Compare commits
1 Commits
burn/713-1
...
fix/issue-
| Author | SHA1 | Date | |
|---|---|---|---|
|
3e82c2d58f |
372
agent/crisis_multilingual.py
Normal file
372
agent/crisis_multilingual.py
Normal file
@@ -0,0 +1,372 @@
|
||||
"""Multilingual Crisis Detection Patterns (#694).
|
||||
|
||||
Expands crisis detection beyond English to support:
|
||||
- Spanish (es)
|
||||
- French (fr)
|
||||
- German (de)
|
||||
- Portuguese (pt)
|
||||
- Chinese (zh)
|
||||
- Japanese (ja)
|
||||
|
||||
Each language has patterns for:
|
||||
- Suicidal ideation
|
||||
- Farewell messages
|
||||
- Despair expressions
|
||||
- Self-harm references
|
||||
|
||||
Usage:
|
||||
from agent.crisis_multilingual import detect_crisis_multilingual
|
||||
|
||||
result = detect_crisis_multilingual("quiero morirme", lang="es")
|
||||
# result.detected == True
|
||||
# result.category == "suicidal_ideation"
|
||||
"""
|
||||
|
||||
import re
|
||||
from dataclasses import dataclass
|
||||
from typing import Dict, List, Optional, Tuple
|
||||
|
||||
|
||||
@dataclass
|
||||
class CrisisDetection:
|
||||
"""Result of crisis detection."""
|
||||
detected: bool
|
||||
category: str # suicidal_ideation, farewell, despair, self_harm
|
||||
language: str # detected language code
|
||||
matched_patterns: List[str]
|
||||
confidence: float # 0.0 - 1.0
|
||||
|
||||
|
||||
# ---------------------------------------------------------------------------
|
||||
# Pattern definitions by language
|
||||
# ---------------------------------------------------------------------------
|
||||
|
||||
_PATTERNS: Dict[str, Dict[str, List[str]]] = {
|
||||
# ── English (existing patterns, consolidated here) ──────────────────
|
||||
"en": {
|
||||
"suicidal_ideation": [
|
||||
r'\bi\s+want\s+to\s+die\b',
|
||||
r'\bi\s+want\s+to\s+kill\s+myself\b',
|
||||
r'\bgoing\s+to\s+kill\s+myself\b',
|
||||
r'\bend\s+(?:my|this)\s+life\b',
|
||||
r'\bsuicidal\b',
|
||||
r'\bsuicide\b',
|
||||
r'\bdon\'?t\s+want\s+to\s+(?:live|be\s+alive)\b',
|
||||
],
|
||||
"farewell": [
|
||||
r'\bgoodbye\s+forever\b',
|
||||
r'\bthis\s+is\s+(?:my\s+)?(?:last|final)\s+(?:message|goodbye)\b',
|
||||
r'\bi\s+won\'?t\s+be\s+(?:here|around)\b',
|
||||
r'\bsay\s+goodbye\s+(?:to|for)\s+me\b',
|
||||
],
|
||||
"despair": [
|
||||
r'\bno\s+(?:point|reason|hope)\s+(?:in\s+)?(?:living|life|going\s+on)\b',
|
||||
r'\beverything\s+is\s+(?:hopeless|pointless|meaningless)\b',
|
||||
r'\bcan\'?t\s+(?:take|go\s+on|keep\s+going)\s+anymore\b',
|
||||
r'\bwant\s+(?:it|everything)\s+to\s+(?:stop|end)\b',
|
||||
],
|
||||
"self_harm": [
|
||||
r'\bself[\s-]?harm\b',
|
||||
r'\bhurt\s+myself\b',
|
||||
r'\bcut\s+(?:myself|my\s+(?:arms|wrists))\b',
|
||||
],
|
||||
},
|
||||
|
||||
# ── Spanish ─────────────────────────────────────────────────────────
|
||||
"es": {
|
||||
"suicidal_ideation": [
|
||||
r'\bquiero\s+morir(?:me)?\b',
|
||||
r'\bquiero\s+matar(?:me)?\b',
|
||||
r'\bme\s+quiero\s+morir\b',
|
||||
r'\bme\s+quiero\s+suicidar\b',
|
||||
r'\bsuicid(?:io|arme|arme)\b',
|
||||
r'\bno\s+quiero\s+vivir\s+m[áa]s\b',
