AGENT SKILL

telegram-delivery-diagnostics

Diagnose and fix Telegram delivery issues for Hermes cron jobs — delivery pipeline tracing, DNS/network diagnostics, adapter state verification, and known failure patterns.

License 许可
MIT
最近更新
2026-11-14
Stars
★ 4
Forks
1

01 前言 Front Matter

name
telegram-delivery-diagnostics
author
Steadfaste Fleet
version
1.0.0
category
devops
platforms
[ "linux", "macos" ]
description
Diagnose and fix Telegram delivery issues for Hermes cron jobs — delivery pipeline tracing, DNS/network diagnostics, adapter state verification, and known failure patterns.

02 资源文件 1

03 SKILL.md 正文

Telegram Delivery Diagnostics

Diagnose why a Hermes cron job's output didn't reach the user on Telegram.

When to use

  • User says "I didn't get my news/brief/alert today"
  • Cron job logs delivered but user didn't see it
  • Gateway.log shows Telegram DNS failures or fallback IP flapping
  • After a network outage or DNS resolver change
  • When adding a new cron with Telegram delivery target

Architecture — How Cron Deliveries Work

Cron job finishes
  → scheduler.py queues delivery via gateway's event loop (future)
  → DeliveryRouter._deliver_to_platform()
  → Telegram adapter.send()
     → self._bot.send_message() (python-telegram-bot)
        → httpx client with TelegramFallbackTransport
           → Primary DNS: api.telegram.org
           → Fallback: 149.154.166.110, 149.154.167.220
     → 3 retry attempts per chunk on NetworkError
  → Returns SendResult(success=True, message_id=...)
  → Scheduler logs: "delivered to telegram:CHAT_ID via live adapter"

Two delivery paths

Path When Mechanism
Live adapter Default for all crons Goes through gateway's adapter (adapter.send()). Returns success/failure.
Standalone Live adapter unavailable or timed out Direct HTTP to Telegram API via _send_to_platform(). Bypasses gateway entirely.

Key configuration points

Layer What to check File
Bot token TELEGRAM_BOT_TOKEN ~/.hermes/.env
Home channel TELEGRAM_HOME_CHANNEL (should match user chat_id) ~/.hermes/.env
Allowed users TELEGRAM_ALLOWED_USERS ~/.hermes/.env
Platform config display.platforms.telegram.streaming: true ~/.hermes/config.yaml
IPv4 preference network.force_ipv4: true ~/.hermes/config.yaml
Fallback IPs TELEGRAM_FALLBACK_IPS env var (optional, overrides defaults) ~/.hermes/.env or config
Cron delivery target Format: telegram:CHAT_ID Cron job definition

Verification checklist

When a user reports missing delivery, run through these in order:

1. Check cron ran successfully

cronjob action='list' --filter <job-name>
# Check: last_status == "ok", last_delivery_error == null

2. Check delivery log in agent.log

grep "<JOB_ID>.*delivered" ~/.hermes/logs/agent.log
# Should say "delivered to telegram:<TELEGRAM_HOME_CHANNEL> via live adapter"

3. Check gateway delivery log

grep -a "<JOB_ID>" ~/.hermes/logs/gateway.log
# Should show "Cron output preserved for chunking adapter" if message > 4000 chars
# NO error after this entry means send was successful

4. Check Telegram API connectivity

# Test directly (bypasses gateway)
TOKEN=$(grep "^TELEGRAM_BOT_TOKEN=" ~/.hermes/.env | cut -d= -f2-)
curl -s --max-time 10 "https://api.telegram.org/bot${TOKEN}/getMe"
# Expect: {"ok":true,"result":{"id":...,"is_bot":true,...}}

# Test sendMessage
curl -s --max-time 10 -X POST "https://api.telegram.org/bot${TOKEN}/sendMessage" \
  -d "chat_id=${TELEGRAM_HOME_CHANNEL}" -d "text=🧪 Diagnostics test"

5. Check DNS resolution

# System DNS
getent hosts api.telegram.org
# Should return an IPv4 address (149.154.166.110)
# If IPv6 (2001:67c:4e8:f004::9) appears, IPv6 may be failing

# Check if force_ipv4 patch is active
python3 -c "
import socket
print('Patched:', getattr(socket.getaddrinfo, '_hermes_ipv4_patched', False))
"

# Full DNS diagnostics
for i in 1 2 3 4 5; do
  dig +short api.telegram.org @192.168.1.1 2>/dev/null
done
# If results are inconsistent → DNS flapping

6. Check gateway Telegram connection state

grep -a "Telegram" ~/.hermes/logs/gateway.log | grep "07-08" | tail -10
# Look for: 
# - "sticky fallback IP" → DNS was failing, using fallback
# - "sticky fallback IP failed" → fallback ALSO failed
# - "polling resumed" → recovered
# - "polling reconnect failed" → still down

Known failure patterns

Pattern A: DNS flapping — systemd-resolved intermittent failures

Symptom: [Errno -2] Name or service not known in errors.log, followed by automatic fallback IP usage. Cron deliveries may succeed or fail depending on timing.

Root cause: System DNS resolver (192.168.1.1 via systemd-resolved) intermittently fails to resolve api.telegram.org. The TelegramFallbackTransport handles this by retrying fallback IPs 149.154.166.110, 149.154.167.220.

Impact:

  • Send path has its OWN transport (separate from polling)
  • Each send gets 3 retries with fallback IPs
  • During periods when BOTH DNS and fallback IPs fail, sends fail
  • The force_ipv4: true config can't help when DNS itself is unreachable

Fix:

  1. (Preferred) Pin IPv4 in /etc/hosts:

    echo "149.154.166.110 api.telegram.org" | sudo tee -a /etc/hosts
    

    This bypasses system DNS entirely for api.telegram.org.

