Spawned VMs not getting IP address

[Abdi Ibrahim]

Hi All,
I'm fairly new to the OpenStack community but I can come up to speed very quickly.... so bare with me if I do not or mis understand something.

I started building a lab as a PoC (Prof of Concept) for the OpenStack Folsom Edition based on Cisco documentation - link. It's pretty much a document that Cisco put together based/inspired on/from: Bilel Msekni OpenStack Folsom Install Guide - link, Emilien Macchi's Folsom guide - link and the OpenStack Offical Core document.

Th environment is configured as follow:
- 3 x HP DL380 G6 (12 Cores, 48 GB RAM, 12x600GB of local SAS disks and 4 x1Gbps NIC )
- 1 Control Node: All nova core/shared and API services running
- 1 Network Node (Quantum): dhcp, L3, ovs-agent and ovs-plugin
- 1 Compute Node: nova compute
- OS Platform: Ubuntu 12.04 TLS

I've attached an overview diagram of the deployment.

Everything works perfectly as per design with one major exception: I'm able to spawn VMs but they're not getting any IPs. I'm using GRE in tunnel mode for th L2 configuration. I choose this mode because it's far simpler to implement and have no limitation on VLANs. I've looked in all the logs and I cannot pinpoint anything as to what's causing this issue. I create 2 images in Glance (one for Cirros and the other for Ubuntu cloud image) and both images are producing the same issue: VM is spawned and during the boot time they issue a dhcp discover and no lease is offered (at least from the logs).

I can appreciate if you give me some direction on where I should start looking or if someone have ever came across this issue before. let me know if you need logs or config files for my deployment.

Thank you,
Abdi

[subconscious (subconscious)]

I see you mention that you followed the Cisco document for your PoC. Did you also use the Cisco folsom packages?

[subconscious (subconscious)]

Also, I should mention that if you are using the Cisco packages, there is a new release that we put out on this past Friday. You may want to upgrade.

[Abdi Ibrahim (abdi-w)]

Yes, I followed the documentation steps including using the COE (Cisco OpenStack Edition) packages in my build. I'll try to give an upgrade this afternoon and will post result. I need to address this issue before I can proceed show casing the OpenStack cababilities to my peers before we can assess it's production readiness.

Thank Don... will get back to you.

Abdi

[Abdi Ibrahim (abdi-w)]

One more thing Don: Is there a specific way of upgrading the COE packages besides just issuing the "apt-get update && apt-get dist-upgrade -y" command. COE repository is already in my package source list file.

Thanks,
Abdi

[subconscious (subconscious)]

Abdi, my last reply was somewhat incorrect. Since you are upgrading you do not have to run the puppet-modules.sh script. Rather you will want to delete the /etc/puppet/modules directory. The puppet modules are now delivered via at and located in /usr/share/puppet/modules. I don't recall if the package makes a symlink from the new directory to /etc/puppet/modules, but if it does not, then you will need to create one.

[Abdi Ibrahim (abdi-w)]

Hey Don,
I actually use the manual build approach where I followed the entire deployment using the step by step provided by the document. The rational is as the guide also noted for me to be very familiar with all of the OpenStack components; specifically the Quantum service and OVS plugin with GRE in tunnel mode.

[Abdi Ibrahim (abdi-w)]

Well just that it's up there here is the link for the guide:
http://docwiki.cisco.com/wiki/Cisco_OpenStack_Edition:_Folsom_Manual_Install

[Abdi Ibrahim (abdi-w)]

Hi Don,
I've rebuilt the entire installation and cluster but still VMs created by the tenants are not getting IP addresses from the DHCP agent. there is either something fundamentally wrong with the document, assumption been made in that I'm missing or there is a bug somewhere.

I can not be the only one seeing this issue... I followed the documentation 3 - 5 times and always produced the same results as far as spawning VMs.

I'm suspecting that there is an issue in the networking processes of the Quantum: the server, the agents (l3, dhcp and ovs-plugin) or the step produced by the documentation.

Again, I'm following the manual install steps document/guide.

I would appreciate if someone out there in your community can provide some input here.

Thank you in advance,
Abdi

[Abdi Ibrahim (abdi-w)]

it's been a week since my last update but here is the final one:
I fixed this issue by rebuilding the cluster using the source files from Ubuntu's cloud archive repository instead of the COE sources. followed the same manual build procedure but with "Per Tenant Routers with Private Networks" in the Quantum configuration.

Quantum is configured with OVS + GRE with Tunneling.

VMs are getting Private Tenant IP addresses and after contacting the Metadata server running in my controller node they're also getting their private ssh keys injected. After associating/assigning floating IP address to the instances I'm also able to connect them from the public network.

Made a lots of good and positive progress... thank you for your inputs thus far.

Abdi