| import asyncio
|
| import websockets
|
| import json
|
| import os
|
| import uuid
|
| import time
|
| import jwt
|
| from typing import Dict, Any, Optional, List
|
| import numpy as np
|
| from fastapi import FastAPI, WebSocket, HTTPException, Depends, Request, Response
|
| from fastapi.responses import HTMLResponse, JSONResponse
|
| from fastapi.middleware.cors import CORSMiddleware
|
| from fastapi.security import HTTPBearer, HTTPAuthorizationCredentials
|
| from datetime import datetime, timedelta
|
| import hashlib
|
| import gzip
|
| import base64
|
| import logging
|
| from pydantic import BaseModel
|
| from http_storage import LocalStorage
|
| from config import get_db_url
|
|
|
|
|
| logging.basicConfig(
|
| level=logging.INFO,
|
| format='%(asctime)s - %(levelname)s - %(message)s'
|
| )
|
|
|
|
|
| app = FastAPI(
|
| title="Virtual GPU Server",
|
| description="HTTP and WebSocket API for Virtual GPU v2",
|
| version="2.0.0"
|
| )
|
|
|
|
|
| app.add_middleware(
|
| CORSMiddleware,
|
| allow_origins=["*"],
|
| allow_credentials=True,
|
| allow_methods=["*"],
|
| allow_headers=["*"],
|
| )
|
|
|
|
|
| JWT_SECRET = "virtual_gpu_secret_key_2025"
|
| JWT_ALGORITHM = "HS256"
|
| JWT_EXPIRATION_HOURS = 24
|
|
|
|
|
| security = HTTPBearer()
|
|
|
|
|
| class SessionCreateRequest(BaseModel):
|
| client_id: Optional[str] = None
|
| resource_limits: Optional[Dict[str, Any]] = None
|
|
|
| class SessionResponse(BaseModel):
|
| session_token: str
|
| session_id: str
|
| expires_at: datetime
|
|
|
| class VRAMWriteRequest(BaseModel):
|
| data: List[Any]
|
| metadata: Optional[Dict[str, Any]] = None
|
| model_size: Optional[int] = None
|
|
|
| class VRAMResponse(BaseModel):
|
| status: str
|
| message: Optional[str] = None
|
| data: Optional[List[Any]] = None
|
| metadata: Optional[Dict[str, Any]] = None
|
| source: Optional[str] = None
|
|
|
| class StateRequest(BaseModel):
|
| data: Dict[str, Any]
|
| timestamp: Optional[float] = None
|
|
|
| class StateResponse(BaseModel):
|
| status: str
|
| message: Optional[str] = None
|
| data: Optional[Dict[str, Any]] = None
|
| source: Optional[str] = None
|
|
|
| class CacheRequest(BaseModel):
|
| data: Any
|
| ttl: Optional[int] = None
|
|
|
| class CacheResponse(BaseModel):
|
| status: str
|
| message: Optional[str] = None
|
| data: Optional[Any] = None
|
| source: Optional[str] = None
|
|
|
| class ModelLoadRequest(BaseModel):
|
| model_data: Optional[Dict[str, Any]] = None
|
| model_path: Optional[str] = None
|
| model_hash: Optional[str] = None
|
|
|
| class ModelInferenceRequest(BaseModel):
|
| input_data: List[Any]
|
| batch_size: Optional[int] = None
|
|
|
| class ErrorResponse(BaseModel):
|
| status: str
|
| error_code: str
|
| message: str
|
| details: Optional[Dict[str, Any]] = None
|
| retry_after: Optional[int] = None
|
| request_id: str
|
|
|
| class VirtualGPUServer:
|
| def __init__(self):
|
|
|
| self.storage = LocalStorage(db_url=get_db_url())
|
| if not self.storage.wait_for_connection(timeout=30):
|
| raise RuntimeError("Could not connect to storage backend")
|
|
|
|
|
| self.model_registry = {}
|
| self.vram_path.mkdir(parents=True, exist_ok=True)
|
| self.state_path.mkdir(parents=True, exist_ok=True)
|
| self.cache_path.mkdir(parents=True, exist_ok=True)
|
| self.models_path.mkdir(parents=True, exist_ok=True)
|
|
|
|
|
| self.vram_cache: Dict[str, Any] = {}
|
| self.state_cache: Dict[str, Any] = {}
|
| self.memory_cache: Dict[str, Any] = {}
|
| self.model_cache: Dict[str, Any] = {}
|
|
|
|
|
| self.http_sessions: Dict[str, Dict[str, Any]] = {}
|
|
|
|
|
| self.active_connections: Dict[str, WebSocket] = {}
|
| self.active_sessions: Dict[str, Dict[str, Any]] = {}
|
| self.heartbeat_interval = 5
|
| self.connection_timeout = 30
|
|
|
|
|
| self.ops_counter = 0
|
| self.start_time = time.time()
|
| self.request_counter = 0
|
|
|
| def _make_json_serializable(self, obj):