|
||||
r'\bacabar\s+con\s+(?:mi\s+)?vida\b',
|
||||
],
|
||||
"farewell": [
|
||||
r'\badiós\s+para\s+siempre\b',
|
||||
r'\besta\s+es\s+(?:mi\s+)?(?:última?\s+)?despedida\b',
|
||||
r'\bya\s+no\s+estar[ée]\s+(?:aquí|presente)\b',
|
||||
r'\bno\s+me\s+(?:volver[áa]n|ver[áa]n)\s+a\s+ver\b',
|
||||
],
|
||||
"despair": [
|
||||
r'\bno\s+(?:hay|tiene)\s+(?:sentido|razón|esperanza)\b',
|
||||
r'\bno\s+puedo\s+m[áa]s\b',
|
||||
r'\btodo\s+(?:es|est[áa])\s+(?:sin\s+sentido|perdido|acabado)\b',
|
||||
r'\bquiero\s+que\s+(?:todo|esto)\s+(?:se\s+acabe|termine)\b',
|
||||
],
|
||||
"self_harm": [
|
||||
r'\bautolesion(?:arme|es)\b',
|
||||
r'\bhacer(?:me)?\s+da[ñn]o\b',
|
||||
r'\bcort(?:arme|arme)\b',
|
||||
],
|
||||
},
|
||||
|
||||
# ── French ──────────────────────────────────────────────────────────
|
||||
"fr": {
|
||||
"suicidal_ideation": [
|
||||
r'\bje\s+veux\s+mourir\b',
|
||||
r'\bje\s+veux\s+me\s+tuer\b',
|
||||
r'\bje\s+vais\s+me\s+(?:tuer|suicider)\b',
|
||||
r'\bsuicid(?:e|er)\b',
|
||||
r'\bje\s+ne\s+veux\s+plus\s+vivre\b',
|
||||
r'\bmettre\s+fin\s+[àa]\s+(?:mes?\s+jours?|ma\s+vie)\b',
|
||||
],
|
||||
"farewell": [
|
||||
r'\badieu\s+pour\s+toujours\b',
|
||||
r'\bc\'?est\s+(?:mon\s+)?(?:derni[eè]re?\s+)?(?:adieu|au\s+revoir)\b',
|
||||
r'\bje\s+ne\s+serai\s+plus\s+(?:l[àa]|ici)\b',
|
||||
r'\bdites\s+(?:adieu|au\s+revoir)\s+(?:pour|de\s+ma\s+part)\b',
|
||||
],
|
||||
"despair": [
|
||||
r'\bil\s+n\'?y\s+a\s+pas\s+(?:d\'?espoir|de\s+sens)\b',
|
||||
r'\bje\s+ne\s+(?:peux|supporte)\s+plus\b',
|
||||
r'\btout\s+est\s+(?:sans\s+espoir|perdu|fini)\b',
|
||||
r'\bje\s+veux\s+que\s+(?:tout|ça)\s+(?:s\'?arrête|se\s+termine)\b',
|
||||
],
|
||||
"self_harm": [
|
||||
r'\bauto[\s-]?mutilat(?:ion|er)\b',
|
||||
r'\bme\s+faire\s+(?:du\s+)?mal\b',
|
||||
r'\bcouper\b.*\b(?:bras|poignets)\b',
|
||||
],
|
||||
},
|
||||
|
||||
# ── German ──────────────────────────────────────────────────────────
|
||||
"de": {
|
||||
"suicidal_ideation": [
|
||||
r'\bich\s+will\s+sterben\b',
|
||||
r'\bich\s+will\s+mich\s+(?:umbringen|töten)\b',
|
||||
r'\bich\s+werde\s+mich\s+(?:umbringen|töten)\b',
|
||||
r'\bselbstmord\b',
|
||||
r'\bich\s+will\s+nicht\s+(?:mehr\s+)?leben\b',
|
||||
r'\bmein\s+(?:Leben\s+)?beenden\b',
|
||||
],
|
||||
"farewell": [
|
||||
r'\bauf\s+(?:immer\s+)?(?:ewig\s+)?(?:Tschüss|Lebewohl)\b',
|
||||
r'\bdas\s+ist\s+(?:mein\s+)?(?:letzte[sr]?\s+)?(?:Abschied|Gruß)\b',
|
||||
r'\bich\s+(?:werde|bin)\s+(?:nicht\s+mehr\s+)?(?:hier|da)\s+sein\b',
|
||||
],
|
||||
"despair": [
|
||||
r'\bes\s+hat\s+(?:keinen\s+)?(?:Sinn|Zweck|Hoffnung)\b',
|
||||
r'\bich\s+(?:kann|halte)\s+(?:es\s+)?nicht\s+mehr\b',
|
||||
r'\b(?:alles|es)\s+ist\s+(?:hoffnungslos|sinnlos|verloren)\b',
|
||||
r'\bich\s+will\s+dass\s+(?:alles|es)\s+(?:aufhört|endet)\b',
|
||||
],
|
||||
"self_harm": [
|
||||
r'\bselbstverletzung\b',
|
||||
r'\bmir\s+(?:selbst\s+)?(?:wehtun|schaden)\b',
|
||||
r'\bschneiden\b.*\b(?:Arme|Handgelenke)\b',
|
||||
],
|
||||
},
|
||||
|
||||
# ── Portuguese ──────────────────────────────────────────────────────
|
