I work on container infrastructure at Google. Most of that work goes upstream, into containerd and Kubernetes.

I'm a reviewer for containerd, and took on the security advisor role there during the wave of AI-generated vulnerability reports. Most of my week goes to triaging what comes in and patching what holds up.

I led GKE's migration to containerd 2.0, which removed support for the CRI v1alpha2 API and Docker Schema 1 images. I built the detection path from node to customer console: every cluster reaching Kubernetes 1.33 passes through it, and the ones still using those features have their upgrades paused until they migrate off. For customers who couldn't tell which workload was still calling the deprecated API, I wrote an eBPF tracer that resolves the calls back to the container, which also flagged widely used observability and security agents needing updates. The hardest part was coordination: getting every GKE service still on the old API released before the cutoff.

I co-authored writable cgroups, which lets AI/ML frameworks like Ray set resource limits on their child processes without running as privileged. Earlier, in Kubernetes SIG-Storage, I authored the ReadWriteOncePod access mode (a volume restricted to one pod, stable in 1.29).

My wife and I compete in ballroom dance.

I'm in the Bay Area. Happy to talk containerd or Kubernetes.