News
Justin Garrison: lessons from building and shipping Disney+
At Edgecase 2025, Justin Garrison told a story that captivated the entire audience. Currently Head of Product at Sidero Labs, Justin was one of the people who built an entirely new streaming platform for Disney. His storytelling showed he had learned from the best storytelling company in the world.
Justin apologized right at the start of his talk because it isn't about edge computing or even about his current employer, Sidero Labs. Sticking to the theme of the conference, he expressed the hope that his story encourages audience members to apply the lessons learned in their own timelines.
It starts with someone who believes in you
“Once upon a time”, that’s how this Disney story starts. In December 2018, it was Justin's first day on the job. He had something going for him at Disney: a manager who believed in him. The management knew that he could do it. And they needed that, because Disney+ has to be stable. It can’t be down.
Although he had written a book about the cloud, Justin himself had never done production things in the cloud environment: “And somehow, through the interview process, I convinced people to let me run cloud infrastructure for them in a large-scale environment. My manager was probably the best manager I've ever had in my life, not because of anything specific management did, but because they believed in me. Because every time I would ask questions, they believed that I'd come up with the right answer, and they knew that I could do something that I did not think I was capable of. And so I would go to them for help.”
11 months to build Disney+
Justin was one of four people in the team. Two people had experience building streaming infrastructure for years. But these two colleagues told Justin that they would be gone in two weeks, so he had a lot of knowledge to absorb: “It was the first time and only time so far in my career that I felt like a complete imposter. I did not belong. This was not something that I knew how to do.”
“At that point, I froze. I'm just like, I don't know where to start now. I would talk to my manager, and he told me just to ask the rest of the team what they think will break. Because it turns out that a lot of people running the systems all day, every day, usually know what's going to break. I thought that was a great idea. I was in charge of infrastructure and observability. And we had a team that did CI/CD, Jenkins. And we had a team that did a deployment API. Great. I'm just going to interview everyone and ask them, What do you think is going to break? And nine out of ten people told me Elasticsearch was going to break. If you've ever run Elasticsearch, you probably might just assume that it's going to break, too. So now I knew where to start. We eventually picked Datadog to help scale Elasticsearch.”
Another person said that cluster maintenance is going to break. Deploying and upgrading clusters was done from a single Ruby script and just rotated out an auto-scaling group. So, how to solve the cluster maintenance problem? Justin and his colleague both had Kubernetes experience and Disney+ was built on ECS. Managing cloud formation was the way to go, so they said: what if we built a cluster API for ECS?
They did that, but it broke, and nobody knew why. Workloads, timestamps, logs, snapshots: Justin analyzed them all. The solution needed to be in place in two months. In this case, delete the step of logging into SysOps in the process. Another lesson learned.
Scaling and more scaling
When Disney+ wanted to scale up its hosting, Amazon told them they had no more physical servers available, not even virtual servers. This was a week before everything should be ready, the public launch of the streaming service. So they decided to roll out the clusters one at a time to see if it doesn’t break. That worked!
The internal Slack channel had a list of who was responsible for what task during the launch. Sounds simple, but it worked.
Justin could see the subscriptions coming in. Right before he went to sleep, he made an ITTT for a Disney+ hashtag on Twitter to alert him. That exploded during the night. With 10 million sign-ups in a day, it was the fastest-growing paid service in the world. But troubleshooting was still daily work. For example, only three VMs got all this traffic, which forced immediate scaling challenges, and implementations were done as soon as possible.
The price to fix things
An important lesson that Justin learned is: you can tell management whatever price it costs. They will pay it. So always give large estimates for what you need. Because if you go back to them during the process to tell them you need more budget, they get mad. This is the best advice Justin ever got from a manager, he says.
Talking about value for money: the millions that Disney spent on building Disney+ were very well spent. The trust that the service would not be going down led to a 60 billion dollar stock value increase for Disney in the very first weeks after the launch.
The bottom line of Justin’s talk: trust in yourself only comes once you’ve done it. The manager who started it all by trusting Justin was the catalyst for this ROI. Justin is still thankful for believing in him.
