First Impressions
In December of 2011, StackOps released version 0.3 of their OpenStack distribution, following the Diablo stable release that occurred in late September. I recently went through the process of using the StackOps Smart Installer to get a single node instance up and running, and wanted to share my experiences.
First, I love the general idea of simplifying the installation and configuration of the OpenStack components. At least for a proof of concept, this is a great tool to get you started with the platform. Unfortunately, after both using the installer and reading through all of the StackOps documentation on the subject, I'm left with concerns about the whole Smart Installer model itself. I'd love for StackOps to rethink some of their base assumptions and decisions, so that they can move forward with a more compelling model.
Basically, the Smart Installer concept is to start with the StackOps distribution ISO image (or a bootable USB drive with the same material on it), perform a fairly standard Ubuntu Server 10.04 LTS setup process, and then use their Smart Installer system to get the node up and running.
My first general concern is directly tied to the way that the Smart Installer works. The newly installed server hosts an installation control process, via a web server listening on TCP 8888 and running on the server itself. When a user hits that URL, some information about the server is passed off to the StackOps hosted Smart Installer landing page. Here, you're asked to register and then log into their site. While I'm sure that StackOps would say that they treat information about the registered systems and users with care, I got that nagging feeling that exposing the configuration details of a production OpenStack platform to a third party might not be the best thing for a provider to be doing. I do like the fact that they make it clear for how users can opt-in/out of receiving commercial sales contact and email from the company when registering. They also make it possible to perform the installation without registering a user account, which theoretically means that your server configuration data isn't stored permanently. I'm just not sure that these options are enough to get me past the idea that the system data has to be passed to the distribution vendor.
Another general concern with the Smart Installer approach is scalability. I wasn't able to determine how the installation process would work when done at any significant scale. For example, assume that you have a compute pod with 80 hosts. That's allot of website navigation to get all 80 configured. It seems like Dell's Crowbar system is much more useful for large scale deployments. I'd love to find out that StackOps has a similar approach!
Beyond the major concerns of data security and installation process scalability, there are a few minor issues that I ran into while performing the installation:
Installation Progress Tracking
When you get to the end of the configuration data input process, the actual installation and configuration of OpenStack is like a black hole process. First off, when you click the "Start deployment now!" button, there is no visual indicator anywhere on the page to tell you where the installation process is in it's steps, or even if the process has actually started! The button itself doesn't even reflect the fact that it was clicked (I tested on Chrome, so perhaps this works on other browsers?). The only indication you get that something is happening is what appears to be a very long page loading process.
That is, until you get an error...
Error Reporting
The Smart Installer configuration process makes it clear that OpenStack Horizon support is experimental, but it's enabled by default. Unfortunately it doesn't seem to work at all.
My issue isn't so much about this experimental feature, but the fact that error conditions are so poorly reported to the user when they DO occur (I ran into volume manager errors as well, but that was my fault for not reading the documentation thoroughly enough). When a problem happens, you see this:
That's a pretty ugly user error report, with no obvious way to deal with the condition. The trick to fixing this is to use the browser back button to go backwards through the configuration process, and then uncheck the option to install Horizon. You can then for forward again, re-filling in the required data for the screens that you went past on your way back.
Online Help / Support Confusions
Gor the actual Horizon installation issue, I reported it on the StackOps Forum. I'm not sure what to make of the forum though, since it's pretty empty in there! The other place that I see StackOps directing users is to the StackOps environment on Get Satisfaction. I couldn't find any reference to the problem on that site.
I did end up finding it in the StackOps Jira tracking system (which I have to say is nice to be able to access as a user): STACKOPSDISTRO-51 You have to use your installer.stackops.org account to access the Jira Issue.
Thinking About the Future
I believe that the StackOps team is doing a great job of taking the young OpenStack projects and creating a reasonably useful distribution for fast proof of concept environments. I truly respect that they are only on version 0.3 of the StackOps Distro, and I know how long a road it can be to get to a version 1.0. Hopefully, my general concerns about the model can be addressed over time (while still keeping the current, or similar, approach in place for the small scale non-production environments). They should be happy about what they have built, but they shouldn't be satisfied yet. Good luck StackOps team!
