wooo/awoooi
1
1
Fork 0
Code Issues 2 Pull Requests 1 Actions 3 Packages Projects Releases Wiki Activity

feat(api): Phase 15.1 Langfuse LLMOps 整合 + 模型升級

Browse Source 
## 新功能
- Langfuse 自建部署 (192.168.0.110:3100)
- langfuse_client.py - LLM 呼叫追蹤包裝
- OpenClaw 整合 Langfuse trace

## 模型升級 (統帥批准)
- 生產預設: llama3.2:3b → qwen2.5:7b-instruct
- 摘要任務: llama3.2:3b (速度優先)

## 配置更新
- requirements.txt: +langfuse>=2.0.0
- config.py: +LANGFUSE_* 設定
- models.json: 更新 Ollama 模型配置
- K8s: Secret + ConfigMap 更新

## 審查通過
- 模組化檢查 ✅
- 核心測試 31/31 ✅

Co-Authored-By: Claude Opus 4.5 <noreply@anthropic.com>
This commit is contained in:
OG T
2026-03-26 00:32:19 +08:00
parent 31fabe8d61
commit 1ac8965a7a
11 changed files with 727 additions and 31 deletions
Download Patch File Download Diff File Expand all files Collapse all files

168
.agents/skills/02-lewooogo-backend-core.md
View File

@@ -6,6 +6,28 @@
---
## 文件資訊
| 欄位 | 值 |
|------|-----|
| **版本** | v1.4 |
| **建立日期** | 2026-03-20 (台北) |
| **建立者** | Claude Code |
| **最後修改** | 2026-03-26 00:20 (台北) |
| **修改者** | Claude Code |
### 變更紀錄
| 版本 | 日期 | 執行者 | 變更內容 |
|------|------|--------|----------|
| v1.0 | 2026-03-20 | Claude Code | 初始建立 |
| v1.1 | 2026-03-25 | Claude Code | 新增 MCP Tool 實作標準 |
| v1.2 | 2026-03-25 | Claude Code | 加入文件資訊區塊 |
| v1.3 | 2026-03-26 | Claude Code | 🔴🔴🔴 新增積木化強制執行章節 (32 項違規審計後) |
| v1.4 | 2026-03-26 | Claude Code | 📊 新增 Langfuse LLMOps 整合章節 (Phase 15.1) |
---
## 核心約束 (Six Iron Laws)
### 1. Async-First (非同步優先)
@@ -119,6 +141,45 @@ OTEL_SERVICE_NAME=awoooi-api
---
## Langfuse LLMOps (Phase 15.1)
> **用途**: LLM 呼叫追蹤、成本監控、Prompt 版本管理
> **端點**: http://192.168.0.110:3100
### 環境變數
```bash
LANGFUSE_URL=http://192.168.0.110:3100
LANGFUSE_PUBLIC_KEY=pk-lf-xxx # 從 K8s Secret
LANGFUSE_SECRET_KEY=sk-lf-xxx # 從 K8s Secret
```
### 整合模式 (待實作)
```python
from langfuse import Langfuse
langfuse = Langfuse(
public_key=settings.LANGFUSE_PUBLIC_KEY,
secret_key=settings.LANGFUSE_SECRET_KEY,
host=settings.LANGFUSE_URL
)
# 包裝 LLM 呼叫
with langfuse.trace(name="openclaw_decision") as trace:
generation = trace.generation(
name="ollama_call",
model="llama3.2:3b",
input=prompt,
)
result = await _call_ollama(prompt)
generation.end(output=result)
```
**Memory 參考**: `reference_langfuse_credentials.md`
---
## 機密管理 (嚴禁硬編碼)
```python
@@ -315,12 +376,119 @@ packages/lewooogo-data/src/lewooogo_data/
---
## 🧰 MCP Tool 實作規範 (Phase 13.2)
> **目標**: 將 Mock MCP Tool 升級為真實系統連接
> **優先級**: P0 最優先
### MCP Bridge 位置
```
apps/api/src/plugins/mcp/
├── __init__.py
├── mcp_bridge.py # 核心橋樑
└── tools/ # 新增目錄
├── signoz_tool.py # #79 SignOz MCP
├── kubernetes_tool.py # #80 Kubernetes MCP
└── postgres_tool.py # #81 PostgreSQL MCP
```
### Tool 實作模板
```python
from src.plugins.mcp.mcp_bridge import MCPTool, MCPToolResult
class SignOzTool:
"""SignOz 監控查詢 Tool"""
async def query_traces(
self,
service_name: str,
start_time: datetime,
end_time: datetime,
) -> MCPToolResult:
"""查詢服務 Traces
Args:
service_name: 服務名稱 (awoooi-api, awoooi-web)
start_time: 查詢起始時間 (台北時區)
end_time: 查詢結束時間
Returns:
MCPToolResult with trace data
Warning:
SignOz API 有速率限制，避免高頻查詢
"""
