Cluster Reconciliation: Managing Resources Across Multiple Clusters
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Friday, 2020-11-20· at 17:05· Intrado Virtual Event Platform,·
summary
Proposes a concrete design: a cluster inventory API that declares the exact YAML desired per cluster and a cluster reconciler that creates, updates, and garbage-collects those objects; prototype implements a minimal inventory plus reconcilers and exposes painful edge cases (immutable fields, finalizers, defaulted/immutable Service fields, server‑side apply conflicts). Notes an existing but lightly adopted cluster registry for credentials and status. Argues scheduling requires a separate workload API and scheduler and that most work succeeds only after platform‑specific intent metadata is added.
AI-written summary · about
chapters (19)
- 0:00Intro
- 0:57Why Multicluster
- 3:13What kinds of Things
- 3:56What Problems Are We Trying to Solve
- 4:44What Does Failure Look Like?
- 6:48The Kubernetes Object Model
- 8:14Kubernetes Verbs
- 10:05(Serverside) Apply
- 12:20A Service in Kubernetes
- 16:43Reconciliation Challenges
- 18:48Evolution of Cluster Management with Scale
- 19:55Hardcoded Clusters
- 20:51Cluster Groups
- 22:37Cluster Scheduling
- 26:15The Cluster Inventory API
- 26:39The Cluster Reconciler
- 29:24The Cluster Registry
- 30:14The Workload API
- 31:06The Cluster Scheduler
related talks
38mMulticluster Kubernetes, today and tomorrow
29mClusterless - Utilizing a Cluster of Clusters Seamlessly
31mIt's all about reconciliation - anatomy of a kubernetes controller
34mPatterns for building Highly Available Cluster Schedulers
41mIntroducing
31mConnecting Dots: Unified Hybrid Multi-Cluster Auth Experience With SPIFFE and Cluster Inventory API
29mMulti-tenancy vs. Multi-cluster: When Shou...
36mCluster API Deep Dive