Open Stack

Hi,

I just posted the following at http://forums.openstack.org/viewtopic.php?f=15&t=1435, then realized this mailing list might be a better place to ask the question.

In summary, I've cobbled together devstack-based nodes to exercise quantum/openvswitch (when I say cobbled, I mean my result is the combination of information from wiki and from devstack, and elsewhere to create my localrc files, since there is no one definitive template that I could use, and it seems that devstack examples are not current with what is happening on Folsom). One node is a controller, one is a compute node. I can launch using horizon on the controller, VMs launched on the controller are pingable, but ones launched on the compute node are not. The big difference I can see is a missing gateway interface on the controller (on gw-* displayed when I run ifconfig). By inspection of the logs, I can see that the VMs are unable to establish a network, and I think the missing gateway interface may be the root cause for that.

Below are details:

Two hosts, one configured as a controller, the other configured as a compute node.
Each host is dual homed, network for eth0 is connected to the local intranet, network for eth1 is configured as a local net 192.168.3.0
On the controller host, I used devstack with the following localrc (which is an aggregation of stuff I found on the devstack site, and stuff I found recently on the quantum wiki -- it would be nice if complete templates for a controller and compute node supporting devstack and openvswitch were published on the devstack site or the wiki, perhaps since we are not yet at Folsom it makes sense they don't exist, if I get something working, I will share my configuration in the entirety at whatever is the most appropriate place). Anyway, controller host localrc is:

HOST_IP=192.168.3.1
FLAT_INTERFACE=eth1
FIXED_RANGE=10.4.128.0/20
FIXED_NETWORK_SIZE=4096
FLOATING_RANGE=192.168.3.128/25
MULTI_HOST=True
LOGFILE=/opt/stack/logs/stack.sh.log
ADMIN_PASSWORD=password
MYSQL_PASSWORD=password
RABBIT_PASSWORD=password
SERVICE_PASSWORD=password
SERVICE_TOKEN=xyzpdqlazydog
ENABLED_SERVICES=g-api,g-reg,key,n-api,n-cpu,n-net,n-sch,n-vnc,horizon,mysql,rabbit,openstackx,q-svc,quantum,q-agt,q-dhcp
Q_PLUGIN=openvswitch
Q_AUTH_STRATEGY=noauth

If I run stack on this host, I get the following nova.conf:

[DEFAULT]
verbose=True
auth_strategy=keystone
allow_resize_to_same_host=True
root_helper=sudo /usr/local/bin/nova-rootwrap /etc/nova/rootwrap.conf
compute_scheduler_driver=nova.scheduler.filter_scheduler.FilterScheduler
dhcpbridge_flagfile=/etc/nova/nova.conf
fixed_range=10.4.128.0/20
s3_host=192.168.3.1
s3_port=3333
network_manager=nova.network.quantum.manager.QuantumManager
quantum_connection_host=localhost
quantum_connection_port=9696
quantum_use_dhcp=True
libvirt_vif_type=ethernet
libvirt_vif_driver=nova.virt.libvirt.vif.LibvirtOpenVswitchDriver
linuxnet_interface_driver=nova.network.linux_net.LinuxOVSInterfaceDriver
osapi_compute_extension=nova.api.openstack.compute.contrib.standard_extensions
my_ip=192.168.3.1
public_interface=br100
vlan_interface=eth0
flat_network_bridge=br100
flat_interface=eth1
sql_connection=mysql://root:password@localhost/nova?charset=utf8
libvirt_type=kvm
libvirt_cpu_mode=none
instance_name_template=instance-%08x
novncproxy_base_url=http://192.168.3.1:6080/vnc_auto.html
xvpvncproxy_base_url=http://192.168.3.1:6081/console
vncserver_listen=127.0.0.1
vncserver_proxyclient_address=127.0.0.1
api_paste_config=/etc/nova/api-paste.ini
image_service=nova.image.glance.GlanceImageService
ec2_dmz_host=192.168.3.1
rabbit_host=localhost
rabbit_password=password
glance_api_servers=192.168.3.1:9292
force_dhcp_release=True
multi_host=True
send_arp_for_ha=True
logging_context_format_string=%(asctime)s %(color)s%(levelname)s %(name)s [^[[01;36m%(request_id)s ^[[00;36m%(user_name)s %(project_name)s%(color)s] ^[[01;35m%(instance)s%(color)s%(message)s^[[00m
logging_default_format_string=%(asctime)s %(color)s%(levelname)s %(name)s [^[[00;36m-%(color)s] ^[[01;35m%(instance)s%(color)s%(message)s^[[00m
logging_debug_format_suffix=^[[00;33mfrom (pid=%(process)d) %(funcName)s %(pathname)s:%(lineno)d^[[00m
logging_exception_prefix=%(color)s%(asctime)s TRACE %(name)s ^[[01;35m%(instance)s^[[00m
compute_driver=libvirt.LibvirtDriver
firewall_driver=nova.virt.libvirt.firewall.IptablesFirewallDriver
enabled_apis=ec2,osapi_compute,osapi_volume,metadata