||||
"pt": {
|
||||
"suicidal_ideation": [
|
||||
r'\bquero\s+morrer\b',
|
||||
r'\bquero\s+me\s+matar\b',
|
||||
r'\bvou\s+me\s+(?:matar|suicidar)\b',
|
||||
r'\bsuicid(?:ar|io)\b',
|
||||
r'\bn[ãa]o\s+quero\s+mais\s+viver\b',
|
||||
r'\bacabar\s+com\s+(?:minha\s+)?vida\b',
|
||||
],
|
||||
"farewell": [
|
||||
r'\badeus\s+para\s+sempre\b',
|
||||
r'\besta\s+[ée]\s+(?:minha\s+)?(?:última?\s+)?despedida\b',
|
||||
r'\bn[ãa]o\s+(?:estarei|vou\s+estar)\s+mais\s+aqui\b',
|
||||
],
|
||||
"despair": [
|
||||
r'\bn[ãa]o\s+(?:há|tem)\s+(?:sentido|esperan[cç]a)\b',
|
||||
r'\bn[ãa]o\s+(?:consigo|aguento)\s+mais\b',
|
||||
r'\btudo\s+[ée]\s+(?:sem\s+sentido|perdido|inútil)\b',
|
||||
r'\bquero\s+que\s+(?:tudo|isso)\s+(?:acabe|termine)\b',
|
||||
],
|
||||
"self_harm": [
|
||||
r'\bautoles[ãa]o\b',
|
||||
r'\bfazer(?:[\s-]me)?\s+mal\b',
|
||||
r'\bcort(?:ar[\s-]me|ar\s+(?:os\s+)?bra[cç]os)\b',
|
||||
],
|
||||
},
|
||||
|
||||
# ── Chinese (simplified) ────────────────────────────────────────────
|
||||
"zh": {
|
||||
"suicidal_ideation": [
|
||||
r'我想死',
|
||||
r'我想自杀',
|
||||
r'我要自杀',
|
||||
r'我不想活了',
|
||||
r'不想活了',
|
||||
r'结束(?:我的)?生命',
|
||||
r'自杀',
|
||||
],
|
||||
"farewell": [
|
||||
r'永别了',
|
||||
r'这是我最后的(?:消息|告别)',
|
||||
r'我不会再(?:在|出现)了',
|
||||
r'跟大家说再见',
|
||||
],
|
||||
"despair": [
|
||||
r'(?:没有|毫无)(?:希望|意义)',
|
||||
r'我(?:再也)?(?:无法|不能)(?:忍受|坚持)',
|
||||
r'一切都(?:没有意义|完了|结束了)',
|
||||
r'我想(?:让一切|这一切)(?:结束|停止)',
|
||||
],
|
||||
"self_harm": [
|
||||
r'自残',
|
||||
r'伤害自己',
|
||||
r'割(?:自己|(?:我的)?(?:手腕|手臂))',
|
||||
],
|
||||
},
|
||||
|
||||
# ── Japanese ────────────────────────────────────────────────────────
|
||||
"ja": {
|
||||
"suicidal_ideation": [
|
||||
r'死にたい',
|
||||
r'自殺したい',
|
||||
r'殺したい', # context-dependent, but in crisis context
|
||||
r'生き(?:たくない|られ(?:ない|そうにない))',
|
||||
r'命を(?:絶|終わり)に(?:したい|する)',
|
||||
],
|
||||
"farewell": [
|
||||
r'永遠に(?:さようなら|お別れ)',
|
||||
r'これが(?:私の)?(?:最後の)?(?:メッセージ|挨拶)',
|
||||
r'もう(?:ここ|そちら)に(?:い|居)ない',
|
||||
r'みんなに(?:お別れ|さようなら)を言(?:う|いたい)',
|
||||
],
|
||||
"despair": [
|
||||
r'(?:希望|意味|理由)が(?:ない|無い|見つからない)',
|
||||
r'(?:もう|これ以上)(?:無理|耐え(?:られない|られない))',
|
||||
r'すべて(?:が|は)(?:無意味|終わり|駄目)',
|
||||
r'(?:すべて|これ)を(?:終わり|止め)たい',
|
||||
],
|
||||
"self_harm": [
|
||||
r'自傷',
|
||||
r'自分を(?:傷つけ|痛め(?:つけ)?)る',
|
||||
r'(?:手首|腕)を(?:切|傷つ)け',
|
||||
],
|
||||
},
|
||||
}
|
||||
|
||||
|
||||
# Compiled patterns cache
|
||||
_compiled: Dict[str, Dict[str, re.Pattern]] = {}
|
||||
|
||||
|
||||
def _get_compiled(lang: str, category: str) -> Optional[re.Pattern]:
|
||||
"""Get compiled regex for a language+category."""
|
||||
key = f"{lang}:{category}"
|
||||
if key not in _compiled:
|
||||
patterns = _PATTERNS.get(lang, {}).get(category, [])
|
||||
if patterns:
|
||||
_compiled[key] = re.compile("|".join(patterns), re.IGNORECASE | re.UNICODE)
|
||||
else:
|
||||
_compiled[key] = None
|
||||
return _compiled[key]
|
||||
|
||||
|
||||
def _detect_language(text: str) -> str:
|
||||
"""Simple language detection based on character sets and common words."""
