Summary

Provider auth pre-warming blocks the Node.js event loop for 60-90 seconds during gateway startup, causing all concurrent I/O operations (HTTP API calls, MCP server startup, WebSocket connections) to time out. Affects all tested remote model providers.

Environment

• OS: Windows 10 Pro 22H2 (19045)
• OpenClaw: 2026.5.22 (a374c3a)
• Node.js: v24.14.1
• Shell: bash (via VS Code)

Reproduction Steps

1. Start OpenClaw gateway with any remote model provider (e.g., DashScope, CTYun).
2. Observe logs during the first 2 minutes of startup.

Reproduction rate: 100% across multiple restarts and model switches.

Tested Providers (All Affected)

Model APIs respond quickly via direct curl/PowerShell (0.2-1.0s). The issue is in OpenClaw's auth pre-warming, not upstream APIs.

Cascade Failure Chain

Gateway Start
  |
  └─ Auth Pre-Warming (event loop blocked 60-90s)
       ├─ [TIMEOUT] Feishu tenant_access_token API (30s)
       ├─ [TIMEOUT] Feishu bot identity ping (30s, 5 retries over 15 min)
       ├─ [TIMEOUT] MCP server startup (30s, no tools for agent)
       ├─ [DELAY] Health check: 3-22s (normal <0.01s)
       ├─ [DELAY] Control UI: agents.list 14-16s, models.list 16-23s
       ├─ [DELAY] Message response: 80-126s (normal <10s)
       └─ [CPU] 97% utilization, 666MB memory for gateway process

Key Log Evidence

provider auth state pre-warmed in 90402ms eventLoopMax=70732.7ms
liveness warning: reasons=event_loop_delay,event_loop_utilization,cpu 
   eventLoopDelayP99Ms=22968 eventLoopUtilization=0.973 cpuCoreRatio=0.967

feishu[default]: bot info probe timed out after 30000ms; continuing startup
   (repeats 5 times over 15 minutes before bot identity resolves)

failed to start server "windows-automation" (...): MCP server connection timed out after 30000ms

stalled session: age=162s classification=stalled_agent_run

Feishu APIs (tenant_access_token, bot ping) respond in ~0.2s when tested directly with curl, but time out (30s) when called from the Node.js process during auth pre-warming.

Message Response Timeline (Real Example)

Per-message overhead (post-auth-warming):

• tool-policy: 2.3s
• image-tool: 1.2s
• plugin-tools: 1.6s
• system-prompt: 3.3-5.1s
• session-resource-loader: 4.1-7.2s

Total overhead before model call: 15-17s

Workaround

• Wait 2-3 minutes after gateway start before sending messages.
• Feishu bot identity may need 5-15 minutes to recover via background retry.

Expected Behavior

Auth pre-warming should be non-blocking (async) or use worker threads to avoid starving the event loop. A 60-90s synchronous block in the main event loop makes the gateway unusable during startup and degrades reliability permanently.

Affected Components

Additional Context

• Issue persists across multiple restarts and different model providers.
• No error logs from the model APIs themselves; the calls succeed once the event loop becomes free.
• The problem appears to be architectural: the gateway's startup sequence does not follow Node.js non-blocking design principles.