|
| """Convert non-JSON-serializable objects to serializable format"""
|
| if isinstance(obj, dict):
|
| return {k: self._make_json_serializable(v) for k, v in obj.items()}
|
| elif isinstance(obj, list):
|
| return [self._make_json_serializable(i) for i in obj]
|
| elif isinstance(obj, tuple):
|
| return list(obj)
|
| elif isinstance(obj, (np.ndarray, np.generic)):
|
| return obj.tolist()
|
| elif isinstance(obj, (Path, uuid.UUID)):
|
| return str(obj)
|
| elif hasattr(obj, '__dict__'):
|
|
|
| return self._make_json_serializable(obj.__dict__)
|
| elif isinstance(obj, (int, float, str, bool, type(None))):
|
| return obj
|
| else:
|
|
|
| return str(obj)
|
|
|
| def create_session_token(self, session_id: str, client_id: str = None, resource_limits: Dict[str, Any] = None) -> str:
|
| """Create a JWT session token"""
|
| payload = {
|
| "session_id": session_id,
|
| "client_id": client_id or "anonymous",
|
| "resource_limits": resource_limits or {},
|
| "created_at": time.time(),
|
| "expires_at": time.time() + (JWT_EXPIRATION_HOURS * 3600)
|
| }
|
| return jwt.encode(payload, JWT_SECRET, algorithm=JWT_ALGORITHM)
|
|
|
| def verify_session_token(self, token: str) -> Dict[str, Any]:
|
| """Verify and decode a JWT session token"""
|
| try:
|
| payload = jwt.decode(token, JWT_SECRET, algorithms=[JWT_ALGORITHM])
|
| if payload["expires_at"] < time.time():
|
| raise HTTPException(status_code=401, detail="Session token expired")
|
| return payload
|
| except jwt.InvalidTokenError:
|
| raise HTTPException(status_code=401, detail="Invalid session token")
|
|
|
| def generate_request_id(self) -> str:
|
| """Generate a unique request ID"""
|
| self.request_counter += 1
|
| return f"req_{int(time.time())}_{self.request_counter}"
|
|
|
| def compress_data(self, data: bytes) -> bytes:
|
| """Compress data using gzip"""
|
| return gzip.compress(data)
|
|
|
| def decompress_data(self, data: bytes) -> bytes:
|
| """Decompress gzip data"""
|
| return gzip.decompress(data)
|
|
|
| async def handle_vram_operation(self, operation: dict) -> dict:
|
| """Handle VRAM read/write operations (preserved from WebSocket implementation)"""
|
| try:
|
| op_type = operation.get('type')
|
| if not op_type:
|
| raise ValueError("Missing operation type")
|
|
|
| block_id = operation.get('block_id')
|
| if not block_id:
|
| raise ValueError("Missing block_id")
|
|
|
| data = operation.get('data')
|
| if data and isinstance(data, (dict, list)):
|
| data = self._make_json_serializable(data)
|
|
|
| if op_type == 'write':
|
| if data is None:
|
| raise ValueError("Missing data for write operation")
|
|
|
|
|
| arr_data = np.array(data)
|
|
|
| metadata = operation.get('metadata', {})
|
| metadata.update({
|
| 'dtype': str(arr_data.dtype),
|
| 'shape': arr_data.shape
|
| })
|
|
|
|
|
| self.storage.store_tensor(block_id, arr_data, metadata)
|
|
|
| self.vram_cache[block_id] = arr_data
|
|
|
| return {'status': 'success', 'message': f'Block {block_id} written'}
|
|
|
| if op_type == 'read':
|
| if block_id in self.vram_cache:
|
|
|
| metadata = self.storage.get_tensor_metadata(block_id)
|
|
|
| return {
|
| 'status': 'success',
|
| 'data': self.vram_cache[block_id] if isinstance(self.vram_cache[block_id], list) else self.vram_cache[block_id].tolist(),
|
| 'metadata': metadata,
|
| 'source': 'cache'
|
| }
|
|
|
|
|
| result = self.storage.load_tensor(block_id)
|
| if result is not None:
|
| data, metadata = result
|
| self.vram_cache[block_id] = data
|
|
|
|
|
| metadata_path = self.vram_path / f"{block_id}_metadata.json"
|
| metadata = {}
|
| if metadata_path.exists():
|
| with open(metadata_path, 'r') as f:
|
| metadata = json.load(f)
|
|
|
| return {
|
| 'status': 'success',
|
| 'data': data.tolist(),
|
| 'metadata': metadata,
|
| 'source': 'disk'
|
| }
|
| return {'status': 'error', 'message': 'Block not found'}
|
|
|
