hmo
cb334ddd54
fix(bot+auto_heal): stop auto_heal from killing bot during slow LLM calls
...
Root cause of 'no response': auto_heal restart_zhiwei_bot fired whenever
inbound>0 + outbound=0 + errors>0 — which is exactly the normal state while
the bot waits for a slow LLM call (agent tool-use turns take 1-10 min).
Restart killed the in-flight LLM call -> user never got reply -> next cron
cycle saw same state -> restart again. Death loop (fired 22:30,22:35,22:40,
22:45,22:50,23:35).
Fixes:
- CALL_HERMES_TIMEOUT 180s -> 600s (agent tool calls need minutes)
- health(): parse last_inbound/outbound timestamps from journal
- auto_heal: restart bot ONLY if last inbound >600s old with no outbound
since (truly stuck), plus 600s bot-restart cooldown
- slow-but-normal state now logs bot_busy_not_stuck instead of killing
2026-07-19 23:43:28 +08:00
hmo
e0f47f9349
fix(xmpp_monitor): exclude key7(kimi) from auto-heal switch candidates
...
zhiwei must stay on ocg-key6 (deepseek-v4-flash) per user decision.
key7 is kimi — different model provider, incompatible model name.
If best_key returns key7, auto_heal logs skip_switch instead of switching.
2026-07-19 22:09:15 +08:00
hmo
9509b80d8b
feat(xmpp_monitor): dynamic key switching + coalesce auto-heal
...
- xmpp_logger.py:
- Add switch_key(key_id): switch Hermes model.provider via sed + systemctl restart
- Add current_provider() reading model: block correctly (not just first ' provider:' line)
- Add KEY_TO_PROVIDER mapping (AgentsMeeting key_id -> Hermes provider name)
- Add RESTART_COOLDOWN_FILE/SEC = 180s to prevent restart loops
- auto_heal(): detect HTTP 429 / Weekly usage limit -> call best_key() -> switch_key()
- auto_heal(): detect timeout/error -> async systemctl restart (Popen, not blocking)
- _verify_llm(): bump timeout 25s -> 90s (cold-start gateway takes 20-40s)
- health(): urlopen timeout 8s -> 90s (match verify window)
- Use sudo NOPASSWD (hmo ALL=(ALL) NOPASSWD: ALL already configured)
- agents_health_check.py:
- Replace inline pkill+Popen restart logic (caused multiple instances) with systemctl
- Add RESTART_COOLDOWN_FILE state to skip restart within 3 min of last
- Call xmpp_logger.auto_heal() at end of every cron cycle
- Both inline restart and auto_heal restarts share cooldown file
- scripts/key_status.py: Reports weekly/monthly/rolling status of all 6 OCG keys
- scripts/test_llm.py: 90s timeout test (was 15s, gateway cold-start >= 30s)
- scripts/test_production.py: smoke test on /home/hmo/MoFin/ (hardlinked to web-dashboard)
Fixes:
- Old assumption 'Cloudflare blocks Python User-Agent' was WRONG.
True cause was HTTP 429 Weekly usage limit on key5 (12hr reset window).
Hermes silently ignored providers.X.headers config keys; only model.default_headers works.
Config already has model.default_headers: User-Agent: curl/8.5.0 (defense in depth).
- Multiple gateway instances were caused by 3 competing systemd units
(hermes-gateway@.service template + hermes-gateway-zhiwei.service named).
Masked the template unit @position-analyst and @zhiwei so only the named one wins.
2026-07-19 20:12:47 +08:00
hmo
bff246313b
feat: auto-heal checks key availability before restarting gateway
2026-07-19 14:46:46 +08:00
hmo
f791d8dee7
feat: API key availability monitoring — auto-select best key from AgentsMeeting
2026-07-19 14:41:18 +08:00
hmo
6398f595b2
feat: auto-heal pipeline — detect+restart gateway/bot/ejabberd on failure
2026-07-19 14:18:45 +08:00
hmo
0802046d5c
feat: Hermes Gateway + LLM provider monitoring, root cause visible in health tab
2026-07-19 13:25:37 +08:00
hmo
50ae4e1cc9
feat: XMPP bot journal monitoring — detect inbound/outbound/errors, smooth health refresh
2026-07-19 13:15:37 +08:00
hmo
b41bed7959
feat: XMPP observability — logger, monitor endpoints, clean dead xiaoguo refs
2026-07-19 12:47:15 +08:00
hmo