openclaw - 💡(How to fix) Fix [Bug]: Telegram channel: fetch-timeout 10000ms on sendChatAction/sendMessage (undici, Windows, Node 24) [2 comments, 3 participants]

Bug type

Regression (worked before, now fails)

Beta release blocker

No

Summary

On Windows with Node.js 24, the OpenClaw Telegram channel fails to deliver agent replies because undici HTTP requests to api.telegram.org exceed the hardcoded 10000ms fetch timeout, even though the Telegram API itself responds in <200ms when called via native https.

Steps to reproduce

1. Run OpenClaw 2026.4.29 gateway on Windows 11 with Telegram channel configured
2. Send any message to the bot from Telegram
3. Agent processes message (typing indicator appears) but reply never arrives
4. In logs observe: {"subsystem":"fetch-timeout","timeoutMs":10000,"elapsedMs":21675, "url":"https://api.telegram.org/bot.../getMe"} {"subsystem":"telegram/api","message":"telegram sendChatAction failed: Network request for 'sendChatAction' failed!"}

Expected behavior

Agent reply is delivered to Telegram after model generates response.

Actual behavior

fetch-timeout fires at 10000ms. elapsedMs=21675. sendMessage/sendChatAction fail silently. No reply delivered. Native https.request to same URL: 118ms. Only undici is affected.

OpenClaw version

OpenClaw: 2026.4.29

Operating system

Windows 11

Install method

npm global

Model

ollama/qwen3:4b (primary) / ollama/phi4-reasoning:plus (fallback)

Provider / routing chain

openclaw -> ollama (local, 127.0.0.1:11434) -> qwen3:4b

Additional provider/model setup details

Local Ollama instance on Windows 11. No cloud provider. Config: ~/.openclaw-sentinel/openclaw.json models.providers.ollama.baseUrl = http://127.0.0.1:11434 models.providers.ollama.api = ollama models.providers.ollama.timeoutSeconds = 300 GPU: AMD Radeon 8060S (Strix Halo) — 64GB UMA VRAM Model runs 100% GPU. Generation is not the bottleneck.

Logs, screenshots, and evidence

Key log lines from C:\tmp\openclaw\openclaw-2026-05-02.log:

{"subsystem":"fetch-timeout","timeoutMs":10000,"elapsedMs":21675,
 "operation":"fetchWithTimeout",
 "url":"https://api.telegram.org/bot.../getMe",
 "message":"fetch timeout reached; aborting operation"}

{"subsystem":"telegram/api",
 "message":"telegram sendChatAction failed: Network request for 'sendChatAction' failed!"}

{"subsystem":"agent/embedded","message":"[trace:embedded-run] prep stages:
 totalMs=50284 stages=core-plugin-tools:21825ms,system-prompt:11549ms,stream-setup:12066ms"}

DIAGNOSTIC: Native Node.js https.request to api.telegram.org = 118ms.
undici (used by OpenClaw) = 21675ms on same endpoint.
Bug is isolated to undici fetch layer, not network or Telegram API.

Impact and severity

Affected: ALL Windows users using Telegram channel with local Ollama provider. Severity: CRITICAL — Telegram channel completely non-functional on Windows. Frequency: Always (100% reproducible). Consequence: Agent generates reply but sendMessage/sendChatAction never delivered. Typing indicator appears in Telegram but no reply arrives. Workaround: None within OpenClaw. Requires external bot implementation.

Additional information

Root cause hypothesis

undici 8.1.0 (bundled with OpenClaw 2026.4.29) appears to hang on HTTPS connections to api.telegram.org on Windows / Node.js 24.15.0.

Native Node.js https.request to the same URL: 118ms ✅ undici fetch to the same URL: 21675ms (timeout) ❌

Possible causes:

1. undici attempting HTTP/2 (ALPN negotiation) which Telegram API rejects, causing fallback delay > 10s hardcoded timeout
2. undici connection pool deadlock on Windows with Node 24 + undici 8.1.0
3. TLS session resumption failure specific to api.telegram.org on Windows

What was tried

• NODE_OPTIONS=--dns-result-order=ipv4first → no effect
• UNDICI_NO_ORIGIN=1 → no effect
• DNS resolves correctly in 73ms (IPv4: 149.154.166.110)
• Telegram API token is valid (PowerShell Invoke-RestMethod: 200ms)

Secondary issue: scope upgrade loop

On every gateway restart, telegram/native-approvals subsystem generates a new device pairing request (new UUID each time). Gateway cannot self-approve, causing PAIRING_REQUIRED loop that eventually kills the process.

Log evidence: {"subsystem":"telegram/native-approvals", "message":"connect error: scope upgrade pending approval (requestId: b7aef8f7-9b26-4171-8290-105298f9b4cf)"} {"message":"gateway closed (1008): pairing required: device is asking for more scopes than currently approved"}

System specs

• CPU: AMD Ryzen AI Max+ 395 (Strix Halo)
• GPU: AMD Radeon 8060S — 64GB UMA VRAM
• RAM: 128GB LPDDR5X 8000 MT/s
• OS: Windows 11
• Node.js: 24.15.0
• OpenClaw: 2026.4.29 (a448042)
• undici: 8.1.0
• grammy: 1.42.0

Suggested fix

Increase fetch timeout for Telegram API calls from 10000ms to ≥ 60000ms, or use native Node.js https instead of undici for Telegram channel requests.