magpie-gemini-web 0.3.3 → 0.4.1

package/native-task/bridge/gemini_native_bridge/server.py

@@ -0,0 +1,441 @@
+"""HTTP server implementing local OpenAI-compatible endpoints."""
+from __future__ import annotations
+
+import hmac
+import json
+import re
+import socket
+import threading
+import time
+import urllib.parse
+from http.server import BaseHTTPRequestHandler, ThreadingHTTPServer
+from typing import Any
+
+from .config import RuntimeConfig
+from .contract import BridgeError, MAX_REQUEST_BYTES, json_loads, parse_request
+from .protocol import MODEL_ID
+from .state import Manager
+
+DIGITS = re.compile(r"^[0-9]+$")
+
+
+class BridgeHandler(BaseHTTPRequestHandler):
+    protocol_version = "HTTP/1.1"
+
+    def setup(self):
+        super().setup()
+        if hasattr(self, "connection") and hasattr(self.connection, "settimeout"):
+            self.connection.settimeout(15.0)
+
+    def log_message(self, format: str, *args: Any):
+        pass
+
+    @property
+    def server_instance(self) -> BridgeServer:
+        return self.server  # type: ignore
+
+    def _send_error_json(self, status: int, code: str, message: str, sign_in_kept: bool = True):
+        body = json.dumps({
+            "error": {
+                "code": code,
+                "type": "invalid_request_error" if status < 500 else "api_error",
+                "message": message,
+                "param": None,
+            }
+        }).encode("utf-8")
+        try:
+            self.send_response(status)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(body)))
+            self.send_header("Connection", "close")
+            if sign_in_kept:
+                self.send_header("X-Magpie-Sign-In", "kept")
+            else:
+                self.send_header("X-Magpie-Sign-In", "expired")
+            self.end_headers()
+            self.wfile.write(body)
+            self.wfile.flush()
+        except (BrokenPipeError, ConnectionResetError, OSError):
+            pass
+
+    def _check_auth(self) -> bool:
+        auth_hdrs = self.headers.get_all("Authorization")
+        if not auth_hdrs or len(auth_hdrs) > 1:
+            self._send_error_json(401, "unauthorized", "Missing or duplicate Authorization header.", sign_in_kept=False)
+            return False
+
+        auth_hdr = auth_hdrs[0]
+        if not auth_hdr.startswith("Bearer "):
+            self._send_error_json(401, "unauthorized", "Authorization header must use Bearer scheme.", sign_in_kept=False)
+            return False
+
+        token = auth_hdr[7:].strip()
+        expected = self.server_instance.config.api_key
+        if not hmac.compare_digest(token.encode("utf-8"), expected.encode("utf-8")):
+            self._send_error_json(401, "unauthorized", "Invalid local API key.", sign_in_kept=False)
+            return False
+        return True
+
+    def _read_body_json(self) -> dict | None:
+        te_hdrs = self.headers.get_all("Transfer-Encoding")
+        if te_hdrs:
+            self._send_error_json(400, "unsupported_transfer_encoding", "Transfer-Encoding is not supported.")
+            return None
+
+        cl_hdrs = self.headers.get_all("Content-Length")
+        if not cl_hdrs:
+            self._send_error_json(400, "missing_content_length", "Content-Length header is required.")
+            return None
+        if len(cl_hdrs) > 1:
+            self._send_error_json(400, "duplicate_content_length", "Duplicate Content-Length headers are not permitted.")
+            return None
+
+        cl_str = cl_hdrs[0].strip()
+        if len(cl_str) > 7:
+            self._send_error_json(400, "request_too_large", f"Content-Length exceeds {MAX_REQUEST_BYTES} bytes limit.")
+            return None
+
+        if not DIGITS.fullmatch(cl_str):
+            self._send_error_json(400, "invalid_content_length", "Content-Length must be an unsigned integer.")
+            return None
+
+        length = int(cl_str)
+        if length > MAX_REQUEST_BYTES:
+            self._send_error_json(400, "request_too_large", f"Content-Length exceeds {MAX_REQUEST_BYTES} bytes limit.")