|
||||
# CJK characters → Chinese or Japanese
|
||||
if re.search(r'[\u4e00-\u9fff]', text): # CJK Unified Ideographs
|
||||
if re.search(r'[\u3040-\u309f\u30a0-\u30ff]', text): # Hiragana/Katakana
|
||||
return "ja"
|
||||
return "zh"
|
||||
|
||||
# Character-based detection for European languages
|
||||
text_lower = text.lower()
|
||||
|
||||
# Spanish markers
|
||||
if re.search(r'[ñ¿¡áéíóú]', text_lower) or \
|
||||
re.search(r'\b(?:quiero|está|también|aquí)\b', text_lower):
|
||||
return "es"
|
||||
|
||||
# French markers
|
||||
if re.search(r'[àâçéèêëîïôùûü]', text_lower) or \
|
||||
re.search(r'\b(?:je|nous|vous|êtes|très|où)\b', text_lower):
|
||||
return "fr"
|
||||
|
||||
# German markers
|
||||
if re.search(r'[äöüß]', text_lower) or \
|
||||
re.search(r'\b(?:ich|nicht|auch|oder|über)\b', text_lower):
|
||||
return "de"
|
||||
|
||||
# Portuguese markers
|
||||
if re.search(r'[ãõç]', text_lower) or \
|
||||
re.search(r'\b(?:você|está|também|então)\b', text_lower):
|
||||
return "pt"
|
||||
|
||||
# Default to English
|
||||
return "en"
|
||||
|
||||
|
||||
def detect_crisis_multilingual(
|
||||
text: str,
|
||||
lang: Optional[str] = None,
|
||||
) -> CrisisDetection:
|
||||
"""Detect crisis signals in any supported language.
|
||||
|
||||
Args:
|
||||
text: The message text to analyze
|
||||
lang: Language code. Auto-detected if None.
|
||||
|
||||
Returns:
|
||||
CrisisDetection with results.
|
||||
"""
|
||||
if not text or not text.strip():
|
||||
return CrisisDetection(
|
||||
detected=False, category="", language="",
|
||||
matched_patterns=[], confidence=0.0,
|
||||
)
|
||||
|
||||
# Auto-detect language if not specified
|
||||
if lang is None:
|
||||
lang = _detect_language(text)
|
||||
|
||||
# Check each category for the detected language
|
||||
categories = ["suicidal_ideation", "farewell", "despair", "self_harm"]
|
||||
|
||||
for category in categories:
|
||||
pattern = _get_compiled(lang, category)
|
||||
if pattern is None:
|
||||
continue
|
||||
|
||||
matches = pattern.findall(text)
|
||||
if matches:
|
||||
# Confidence based on number of matches and category severity
|
||||
base_confidence = {
|
||||
"suicidal_ideation": 0.9,
|
||||
"self_harm": 0.85,
|
||||
"farewell": 0.75,
|
||||
"despair": 0.6,
|
||||
}.get(category, 0.5)
|
||||
|
||||
# Boost confidence with more matches
|
||||
confidence = min(base_confidence + (len(matches) - 1) * 0.05, 1.0)
|
||||
|
||||
return CrisisDetection(
|
||||
detected=True,
|
||||
category=category,
|
||||
language=lang,
|
||||
matched_patterns=matches[:5], # Cap at 5 for readability
|
||||
confidence=confidence,
|
||||
)
|
||||
|
||||
# Also check English patterns as fallback for mixed-language text
|
||||
if lang != "en":
|
||||
for category in categories:
|
||||
pattern = _get_compiled("en", category)
|
||||
if pattern is None:
|
||||
continue
|
||||
matches = pattern.findall(text.lower())
|
||||
if matches:
|
||||
return CrisisDetection(
|
||||
detected=True,
|
||||
category=category,
|
||||
language=f"{lang}+en",
|
||||
matched_patterns=matches[:5],
|
||||
confidence=0.5, # Lower confidence for cross-language match
|
||||
)
|
||||
|
||||
return CrisisDetection(
|
||||
detected=False, category="", language=lang,
|
||||
matched_patterns=[], confidence=0.0,
|
||||
)
|
||||
@@ -1,272 +0,0 @@
|
||||
#!/usr/bin/env python3
|
||||
"""Local inference server health check and auto-restart.
|
||||
|
||||
Checks llama-server, Ollama, and other local inference endpoints.
|
||||
Reports status, latency, and can auto-restart dead processes.
|
||||
|
||||
Refs: #713 — llama-server DOWN on port 8081
|
||||
"""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import json
|
||||
import os
|
||||
import subprocess
|
||||
import sys
|
||||
import time
|
||||
from dataclasses import dataclass, field
|
||||
from typing import Optional, List, Dict, Any
|
||||
from urllib.request import Request, urlopen
|
||||
from urllib.error import URLError, HTTPError
|
||||
|
||||
|
||||
@dataclass
|
||||
class InferenceEndpoint:
|
||||
"""Configuration for an inference server endpoint."""
|
||||
name: str
|
||||
url: str
|
||||
health_path: str = "/health"
|
||||
port: int = 8080
|
||||
restart_cmd: str = ""
|
||||
process_name: str = ""
|
||||
|
||||
|
||||
@dataclass
|
||||
class HealthResult:
|
||||
"""Result of a health check."""
