🔴 2026.5.22 Regression: Provider Auth Pre-warm blocks event loop (310s), Gateway becomes unusable

Reporter: 小龙 (OpenClaw user)
Date: 2026-05-25
Environment: macOS 26.4.1 (MacBook Pro, 16GB RAM, x64), Node.js 24.15.0
Upgrade path: 2026.5.20 → 2026.5.22 → downgraded to 2026.5.20

Problem Description

After upgrading to 2026.5.22, the Gateway's Provider Auth pre-warm mechanism has a severe performance regression. Pre-warm time jumped from milliseconds (normal) to 179-310 seconds, and the entire Node.js event loop is blocked during this process (eventLoopMax reaching 64 seconds), causing all WebSocket requests to queue up.

Reproduction Conditions

• System: macOS (MacBook Pro)
• Node.js 24.15.0
• Multiple Model Providers configured (bailian, xai, openai, ymhss)
• Upgrade: npm install -g [email protected] + Gateway auto-reload

Impact

Key Logs

# First startup after upgrade (SIGUSR1 drain + auto-reload)
2026-05-25T03:58:35.616+08:00 [gateway] provider auth state pre-warmed in 310057ms eventLoopMax=64458.1ms

# After hard restart (kill + fresh start) — issue persists
2026-05-25T04:25:08.080+08:00 [gateway] provider auth state pre-warmed in 179363ms eventLoopMax=43251.7ms

# Memory pressure warning
2026-05-25T03:55:29.702+08:00 [diagnostics/memory] memory pressure: level=warning reason=heap_threshold rssBytes=1390784512 heapUsedBytes=1091051424 thresholdBytes=1073741824

Analysis

1. The issue reproduces on both cold start and hot reload — not a transient state
2. Affects all users with multiple Provider configurations
3. The auth pre-warm logic appears to be a breaking change between 5.20 and 5.22
4. Likely related to a change from lightweight token verification to full model list enumeration

Suggested Fixes

Short-term

1. Publish a known issues notice for 2026.5.22, recommend users hold upgrade
2. Add openclaw rollback <version> command that auto-handles version markers
3. Add provider auth pre-warm time check to openclaw doctor

Long-term

1. Provider auth pre-warm should be async — must not block the event loop
2. Add pre-warm timeout mechanism (>30s should be treated as abnormal)
3. Automated downgrade support
4. Add openclaw upgrade --dry-run pre-check before actual upgrade

Appendix: Environment Details

• OS: macOS 26.4.1 (MacBook Pro, x64)
• RAM: 16GB
• Node.js: 24.15.0
• Upgraded from: 2026.5.20
• Upgraded to: 2026.5.22
• Current: 2026.5.20 (downgraded)
• Model Providers: bailian (qwen3.6-plus), xai (grok), openai (gpt-5.x), ymhss (claude-opus-4-6-thinking)
• Plugins: 12 (acpx, bonjour, browser, canvas, codex, device-pair, file-transfer, memory-core, ocnmps-router, phone-control, talk-voice, telegram)