+            return None
+
+        try:
+            raw = self.rfile.read(length)
+            if len(raw) != length:
+                self._send_error_json(400, "truncated_body", "Request body was truncated.")
+                return None
+            parsed = json_loads(raw.decode("utf-8"))
+            if parsed is None or not isinstance(parsed, dict):
+                self._send_error_json(400, "invalid_json", "Request body must be valid JSON.")
+                return None
+            return parsed
+        except UnicodeError:
+            self._send_error_json(400, "invalid_utf8", "Request body must be valid UTF-8.")
+            return None
+        except Exception:
+            self._send_error_json(400, "invalid_json", "Request body must be valid JSON.")
+            return None
+
+    def do_GET(self):
+        if not self._check_auth():
+            return
+
+        parsed = urllib.parse.urlparse(self.path)
+        path = parsed.path
+
+        if path == "/health":
+            body = json.dumps({
+                "service": "gemini-native-task",
+                "version": "0.1.0",
+                "instance_id": self.server_instance.config.instance_id,
+                "active_tasks": 1 if self.server_instance.manager._active_task_id else 0,
+            }).encode("utf-8")
+            self.send_response(200)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(body)))
+            self.send_header("Connection", "close")
+            self.end_headers()
+            self.wfile.write(body)
+            self.wfile.flush()
+            return
+
+        if path == "/v1/models":
+            body = json.dumps({
+                "object": "list",
+                "data": [{
+                    "id": MODEL_ID,
+                    "object": "model",
+                    "created": self.server_instance.created_time,
+                    "owned_by": "gemini-native-task",
+                }],
+            }).encode("utf-8")
+            self.send_response(200)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(body)))
+            self.send_header("Connection", "close")
+            self.end_headers()
+            self.wfile.write(body)
+            self.wfile.flush()
+            return
+
+        if path.startswith("/v1/native/sessions/"):
+            raw_session = path[len("/v1/native/sessions/"):]
+            session_id = urllib.parse.unquote(raw_session)
+            try:
+                status_data = self.server_instance.manager.status(session_id)
+                body = json.dumps(status_data).encode("utf-8")
+                self.send_response(200)
+                self.send_header("Content-Type", "application/json")
+                self.send_header("Content-Length", str(len(body)))
+                self.send_header("Connection", "close")
+                self.end_headers()
+                self.wfile.write(body)
+                self.wfile.flush()
+            except BridgeError as be:
+                self._send_error_json(be.status, be.code, be.message)
+            return
+
+        self._send_error_json(404, "not_found", f"Path {path} not found.")
+
+    def do_POST(self):
+        if not self._check_auth():
+            return
+
+        parsed = urllib.parse.urlparse(self.path)
+        path = parsed.path
+
+        if path == "/v1/native/shutdown":
+            body_dict = self._read_body_json()
+            if body_dict is None:
+                return
+
+            body = json.dumps({"status": "shutting_down"}).encode("utf-8")
+            self.send_response(202)
+            self.send_header("Content-Type", "application/json")
+            self.send_header("Content-Length", str(len(body)))
+            self.send_header("Connection", "close")
+            self.end_headers()
+            self.wfile.write(body)
+            self.wfile.flush()
+            threading.Thread(target=self.server_instance.stop, daemon=True).start()
+            return
+
+        if path.startswith("/v1/native/sessions/") and path.endswith("/cancel"):
+            raw_session = path[len("/v1/native/sessions/"):-len("/cancel")]
+            session_id = urllib.parse.unquote(raw_session)
+            body_dict = self._read_body_json()
+            if body_dict is None:
+                return
+
+            try:
+                res = self.server_instance.manager.cancel(session_id)
+                body = json.dumps(res).encode("utf-8")
+                self.send_response(202)
+                self.send_header("Content-Type", "application/json")
+                self.send_header("Content-Length", str(len(body)))
+                self.send_header("Connection", "close")
+                self.end_headers()
+                self.wfile.write(body)
+                self.wfile.flush()
+            except BridgeError as be:
+                self._send_error_json(be.status, be.code, be.message)
+            return
+
+        if path == "/v1/chat/completions":
+            gemini_sessions = self.headers.get_all("X-Gemini-Session") or []
+            magpie_sessions = self.headers.get_all("X-Magpie-Session") or []
+
+            if len(gemini_sessions) > 1 or len(magpie_sessions) > 1:
+                self._send_error_json(400, "duplicate_session_header", "Duplicate session headers provided.")
