OpenClaw Repo: Backpressure Overhaul, Scheduler Telemetry, and a Redis Pool Leak

Tools

Merged: Unified Backpressure Strategy Across Core

This PR lands a single, consistent approach to handling load backpressure. Before this, different components had their own ad-hoc solutions, which made behavior unpredictable under stress. For builders, this means more reliable performance when your instance is getting hammered.

Fix: Redis Connection Pool Leak Under High Churn

A fix for a sneaky bug where Redis connections weren't being properly returned to the pool under specific error conditions. Left unchecked, this would slowly starve your instance of connections. It's the kind of silent leak that eats your uptime on a weekend.

Analysis

Proposal: Embed Scheduler Telemetry for Runtime Insights

A proposal to expose detailed scheduler metrics directly from the runtime. This is about giving you the data to understand why task scheduling is slow or uneven, without guesswork. It's a big deal for anyone trying to debug performance bottlenecks in production.

Discussion: Context Cancellation Propagation in Long-Running Hooks

An issue highlighting that cancellation signals don't always propagate cleanly through deep chains of async hooks. This can leave orphaned processes or hold locks longer than necessary. It's a deep reliability issue that's finally getting proper attention.

News

Hot Path: Plugin Lock Ordering Fix for Concurrent Access

A fix for a race condition where two plugins could acquire locks in different orders, potentially leading to a deadlock. This is the kind of hard-won concurrency bug that only shows up under real load. A critical stability patch you want in your next build.