Configure node autonomy
In a cloud-edge environment, network disconnections between the cloud and edge are common. OpenYurt provides multiple levels of node autonomy capabilities to ensure that edge services remain stable and operational during such disconnections.
This document describes three key autonomy features:
1. Self-Healing from Restarts During Disconnection
Background: In native Kubernetes, if an edge node or its kubelet component restarts while disconnected from the cloud, it cannot recover the running pods because their definitions are stored in memory and cannot be fetched from the API server.
OpenYurt's Solution: OpenYurt's core edge component, YurtHub, solves this problem. YurtHub acts as a local proxy and cache for all API server requests on the node. When the node is online, it caches necessary data (like Pod specs). If the node or kubelet restarts during a network disconnection, kubelet can retrieve the required data from YurtHub's local cache, enabling it to successfully recover and restart the business containers.
How to Use:
This self-healing capability is enabled by default on any edge node where YurtHub is running. Nodes joined to the cluster using yurtadm join will have YurtHub installed automatically.
2. Preventing Pod Eviction During Disconnection
Background: Native Kubernetes will evict (delete and reschedule) pods from a node if the kubelet fails to report its heartbeat for a certain period (e.g., 5 minutes). In edge scenarios, this is undesirable as network disconnections are common and do not necessarily mean the node is faulty.
OpenYurt provides two mechanisms to prevent unwanted pod evictions.
2.1 Per-Node Autonomy Duration
This is the recommended approach for most edge nodes. It allows you to configure a specific eviction timeout for each edge node, overriding the cluster's global timeout.
How to Use:
To enable this feature, add the node.openyurt.io/autonomy-duration annotation to your edge node.
# First, ensure the node is labeled as an edge node
kubectl label node my-edge-node openyurt.io/is-edge-worker=true
# Recommended: Set a 24-hour autonomy duration
# This means the control plane will wait 24 hours before evicting pods from this node.
kubectl annotate node my-edge-node node.openyurt.io/autonomy-duration="24h"
# To achieve indefinite autonomy (pods will never be evicted due to heartbeat loss)
kubectl annotate node my-edge-node node.openyurt.io/autonomy-duration="0"
The duration format can be found here.