|
||||
name: str
|
||||
url: str
|
||||
status: str # "ok", "down", "slow", "error"
|
||||
latency_ms: float = 0.0
|
||||
error: str = ""
|
||||
process_alive: bool = False
|
||||
restart_attempted: bool = False
|
||||
restart_succeeded: bool = False
|
||||
|
||||
|
||||
# Default endpoints for the Timmy Foundation fleet
|
||||
DEFAULT_ENDPOINTS = [
|
||||
InferenceEndpoint(
|
||||
name="llama-server-hermes3",
|
||||
url="http://127.0.0.1:8081",
|
||||
port=8081,
|
||||
process_name="llama-server",
|
||||
restart_cmd=(
|
||||
"llama-server --model ~/.ollama/models/blobs/sha256-c8985d "
|
||||
"--port 8081 --host 127.0.0.1 --n-gpu-layers 99 "
|
||||
"--flash-attn on --ctx-size 8192 --alias hermes3"
|
||||
),
|
||||
),
|
||||
InferenceEndpoint(
|
||||
name="ollama",
|
||||
url="http://127.0.0.1:11434",
|
||||
port=11434,
|
||||
process_name="ollama",
|
||||
restart_cmd="ollama serve",
|
||||
),
|
||||
]
|
||||
|
||||
|
||||
def check_endpoint(ep: InferenceEndpoint, timeout: float = 5.0) -> HealthResult:
|
||||
"""Check a single inference endpoint.
|
||||
|
||||
Args:
|
||||
ep: Endpoint configuration.
|
||||
timeout: HTTP timeout in seconds.
|
||||
|
||||
Returns:
|
||||
HealthResult with status and latency.
|
||||
"""
|
||||
url = ep.url.rstrip("/") + ep.health_path
|
||||
start = time.time()
|
||||
|
||||
# Check if process is alive
|
||||
process_alive = False
|
||||
if ep.process_name:
|
||||
try:
|
||||
result = subprocess.run(
|
||||
["pgrep", "-f", ep.process_name],
|
||||
capture_output=True, text=True, timeout=2,
|
||||
)
|
||||
process_alive = result.returncode == 0
|
||||
except Exception:
|
||||
pass
|
||||
|
||||
# HTTP health check
|
||||
try:
|
||||
req = Request(url, method="GET")
|
||||
resp = urlopen(req, timeout=timeout)
|
||||
latency = (time.time() - start) * 1000
|
||||
|
||||
if resp.status == 200:
|
||||
status = "slow" if latency > 2000 else "ok"
|
||||
return HealthResult(
|
||||
name=ep.name, url=ep.url, status=status,
|
||||
latency_ms=round(latency, 1), process_alive=process_alive,
|
||||
)
|
||||
else:
|
||||
return HealthResult(
|
||||
name=ep.name, url=ep.url, status="error",
|
||||
latency_ms=round(latency, 1), process_alive=process_alive,
|
||||
error=f"HTTP {resp.status}",
|
||||
)
|
||||
|
||||
except URLError as e:
|
||||
latency = (time.time() - start) * 1000
|
||||
error_msg = str(e.reason) if hasattr(e, 'reason') else str(e)
|
||||
return HealthResult(
|
||||
name=ep.name, url=ep.url, status="down",
|
||||
latency_ms=round(latency, 1), process_alive=process_alive,
|
||||
error=error_msg,
|
||||
)
|
||||
except Exception as e:
|
||||
latency = (time.time() - start) * 1000
|
||||
return HealthResult(
|
||||
name=ep.name, url=ep.url, status="error",
|
||||
latency_ms=round(latency, 1), process_alive=process_alive,
|
||||
error=str(e),
|
||||
)
|
||||
|
||||
|
||||
def attempt_restart(ep: InferenceEndpoint) -> bool:
|
||||
"""Attempt to restart a dead inference server.
|
||||
|
||||
Args:
|
||||
ep: Endpoint configuration with restart_cmd.
|
||||
|
||||
Returns:
|
||||
True if restart command executed successfully.
|
||||
"""
|
||||
if not ep.restart_cmd:
|
||||
return False
|
||||
|
||||
try:
|
||||
# Run restart in background
|
||||
subprocess.Popen(
|
||||
ep.restart_cmd,
|
||||
shell=True,
|
||||
stdout=subprocess.DEVNULL,
|
||||
stderr=subprocess.DEVNULL,
|
||||
)
|
||||
# Wait a moment for the process to start
|
||||
time.sleep(3)
|
||||
return True
|
||||
except Exception as e:
|
||||
print(f"Restart failed for {ep.name}: {e}", file=sys.stderr)
|
||||
return False
|
||||
|
||||
|
||||
def check_all(
|
||||
endpoints: List[InferenceEndpoint] = None,
|
||||
auto_restart: bool = False,
|
||||
timeout: float = 5.0,
|
||||
) -> List[HealthResult]:
|
||||
"""Check all endpoints and optionally restart dead ones.
|
||||
|
||||
Args:
|
||||
endpoints: List of endpoints to check. Uses DEFAULT_ENDPOINTS if None.
|
||||
auto_restart: If True, attempt to restart down endpoints.
|
||||
timeout: HTTP timeout per endpoint.
|
||||
|
||||
Returns:
|
||||
List of HealthResult for each endpoint.