# 實作邏輯...
```
### Tool 實作鐵律
| 規則 | 說明 |
|------|------|
| Privacy Shield | 所有 Tool 調用必須經過 RehydrationEngine |
| 錯誤處理 | 外部系統失敗不影響主流程 |
| 日誌脫敏 | 禁止 log 敏感參數 (IP/Token) |
| 超時設定 | 每個 Tool 必須設定合理 timeout |
| 台北時區 | 所有時間參數必須使用 +8 時區 |
### Tool 狀態
| Tool | 狀態 | 說明 |
|------|------|------|
| Kubernetes | 🟡 Mock | #80 待實作真實 kubectl |
| Database | 🟡 Mock | #81 待連接 PostgreSQL |
| **SignOz** | ❌ 缺失 | #79 **P0** 查詢 Trace/Logs |
| **維運手冊 RAG** | ❌ 缺失 | #84 Markdown 向量化 |
---
## 🔴🔴🔴 積木化強制執行 (2026-03-26 審計後新增)
> **審計結果**: 32 項違規，3 週開發工時浪費
> **Memory**: `feedback_lewooogo_modular_enforcement.md`
### 修改前必問 5 題
1. **這個邏輯是否已存在於 packages/？** → 用 packages，禁止重寫
2. **Router 是否只做 HTTP 轉發？** → 禁止 Router 直接存取 Redis/DB
3. **Service 是否依賴 Interface？** → 必須用 Protocol/ABC
4. **是否可被其他模組重用？** → 可重用邏輯放 packages/
5. **是否遵循依賴注入？** → 禁止 global singleton
### 禁止清單 (Router 層)
```python
# ❌ 禁止
from src.core.redis_client import get_redis # 應透過 Service
from src.db.base import get_session # 應透過 Repository
LUA_SCRIPT = """...""" # 應放在 Repository
```
### 正確架構層次
```
api/v1/*.py (Router) → services/*.py (Service) → packages/lewooogo-*/ (積木)
```
### 違規時
1. **停止修改，通報統帥**
2. **必須修改違規檔案 → 同時修復違規**
3. **新增功能 → 放在正確層次**
---
## 參考文檔
- `apps/api/src/core/config.py`: 設定中心
- `apps/api/src/main.py`: FastAPI 應用入口
- `apps/api/src/plugins/mcp/mcp_bridge.py`: MCP Bridge 核心
- `packages/lewooogo-data/`: 記憶體 Provider 積木
- `packages/lewooogo-brain/`: AI 引擎積木
- `memory/feedback_lewooogo_modular_enforcement.md`: 積木化強制執行鐵律
- ADR-001: MCP Protocol 採用
- ADR-005: BFF 閘道架構
- ADR-006: AI 備援策略
- ADR-008: Python 模組化獨立積木架構

8
apps/api/models.json
View File

@@ -3,7 +3,7 @@
"name": "OpenClaw AI Router Configuration",
"version": "1.0.0",
"description": "AI 模型路由與備援設定 (ADR-006)",
"updated_at": "2026-03-21",
"updated_at": "2026-03-26",
"default_provider": "ollama",
"fallback_order": ["ollama", "gemini", "claude"],
@@ -16,9 +16,9 @@
"endpoint": "http://192.168.0.188:11434",
"api_path": "/api/generate",
"models": {
"default": "llama3.2:3b",
"rca": "llama3.2:3b",
"summary": "llama3.2:1b"
"default": "qwen2.5:7b-instruct",
"rca": "qwen2.5:7b-instruct",
"summary": "llama3.2:3b"
},
"options": {
"temperature": 0.1,

6
apps/api/requirements.txt
View File

@@ -36,6 +36,12 @@ opentelemetry-instrumentation-fastapi>=0.41b0
opentelemetry-instrumentation-httpx>=0.41b0
opentelemetry-instrumentation-logging>=0.41b0
