Summary

• Problem: When one Telegram message handler takes a long time (e.g., a 10-minute agent timeout), ALL other messages are blocked until it completes
• Root Cause: grammy's default bot.start() processes updates sequentially
• Fix: Use @grammyjs/runner to process updates concurrently

The Problem

When using grammy's default long polling with bot.start(), updates are processed one at a time:

// From grammy source:
// handle updates sequentially (!)
await this.handleUpdate(update);

This means if one message handler hangs (e.g., waiting for an LLM response that times out after 10 minutes), every other message in every other chat waits in a queue behind it. The entire bot becomes unresponsive.

The Solution

The @grammyjs/runner plugin is specifically designed to solve this. It processes updates concurrently, so multiple conversations can be handled in parallel. A slow or stuck handler in one chat doesn't block handlers in other chats.

Changes

1. Added @grammyjs/runner dependency
2. Replaced bot.start() with run(bot) in monitor.ts
3. Updated abort signal handling to work with the runner
4. Added tests verifying the runner is used and responds to abort signals

Test plan

• All existing telegram tests pass (66 tests)
• New tests verify runner is invoked for long polling
• New tests verify abort signal stops the runner
• Manual testing: send multiple messages while one is processing slowly

🤖 Generated with Claude Code