在做任何读（get、list）操作时，不能假设读到 的是最新版本，也不能假设一次 reconcile 中始 终会读到同一个版本 避免写后读 同一个 controller，在一次 reconcile 中，避免写 （create、update、delete）完一个对象后马上去 读（get、list）最新版本，等controller-runtime 触发下一次 reconcile 遵循惯例开发模式 即 controller 用读 stale 对象。 如下图所示，某个版本为3（gen=3）的集群（Cluster）中有一个 stale 对象，即 StatefulSet (gen=2)。 Controller-runtime 的 Reconcile 过程是一个 EDA 模型，当该 stale 对象的更新到达 cache 时，controller-runtime 会发送一个事件（Event）给到 owner controller（也就是我们的 controller-type OwnerReference of the given type. // Use Owns(object, builder.MatchEveryOwner) to reconcile all owners. // // By default, this is the equivalent of calling // Watches(object, handler.Enqu

Message Queue Dynamic Quota Operator Cluster _1 Cluster_1 Quota for Biz A etcd DynamicQuota reconcile Prometheus +Grafana Commit workload and DynamicQuota, PriorityClass, Workload Annotations Dynamic adjustment of offline jobs quota. Ø Offline job queue manager. Ø DynamicQuota-Operator to reconcile business quota. Ø ValidatingWebhook to validate pod add request. DynamicQuota Large-scale and

deleted Resource Item Resource status Resource Spec: running list/watch resource change Reconcile resource status to spec © 2017 Rancher Labs, Inc. API Aggregation • What API aggregation provides

control loop container VIP volume W I S B : W h a t i t s h o u l d b e Converge & Reconcile WIRI: What it really is Kubernetes Core concept of Kubernetes - Declarative magic What is an

nivenly.com” So how does it work ? ? ? Kops 1.4 State Store Unmarshal Tasks Map Compare Reconcile Kops 1.4 tasks := []*Task We have to track our work Where do tasks come from ? ? ? Kops

Oriented Programming Core of API “OO” 1.API objects stores in etcd 2.Control loops (Sync Loop) to reconcile API objects Example kubelet SyncLoop kubelet SyncLoop proxy proxy 1 Pod created etcd scheduler