# ==========================================================================
# Langfuse (LLMOps Observability)
# Phase 15.1: LLM 呼叫追蹤、成本監控、Prompt 版本管理
# ==========================================================================
langfuse>=2.0.0
# Development
pytest>=7.4.0
pytest-asyncio>=0.23.0

26
apps/api/src/core/config.py
View File

@@ -157,6 +157,28 @@ class Settings(BaseSettings):
description="Trace sampling rate (1.0 = 100%)",
)
# ==========================================================================
# Langfuse LLMOps (Phase 15.1)
# LLM 呼叫追蹤、成本監控、Prompt 版本管理
# 端點: http://192.168.0.110:3100 (DevOps 金庫)
# ==========================================================================
LANGFUSE_ENABLED: bool = Field(
default=True,
description="Enable Langfuse LLM observability",
)
LANGFUSE_URL: str = Field(
default="http://192.168.0.110:3100",
description="Langfuse self-hosted URL",
)
LANGFUSE_PUBLIC_KEY: str = Field(
default="",
description="Langfuse public key (from K8s Secret)",
)
LANGFUSE_SECRET_KEY: str = Field(
default="",
description="Langfuse secret key (from K8s Secret)",
)
# ==========================================================================
# AI Fallback Strategy (ADR-006)
# Order: Ollama (local) -> Gemini (cloud) -> Claude (cloud)
@@ -224,8 +246,8 @@ class Settings(BaseSettings):
description="OpenClaw AI Agent service URL",
)
OPENCLAW_DEFAULT_MODEL: str = Field(
default="llama3.2:3b",
description="Default Ollama model for RCA analysis",
default="qwen2.5:7b-instruct",
description="Default Ollama model for RCA analysis (7B params, better Chinese)",
)
OPENCLAW_TIMEOUT: int = Field(
default=90,

285
apps/api/src/services/langfuse_client.py Normal file
View File

@@ -0,0 +1,285 @@
"""
Langfuse LLMOps Client - Phase 15.1
===================================
LLM 呼叫追蹤、成本監控、Prompt 版本管理
Phase 15.1 (2026-03-26)
端點: http://192.168.0.110:3100 (DevOps 金庫)
Features:
- 自動追蹤所有 LLM 呼叫 (Ollama/Gemini/Claude)
- 成本估算與監控
- Prompt 版本管理
- 與 OTEL Trace 整合
Usage:
from src.services.langfuse_client import get_langfuse, langfuse_trace
# 方法 1: Context Manager
async with langfuse_trace("openclaw_decision") as trace:
result = await call_llm(prompt)
trace.generation(
name="ollama_call",
model="qwen2.5:7b-instruct",
input=prompt,
output=result,
)
# 方法 2: 裝飾器
@langfuse_observe(name="analyze_incident")
async def analyze_incident(incident_id: str):
...