| return {'status': 'error', 'message': f'Unknown operation type: {op_type}'}
|
|
|
| except ValueError as e:
|
| return {'status': 'error', 'message': str(e)}
|
| except Exception as e:
|
| return {'status': 'error', 'message': f'Operation failed: {str(e)}'}
|
|
|
| async def handle_state_operation(self, operation: dict) -> dict:
|
| """Handle GPU state operations (preserved from WebSocket implementation)"""
|
| op_type = operation.get('type')
|
| component = operation.get('component')
|
| state_id = operation.get('state_id')
|
| state_data = operation.get('data')
|
|
|
| if op_type == 'save':
|
|
|
| success = self.storage.store_state(component, state_id, state_data)
|
| if success:
|
|
|
| self.state_cache[f"{component}:{state_id}"] = state_data
|
| return {'status': 'success', 'message': f'State {state_id} saved'}
|
| else:
|
| return {'status': 'error', 'message': 'Failed to save state'}
|
|
|
| elif op_type == 'load':
|
| cache_key = f"{component}:{state_id}"
|
| if cache_key in self.state_cache:
|
| return {
|
| 'status': 'success',
|
| 'data': self.state_cache[cache_key],
|
| 'source': 'cache'
|
| }
|
|
|
|
|
| state_data = self.storage.load_state(component, state_id)
|
| if state_data:
|
| self.state_cache[cache_key] = state_data
|
| return {
|
| 'status': 'success',
|
| 'data': state_data,
|
| 'source': 'database'
|
| }
|
|
|
| return {'status': 'error', 'message': 'State not found'}
|
|
|
| async def handle_cache_operation(self, operation: dict) -> dict:
|
| """Handle cache operations using database storage"""
|
| op_type = operation.get('type')
|
| key = operation.get('key')
|
| data = operation.get('data')
|
| ttl = operation.get('ttl')
|
|
|
| if op_type == 'set':
|
|
|
| self.memory_cache[key] = data
|
|
|
|
|
| metadata = {
|
| 'created_at': time.time(),
|
| 'ttl': ttl
|
| }
|
| if isinstance(data, (dict, list)):
|
| data = json.dumps(data)
|
| success = self.storage.store_in_cache(key, data, metadata)
|
|
|
| if success:
|
| return {'status': 'success', 'message': f'Cache key {key} set'}
|
| return {'status': 'error', 'message': 'Failed to store in cache'}
|
|
|
| elif op_type == 'get':
|
|
|
| if key in self.memory_cache:
|
| return {
|
| 'status': 'success',
|
| 'data': self.memory_cache[key],
|
| 'source': 'memory'
|
| }
|
|
|
|
|
| result = self.storage.load_from_cache(key)
|
| if result:
|
| data, metadata = result
|
| if isinstance(data, bytes):
|
| try:
|
| data = json.loads(data)
|
| except:
|
| pass
|
|
|
| self.memory_cache[key] = data
|
| return {
|
| 'status': 'success',
|
| 'data': data,
|
| 'metadata': metadata,
|
| 'source': 'database'
|
| }
|
|
|
| return {'status': 'error', 'message': 'Cache key not found'}
|
|
|
| def get_stats(self) -> dict:
|
| """Get server statistics"""
|
| current_time = time.time()
|
| uptime = current_time - self.start_time
|
| ops_per_second = self.ops_counter / uptime if uptime > 0 else 0
|
|
|
| return {
|
| 'uptime': uptime,
|
| 'total_operations': self.ops_counter,
|
| 'ops_per_second': ops_per_second,
|
| 'active_connections': len(self.active_connections),
|
| 'active_http_sessions': len(self.http_sessions),
|
| 'vram_cache_size': len(self.vram_cache),
|
| 'state_cache_size': len(self.state_cache),
|
| 'memory_cache_size': len(self.memory_cache),
|
| 'model_cache_size': len(self.model_cache)
|
| }
|
|
|
|
|
| server = VirtualGPUServer()
|
|
|
|
|
| def get_current_session(credentials: HTTPAuthorizationCredentials = Depends(security)) -> Dict[str, Any]:
|
| return server.verify_session_token(credentials.credentials)
|
|
|
|
|
|
|
| @app.post("/api/v1/sessions", response_model=SessionResponse)
|
| async def create_session(request: SessionCreateRequest):
|
| """Create a new HTTP session"""
|
| session_id = str(uuid.uuid4())
|
| client_id = request.client_id or "anonymous"
|
|
|
|
|
| token = server.create_session_token(session_id, client_id, request.resource_limits)
