OpenClaw Signals: Worker Rebalancing, Health Probes, and the Resilience Layer

Tools

Worker Load Rebalancing for Multi-Node Clusters

This PR introduces automatic task redistribution when worker nodes become unevenly loaded. If you're running OpenClaw at scale, this is the kind of infra that prevents cascading slowdowns without manual intervention.

Liveness and Readiness Probe Endpoints

Adds standard health check endpoints that orchestrators like Kubernetes expect out of the box. Simple but critical for anyone deploying OpenClaw in containerized environments.

Dedicated Resilience Layer with Configurable Retry Strategies

Extracts retry logic into its own module with pluggable backoff strategies. This is the kind of modular refactoring that makes the stack easier to reason about when things break at 3am.

Graceful Shutdown Hooks for Task Draining

Ensures in-flight tasks complete before a worker exits, preventing data loss during deployments. Every serious task runner needs this, and it's surprising it took this long to land.

Analysis

Discussion: Standardizing Error Classification Across Plugins

A proposal to define a shared error taxonomy so plugins can signal transient vs permanent failures consistently. This would unlock smarter retry decisions across the whole ecosystem.

Debate: Should Workers Self-Report Capacity Metrics?

A thread exploring whether workers should expose their own resource utilization to the scheduler. This directly connects to the rebalancing work and could make scheduling decisions way more intelligent.