+                return
+
+            sess_gemini = gemini_sessions[0] if gemini_sessions else None
+            sess_magpie = magpie_sessions[0] if magpie_sessions else None
+
+            if sess_gemini and sess_magpie and sess_gemini != sess_magpie:
+                self._send_error_json(400, "session_header_conflict", "X-Gemini-Session and X-Magpie-Session must match.")
+                return
+
+            session_id = sess_gemini or sess_magpie
+            if not session_id:
+                self._send_error_json(400, "missing_session", "Provide a stable session ID via X-Gemini-Session header.")
+                return
+
+            body_dict = self._read_body_json()
+            if body_dict is None:
+                return
+
+            try:
+                comp_req = parse_request(body_dict)
+                ticket = self.server_instance.manager.submit(session_id, comp_req)
+            except BridgeError as be:
+                self._send_error_json(be.status, be.code, be.message)
+                return
+            except Exception:
+                self._send_error_json(409, "submission_failed", "Failed to submit completion request.")
+                return
+
+            if not comp_req.stream:
+                try:
+                    result = ticket.result(timeout=self.server_instance.config.task_timeout)
+                    resp_bytes = json.dumps(result, ensure_ascii=False).encode("utf-8")
+                    self.send_response(200)
+                    self.send_header("Content-Type", "application/json")
+                    self.send_header("Content-Length", str(len(resp_bytes)))
+                    self.send_header("Connection", "close")
+                    self.end_headers()
+                    self.wfile.write(resp_bytes)
+                    self.wfile.flush()
+                except BridgeError as be:
+                    self._send_error_json(be.status, be.code, be.message)
+                except TimeoutError:
+                    self.server_instance.manager.abandon_delivery(session_id, ticket.task_id, comp_req.digest)
+                    self._send_error_json(409, "task_timeout", "Task timed out waiting for completion.")
+                except (BrokenPipeError, ConnectionResetError, OSError):
+                    self.server_instance.manager.abandon_delivery(session_id, ticket.task_id, comp_req.digest)
+                except Exception:
+                    self._send_error_json(409, "execution_failed", "Task execution failed.")
+                return
+
+            # Streaming response (SSE)
+            try:
+                self.send_response(200)
+                self.send_header("Content-Type", "text/event-stream")
+                self.send_header("Cache-Control", "no-cache")
+                self.send_header("Connection", "close")
+                self.send_header("X-Magpie-Sign-In", "kept")
+                self.end_headers()
+            except (BrokenPipeError, ConnectionResetError, OSError):
+                self.server_instance.manager.abandon_delivery(session_id, ticket.task_id, comp_req.digest)
+                return
+
+            disconnected = False
+            stream_deadline = time.monotonic() + self.server_instance.config.task_timeout
+            interval = self.server_instance.heartbeat_interval
+
+            while not ticket.event.wait(timeout=interval):
+                if time.monotonic() >= stream_deadline:
+                    self.server_instance.manager.abandon_delivery(session_id, ticket.task_id, comp_req.digest)
+                    try:
+                        err_chunk = f"data: {json.dumps({'error': {'code': 'task_timeout', 'type': 'api_error', 'message': 'Task timed out.'}})}\n\ndata: [DONE]\n\n".encode("utf-8")
+                        self.wfile.write(err_chunk)
+                        self.wfile.flush()
+                    except (BrokenPipeError, ConnectionResetError, OSError):
+                        pass
+                    return
+
+                try:
+                    self.wfile.write(b": keepalive\n\n")
+                    self.wfile.flush()
+                except (BrokenPipeError, ConnectionResetError, OSError):
+                    disconnected = True
+                    break
+
+            if disconnected:
+                self.server_instance.manager.abandon_delivery(session_id, ticket.task_id, comp_req.digest)
+                return
+
+            if ticket.error is not None:
+                err_chunk = f"data: {json.dumps(ticket.error.body())}\n\ndata: [DONE]\n\n".encode("utf-8")
+                try:
+                    self.wfile.write(err_chunk)