|
|
|
|
|
| server.http_sessions[session_id] = {
|
| 'session_id': session_id,
|
| 'client_id': client_id,
|
| 'created_at': time.time(),
|
| 'resource_limits': request.resource_limits or {},
|
| 'ops_count': 0
|
| }
|
|
|
| expires_at = datetime.fromtimestamp(time.time() + (JWT_EXPIRATION_HOURS * 3600))
|
|
|
| return SessionResponse(
|
| session_token=token,
|
| session_id=session_id,
|
| expires_at=expires_at
|
| )
|
|
|
| @app.post("/api/v1/vram/blocks/{block_id}", response_model=VRAMResponse)
|
| async def write_vram_block(
|
| block_id: str,
|
| request: VRAMWriteRequest,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Write tensor data to VRAM block"""
|
| try:
|
| operation = {
|
| 'operation': 'vram',
|
| 'type': 'write',
|
| 'block_id': block_id,
|
| 'data': request.data,
|
| 'metadata': request.metadata or {},
|
| 'model_size': request.model_size
|
| }
|
|
|
| result = await server.handle_vram_operation(operation)
|
| server.ops_counter += 1
|
|
|
| if result['status'] == 'success':
|
| return VRAMResponse(
|
| status=result['status'],
|
| message=result['message']
|
| )
|
| else:
|
| raise HTTPException(status_code=400, detail=result['message'])
|
|
|
| except Exception as e:
|
| request_id = server.generate_request_id()
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"VRAM write operation failed: {str(e)}"
|
| )
|
|
|
| @app.get("/api/v1/vram/blocks/{block_id}", response_model=VRAMResponse)
|
| async def read_vram_block(
|
| block_id: str,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Read tensor data from VRAM block"""
|
| try:
|
| operation = {
|
| 'operation': 'vram',
|
| 'type': 'read',
|
| 'block_id': block_id
|
| }
|
|
|
| result = await server.handle_vram_operation(operation)
|
| server.ops_counter += 1
|
|
|
| if result['status'] == 'success':
|
| return VRAMResponse(
|
| status=result['status'],
|
| data=result.get('data'),
|
| metadata=result.get('metadata'),
|
| source=result.get('source')
|
| )
|
| else:
|
| raise HTTPException(status_code=404, detail=result['message'])
|
|
|
| except HTTPException:
|
| raise
|
| except Exception as e:
|
| request_id = server.generate_request_id()
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"VRAM read operation failed: {str(e)}"
|
| )
|
|
|
| @app.delete("/api/v1/vram/blocks/{block_id}")
|
| async def delete_vram_block(
|
| block_id: str,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Delete tensor data from VRAM block"""
|
| try:
|
|
|
| if block_id in server.vram_cache:
|
| del server.vram_cache[block_id]
|
|
|
|
|
| file_path = server.vram_path / f"{block_id}.npy"
|
| metadata_path = server.vram_path / f"{block_id}_metadata.json"
|
|
|
| if file_path.exists():
|
| file_path.unlink()
|
| if metadata_path.exists():
|
| metadata_path.unlink()
|
|
|
| server.ops_counter += 1
|
| return {"status": "success", "message": f"Block {block_id} deleted"}
|
|
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"VRAM delete operation failed: {str(e)}"
|
| )
|
|
|
| @app.post("/api/v1/state/{component}/{state_id}", response_model=StateResponse)
|
| async def save_state(
|
| component: str,
|
| state_id: str,
|
| request: StateRequest,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Save component state"""
|
| try:
|
| operation = {
|
| 'operation': 'state',
|
| 'type': 'save',
|
| 'component': component,
|
| 'state_id': state_id,
|
| 'data': request.data
|
| }
|
|
|
| result = await server.handle_state_operation(operation)
|
| server.ops_counter += 1
|
|
|
| if result['status'] == 'success':
|
| return StateResponse(
|
| status=result['status'],
|
| message=result['message']
|
| )
|
| else:
|
| raise HTTPException(status_code=400, detail=result['message'])
|
|
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"State save operation failed: {str(e)}"
|
| )
|
|
|
| @app.get("/api/v1/state/{component}/{state_id}", response_model=StateResponse)
