The Divided States of Cloud Computing

As teams move more workloads to the cloud, it’s easy to celebrate how disposable our infrastructure has become: a network, a cluster, a fleet of servers can be torn down and rebuilt from code in minutes. But there is one thing infrastructure as code can never bring back, and that’s the state inside the systems we run. The rows in a database, the messages on a queue, the volumes behind a stateful workload, the objects in a bucket (if you don’t invest time in a versioning strategy): this is the state you genuinely cannot recreate, and it’s divided across more places than most teams realise.

In this talk, I’ll learnings around how to manage and protect application state on AWS, the part of the system that survives long after the code that created it.

We’ll look at where your state actually lives, and how to protect each kind, including:

• Relational and transactional data in managed database clusters such as Aurora and RDS
• Streaming and messaging state in systems like Kafka, and what “backup” even means there
• Persistent data behind Kubernetes workloads, living on block and file storage long after pods come and go
• Objects and media in object storage, and the difference between versioning, replication, and a true offsite copy
• Why “managed by Terraform” is not the same as “safe”, and how to protect a resource’s data independently of the code that provisioned it

This session focuses on practical implementation, offering insights into what worked well, what didn’t, and how technical expectations can sometimes differ from operational reality. Whether you’re just starting to run stateful workloads on AWS, or already operating them at scale across regions, this talk will provide tangible strategies, architectural patterns, and hard-earned lessons to protect the data your infrastructure can’t bring back.