text
stringlengths
0
2.67k
**Martin Jackson:** Alright, so the way we did this - we enlisted a lot of help. There were, I don't know, three, four, five of us full-time on the configuration management team. We had a bunch of people from the stores team that were helping. We had a bunch of people from the DCs team, the distribution center team, be...
Now, one of the things to understand about the Walmart architecture at the time is that there was a very well-established protocol for what you have to do when you deploy one application to every store. Everybody understood that process, people knew what to expect, and there's a cadence. One store first, then 5, then 1...
**Autumn Nash:** So did you do this with no scheduled downtime, being that some Walmarts are 24 hours?
**Martin Jackson:** Oh, yeah, there was no need for downtime for any of this. I mean, we didn't schedule any downtime...
**Autumn Nash:** Well, that's what I'm saying. Did you have a window, like, if it all went wrong? Because this is the first time that you were doing it on that scale. Oh, okay. So not at night?
**Martin Jackson:** Right. Night time is when all the important, business-critical end of day functions and things like that happen.
**Autumn Nash:** That makes sense.
**Martin Jackson:** Cash wrap, and all those kinds of things. So it's really attractive to say "Yeah, we'll do this at one in the morning." Like, "Y'all want to impact our cash reconciliation...?"
**Autumn Nash:** It doesn't sound good.
**Martin Jackson:** "No, we don't want to impact your cash reconciliation. No." So there were two aspects of this that made it work for us. The first was that we made -- first of all, we rolled out Puppet in enforcing mode, everywhere, day one. Had to be. Because you're not providing any kind of value, and you've got n...
So we were deploying Puppet, and the only thing that it was managing in full enforcing mode was NTP. So our goal was to get the Puppet catalog that manages NTP and Puppet itself onto every store server by lockdown.
**Autumn Nash:** So when you were doing this, you said that you pulled people from other teams. What was the prep for that? Did you have to then train those other teams to be able to help you with this, or did you kind of just make it as easy as possible and tell them to press button? What was the prep to get these oth...
**Martin Jackson:** Oh, we were absolutely making it up as we were going.
**Autumn Nash:** So how do you then teach other departments to do this for this big rollout? What was your plan for if it went wrong? What was your, I guess, backup?
**Martin Jackson:** Well, you know, when Cortez reached the new world, he burned the ships...
**Autumn Nash:** You were like "YOLO."
**Martin Jackson:** There wasn't much in the way of a backup plan. The simple thing would have been just to use one of the other mass orchestration systems we had to remove every last trace of Puppet from all the servers that it was on. That was pretty straightforward. And so we didn't put much more thought into it tha...
But getting these other people involved... We had an incredible esprit de corps on the configuration management team. We very much believed that what we were doing was going to fundamentally change the way people interacted with systems. We saw the change that was happening in the world, and that -- you know, everybody...
\[42:06\] These days you can name any 10 groups that are running a hundred thousand servers. It's not that big of a deal. For us at the time it sure was, now it isn't. And it was important to us to kind of be part of being part of that hinge, where everybody said "Hey, maybe we can do this at this scale, making people'...
So all of these guys had spent most of their careers doing these 2, 3 AM calls and saying "Hey, we have a system that's going to make this easier, for you and for everybody who follows. And if we get this done, we will be legendary." It was not hard to get people to sign on to that.
**Autumn Nash:** Yeah, for sure. It makes your life better.
**Justin Garrison:** So you made it? So you deployed all 5,000 stores in two months?
**Martin Jackson:** Yup.
**Autumn Nash:** These retro episodes are so cool, because for instance WebMD - I didn't know what... What did they end up buying, or what were they supposed to be -- I think they were supposed to be one of the biggest Walmarts of medical, or something... Yeah, \[unintelligible 00:43:34.01\] And then who would have tho...
**Martin Jackson:** We were very happy to not be on the front page because of that stuff.
**Justin Garrison:** Yeah. You don't want to make news there.
**Martin Jackson:** Well, we didn't want to -- it was something that we knew. And we could walk around and say "Yeah, we know that we've done these things." And if you're interested in sort of the overall history of it, everything changed at Walmart the day that Amazon decided that it was going to be profitable.
