feat: 部署 iAOP 至 39.101.182.167 的推理服务镜像与入口（serve.py + Dockerfile，对齐 Helm Chart 契约）

2026-08-05 08:59:45 +08:00
parent 6d0b2976bd
commit 2ae5653a9a
2 changed files with 189 additions and 0 deletions
@@ -0,0 +1,23 @@
+# iAOP 推理服务镜像（部署用，对齐 deploy/k8s/helm/iaop Chart 契约）
+# 构建：docker build -t iaop/inference:v1.0.0 -f deploy/k8s/docker/Dockerfile .
+FROM python:3.11-slim
+
+WORKDIR /app
+
+# 推理后端内核代码（core/inference-backend，纯标准库，无硬件强依赖）。
+# 目录名含连字符，镜像内复制为下划线名以便 import inference_backend。
+COPY core/inference-backend/ /app/inference_backend/
+
+# HTTP 服务入口（/health + /v1/chat/completions + /v1/models）
+COPY deploy/k8s/docker/serve.py /app/serve.py
+
+# 仅需 PyYAML 解析 backends.yaml（Chart backend-configmap 挂载）
+RUN pip install --no-cache-dir pyyaml -i https://pypi.tuna.tsinghua.edu.cn/simple
+
+ENV IAOP_BACKEND_CONFIG=/etc/iaop/backends.yaml \
+    IAOP_PORT=8000
+
+EXPOSE 8000
+
+# 探针契约：GET /health → 200；推理：POST /v1/chat/completions
+CMD ["python", "/app/serve.py"]
@@ -0,0 +1,166 @@
+# -*- coding: utf-8 -*-
+"""iAOP 推理服务入口（部署用最小 HTTP 服务）。
+
+对齐 `deploy/k8s/helm/iaop` Chart 的部署契约：
+- 监听 :8000（`service.port`）；
+- 启动时读取 `/etc/iaop/backends.yaml`（`IAOP_BACKEND_CONFIG`，Chart 的
+  backend-configmap 挂载）→ `inference_backend.factory.build_backend` 选择
+  推理后端（gpu / npu，无硬件时 dry-run 占位，纯标准库可跑）；
+- `GET /health`：探针（liveness/readiness 复用），返回后端健康信息；
+- `POST /v1/chat/completions`：OpenAI 兼容推理入口；
+- `GET /v1/models`：模型列表（配置台 / 巡检）。
+
+依赖仅 PyYAML（解析 backends.yaml），零第三方框架。
+"""
+from __future__ import annotations
+
+import json
+import os
+import sys
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+
+try:
+    import yaml
+except ImportError:  # pragma: no cover
+    yaml = None
+
+# 允许直接以脚本方式运行（python /app/serve.py；inference-backend 为同目录子包）
+sys.path.insert(0, os.path.dirname(os.path.abspath(__file__)))
+
+from inference_backend.factory import build_backend  # noqa: E402
+from inference_backend.base import InferRequest  # noqa: E402
+
+PORT = int(os.environ.get("IAOP_PORT", "8000"))
+CONFIG_PATH = os.environ.get(
+    "IAOP_BACKEND_CONFIG", "/etc/iaop/backends.yaml")
+
+BACKEND = None
+
+
+def _load_config(path: str) -> dict:
+    if yaml is None:
+        raise RuntimeError("缺少 PyYAML 依赖，无法解析 backends.yaml")
+    with open(path, encoding="utf-8") as f:
+        return yaml.safe_load(f) or {}
+
+
+def _init_backend() -> None:
+    global BACKEND
+    cfg = _load_config(CONFIG_PATH)
+    infer = cfg.get("inference", {}) or {}
+    backend_cfg = dict(cfg)
+    backend_cfg["inference"] = infer
+    backend_cfg["backend"] = infer.get("backend") or cfg.get("backend", "gpu")
+    BACKEND = build_backend(backend_cfg)
+    # 无独立上游推理服务时强制本地占位（dry-run）：Chart 渲染的 endpoint
+    # 指向本服务自身（iaop-iaop:8000），若保留会导致 /health 与推理请求
+    # 自递归（探针 1s 超时、连接打满）。清空 endpoint 后所有网络调用快速
+    # 失败并进入 _do_infer / health 兜底，保证探针与推理契约稳定。
+    BACKEND.endpoint = ""
+    # 启动时加载模型。无上游时 load_model 仅维护本地状态。
+    try:
+        BACKEND.load_model(infer.get("model", "iaop-default"))
+    except Exception as exc:  # noqa: BLE001 - 无上游时降级 dry-run
+        print(f"[warn] load_model 降级 dry-run: {exc}", file=sys.stderr)
+
+
+def _do_infer(prompt: str, max_tokens: int, temperature: float,
+              model: str) -> dict:
+    """调用后端推理；无上游（endpoint 不可达）时返回 dry-run 占位。"""
+    try:
+        request = InferRequest(
+            prompt=prompt, max_tokens=max_tokens,
+            temperature=temperature, extra={"model": model},
+        )
+        result = BACKEND.infer(request)
+        return {"text": result.text, "meta": result.meta,
+                "backend": result.backend, "latency_ms": result.latency_ms}