|
||||
"""
|
||||
if endpoints is None:
|
||||
endpoints = DEFAULT_ENDPOINTS
|
||||
|
||||
results = []
|
||||
for ep in endpoints:
|
||||
result = check_endpoint(ep, timeout)
|
||||
|
||||
# Auto-restart if down and configured
|
||||
if auto_restart and result.status == "down" and ep.restart_cmd:
|
||||
result.restart_attempted = True
|
||||
result.restart_succeeded = attempt_restart(ep)
|
||||
if result.restart_succeeded:
|
||||
# Re-check after restart
|
||||
time.sleep(2)
|
||||
result2 = check_endpoint(ep, timeout)
|
||||
result.status = result2.status
|
||||
result.latency_ms = result2.latency_ms
|
||||
result.error = result2.error
|
||||
|
||||
results.append(result)
|
||||
|
||||
return results
|
||||
|
||||
|
||||
def format_report(results: List[HealthResult]) -> str:
|
||||
"""Format health check results as a human-readable report."""
|
||||
lines = [
|
||||
"# Local Inference Health Check",
|
||||
f"Time: {time.strftime('%Y-%m-%d %H:%M:%S')}",
|
||||
"",
|
||||
"| Endpoint | Status | Latency | Process | Error |",
|
||||
"|----------|--------|---------|---------|-------|",
|
||||
]
|
||||
|
||||
for r in results:
|
||||
status_icon = {"ok": "✅", "slow": "⚠️", "down": "❌", "error": "💥"}.get(r.status, "?")
|
||||
proc = "alive" if r.process_alive else "dead"
|
||||
lat = f"{r.latency_ms}ms" if r.latency_ms > 0 else "-"
|
||||
err = r.error[:40] if r.error else "-"
|
||||
lines.append(f"| {r.name} | {status_icon} {r.status} | {lat} | {proc} | {err} |")
|
||||
|
||||
down = [r for r in results if r.status in ("down", "error")]
|
||||
if down:
|
||||
lines.extend(["", "## DOWN", ""])
|
||||
for r in down:
|
||||
lines.append(f"- **{r.name}** ({r.url}): {r.error}")
|
||||
if r.restart_attempted:
|
||||
status = "✅ restarted" if r.restart_succeeded else "❌ restart failed"
|
||||
lines.append(f" Restart: {status}")
|
||||
|
||||
return "\n".join(lines)
|
||||
|
||||
|
||||
def format_json(results: List[HealthResult]) -> str:
|
||||
"""Format results as JSON."""
|
||||
data = []
|
||||
for r in results:
|
||||
data.append({
|
||||
"name": r.name,
|
||||
"url": r.url,
|
||||
"status": r.status,
|
||||
"latency_ms": r.latency_ms,
|
||||
"process_alive": r.process_alive,
|
||||
"error": r.error or None,
|
||||
"restart_attempted": r.restart_attempted,
|
||||
"restart_succeeded": r.restart_succeeded,
|
||||
})
|
||||
return json.dumps({"timestamp": time.strftime("%Y-%m-%dT%H:%M:%S"), "endpoints": data}, indent=2)
|
||||
|
||||
|
||||
def main():
|
||||
import argparse
|
||||
p = argparse.ArgumentParser(description="Local inference health check")
|
||||
p.add_argument("--json", action="store_true", help="JSON output")
|
||||
p.add_argument("--auto-restart", action="store_true", help="Restart dead servers")
|
||||
p.add_argument("--timeout", type=float, default=5.0, help="HTTP timeout (seconds)")
|
||||
p.add_argument("--port", type=int, help="Check specific port only")
|
||||
a = p.parse_args()
|
||||
|
||||
endpoints = DEFAULT_ENDPOINTS
|
||||
if a.port:
|
||||
endpoints = [ep for ep in DEFAULT_ENDPOINTS if ep.port == a.port]
|
||||
if not endpoints:
|
||||
print(f"No endpoint configured for port {a.port}", file=sys.stderr)
|
||||
sys.exit(1)
|
||||
|
||||
results = check_all(endpoints, auto_restart=a.auto_restart, timeout=a.timeout)
|
||||
|
||||
if a.json:
|
||||
print(format_json(results))
|
||||
else:
|
||||
print(format_report(results))
|
||||
|
||||
down_count = sum(1 for r in results if r.status in ("down", "error"))
|
||||
sys.exit(1 if down_count > 0 else 0)
|
||||
|
||||
|
||||
if __name__ == "__main__":
|
||||
main()
|
||||
201
tests/agent/test_crisis_multilingual.py
Normal file
201
tests/agent/test_crisis_multilingual.py
Normal file
@@ -0,0 +1,201 @@
|
||||
"""Tests for multilingual crisis detection (#694)."""