"""
from contextlib import asynccontextmanager
from functools import wraps
from typing import Any, Callable
import structlog
from src.core.config import settings
logger = structlog.get_logger(__name__)
# Langfuse client singleton
_langfuse_client = None
def get_langfuse():
"""
取得 Langfuse client singleton
Returns:
Langfuse client 或 None (如果未啟用或未配置)
"""
global _langfuse_client
if not settings.LANGFUSE_ENABLED:
return None
if not settings.LANGFUSE_PUBLIC_KEY or not settings.LANGFUSE_SECRET_KEY:
logger.warning(
"langfuse_not_configured",
message="Langfuse enabled but keys not set",
)
return None
if _langfuse_client is None:
try:
from langfuse import Langfuse
_langfuse_client = Langfuse(
public_key=settings.LANGFUSE_PUBLIC_KEY,
secret_key=settings.LANGFUSE_SECRET_KEY,
host=settings.LANGFUSE_URL,
)
logger.info(
"langfuse_initialized",
host=settings.LANGFUSE_URL,
)
except Exception as e:
logger.error(
"langfuse_init_failed",
error=str(e),
)
return None
return _langfuse_client
class LangfuseTraceContext:
"""Langfuse Trace Context for tracking LLM calls"""
def __init__(self, name: str, metadata: dict[str, Any] | None = None):
self.name = name
self.metadata = metadata or {}
self.trace = None
self._client = get_langfuse()
def __enter__(self):
if self._client:
try:
self.trace = self._client.trace(
name=self.name,
metadata=self.metadata,
)
except Exception as e:
logger.warning("langfuse_trace_start_failed", error=str(e))
return self
def __exit__(self, exc_type, exc_val, exc_tb):
# Langfuse auto-flushes, no explicit close needed
pass
def generation(
self,
name: str,
model: str,
input: str | dict[str, Any],
output: str | dict[str, Any] | None = None,
usage: dict[str, int] | None = None,
metadata: dict[str, Any] | None = None,
):
"""
記錄一次 LLM generation
Args:
name: Generation 名稱 (e.g., "ollama_call", "gemini_fallback")
model: 模型名稱 (e.g., "qwen2.5:7b-instruct", "gemini-1.5-flash")
input: 輸入 prompt
output: 輸出結果
usage: Token 使用量 {"input": x, "output": y}
metadata: 額外 metadata
"""
if not self.trace:
return None
try:
gen = self.trace.generation(
name=name,
model=model,
input=input,
output=output,
usage=usage,
metadata=metadata or {},
)
return gen
except Exception as e:
logger.warning(
"langfuse_generation_failed",
error=str(e),
name=name,
model=model,
)
return None
def span(self, name: str, metadata: dict[str, Any] | None = None):
"""
記錄一個 span (非 LLM 操作)
Args:
name: Span 名稱
metadata: 額外 metadata
"""
if not self.trace:
return None
try:
return self.trace.span(name=name, metadata=metadata or {})
except Exception as e:
logger.warning("langfuse_span_failed", error=str(e), name=name)
return None
def score(
self,
name: str,
value: float,
comment: str | None = None,
):
"""
記錄評分 (用於 Prompt 品質追蹤)
Args:
name: 評分名稱 (e.g., "response_quality", "format_compliance")
value: 分數 (0.0 - 1.0)
comment: 評論
"""
if not self.trace:
return
try:
self.trace.score(
name=name,
value=value,
comment=comment,
)
except Exception as e:
logger.warning(
"langfuse_score_failed",
error=str(e),
name=name,
)
def langfuse_trace(name: str, metadata: dict[str, Any] | None = None):
"""
Langfuse trace context manager
Usage:
with langfuse_trace("openclaw_decision") as trace:
result = await call_llm(prompt)
trace.generation(name="ollama", model="qwen2.5:7b-instruct", ...)
"""
return LangfuseTraceContext(name=name, metadata=metadata)
@asynccontextmanager
async def langfuse_trace_async(name: str, metadata: dict[str, Any] | None = None):
"""
Async version of langfuse_trace
Usage:
async with langfuse_trace_async("openclaw_decision") as trace:
result = await call_llm(prompt)
"""
ctx = LangfuseTraceContext(name=name, metadata=metadata)
ctx.__enter__()
try:
yield ctx
finally:
ctx.__exit__(None, None, None)
def langfuse_observe(
name: str | None = None,
metadata: dict[str, Any] | None = None,
):
"""
Langfuse 裝飾器 - 自動追蹤函數執行
Usage:
@langfuse_observe(name="analyze_incident")
async def analyze_incident(incident_id: str):
...
