openstack team mailing list archive

[Ben <ben.div@xxxxxxxxx>]

Hi the list,

I had issues with nova-scheduler, but after upgrading ubuntu packages, 
the problem have disapeared.

Now, I have another well know problem : on a multinode setup, I have a 
connection refused when the instance tries to get metadata from nova-api :

2012-08-29 22:00:05,968 - util.py[WARNING]: 
'http://169.254.169.254/2009-04-04/meta-data/instance-id' failed 
[0/120s]: url error [[Errno 111] Connection refused]

If I watch to iptables, I can see that 169.254.169.254 is rerouted to 
itself (.52, the compute node) instead of the controller node (.50)

# iptables -t nat -L -v | grep -n3 169.254.169.254

45-Chain nova-network-PREROUTING (1 references)
46- pkts bytes target     prot opt in     out     source 
destination
47:   17  1020 DNAT       tcp  --  any    any     anywhere 
169.254.169.254      tcp dpt:http to:94.xx.xx.52:8775

I tried these option in nova.conf but it didn't solved the problem
--dmz_cidr=94.xx.xx.50/32
--ec2_dmz_host=94.xx.xx.50

How can I force nova-compute to reroute the metadata request on the 
controller node instead of itself ?

Another non-related question : my nova.conf have its options prefixed 
with -- (example --multi_host=true)
I saw some nova.conf with no prefix, just the name of the option 
(example multi_host=true).
What is the good format ?


Thank you,

Ben