This is a little late coming but figured I should post anyways. ChefConf 2012 was an amazing event and absolutely surprising considering it’s the first conference Opscode has put on. Here are a few things that I noticed:
The attitude and culture around the people there was incredible. Everyone was focused on the needs of the business that they worked for. Sometimes it meant making developers more productive by being to provide consistent environments or by lessening the pain operations have to go through in order to release software created by development teams.
It inspired me to see so many people in the business to help others. There was not a lot of ego, but there was definitely a sense of pride in the work that many of the people did there.
I was hoping to find some excited chefs out there that would love to work for ThoughtWorks but it seemed everyone was doing the same. Attempts to recruit me occurred as I was trying to recruit them! This is definitely a hot part of the software industry.
What made matters worse is that nearly everyone is extremely satisfied with their jobs. Pretty hard to recruit people that are quite comfortable with their positions and projects.
Configuration management of a single node seems to nearly be a solved issue. A running theme amongst many of the sessions was that orchesration. There were many discussions but it didn’t sound like it’s been solved in a uniform way. Maybe orchestration is something that has to be done custom depending on your environment. While listening to the conversations it appeared that many had different requirements for how they do the orchestration. I’ve written plenty of orchestration code and was relieved that I wasn’t the only person that found it hard to do. This subject will my “next level” in understanding infrastructure automation processes.
Sessions of Note
Transformation to Continuous Delivery at Ancestry
Since I have a lot Windows background, I was really interested in the two talks given by the folks from Ancestry.com. The first one was a high level summary of Ancenstry’s path to Continuous Delivery. John Esser articulated the principles well and was pleasantly surprised to see him promote ThoughtWorks Studios’ product Go.
The main takeaway I got from the talk that the transformation seemed to be more of a culture shift, rather than a technical one.
DevOps + Chef + Windows at Ancestry.com
Next was a more technical discussion by Dan Gilmer which reminded me of all the lessons I’ve learned in the Windows automation world. I had an excellent chat with him and shared war stories that were all to familiar to him. I also got a kick out of his pride in working for days without ever having to Remote Desktop into a machine.
This is a problem domain that seems to be repeatably solved. Hopefully some OSS will come out of this. I would like to dive deeper into how they are driving out infrastructure using data_bags rather than roles.
Opscode Chef State of the Union
Adam Jacobs keynote was a good reminder that Chef is about providing us with infrastructure primitives. As much as we like to standardize, the real world isn’t ever going to reach a global convergence. Every site is different and they decided they didn’t want to be the dictators on how you setup your infrastructure. I think this is great because it allows Chef users to define the opinionated stacks that they require. It’s a good separation of concerns in my opinion and a concept that I will try to pull into future projects.
Test-Driven Development for Chef Practitioners
Jim Hopp from Lookout had an awesome slide deck (hand drawn with paper and crayola markers) that discussed Test Driven Development of cookbooks. I look forward to trying out the concepts he discussed in my next project. For developers the concepts might seem pretty standard but the ideas are very new for operations minded folk.
CycleCloud + Chef = 50,000-core Utility Supercomputer for Science
Lastly, CycleCloud’s Jason Stowe impressed everyone by discussing how they were able to spin up $20 million dollars worth of infrastructure using AWS and only paid $5k/hr for 3 hours use. Not only did the massive infrastructure speed things up for them, it actually produced better results when performing protein analysis. It begs the question of what other things are we missing because we are yet to through large enough computing systems at them.
The couple days were inspiring by I had to keep my energy in check because I learned that certain things are not necessary for small projects with just a handful of servers. I’ve been using chef-solo for all my machines because the size of the environments I’m dealing with can easily be tracked in simpler ways. I do look forward to the day when a chef server is necessary to deal with all the infrastructure. That’s where metrics like 1 engineer per 1000 machines is possible.