**Autumn Nash:** That's interesting.
**Martin Jackson:** The day that Amazon decided it was going to be profitable, everybody said "Okay, stop. We need to do something now." For years it was "Oh, they sell books. Oh, they're never going to be profitable. Are they even in retail?" And then Amazon decided it was going to be profitable, and a hush fell over ...
**Justin Garrison:** One thing about this episode is it's like an unintentional retro episode, but also it's completely modern, right? This isn't something that happened in the nineties, or before. This is 2010, 2014, 2015 timeframe, still happening today. Looking back at everything that you did and accomplished with t...
**Martin Jackson:** Oh...
**Justin Garrison:** It's a big question to put you on the spot for.
**Martin Jackson:** I've had an awful long time to think about all of this, and the adoption of Puppet, and the usage of modern infrastructure as code systems... We paved the way for a bunch of other modern orchestration systems, and more importantly, we gave the old architecture the bridge into the future that it need...
\[46:07\] There's a lot to say -- the company needed that stuff to keep running. And if it hadn't been able to keep managing it, things would be very different. But the one thing that I wish we could have done better in those days was to prioritize and facilitate self-service for infrastructure as code, without a doubt...
**Justin Garrison:** Enable other people to -- yeah.
**Martin Jackson:** Right. Because I love Puppet, a lot. I have friends that work for Puppet, I have friends that used to work for Puppet... A few of my friends at Puppets are still there. Most of them have moved on since the acquisition. But one of the hardest things about adopting Puppet at scale is that if you use t...
But the sort of Achilles heel of a centralized Puppet setup like that is that everything that happens in Puppet is based on this classification process, which is where the node sends in its facts, and then the Puppet server evaluates those facts and then decides which classes it's going to send down, and effectively wh...
**Justin Garrison:** That's a long chain of events.
**Martin Jackson:** It is. And a lot of other infrastructure-as-code systems are much better about facilitating self-service in that way.
**Justin Garrison:** Well, Martin, this has been a blast, and it's not even a look back in history, it's just a look back in the last 5 to 10 years. This is just how things are for a lot of people still today. There's still a lot of infrastructure and organizations that are going through this type of migration, and loo...
**Martin Jackson:** Oh. Well, I have a Twitter account that I almost never post from, that's Mjolnir40k. And LinkedIn, I suppose, is one of my main online presences. But I speak at a bunch of Red Hat events, and I'm usually at Automates, or Summit, or those kinds of things.
**Justin Garrison:** Cool. I'll put a link to some of those in the show notes, and thank you so much. This was great.
**Martin Jackson:** Alright. Thank you very much.
**Break**: \[49:24\]
**Justin Garrison:** Thank you so much, Martin, for coming on the show and talking to us about what it was like for you at Walmart, deploying at this scale, and making these foundational changes in a very diverse infrastructure. It was awesome, and also a little bit scary for me to remember some of the things that I we...
**Autumn Nash:** Also, I really love the vibe of the retro episodes. I love the fact that so much of what we did before is still so relevant.
**Justin Garrison:** And having the time to look back longer term - you're disconnected from the problem, you've thought about it, and you thought about all the things you did right, and some of the things you did wrong, and you can just say "Actually, we did the best we could. With the knowledge we had and the tools w...
**Autumn Nash:** It also gives them enough of a layer into a little bit less -- I don't know. Sometimes you need to know a little bit more of what you're doing because it helps you to be able to solve problems and see where those things intersect. We want to abstract problems away, but you don't want to abstract them s...
**Justin Garrison:** Yeah. The layer below is sometimes important. And I went through multiple companies where we tried to make self-service configuration management for application developers a thing, and it was always the wrong layer, because it was not an application development tool.
**Autumn Nash:** Because you don't want to give -- sometimes developers are like "I want control of everything." And we're like "Bro, you don't need that." And then sometimes we're like "But we need you to not just do hood rat things on the top layer, because you don't understand the bottom layer. We need you to -- the...