|
| async def load_state(
|
| component: str,
|
| state_id: str,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Load component state"""
|
| try:
|
| operation = {
|
| 'operation': 'state',
|
| 'type': 'load',
|
| 'component': component,
|
| 'state_id': state_id
|
| }
|
|
|
| result = await server.handle_state_operation(operation)
|
| server.ops_counter += 1
|
|
|
| if result['status'] == 'success':
|
| return StateResponse(
|
| status=result['status'],
|
| data=result.get('data'),
|
| source=result.get('source')
|
| )
|
| else:
|
| raise HTTPException(status_code=404, detail=result['message'])
|
|
|
| except HTTPException:
|
| raise
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"State load operation failed: {str(e)}"
|
| )
|
|
|
| @app.post("/api/v1/cache/{key}", response_model=CacheResponse)
|
| async def set_cache(
|
| key: str,
|
| request: CacheRequest,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Set cache value"""
|
| try:
|
| operation = {
|
| 'operation': 'cache',
|
| 'type': 'set',
|
| 'key': key,
|
| 'data': request.data
|
| }
|
|
|
| result = await server.handle_cache_operation(operation)
|
| server.ops_counter += 1
|
|
|
| if result['status'] == 'success':
|
| return CacheResponse(
|
| status=result['status'],
|
| message=result['message']
|
| )
|
| else:
|
| raise HTTPException(status_code=400, detail=result['message'])
|
|
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"Cache set operation failed: {str(e)}"
|
| )
|
|
|
| @app.get("/api/v1/cache/{key}", response_model=CacheResponse)
|
| async def get_cache(
|
| key: str,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Get cache value"""
|
| try:
|
| operation = {
|
| 'operation': 'cache',
|
| 'type': 'get',
|
| 'key': key
|
| }
|
|
|
| result = await server.handle_cache_operation(operation)
|
| server.ops_counter += 1
|
|
|
| if result['status'] == 'success':
|
| return CacheResponse(
|
| status=result['status'],
|
| data=result.get('data'),
|
| source=result.get('source')
|
| )
|
| else:
|
| raise HTTPException(status_code=404, detail=result['message'])
|
|
|
| except HTTPException:
|
| raise
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"Cache get operation failed: {str(e)}"
|
| )
|
|
|
| def sanitize_filename(name: str) -> str:
|
| """
|
| Sanitize a string for safe file system usage.
|
| Replaces slashes with double underscores.
|
| """
|
| return name.replace('/', '__')
|
|
|
| @app.post("/api/v1/models/{model_name:path}/load")
|
| async def load_model(
|
| model_name: str,
|
| request: ModelLoadRequest,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Load AI model"""
|
| try:
|
|
|
| logging.info(f"Received model load request for: {model_name}")
|
|
|
|
|
| safe_name = sanitize_filename(model_name)
|
|
|
| if not request.model_data:
|
| raise HTTPException(
|
| status_code=400,
|
| detail="model_data is required and must include architecture configuration"
|
| )
|
|
|
|
|
| required_fields = ['num_sms', 'tensor_cores_per_sm', 'cuda_cores_per_sm']
|
| missing_fields = [field for field in required_fields if field not in request.model_data]
|
| if missing_fields:
|
| raise HTTPException(
|
| status_code=400,
|
| detail=f"Missing required model configuration fields: {missing_fields}"
|
| )
|
|
|
|
|
| model_info = {
|
| 'model_name': model_name,
|
| 'model_data': request.model_data,
|
| 'model_path': request.model_path,
|
| 'model_hash': request.model_hash,
|
| 'loaded_at': time.time(),
|
| 'session_id': session['session_id'],
|
| 'architecture': {
|
| 'num_sms': request.model_data['num_sms'],
|
| 'tensor_cores_per_sm': request.model_data['tensor_cores_per_sm'],
|
| 'cuda_cores_per_sm': request.model_data['cuda_cores_per_sm'],
|
| 'vram_allocation': request.model_data.get('vram_allocation', 'dynamic'),
|
| 'compute_capability': request.model_data.get('compute_capability', '8.0')
|
| }
|