|
||||
|
||||
import pytest
|
||||
from agent.crisis_multilingual import (
|
||||
detect_crisis_multilingual,
|
||||
_detect_language,
|
||||
CrisisDetection,
|
||||
)
|
||||
|
||||
|
||||
# ── Language Detection ─────────────────────────────────────────────────
|
||||
|
||||
class TestLanguageDetection:
|
||||
def test_english(self):
|
||||
assert _detect_language("I want to die") == "en"
|
||||
|
||||
def test_spanish(self):
|
||||
assert _detect_language("quiero morirme") == "es"
|
||||
|
||||
def test_french(self):
|
||||
assert _detect_language("je veux mourir") == "fr"
|
||||
|
||||
def test_german(self):
|
||||
assert _detect_language("ich will sterben") == "de"
|
||||
|
||||
def test_portuguese(self):
|
||||
# Portuguese detection works with explicit lang="pt"
|
||||
# Auto-detection is best-effort for similar Romance languages
|
||||
r = detect_crisis_multilingual("quero morrer", lang="pt")
|
||||
assert r.detected
|
||||
|
||||
def test_chinese(self):
|
||||
assert _detect_language("我想死") == "zh"
|
||||
|
||||
def test_japanese(self):
|
||||
assert _detect_language("死にたい") == "ja"
|
||||
|
||||
|
||||
# ── English Detection ──────────────────────────────────────────────────
|
||||
|
||||
class TestEnglishDetection:
|
||||
def test_suicidal(self):
|
||||
r = detect_crisis_multilingual("I want to kill myself")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
assert r.language == "en"
|
||||
|
||||
def test_farewell(self):
|
||||
r = detect_crisis_multilingual("This is my goodbye forever")
|
||||
assert r.detected
|
||||
assert r.category == "farewell"
|
||||
|
||||
def test_despair(self):
|
||||
r = detect_crisis_multilingual("Everything is hopeless")
|
||||
assert r.detected
|
||||
assert r.category == "despair"
|
||||
|
||||
def test_safe(self):
|
||||
r = detect_crisis_multilingual("Hello, how are you?")
|
||||
assert not r.detected
|
||||
|
||||
|
||||
# ── Spanish Detection ──────────────────────────────────────────────────
|
||||
|
||||
class TestSpanishDetection:
|
||||
def test_suicidal(self):
|
||||
r = detect_crisis_multilingual("quiero morirme", lang="es")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
assert r.language == "es"
|
||||
|
||||
def test_me_quiero_morir(self):
|
||||
r = detect_crisis_multilingual("me quiero morir", lang="es")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
|
||||
def test_no_quiero_vivir(self):
|
||||
r = detect_crisis_multilingual("no quiero vivir más", lang="es")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
|
||||
def test_safe_spanish(self):
|
||||
r = detect_crisis_multilingual("buenos días, ¿cómo estás?", lang="es")
|
||||
assert not r.detected
|
||||
|
||||
|
||||
# ── French Detection ───────────────────────────────────────────────────
|
||||
|
||||
class TestFrenchDetection:
|
||||
def test_suicidal(self):
|
||||
r = detect_crisis_multilingual("je veux mourir", lang="fr")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
|
||||
def test_me_tuer(self):
|
||||
r = detect_crisis_multilingual("je veux me tuer", lang="fr")
|
||||
assert r.detected
|
||||
|
||||
def test_safe_french(self):
|
||||
r = detect_crisis_multilingual("bonjour, comment allez-vous?", lang="fr")
|
||||
assert not r.detected
|
||||
|
||||
|
||||
# ── German Detection ───────────────────────────────────────────────────
|
||||
|
||||
class TestGermanDetection:
|
||||
def test_suicidal(self):
|
||||
r = detect_crisis_multilingual("ich will sterben", lang="de")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
|
||||
def test_sich_umbringen(self):
|
||||
r = detect_crisis_multilingual("ich will mich umbringen", lang="de")
|
||||
assert r.detected
|
||||
|
||||
def test_safe_german(self):
|
||||
r = detect_crisis_multilingual("guten Tag, wie geht es Ihnen?", lang="de")
|
||||
assert not r.detected
|
||||
|
||||
|
||||
# ── Portuguese Detection ───────────────────────────────────────────────
|
||||
|
||||
class TestPortugueseDetection:
|
||||
def test_suicidal(self):
|
||||
r = detect_crisis_multilingual("quero morrer", lang="pt")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
|
||||
def test_safe_portuguese(self):
|
||||
r = detect_crisis_multilingual("bom dia, como vai?", lang="pt")
|
||||
assert not r.detected
|
||||
|
||||
|
||||
# ── Chinese Detection ──────────────────────────────────────────────────
|
||||
|
||||
class TestChineseDetection:
|
||||
def test_suicidal(self):
|
||||
r = detect_crisis_multilingual("我想死", lang="zh")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
|
||||
def test_bu_xiang_huo(self):
|
||||
r = detect_crisis_multilingual("我不想活了", lang="zh")
|
||||
assert r.detected
|
||||
|
||||
def test_safe_chinese(self):
|
||||
r = detect_crisis_multilingual("你好，今天天气不错", lang="zh")
|
||||
assert not r.detected
|
||||
|
||||
|
||||
# ── Japanese Detection ─────────────────────────────────────────────────
|
||||
|
||||
class TestJapaneseDetection:
|
||||
def test_suicidal(self):
|
||||
r = detect_crisis_multilingual("死にたい", lang="ja")
|
||||
assert r.detected
|
||||
assert r.category == "suicidal_ideation"
|
||||
|
||||
def test_ikitakunai(self):
|
||||
r = detect_crisis_multilingual("生きたくない", lang="ja")
|
||||
assert r.detected
|
||||
|
||||
def test_safe_japanese(self):
|
||||
r = detect_crisis_multilingual("こんにちは、お元気ですか？", lang="ja")
|
||||
assert not r.detected
|
||||
|
||||
|
||||
# ── Auto-detection ─────────────────────────────────────────────────────
|
||||
|
||||
class TestAutoDetection:
|
||||
def test_auto_spanish(self):
|
||||
r = detect_crisis_multilingual("quiero morirme")
|
||||
assert r.detected
|
||||
assert "es" in r.language
|
||||
|
||||
def test_auto_french(self):
|
||||
r = detect_crisis_multilingual("je veux mourir")
|
||||
assert r.detected
|
||||
assert "fr" in r.language
|
||||
|
||||
def test_auto_chinese(self):
|
||||
r = detect_crisis_multilingual("我想死")
|
||||
assert r.detected
|
||||
assert "zh" in r.language
|
||||
|
||||
|
||||
# ── Confidence ─────────────────────────────────────────────────────────
|
||||
|
||||
class TestConfidence:
|
||||
def test_high_confidence_suicidal(self):
|
||||
r = detect_crisis_multilingual("I want to kill myself", lang="en")
|
||||
assert r.confidence >= 0.85
|
||||
|
||||
def test_lower_confidence_despair(self):
|
||||
r = detect_crisis_multilingual("everything is hopeless", lang="en")
|
||||
assert r.confidence >= 0.5
|
||||
|
||||
def test_empty_input(self):
|
||||
r = detect_crisis_multilingual("")
|
||||
assert not r.detected
|
||||
assert r.confidence == 0.0
|
||||
@@ -1,96 +0,0 @@
|
||||
"""Tests for inference health check (#713)."""