  2. (Already set) Ensure network.force_ipv4: true in ~/.hermes/config.yaml. This monkey-patches socket.getaddrinfo to prefer AF_INET (IPv4) before IPv6.

  3. (Belt-and-suspenders) Set TELEGRAM_FALLBACK_IPS in .env to ensure fallback is always configured.

Pattern B: IPv6 connectivity failures

Symptom: api.telegram.org resolves to BOTH 149.154.166.110 (IPv4) and 2001:67c:4e8:f004::9 (IPv6). ping6 to the IPv6 address fails, but Python's httpx may attempt IPv6 first and time out.

Diagnosis:

ping6 -c1 -W2 2001:67c:4e8:f004::9
# If FAIL → IPv6 is broken on this network
resolvectl query api.telegram.org
# Shows both addresses

Fix: Same as Pattern A — /etc/hosts IPv4 pinning eliminates IPv6 resolution entirely.

Pattern C: Telegram-side delivery gap (API accepts but doesn't deliver)

Symptom: All system logs show successful delivery (adapter.send() returned success, message_ids received), but user reports messages never arrived. This is rare but documented on Telegram's side.

Diagnosis: Only diagnosable by exclusion — if all system-side checks pass and the user confirms a test message sent NOW arrives, the earlier messages were likely lost in a Telegram-side transient delivery gap.

Fix: No server-side fix. If it happens, note the time and check Telegram status pages.

Pattern D: Adapter not initialized for cron delivery

Symptom: send_result is None or "Not connected", cron falls through to standalone path.

Diagnosis: Check gateway.log for Telegram adapter initialization errors:

grep -a "Not connected\|send_path_degraded\|adapter.*failed" ~/.hermes/logs/gateway.log

Fix: Restart the gateway: hermes restart.

Pattern E: Thread/topic delivery goes to wrong lane

Symptom: Private DM topic deliveries end up in the General topic instead of the named thread.

When it happens: Cron deliveries with attach_to_session: true create named private DM topics. If the topic creation fails or the thread_id is wrong, messages fall through to the wrong lane.

Fix: Check thread_id in the cron's origin metadata. For regular DM deliveries (no thread), ensure deliver: telegram:CHAT_ID has no thread_id component.

Log reference

Where to look

Log Path What it tells you
Cron scheduler ~/.hermes/logs/agent.log Job ran, delivered status, delivery errors
Gateway delivery ~/.hermes/logs/gateway.log Output received, adapter.send() called, Telegram connectivity
Errors ~/.hermes/logs/errors.log DNS failures, Telegram network errors, adapter connection issues

Key log patterns

Successful delivery (normal):

INFO gateway.delivery: Cron output preserved for chunking adapter (8050 chars)
INFO cron.scheduler: Job 'JOB_ID': delivered to telegram:<TELEGRAM_HOME_CHANNEL> via live adapter

Successful delivery (small message — under 4000 chars, no chunking):

INFO cron.scheduler: Job 'JOB_ID': delivered to telegram:<TELEGRAM_HOME_CHANNEL> via live adapter

(No "Cron output preserved" log — message was small enough to send directly.)

DNS failure with fallback recovery:

WARNING plugins.platforms.telegram.telegram_network: Primary api.telegram.org connection failed
WARNING plugins.platforms.telegram.telegram_network: ... path unreachable; using sticky fallback IP 149.154.166.110

Total connectivity loss (both DNS and fallback fail):

WARNING plugins.platforms.telegram.telegram_network: Primary api.telegram.org connection failed
WARNING plugins.platforms.telegram.telegram_network: Fallback IP 149.154.166.110 failed
WARNING plugins.platforms.telegram.telegram_network: Sticky fallback IP ... failed; resetting to primary DNS path

Delivery timeout — assumed delivered (in flight):

WARNING cron.scheduler: Job 'JOB_ID': live adapter send to telegram:CHAT_ID timed out after 60s; already dispatched, assuming delivered

Delivery timeout — standalone fallback:

WARNING cron.scheduler: Job 'JOB_ID': live adapter send ... timed out before the coroutine was dispatched, falling back to standalone

Live adapter routine responses to user-initiated messages:

INFO gateway.platforms.base: [Telegram] Sending response (N chars) to CHAT_ID
INFO hermes_plugins.telegram_platform.adapter: [Telegram] Flushing text batch agent:main:telegram:dm:CHAT_ID (N chars)

Pitfalls

  • "Delivered via live adapter" ≠ proof the user saw it. The log only proves Telegram's API returned a 200. Telegram can accept messages and fail to deliver them to the user's client.
  • gateway.log is a binary file (contains control characters). Always use grep -a when searching it.
  • force_ipv4 doesn't fix all DNS issues. The monkey-patch prefers IPv4 but can't help when the DNS server itself is unreachable ([Errno -2]). Use /etc/hosts pinning for that case.
  • Cron deliveries use a different log path than interactive responses. Interactive messages show Sending response in gateway.log. Cron deliveries go through adapter.send() directly and don't produce that log line. Don't assume a missing "Sending response" means failure.
  • splits_long_messages = True means long crons are chunked. The chunking adapter (splits_long_messages=True) means messages over 4000 chars are split. Each chunk gets its own send_message API call. If one chunk fails after all 3 retries, the ENTIRE send fails.
  • Fallback IPs can also fail. The seed fallback IPs (149.154.166.110, 149.154.167.220) are stable Telegram Bot API endpoints but can be unreachable from certain geographic regions during network events.
  • Two transports, one adapter. The Telegram adapter creates TWO TelegramFallbackTransport instances — one for requests (sending) and one for get_updates (polling). Each has its own primary transport, fallback transports, and sticky IP. The send path's transport is independent of the polling path's connectivity.

04 同仓库更多 Skills