| }
|
|
|
| server.model_cache[model_name] = model_info
|
|
|
|
|
| model_file = server.models_path / f"{safe_name}.json"
|
| model_data_file = server.models_path / f"{safe_name}.data"
|
| logging.info(f"Storing model info at: {model_file}")
|
|
|
|
|
| with open(model_file, 'w') as f:
|
| json.dump(model_info, f)
|
|
|
|
|
| if request.model_data.get('weights') or request.model_data.get('parameters'):
|
| logging.info(f"Storing model data at: {model_data_file}")
|
| with open(model_data_file, 'w') as f:
|
| json.dump(request.model_data, f)
|
|
|
| server.ops_counter += 1
|
| return {
|
| "status": "success",
|
| "message": f"Model {model_name} loaded successfully",
|
| "model_info": {
|
| "name": model_name,
|
| "architecture": model_info['architecture'],
|
| "loaded_at": model_info['loaded_at']
|
| }
|
| }
|
|
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"Model load operation failed: {str(e)}"
|
| )
|
|
|
| @app.post("/api/v1/models/{model_name:path}/inference")
|
| async def run_inference(
|
| model_name: str,
|
| request: ModelInferenceRequest,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Run model inference"""
|
| try:
|
| logging.info(f"Running inference - Raw model name: {model_name}")
|
| safe_name = sanitize_model_name(model_name)
|
| logging.info(f"Running inference - Safe model name: {safe_name}")
|
|
|
|
|
| if model_name not in server.model_cache:
|
|
|
| model_file = server.models_path / f"{safe_name}.json"
|
| if not model_file.exists():
|
| logging.error(f"Model {model_name} not found in cache or filesystem")
|
| raise HTTPException(status_code=404, detail=f"Model {model_name} not loaded")
|
|
|
| logging.info(f"Loading model info from file: {model_file}")
|
| with open(model_file) as f:
|
| model_info = json.load(f)
|
| server.model_cache[model_name] = model_info
|
|
|
|
|
|
|
| result = {
|
| "status": "success",
|
| "output": request.input_data,
|
| "metrics": {
|
| "inference_time": 0.1,
|
| "tokens_processed": len(request.input_data)
|
| },
|
| "model_info": server.model_cache[model_name]
|
| }
|
|
|
| server.ops_counter += 1
|
| logging.info(f"Inference completed successfully for model: {model_name}")
|
| return result
|
|
|
| except HTTPException:
|
| raise
|
| except Exception as e:
|
| logging.error(f"Inference operation failed for {model_name}: {str(e)}")
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"Inference operation failed: {str(e)}"
|
| )
|
|
|
| @app.get("/api/v1/models/{model_name:path}/status")
|
| async def get_model_status(
|
| model_name: str,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Get model status"""
|
| try:
|
| logging.info(f"Checking model status for: {model_name}")
|
|
|
|
|
| if model_name in server.model_cache:
|
| logging.info(f"Model {model_name} found in cache")
|
| return {
|
| "status": "loaded",
|
| "model_info": server.model_cache[model_name]
|
| }
|
|
|
|
|
| safe_name = sanitize_filename(model_name)
|
| model_file = server.models_path / f"{safe_name}.json"
|
| if model_file.exists():
|
| logging.info(f"Model file found: {model_file}")
|
| with open(model_file) as f:
|
| model_info = json.load(f)
|
|
|
| server.model_cache[model_name] = model_info
|
| return {
|
| "status": "loaded",
|
| "model_info": model_info
|
| }
|
|
|
| logging.info(f"Model {model_name} not found in cache or filesystem")
|
| return {
|
| "status": "not_loaded",
|
| "message": f"Model {model_name} is not loaded"
|
| }
|
|
|
| except Exception as e:
|
| logging.error(f"Model status check failed for {model_name}: {str(e)}")
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"Model status check failed: {str(e)}"
|
| )
|
|
|
|
|
| @app.post("/api/v1/chips/{src_chip_id}/transfer/{dst_chip_id}")
|
| async def transfer_between_chips(
|
| src_chip_id: int,
|
| dst_chip_id: int,
|
| request: dict,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Transfer data between GPU chips"""