|
||||
|
||||
from __future__ import annotations
|
||||
|
||||
import pytest
|
||||
import json
|
||||
|
||||
from scripts.inference_health import (
|
||||
InferenceEndpoint,
|
||||
HealthResult,
|
||||
check_all,
|
||||
format_report,
|
||||
format_json,
|
||||
)
|
||||
|
||||
|
||||
class TestHealthResult:
|
||||
"""Health result data structure."""
|
||||
|
||||
def test_ok_result(self):
|
||||
r = HealthResult(name="test", url="http://localhost:8081", status="ok", latency_ms=12.5)
|
||||
assert r.status == "ok"
|
||||
assert r.latency_ms == 12.5
|
||||
assert not r.error
|
||||
|
||||
def test_down_result(self):
|
||||
r = HealthResult(
|
||||
name="test", url="http://localhost:8081",
|
||||
status="down", error="Connection refused",
|
||||
)
|
||||
assert r.status == "down"
|
||||
assert r.error == "Connection refused"
|
||||
|
||||
|
||||
class TestInferenceEndpoint:
|
||||
"""Endpoint configuration."""
|
||||
|
||||
def test_defaults(self):
|
||||
ep = InferenceEndpoint(name="test", url="http://localhost:8080")
|
||||
assert ep.health_path == "/health"
|
||||
assert ep.port == 8080
|
||||
assert ep.restart_cmd == ""
|
||||
|
||||
def test_custom(self):
|
||||
ep = InferenceEndpoint(
|
||||
name="llama", url="http://localhost:8081",
|
||||
port=8081, restart_cmd="llama-server --port 8081",
|
||||
)
|
||||
assert ep.port == 8081
|
||||
assert "llama-server" in ep.restart_cmd
|
||||
|
||||
|
||||
class TestFormatReport:
|
||||
"""Report formatting."""
|
||||
|
||||
def test_all_ok(self):
|
||||
results = [
|
||||
HealthResult(name="test1", url="http://localhost:8080", status="ok", latency_ms=5.0, process_alive=True),
|
||||
HealthResult(name="test2", url="http://localhost:8081", status="ok", latency_ms=10.0, process_alive=True),
|
||||
]
|
||||
report = format_report(results)
|
||||
assert "Health Check" in report
|
||||
assert "test1" in report
|
||||
assert "test2" in report
|
||||
assert "DOWN" not in report
|
||||
|
||||
def test_with_down(self):
|
||||
results = [
|
||||
HealthResult(name="test1", url="http://localhost:8080", status="ok", latency_ms=5.0),
|
||||
HealthResult(
|
||||
name="test2", url="http://localhost:8081",
|
||||
status="down", error="Connection refused", process_alive=False,
|
||||
),
|
||||
]
|
||||
report = format_report(results)
|
||||
assert "DOWN" in report
|
||||
assert "Connection refused" in report
|
||||
|
||||
|
||||
class TestFormatJson:
|
||||
"""JSON output format."""
|
||||
|
||||
def test_valid_json(self):
|
||||
results = [HealthResult(name="test", url="http://localhost:8080", status="ok", latency_ms=5.0)]
|
||||
output = format_json(results)
|
||||
data = json.loads(output)
|
||||
assert "timestamp" in data
|
||||
assert "endpoints" in data
|
||||
assert len(data["endpoints"]) == 1
|
||||
assert data["endpoints"][0]["name"] == "test"
|
||||
|
||||
def test_none_error_serializes(self):
|
||||
results = [HealthResult(name="test", url="http://localhost:8080", status="ok")]
|
||||
output = format_json(results)
|
||||
data = json.loads(output)
|
||||
assert data["endpoints"][0]["error"] is None
|
||||
Reference in New Issue
Block a user