If I run horizon, I can launch vms and ping them. If I look at the logs generated by the VMs, they are able to get a network. Furthermore, I get the following network interface in addition to the tap interfaces generated for each VM:

gw-4f16e8db-20 Link encap:Ethernet HWaddr fa:16:3e:08:e0:2d
inet addr:10.4.128.1 Bcast:10.4.143.255 Mask:255.255.240.0
inet6 addr: fe80::f816:3eff:fe08:e02d/64 Scope:Link
UP BROADCAST RUNNING PROMISC MULTICAST MTU:1500 Metric:1
RX packets:0 errors:0 dropped:0 overruns:0 frame:0
TX packets:6 errors:0 dropped:0 overruns:0 carrier:0
collisions:0 txqueuelen:0
RX bytes:0 (0.0 B) TX bytes:468 (468.0 B)

Now, for the compute node, I use the following:

HOST_IP=192.168.3.2
FLAT_INTERFACE=eth1
FIXED_RANGE=10.4.128.0/20
FIXED_NETWORK_SIZE=4096
FLOATING_RANGE=192.168.3.128/25
MULTI_HOST=1
LOGFILE=/opt/stack/logs/stack.sh.log
ADMIN_PASSWORD=password
MYSQL_PASSWORD=password
RABBIT_PASSWORD=password
SERVICE_PASSWORD=password
SERVICE_TOKEN=xyzpdqlazydog
Q_HOST=192.168.3.1
MYSQL_HOST=192.168.3.1
RABBIT_HOST=192.168.3.1
GLANCE_HOSTPORT=192.168.3.1:9292
ENABLED_SERVICES=n-cpu,rabbit,g-api,n-net,quantum,q-agt
Q_PLUGIN=openvswitch
Q_AUTH_STRATEGY=noauth

The resulting nova.conf is:

[DEFAULT]
verbose=True
auth_strategy=keystone
allow_resize_to_same_host=True
root_helper=sudo /usr/local/bin/nova-rootwrap /etc/nova/rootwrap.conf
compute_scheduler_driver=nova.scheduler.filter_scheduler.FilterScheduler
dhcpbridge_flagfile=/etc/nova/nova.conf
fixed_range=10.4.128.0/20
s3_host=192.168.3.2
s3_port=3333
network_manager=nova.network.quantum.manager.QuantumManager
quantum_connection_host=192.168.3.1
quantum_connection_port=9696
quantum_use_dhcp=True
libvirt_vif_type=ethernet
libvirt_vif_driver=nova.virt.libvirt.vif.LibvirtOpenVswitchDriver
linuxnet_interface_driver=nova.network.linux_net.LinuxOVSInterfaceDriver
osapi_compute_extension=nova.api.openstack.compute.contrib.standard_extensions
my_ip=192.168.3.2
public_interface=br100
vlan_interface=eth0
flat_network_bridge=br100
flat_interface=eth1
sql_connection=mysql://root:password@192.168.3.1/nova?charset=utf8
libvirt_type=kvm
libvirt_cpu_mode=none
instance_name_template=instance-%08x
novncproxy_base_url=http://192.168.3.2:6080/vnc_auto.html
xvpvncproxy_base_url=http://192.168.3.2:6081/console
vncserver_listen=127.0.0.1
vncserver_proxyclient_address=127.0.0.1
api_paste_config=/etc/nova/api-paste.ini
image_service=nova.image.glance.GlanceImageService
ec2_dmz_host=192.168.3.2
rabbit_host=192.168.3.1
rabbit_password=password
glance_api_servers=192.168.3.1:9292
force_dhcp_release=True
multi_host=True
send_arp_for_ha=True
api_rate_limit=False
logging_context_format_string=%(asctime)s %(color)s%(levelname)s %(name)s [^[[01;36m%(request_id)s ^[[00;36m%(user_name)s %(project_name)s%(color)s] ^[[01;35m%(instance)s%(color)s%(message)s^[[00m
logging_default_format_string=%(asctime)s %(color)s%(levelname)s %(name)s [^[[00;36m-%(color)s] ^[[01;35m%(instance)s%(color)s%(message)s^[[00m
logging_debug_format_suffix=^[[00;33mfrom (pid=%(process)d) %(funcName)s %(pathname)s:%(lineno)d^[[00m
logging_exception_prefix=%(color)s%(asctime)s TRACE %(name)s ^[[01;35m%(instance)s^[[00m
compute_driver=libvirt.LibvirtDriver
firewall_driver=nova.virt.libvirt.firewall.IptablesFirewallDriver

I can spin up VMs on this host (usually, when I run horizon on the controller, it is this host upon which the first VM is launched). I get expected IP address in the range 10.4.128.*

Unlike VMs on the host, I cannot ping (from either the controller (less worrisome) or the compute node (very worrisome). I looked at the console log for the VM, it is not getting any network. The other major obvious difference is that there is no inteface gateway device when I do an ifconfig on the compute node.

It is this last point (the lack of a interface gateway) that seems most likely to me to be the issue. Is there something I can run after launching devstack on the controller, before I try to launch VMs, that will cause that gw to be created?

I did some tracebacks in the python code on the controller and it appears the gateway on the controller is being created by the quantum (???) service during its initialization (I grepped around for "gw-") to identify where I should be putting tracebacks. According to what I have read on the net, localrc should not be enabling q-svc on the controller (and this makes sense given I am pointing back at 192.168.3.1 for quantum, as well as other services).

Again, hoping I mostly have the localrc contents right, and that maybe I just need to add some commands to the end of stack,sh to finish it off. Been frustrating seeing the VMs get launched only to not be able to ping them (but damn, that's pretty cool they spin up, don't you think?) I have a lot to learn still (just 2 weeks into this) but kinda stuck on this issue.

Regards,

syd
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20120806/b91a8acd/attachment.html>