"""
def decorator(func: Callable):
trace_name = name or func.__name__
@wraps(func)
async def async_wrapper(*args, **kwargs):
async with langfuse_trace_async(trace_name, metadata) as trace:
# Inject trace into kwargs if function accepts it
if "langfuse_trace" in func.__code__.co_varnames:
kwargs["langfuse_trace"] = trace
return await func(*args, **kwargs)
@wraps(func)
def sync_wrapper(*args, **kwargs):
with langfuse_trace(trace_name, metadata) as trace:
if "langfuse_trace" in func.__code__.co_varnames:
kwargs["langfuse_trace"] = trace
return func(*args, **kwargs)
# Return appropriate wrapper based on function type
import asyncio
if asyncio.iscoroutinefunction(func):
return async_wrapper
return sync_wrapper
return decorator
def flush_langfuse():
"""
手動 flush Langfuse (通常不需要，client 會自動 flush)
用於測試或確保資料送出
"""
client = get_langfuse()
if client:
try:
client.flush()
logger.debug("langfuse_flushed")
except Exception as e:
logger.warning("langfuse_flush_failed", error=str(e))

80
apps/api/src/services/openclaw.py
View File

@@ -33,6 +33,7 @@ from src.core.redis_client import get_redis
from src.models.ai import (
OpenClawDecision,
)
from src.services.langfuse_client import langfuse_trace
from src.services.signoz_client import GoldMetrics, get_signoz_client
from src.utils.timezone import now_taipei_iso
@@ -360,7 +361,7 @@ class OpenClawService:
response = await client.post(
f"{settings.OLLAMA_URL}/api/generate",
json={
"model": "llama3.2:3b", # 使用更大的模型提高品質
"model": "qwen2.5:7b-instruct", # 使用更大的模型提高品質
"prompt": prompt,
"stream": False,
"format": "json", # 強制 JSON 輸出
@@ -823,34 +824,75 @@ class OpenClawService:
若 MOCK_MODE=True，直接回傳模擬結果。
若所有 Provider 失敗，fallback 到 Mock。
Phase 15.1: 整合 Langfuse LLMOps 追蹤
"""
# Mock Mode: 開發測試用
if settings.MOCK_MODE:
logger.info("mock_mode_enabled", using="mock_llm")
return self._generate_mock_response(alert_context or {}, signoz_metrics), "mock", True
for provider in settings.AI_FALLBACK_ORDER:
logger.info("ai_provider_attempt", provider=provider)
# Phase 15.1: Langfuse 追蹤整合
with langfuse_trace(
"openclaw_fallback_chain",
metadata={
"prompt_length": len(prompt),
"fallback_order": settings.AI_FALLBACK_ORDER,
"alert_fingerprint": (alert_context or {}).get("fingerprint", "unknown"),
},
) as trace:
for provider in settings.AI_FALLBACK_ORDER:
logger.info("ai_provider_attempt", provider=provider)
if provider == "ollama":
response, success = await self._call_ollama(prompt)
elif provider == "gemini":
response, success = await self._call_gemini(prompt)
elif provider == "claude":
response, success = await self._call_claude(prompt)
else:
logger.warning("unknown_ai_provider", provider=provider)
continue
start_time = time.time()
model_name = self._get_model_name(provider)
if success:
logger.info("ai_provider_success", provider=provider)
return response, provider, True
if provider == "ollama":
response, success = await self._call_ollama(prompt)
elif provider == "gemini":
response, success = await self._call_gemini(prompt)
elif provider == "claude":
response, success = await self._call_claude(prompt)
else:
logger.warning("unknown_ai_provider", provider=provider)
continue
logger.warning("ai_provider_failed_fallback", provider=provider)
