For the past week or so, in my copious spare time, I’ve been re-building my vLab at work. It’s a cobbled together menagerie of hardware that makes me wish I had a healthier budget so I could spend more time on learning and less on reconfiguring, scrounging and breaking out the baling wire and chewing gum. Dealing with old hardware is distracting and takes your mind off this things that are critical for success. This happened to me. I know plenty of friends in the vCommunity that also have dealt with this. (I hear your heads nodding)

One of the things I’m playing with in in the lab is configuring VMware vCloud Director 5.1 with vCenter 5.1’s SSO functionality. I’m finding that this is one of those times when you really should RTFM and plan ahead more. But that’s ok, I like diving in without docs because then I get to learn more by breaking things and then I have something to share.

Single Sign On

In vSphere 5.1 there is a new feature called Single Sign On. With the new vCenter client now being web based,  SSO now allows VMware to leverage industry standards like SAML so that an admin can log  once to vCenter and be automatically signed on to other resources like vShield Manager and vCloud Director. There’s a great overview of VMware SSO from Justin King here. You can read more about troubleshooting SSO here.

Not Kirk’s Federation

With vCenter and the SSO components up and running I installed the vCloud virtual appliance OVA and proceeded to set up federation between vCenter and vCloud. You can read more about federation in the Wikipedia article, but in a nutshell, it’s a way of linking identities. So, mike@foo.com and mike@bar.com can be linked. A trust relationship is set up so that if I log in from foo.com and hit a web page that needs my bar.com identity I get logged in using my bar.com account without having to provide the credentials.

Back to my lab, when I tried to do this, I ran into some trouble. I kept getting the following error.

Long story short, I had screwed up. I hadn’t been as diligent as I should have been with setting up my systems. The distractions I mentioned earlier had come back to bite me. Having recently added a couple of disk-less ESXi servers, I didn’t finish configuring them. I hadn’t been leveraging NTP as much as I  should.  What I now had was a number of  systems trying to talk to each other but not agreeing on the time. And with SAML, time is critical. You are issued a token and part of that token is a timestamp. If vCloud thinks it is 9pm and vCenter thinks it is 8pm, the token with a short lifetime will come back as already expired. Just what happened to me.

The Time/Space Continuum is restored

I fixed the problem by setting all the ESXi servers to pull from an NTP server. In addition I also set up my vCenter, SQL server and domain controller to pull from NTP. Just like DNS has become a critical infrastructure component that just needs to be there, NTP is now jumped to the “required” list. Valid, consistent time stamps across your datacenter are needed for the operation of your infrastructure. They are also critical for security in things like authentication and log analysis, especially when you are trying to correlate lots of information!

NTP is critical to other vCloud operations as well. My friend Chris Colotti called this out in this blog article.

If you haven’t set up your datacenter with NTP, now’s the time. Oh, and DNS too. It’s time to ditch the hosts files. (yes, there’s still plenty of people using them!!) Going forward, more and more of your datacenter are going to rely on these key components. Ensuring that you use them consistently will help a LOT in troubleshooting large and complex installations.

Thanks for reading!

mike