Lightning Talk: Kubernetes Jobs and the Sidecar Problem
summary
Spotify migrates many stateless services to GKE and hits a concrete problem: Kubernetes Jobs finish their main container while a long‑running sidecar keeps the pod — and thus the Job — from completing. Jobs have been GA ~4.5 years and a prominent GitHub issue is three years old. Current production workaround uses a shared-volume sentinel file where the job writes completion and the sidecar exits — a hack around a metrics relay on port 19000 example. A Kubernetes Enhancement Proposal (KEP) proof-of-concept exists and needs feedback.
AI-written summary · about
chapters (7)
- 0:00Start
- 0:21Kubernetes at Spotify
- 1:25Jobs + Sidecar Problem
- 2:02Sidecars at Spotify
- 2:45Sidecar Problem at Spotify
- 3:02Workaround: Shared Volume Communication
- 3:44Call to Action
related talks
29mLightning Talk: Evolving the Job API to Become Defacto Standard for Batch Worklo...
06mLightning Talk: OpenKruise: Advancing Kubernetes Workload Auto...
27mPragmatic Pod Patterns: Leveraging sidecar containers in Kubernetes
29mThe Great Sidecar Debate
33mMagic of Kubernetes Sidecar Injection: Present and Future - Anton Ovchinnikov
07mLightning Talk: My Job Says 'Running' but Nothing's Running: Kubernetes Status Reality...
10mLightning Talk
25mRelease the Kraken: Bring Sidecar Containers to Next Level