLOATING_RANGE=10.8.234.128/27
MULTI_HOST=1

LOGFILE=/opt/stack/logs/stack.sh.log

DATABASE_PASSWORD=password
RABBIT_PASSWORD=password
SERVICE_TOKEN=password
SERVICE_PASSWORD=password
ADMIN_PASSWORD=password

Q_FLOATING_ALLOCATION_POOL=start=10.8.234.110,end=10.8.234.120
Q_HOST=$SERVICE_HOST
Q_PLUGIN=openvswitch
Q_AUTH_STRATEGY=keystone
MYSQL_HOST=$SERVICE_HOST
RABBIT_HOST=$SERVICE_HOST
GLANCE_HOST_PORT=10.8.234.100:9292

ENABLED_SERVICES=n-cpu,rabbit,neutron,q-agt

Using this config I can see compute1's nova-compute in running ":-)" state through controller's nova-manage service list. But whenever I try to launch an instance, it is unable to retrieve an IP and immediately enters ERROR state. When I shut down compute1 (e.g. using unstack.sh) and wait a while until controller recognizes compute1 being missing, I am once again able to launch instances that get IPs and start as expected.

I would be very thankful for some advice, as this whole networking with Neutron goes truely beyond my expertise.

Regards, Nico

edit retag flag offensive close merge delete