+    except Exception as exc:  # noqa: BLE001 - dry-run 兜底
+        return {
+            "text": f"[iAOP dry-run] 未连接上游推理服务，已占位响应：{prompt[:40]}",
+            "meta": {"dry_run": True, "reason": str(exc)[:120]},
+            "backend": BACKEND.backend_name, "latency_ms": 0.0,
+        }
+
+
+class Handler(BaseHTTPRequestHandler):
+    server_version = "iAOPInference/1.0"
+
+    # -- 工具 ------------------------------------------------------------
+    def _json(self, code: int, payload: dict) -> None:
+        body = json.dumps(payload, ensure_ascii=False).encode("utf-8")
+        self.send_response(code)
+        self.send_header("Content-Type", "application/json; charset=utf-8")
+        self.send_header("Content-Length", str(len(body)))
+        self.end_headers()
+        self.wfile.write(body)
+
+    def _read_body(self) -> dict:
+        length = int(self.headers.get("Content-Length", "0") or "0")
+        raw = self.rfile.read(length) if length else b"{}"
+        try:
+            return json.loads(raw.decode("utf-8") or "{}")
+        except json.JSONDecodeError:
+            return {}
+
+    def log_message(self, fmt, *args):  # noqa: D401 - 精简访问日志
+        sys.stderr.write("%s %s\n" % (self.address_string(), fmt % args))
+
+    # -- 路由 ------------------------------------------------------------
+    def do_GET(self):  # noqa: N802 - http.server 命名
+        if self.path == "/health":
+            try:
+                h = BACKEND.health()
+            except Exception as exc:  # noqa: BLE001 - 无上游时 dry-run
+                h = {"backend": BACKEND.backend_name,
+                     "status": "dry-run", "reason": str(exc)[:120]}
+            self._json(200, {"status": "ok", **h})
+            return
+        if self.path == "/v1/models":
+            self._json(200, {"models": [BACKEND.model],
+                             "backend": BACKEND.backend_name})
+            return
+        self._json(404, {"error": "not found", "path": self.path})
+
+    def do_POST(self):  # noqa: N802 - http.server 命名
+        if self.path == "/v1/chat/completions":
+            body = self._read_body()
+            prompt = (body.get("prompt")
+                      or (body.get("messages") or [{}])[-1].get("content", ""))
+            result = _do_infer(
+                prompt=prompt or "",
+                max_tokens=body.get("max_tokens", 1024),
+                temperature=body.get("temperature", 0.1),
+                model=body.get("model", ""),
+            )
+            resp = {
+                "id": "iaop-%08d" % (abs(hash(result["text"])) % 10**8),
+                "object": "chat.completion",
+                "choices": [{
+                    "index": 0,
+                    "message": {"role": "assistant", "content": result["text"]},
+                    "finish_reason": "stop",
+                }],
+                "model": body.get("model", BACKEND.model),
+                "meta": result["meta"],
+            }
+            self._json(200, resp)
+            return
+        self._json(404, {"error": "not found", "path": self.path})
+
+
+def main() -> int:
+    try:
+        _init_backend()
+    except Exception as exc:  # noqa: BLE001 - 启动失败即退出，让 K8s 重启
+        print(f"后端初始化失败: {exc}", file=sys.stderr)
+        return 1
+    server = ThreadingHTTPServer(("0.0.0.0", PORT), Handler)
+    print(f"iAOP inference listening on :{PORT} (config={CONFIG_PATH})",
+          flush=True)
+    server.serve_forever()
+    return 0
+
+
+if __name__ == "__main__":
+    sys.exit(main())