Bot stays alive but never polls after transient getWebhookInfo failure at startup (zombie process, systemd shows running)

[insign]

Hi, first of all thank you for this great project!

I ran into a case where the bot process looks healthy but never actually polls Telegram, so it stays silent until manually restarted. Sharing details below in case it helps.

Environment

• Package: @grinev/opencode-telegram-bot v0.25.0 (installed via npm)
• Node.js v24.18.0 on linux x64
• Running as systemd system service (opencode-telegram-bot.service, Type=simple, Restart=on-failure, ExecStart=/usr/local/bin/node .../opencode-telegram start)
• OpenCode server on port 4096 (user unit, healthy)

Symptoms

• systemctl status opencode-telegram-bot showed active (running), PID alive, memory/CPU normal.
• Bot did not respond to any Telegram messages (/sessions, /projects, /start stayed as pending updates, pending_update_count grew to 6).
• Manual getUpdates against the same token succeeded with normal updates (no 409 Conflict), proving no active long-poller was running despite the alive process.
• After systemctl restart, the bot logged Bot @OChelioBOT started! and manual getUpdates correctly returned 409 Conflict (single healthy poller). Token and network to api.telegram.org were fine the whole time.

Logs (startup that got stuck, IDs redacted)

[INFO] Starting OpenCode Telegram Bot v0.25.0...
[INFO] Allowed User ID: [REDACTED]
[INFO] [OpenCodeAutoRestart] Enabled: port=4096, intervalSec=300
[INFO] [OpenCodeReady] OpenCode server is ready: reason=auto_restart_startup
[INFO] [SessionCache] Pruned 3 stale directories from cache
[INFO] [PinnedManager] Restoring existing pinned message for session: ses_[REDACTED]
[INFO] [PinnedManager] Loaded 41 file diffs from messages
[INFO] [PinnedManager] Loaded context from history: 392732 tokens, cost: $2.35
[INFO] [Bot] Subscribing to OpenCode events for project: /home/.../codewalk
[INFO] [Attach] Attached to session: session=ses_[REDACTED], directory=/home/.../codewalk
[INFO] [Bot] Restored followed session after OpenCode ready: reason=auto_restart_startup
CLI error: Network request for 'getWebhookInfo' failed!

Note what is missing compared to a healthy startup: no [Bot] Dropping ~N update(s) queued while the bot was offline and no Bot @OChelioBOT started!. On healthy boots those two lines appear right after PinnedManager Restoring.... Earlier boots on the same machine also showed transient Could not clear global commands: HttpError: Network request for 'setMyCommands' failed! right before the same stuck state, so it looks like a transient Telegram API hiccup at startup is enough to trigger it.

Suspected cause (from installed dist/, may have shifted in source)

• dist/app/bootstrap/start-bot-app.js:126: const webhookInfo = await bot.api.getWebhookInfo(); runs before bot.start() with no surrounding try/catch. If it throws, startBotApp() rejects.
• dist/cli.js:192-203: the top-level runCli(...).catch(...) only writes CLI error: ... to stderr and sets process.exitCode = 1. It does not force exit.
• Meanwhile background work (opencodeAutoRestartService.start(), scheduledTaskRuntime, heartbeat setInterval in dist/bot/index.js:85, PinnedManager restore via onReady) keeps the event loop alive, so the process never exits.
• Net effect: bot.start({ drop_pending_updates: true, ... }) at start-bot-app.js:138 is never reached, no polling starts, but systemd still sees active (running) and Restart=on-failure never fires. A zombie that looks healthy.

I did not dig into the TypeScript sources, so paths above are from dist/ — happy to re-check against source if you point me at the right files.

Expected vs actual

• Expected: a Telegram API failure during startup either retries and then starts polling, or fails fast with process exit so the supervisor restarts it.
• Actual: process stays alive with exit code pending but no polling, requiring manual systemctl restart.

Possible fixes (just ideas)

• Wrap the pre-bot.start() Telegram calls (getWebhookInfo, deleteWebhook if needed) with retry + fail-fast process.exit(1) (after flushing logs/settings and stopping background services/timers) so Restart=on-failure works.
• Or ensure startup failure always tears down heartbeatTimer, opencodeAutoRestartService, scheduledTaskRuntime, and listeners before rethrowing, so Node can actually exit when exitCode is set.
• Optionally log a clear FATAL: Telegram polling never started line to make the zombie state obvious.

Workaround for others hitting this

sudo systemctl restart opencode-telegram-bot, then confirm Bot @... started! in journalctl -u opencode-telegram-bot and that a manual getUpdates returns 409 Conflict (healthy single poller).

Happy to provide more logs or test a fix. Thanks again for maintaining this!