|
| try:
|
| data_id = request.get('data_id')
|
| if not data_id:
|
| raise HTTPException(status_code=400, detail="Missing data_id")
|
|
|
|
|
| source_operation = {
|
| 'operation': 'vram',
|
| 'type': 'read',
|
| 'block_id': data_id
|
| }
|
|
|
| source_result = await server.handle_vram_operation(source_operation)
|
| if source_result.get('status') != 'success':
|
| raise HTTPException(status_code=404, detail=f"Source data {data_id} not found")
|
|
|
|
|
| new_data_id = f"{data_id}_chip_{dst_chip_id}"
|
|
|
|
|
| dest_operation = {
|
| 'operation': 'vram',
|
| 'type': 'write',
|
| 'block_id': new_data_id,
|
| 'data': source_result.get('data'),
|
| 'metadata': source_result.get('metadata', {})
|
| }
|
|
|
| dest_result = await server.handle_vram_operation(dest_operation)
|
| if dest_result.get('status') != 'success':
|
| raise HTTPException(status_code=500, detail="Failed to store transferred data")
|
|
|
|
|
| transfer_id = f"transfer_{time.time_ns()}"
|
|
|
| result = {
|
| "status": "success",
|
| "transfer_id": transfer_id,
|
| "src_chip": src_chip_id,
|
| "dst_chip": dst_chip_id,
|
| "data_id": data_id,
|
| "new_data_id": new_data_id
|
| }
|
|
|
| server.ops_counter += 1
|
| return result
|
|
|
| except HTTPException:
|
| raise
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"Chip transfer failed: {str(e)}"
|
| )
|
|
|
| @app.post("/api/v1/sync/barrier/{barrier_id}")
|
| async def create_sync_barrier(
|
| barrier_id: str,
|
| request: dict,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Create synchronization barrier"""
|
| try:
|
| num_participants = request.get('num_participants', 1)
|
|
|
|
|
| barrier_info = {
|
| 'barrier_id': barrier_id,
|
| 'num_participants': num_participants,
|
| 'arrived_participants': 0,
|
| 'created_at': time.time()
|
| }
|
|
|
| server.memory_cache[f"barrier_{barrier_id}"] = barrier_info
|
|
|
| return {
|
| "status": "success",
|
| "barrier_id": barrier_id,
|
| "num_participants": num_participants
|
| }
|
|
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"Barrier creation failed: {str(e)}"
|
| )
|
|
|
| @app.put("/api/v1/sync/barrier/{barrier_id}/wait")
|
| async def wait_sync_barrier(
|
| barrier_id: str,
|
| session: Dict[str, Any] = Depends(get_current_session)
|
| ):
|
| """Wait at synchronization barrier"""
|
| try:
|
| barrier_key = f"barrier_{barrier_id}"
|
| if barrier_key not in server.memory_cache:
|
| raise HTTPException(status_code=404, detail="Barrier not found")
|
|
|
| barrier_info = server.memory_cache[barrier_key]
|
| barrier_info['arrived_participants'] += 1
|
|
|
|
|
| if barrier_info['arrived_participants'] >= barrier_info['num_participants']:
|
|
|
| del server.memory_cache[barrier_key]
|
| return {
|
| "status": "released",
|
| "message": "All participants arrived, barrier released"
|
| }
|
| else:
|
| return {
|
| "status": "waiting",
|
| "arrived": barrier_info['arrived_participants'],
|
| "total": barrier_info['num_participants']
|
| }
|
|
|
| except HTTPException:
|
| raise
|
| except Exception as e:
|
| raise HTTPException(
|
| status_code=500,
|
| detail=f"Barrier wait failed: {str(e)}"
|
| )
|
|
|
|
|
| @app.get("/", response_class=HTMLResponse)
|
| async def handle_index():
|
| """Handle HTTP index request"""
|
| stats = server.get_stats()
|
| html = f"""
|
| <!DOCTYPE html>
|
| <html>
|
| <head>
|
| <title>Virtual GPU Server v2.0</title>
|
| <style>
|
| body {{ font-family: Arial, sans-serif; margin: 40px; }}
|
| table {{ border-collapse: collapse; width: 100%; margin-top: 20px; }}
|
| th, td {{ padding: 12px; text-align: left; border-bottom: 1px solid #ddd; }}
|
| th {{ background-color: #f2f2f2; }}
|
| .stats {{ background-color: #f9f9f9; padding: 20px; border-radius: 5px; }}
|
| .api-info {{ background-color: #e8f4fd; padding: 20px; border-radius: 5px; margin-top: 20px; }}
|
| </style>
|
| </head>
|