latency_ms = (time.time() - start_time) * 1000
# 所有 Provider 失敗時，fallback 到 Mock (優雅降級)
logger.warning("all_providers_failed_using_mock", fallback="mock_llm")
return self._generate_mock_response(alert_context or {}, signoz_metrics), "mock_fallback", True
# Langfuse: 記錄每次 LLM 呼叫
trace.generation(
name=f"{provider}_call",
model=model_name,
input=prompt[:500], # 截斷避免過長
output=response[:500] if success else f"ERROR: {response[:200]}",
metadata={
"success": success,
"latency_ms": round(latency_ms, 2),
"provider": provider,
},
)
if success:
logger.info("ai_provider_success", provider=provider, latency_ms=latency_ms)
# Langfuse: 記錄成功評分
trace.score(name="provider_success", value=1.0, comment=f"Success via {provider}")
return response, provider, True
logger.warning("ai_provider_failed_fallback", provider=provider, latency_ms=latency_ms)
# 所有 Provider 失敗時，fallback 到 Mock (優雅降級)
logger.warning("all_providers_failed_using_mock", fallback="mock_llm")
trace.score(name="provider_success", value=0.0, comment="All providers failed, using mock")
return self._generate_mock_response(alert_context or {}, signoz_metrics), "mock_fallback", True
def _get_model_name(self, provider: str) -> str:
"""取得 provider 對應的模型名稱"""
model_map = {
"ollama": "qwen2.5:7b-instruct",
"gemini": "gemini-1.5-flash",
"claude": "claude-3-haiku-20240307",
}
return model_map.get(provider, provider)
# =========================================================================
# Response Parsing (防禦性解析)

38
docs/LOGBOOK.md
View File

@@ -5,16 +5,42 @@
---
## 📍 當前狀態 (2026-03-26 00:30 台北)
## 📍 當前狀態 (2026-03-26 00:45 台北)
| 項目 | 狀態 |
|------|------|
| **當前 Phase** | **Phase 16 R2 ✅ + R3 ✅ + R4 ✅ 完成** |
| **當前 Phase** | **Phase 15.1 ✅ 全部完成** |
| **Day** | Day 8 |
| **驗證結束** | 2026-03-27 16:04 (48小時後) |
| **重大決策** | ✅ **USE_NEW_ENGINE=true 已啟用** |
| **CI/CD** | ✅ **已修復** (移除自毀指令 + .gitignore 強化) |
| **新規** | ✅ **絞殺者模式** + **封存策略** + **積木化強制執行** |
| **Phase 16** | ✅ R1-R4 全部完成 |
| **重大決策** | ✅ **模型升級 qwen2.5:7b-instruct** |
| **CI/CD** | ✅ **已修復** (31/31 核心測試通過) |
| **LLMOps** | ✅ **Langfuse 整合完成** |
### ✅ 2026-03-26 Phase 15.1 Langfuse 全部完成 (Day 8 凌晨 00:45)
**完成項目**:
| 項目 | 狀態 | 說明 |
|------|------|------|
| 部署 | ✅ | 192.168.0.110:3100 |
| 帳號 | ✅ | admin@awoooi.local |
| API Key | ✅ | pk-lf-xxx / sk-lf-xxx |
| K8s Secret | ✅ | awoooi-secrets patched |
| GitHub Secret | ✅ | LANGFUSE_* 已設定 |
| API 整合 | ✅ | langfuse_client.py + openclaw.py |
| 模型升級 | ✅ | llama3.2:3b → qwen2.5:7b-instruct |
| 架構審查 | ✅ | 模組化檢查通過 |
| 測試驗證 | ✅ | 31/31 核心測試通過 |
**新增檔案**:
- `src/services/langfuse_client.py` (250 行) - Langfuse Client 包裝
- `feedback_model_selection_strategy.md` - 模型選擇策略
**模型配置變更**:
- 預設模型: `qwen2.5:7b-instruct` (7B 參數，品質優先)
- 摘要模型: `llama3.2:3b` (速度優先)
---
### ✅ 2026-03-26 Phase 16 R3 Repository 層整合 (Day 8 深夜 00:30)

69
infra/langfuse/README.md Normal file
View File

@@ -0,0 +1,69 @@
# Langfuse Self-Hosted 部署
> Phase 15.1 LLMOps 觀測平台
## 部署資訊
| 項目 | 值 |
|------|---|
| 主機 | 192.168.0.110 (DevOps 金庫) |
| Port | 3100 |
| URL | http://192.168.0.110:3100 |
| 資料庫 | PostgreSQL 15 (內建) |
## 部署步驟
```bash
# 1. 複製到目標主機
scp -r infra/langfuse/ wooo@192.168.0.110:/opt/langfuse/
# 2. SSH 到目標主機
ssh wooo@192.168.0.110
# 3. 啟動服務
cd /opt/langfuse
