import re import os import mimetypes from wsgiref.util import FileWrapper from django.http import StreamingHttpResponse def file_iterator(file_name, chunk_size=8192, offset=0, length=None): with open(file_name, "rb") as f: f.seek(offset, os.SEEK_SET) remaining = length while True: bytes_length = chunk_size if remaining is None else min(remaining, chunk_size) data = f.read(bytes_length) if not data: break if remaining: remaining -= len(data) yield data def stream_video(request, path): """将视频文件以流媒体的方式响应""" range_header = request.META.get('HTTP_RANGE', '').strip() range_re = re.compile(r'bytes\s*=\s*(\d+)\s*-\s*(\d*)', re.I) range_match = range_re.match(range_header) size = os.path.getsize(path) content_type, encoding = mimetypes.guess_type(path) content_type = content_type or 'application/octet-stream' if range_match: first_byte, last_byte = range_match.groups() first_byte = int(first_byte) if first_byte else 0 last_byte = first_byte + 1024 * 1024 * 8 # 8M 每片,响应体最大体积 if last_byte >= size: last_byte = size - 1 length = last_byte - first_byte + 1 resp = StreamingHttpResponse(file_iterator(path, offset=first_byte, length=length), status=206, content_type=content_type) resp['Content-Length'] = str(length) resp['Content-Range'] = 'bytes %s-%s/%s' % (first_byte, last_byte, size) else: # 不是以视频流方式的获取时,以生成器方式返回整个文件,节省内存 resp = StreamingHttpResponse(FileWrapper(open(path, 'rb')), content_type=content_type) resp['Content-Length'] = str(size) resp['Accept-Ranges'] = 'bytes' return resp
标签:返回,resp,视频流,django,length,byte,first,type,size From: https://www.cnblogs.com/chen729765476/p/17715073.html