| <body>
|
| <h1>Virtual GPU Server v2.0 Status</h1>
|
| <div class="api-info">
|
| <h2>API Information</h2>
|
| <p><strong>HTTP REST API:</strong> Available at /api/v1/</p>
|
| <p><strong>WebSocket API:</strong> Available at /ws (backward compatibility)</p>
|
| <p><strong>API Documentation:</strong> <a href="/docs">/docs</a></p>
|
| </div>
|
| <div class="stats">
|
| <h2>Server Statistics</h2>
|
| <ul>
|
| <li>Uptime: {stats['uptime']:.2f} seconds</li>
|
| <li>Total Operations: {stats['total_operations']}</li>
|
| <li>Operations per Second: {stats['ops_per_second']:.2f}</li>
|
| <li>Active WebSocket Connections: {stats['active_connections']}</li>
|
| <li>Active HTTP Sessions: {stats['active_http_sessions']}</li>
|
| <li>VRAM Cache Size: {stats['vram_cache_size']}</li>
|
| <li>State Cache Size: {stats['state_cache_size']}</li>
|
| <li>Memory Cache Size: {stats['memory_cache_size']}</li>
|
| <li>Model Cache Size: {stats['model_cache_size']}</li>
|
| </ul>
|
| </div>
|
| <h2>Server Files</h2>
|
| <iframe src="/files" style="width: 100%; height: 500px; border: none;"></iframe>
|
| </body>
|
| </html>
|
| """
|
| return HTMLResponse(content=html)
|
|
|
| @app.get("/files", response_class=HTMLResponse)
|
| async def handle_files():
|
| """Handle HTTP files listing request"""
|
| def format_size(size):
|
| for unit in ['B', 'KB', 'MB', 'GB']:
|
| if size < 1024:
|
| return f"{size:.2f} {unit}"
|
| size /= 1024
|
| return f"{size:.2f} TB"
|
|
|
| html = ['<!DOCTYPE html><html><head>',
|
| '<style>',
|
| 'body { font-family: Arial, sans-serif; margin: 20px; }',
|
| 'table { border-collapse: collapse; width: 100%; }',
|
| 'th, td { padding: 12px; text-align: left; border-bottom: 1px solid #ddd; }',
|
| 'th { background-color: #f2f2f2; }',
|
| '</style></head><body>',
|
| '<h2>Server Files</h2>',
|
| '<table><tr><th>Path</th><th>Size</th><th>Last Modified</th></tr>']
|
|
|
| for root, _, files in os.walk(server.base_path):
|
| for file in files:
|
| full_path = Path(root) / file
|
| rel_path = full_path.relative_to(server.base_path)
|
| size = format_size(os.path.getsize(full_path))
|
| mtime = datetime.fromtimestamp(os.path.getmtime(full_path))
|
| html.append(f'<tr><td>{rel_path}</td><td>{size}</td><td>{mtime}</td></tr>')
|
|
|
| html.extend(['</table></body></html>'])
|
| return HTMLResponse(content='\n'.join(html))
|
|
|
|
|
| @app.websocket("/ws")
|
| async def websocket_endpoint(websocket: WebSocket):
|
| await websocket.accept()
|
| session_id = str(uuid.uuid4())
|
| server.active_connections[session_id] = websocket
|
| server.active_sessions[session_id] = {
|
| 'start_time': time.time(),
|
| 'ops_count': 0
|
| }
|
|
|
| try:
|
| while True:
|
| message = await websocket.receive_json()
|
|
|
|
|
| operation_type = message.get('operation')
|
| if operation_type == 'vram':
|
| response = await server.handle_vram_operation(message)
|
| elif operation_type == 'state':
|
| response = await server.handle_state_operation(message)
|
| elif operation_type == 'cache':
|
| response = await server.handle_cache_operation(message)
|
| else:
|
| response = {
|
| 'status': 'error',
|
| 'message': 'Unknown operation type'
|
| }
|
|
|
|
|
| server.ops_counter += 1
|
| server.active_sessions[session_id]['ops_count'] += 1
|
|
|
|
|
| await websocket.send_json(response)
|
|
|
| except Exception as e:
|
| print(f"WebSocket error: {e}")
|
| finally:
|
|
|
| if session_id in server.active_connections:
|
| del server.active_connections[session_id]
|
| if session_id in server.active_sessions:
|
| del server.active_sessions[session_id]
|
|
|
|
|
| if __name__ == "__main__":
|
| import uvicorn
|
| uvicorn.run("virtual_gpu_server_http:app", host="0.0.0.0", port=7860, reload=True)
|
|
|
| @app.get("/api/v1/status")
|
| async def get_status():
|
| """Get server status"""
|
| return {"status": "ok", "message": "Virtual GPU Server is running"}
|
|
|