Catalogue / uber-2024-kubernetes-compute-migration

Uber migrates 3 million cores of shared stateless workloads from Apache Mesos to large Kubernetes clusters

Uber, 2024. Shared stateless container orchestration platform (Kubernetes clusters under Up federation).

Organisation
Uber
Evidence
Reported from production
Source
Migrating Uber’s Compute Platform to Kubernetes: A Technical Journey

Uber Engineering Blog, published 2025-04-24

Records

Decision#d1

Migrate 3 million cores from deprecated Mesos to Kubernetes transparently via Up cluster selection

Decision#d2

Operate large 5,000-7,500-node Kubernetes clusters and tune API server, scheduler, and UI caching for scale

Decision#d3

Add sidecar artifact uploads, batched gradual scaling, and CloneSet in-place updates with image prefetching

Decision#d4

Trigger automatic rollbacks from container-restart heuristics instead of progress deadline timeouts

Failure#f1

Missed deployment events across controller leadership changes delayed rollouts by up to 10 hours

Sign in to read what was chosen, what was rejected and what it cost, each with its place in the source. An account is free.