A laid-off engineer recently exposed the architecture behind Jira and Bitbucket.
So, what does their eight year Envoy proxy infrastructure look like?
As most of us are aware,
Corporate reality is entirely decoupled from engineering reality.
It’s like
You can spend eight years building the exact network infrastructure that scales a 1.79 billion dollar revenue stream, and still get cut in a spreadsheet optimization.
Atlassian laid off 1,600 roles in March.
One of those engineers, Syrakis, published a teardown of his eight-year project: the Open Service Broker.
When I am aggregating technical teardowns and deployment strategies for GritGlean, I am looking for this exact kind of raw, unfiltered production data.
No sanitized marketing copy, just the brutal reality of scaling enterprise infrastructure.
Legacy enterprise infrastructure is notoriously reliant on expensive, proprietary hardware appliances for routing and proxying.
Syrakis and his team ripped that out entirely.
They replaced the hardware bottleneck with 2,000 open-source Envoy proxies distributed across 13 distinct AWS regions.
Moving from a rigid hardware constraint to a distributed software proxy allows for localized failovers and massive horizontal scaling.
Deploying 2,000 Envoy instances is easy.
Managing their state is the nightmare.
To orchestrate the fleet, he built a custom control plane dubbed “Sovereign.”
The architecture relies heavily on DynamoDB for distributed state management and configuration synchronization.
For the execution layer, they deployed Rust sidecars to handle the heavy processing without ballooning the memory footprint.
To shield the entire surface area, they placed AWS CloudFront at the edge for global caching and aggressive DDoS protection.
The code for the Sovereign control plane is now sitting publicly on Bitbucket.
The actual code is only a fraction of the friction.
Syrakis spent a significant portion of the teardown discussing the maintenance reality and the team diplomacy required to execute a migration of this scale.
Ripping out foundational infrastructure while Jira and Confluence are actively serving millions of users requires navigating intense internal politics.
The massive pivot toward generative models is causing structural tension, forcing companies to reallocate headcount away from core infrastructure and into new features.
The takeaway, tbh is brutal.
The code you write belongs to the company, only the system design and the deployment scars belong to you.
Source: http://youtube.com/watch?v=55pTFVoclvE
In case we are meeting for the first time, come over here, it’ll be worth the roller coaster of articles that are gonna come up in the next few weeks.
I swear tracking these updates is a job in itself, lately.
Here’s the list which I’ve built and keep adding on.
And If you need help for analyzing UFC fights, please check out BoutPredict :)