+                    self.wfile.flush()
+                except (BrokenPipeError, ConnectionResetError, OSError):
+                    self.server_instance.manager.abandon_delivery(session_id, ticket.task_id, comp_req.digest)
+                return
+
+            resp = ticket.response
+            if resp is None:
+                err_chunk = f"data: {json.dumps({'error': {'code': 'no_response', 'type': 'api_error', 'message': 'Empty ticket response.'}})}\n\ndata: [DONE]\n\n".encode("utf-8")
+                try:
+                    self.wfile.write(err_chunk)
+                    self.wfile.flush()
+                except (BrokenPipeError, ConnectionResetError, OSError):
+                    self.server_instance.manager.abandon_delivery(session_id, ticket.task_id, comp_req.digest)
+                return
+
+            resp_id = resp["id"]
+            created = resp["created"]
+            choice = resp["choices"][0]
+            finish_reason = choice["finish_reason"]
+            msg = choice["message"]
+
+            chunks = [
+                {
+                    "id": resp_id,
+                    "object": "chat.completion.chunk",
+                    "created": created,
+                    "model": MODEL_ID,
+                    "choices": [{"index": 0, "delta": {"role": "assistant"}, "finish_reason": None}],
+                }
+            ]
+
+            if msg.get("tool_calls"):
+                tc = msg["tool_calls"][0]
+                chunks.append({
+                    "id": resp_id,
+                    "object": "chat.completion.chunk",
+                    "created": created,
+                    "model": MODEL_ID,
+                    "choices": [{
+                        "index": 0,
+                        "delta": {
+                            "tool_calls": [{
+                                "index": 0,
+                                "id": tc["id"],
+                                "type": "function",
+                                "function": {
+                                    "name": tc["function"]["name"],
+                                    "arguments": tc["function"]["arguments"],
+                                },
+                            }]
+                        },
+                        "finish_reason": None,
+                    }],
+                })
+            elif msg.get("content"):
+                chunks.append({
+                    "id": resp_id,
+                    "object": "chat.completion.chunk",
+                    "created": created,
+                    "model": MODEL_ID,
+                    "choices": [{"index": 0, "delta": {"content": msg["content"]}, "finish_reason": None}],
+                })
+
+            chunks.append({
+                "id": resp_id,
+                "object": "chat.completion.chunk",
+                "created": created,
+                "model": MODEL_ID,
+                "choices": [{"index": 0, "delta": {}, "finish_reason": finish_reason}],
+            })
+
+            try:
+                for chunk in chunks:
+                    self.wfile.write(f"data: {json.dumps(chunk, ensure_ascii=False)}\n\n".encode("utf-8"))
+                self.wfile.write(b"data: [DONE]\n\n")
+                self.wfile.flush()
+            except (BrokenPipeError, ConnectionResetError, OSError):
+                self.server_instance.manager.abandon_delivery(session_id, ticket.task_id, comp_req.digest)
+            return
+
+        self._send_error_json(404, "not_found", f"Path {path} not found.")
+
+
+class BridgeServer(ThreadingHTTPServer):
+    def __init__(
+        self,
+        config: RuntimeConfig,
+        manager: Manager | None = None,
+        bind_port: int | None = None,
+        heartbeat_interval: float = 10.0,
+    ):
+        self.config = config
+        self.manager = manager or Manager(config)
+        self.created_time = int(time.time())
+        self.heartbeat_interval = heartbeat_interval
+        self._stopped_event = threading.Event()
+        self._stop_lock = threading.Lock()
+        port = config.port if bind_port is None else bind_port
+        super().__init__(("127.0.0.1", port), BridgeHandler)
+        self.port = self.server_address[1]
+
+    def handle_error(self, request, client_address):
+        # Override to prevent emitting tracebacks to stdout/stderr
+        pass
+
+    def stop(self):
+        with self._stop_lock:
+            if self._stopped_event.is_set():
+                return
+            self._stopped_event.set()
+        try:
+            self.manager.close()
+        finally:
+            self.shutdown()
+            self.server_close()