Dima Kan

What does it take to run thousands of OpenSearch clusters across eight clouds, from AWS, Google Cloud, and Azure to Oracle Cloud, DigitalOcean, OVHcloud, Exoscale, and UpCloud, and keep the on-call rotation quiet? At Aiven we operate open source data services as a product, and the answer looks a lot like the internal platform you are probably building: one control plane, declarative service definitions, and automation that handles the failures so humans don't have to.

This talk walks through how our platform actually works, using managed OpenSearch as the case study: how node failures, version upgrades, and backups run without operator intervention, how we roll changes across a fleet this large without breaking anyone's production, and which golden paths we expose to users instead of raw infrastructure knobs.

Then we look at where this model goes next. We are extending the same platform into a new runtime for data services and a stack for running AI agents, because agents turn out to be platform workloads too: they need state, observability, and guardrails, just like databases do.