docker-compose up -d
# 4. 檢查狀態
docker-compose ps
docker-compose logs -f langfuse
```
## 首次設定
1. 開啟 http://192.168.0.110:3100
2. 註冊管理員帳號
3. 建立 Project: `awoooi-openclaw`
4. 生成 API Key (Public + Secret)
5. 將 API Key 加入 K8s Secret
## API 整合
```python
from langfuse import Langfuse
langfuse = Langfuse(
public_key="pk-xxx",
secret_key="sk-xxx",
host="http://192.168.0.110:3100"
)
```
## 備份
```bash
# 備份資料庫
docker exec langfuse-db pg_dump -U langfuse langfuse > backup.sql
# 還原
cat backup.sql | docker exec -i langfuse-db psql -U langfuse langfuse
```
## 回滾
```bash
cd /opt/langfuse
docker-compose down
# 還原備份後
docker-compose up -d
```

71
infra/langfuse/docker-compose.yml Normal file
View File

@@ -0,0 +1,71 @@
# =============================================================================
# Langfuse Self-Hosted - LLMOps 觀測平台
# Phase 15.1 (2026-03-26)
# 部署位置: 192.168.0.110 (DevOps 金庫)
# =============================================================================
#
# 部署指令:
# scp -r infra/langfuse/ wooo@192.168.0.110:/opt/langfuse/
# ssh wooo@192.168.0.110 "cd /opt/langfuse && docker-compose up -d"
#
# 存取:
# http://192.168.0.110:3100
#
# =============================================================================
services:
langfuse:
image: langfuse/langfuse:2
container_name: langfuse
restart: unless-stopped
ports:
- "3100:3000"
environment:
# === Database ===
DATABASE_URL: postgresql://langfuse:langfuse_secret@langfuse-db:5432/langfuse
# === Auth ===
NEXTAUTH_SECRET: ${NEXTAUTH_SECRET:-langfuse-secret-change-me-in-production}
NEXTAUTH_URL: http://192.168.0.110:3100
# === Security (Langfuse v2 必需) ===
SALT: ${SALT:-langfuse-salt-change-me-in-production}
ENCRYPTION_KEY: ${ENCRYPTION_KEY:-0000000000000000000000000000000000000000000000000000000000000000}
# === Telemetry (可選關閉) ===
TELEMETRY_ENABLED: "false"
# === Self-hosted 模式 ===
LANGFUSE_ENABLE_EXPERIMENTAL_FEATURES: "true"
depends_on:
langfuse-db:
condition: service_healthy
networks:
- langfuse-net
langfuse-db:
image: postgres:15-alpine
container_name: langfuse-db
restart: unless-stopped
environment:
POSTGRES_USER: langfuse
POSTGRES_PASSWORD: langfuse_secret
POSTGRES_DB: langfuse
volumes:
- langfuse-db-data:/var/lib/postgresql/data
healthcheck:
test: ["CMD-SHELL", "pg_isready -U langfuse"]
interval: 5s
timeout: 5s
retries: 5
networks:
- langfuse-net
volumes:
langfuse-db-data:
driver: local
networks:
langfuse-net:
driver: bridge

6
k8s/awoooi-prod/03-secrets.example.yaml
View File

@@ -56,3 +56,9 @@ stringData:
# ============================================================================
JWT_SECRET: "CHANGE_ME_TO_RANDOM_STRING"
JWT_ALGORITHM: "HS256"
# ============================================================================
# Phase 15.1: Langfuse LLMOps (192.168.0.110:3100)
# ============================================================================
LANGFUSE_PUBLIC_KEY: "CHANGE_ME"
LANGFUSE_SECRET_KEY: "CHANGE_ME"

1
k8s/awoooi-prod/04-configmap.yaml
View File

@@ -18,6 +18,7 @@ data:
OPENCLAW_URL: "http://192.168.0.188:8088"
KALI_SCANNER_URL: "http://192.168.0.112:8080"
SIGNOZ_URL: "http://192.168.0.188:3301"
LANGFUSE_URL: "http://192.168.0.110:3100"
# OTEL 可觀測性 (P0 核心神經)
# 注意: gRPC endpoint 不需要 http:// 前綴