问题现象
1、浏览器发起一个下载请求，例如/export
2、服务器接收到请求，相应一个attachment filename=具体文件名的相应，并开始将数据写入HTTPServletResponse的outputStream流中
3、客户端接收到下载文件名响应，用户点击保存后，开始传输数据
4、服务区outputStream流向客户端flush数据
5、客户端接收数据
6、服务器端未发送完数据，浏览器下载文件到一半就提示下载完成。
7、服务器端报：Caused by: java.net.SocketException: Connection reset异常。
问题排查
由于应用属于Web应用，浏览器的请求通过nginx进行转发和负载均衡。发现nginx的请求访问日志，截取了关键信息如下：
2019/05/15 10:36:57 [warn] 198697#0: *14686663 an upstream response is buffered to a temporary file /dev/shm/nginx_temp/proxy/5/24/0000000245 while reading upstream, client: 172.31.127.6, server: sk.jd.com, request: "POST /unionAdgroup/export HTTP/1.1", upstream: "http://127.0.0.1:1601/unionAdgroup/export", host: "sk.jd.com", referrer: "http://sk.jd.com/unionAdgroup/?platform=1&device=-1"
2019/05/15 10:39:17 [warn] 198697#0: *14687121 an upstream response is buffered to a temporary file /dev/shm/nginx_temp/proxy/6/24/0000000246 while reading upstream, client: 172.31.127.6, server: sk.jd.com, request: "POST /unionCreative/export HTTP/1.1", upstream: "http://127.0.0.1:1601/unionCreative/export", host: "sk.jd.com", referrer: "http://sk.jd.com/unionCreative/?platform=1&device=-1"
2019/05/15 10:39:52 [crit] 198697#0: *14687121 pwrite() "/dev/shm/nginx_temp/proxy/6/24/0000000246" failed (28: No space left on device) while reading upstream, client: 172.31.127.6, server: sk.jd.com, request: "POST /unionCreative/export HTTP/1.1", upstream: "http://127.0.0.1:1601/unionCreative/export", host: "sk.jd.com", referrer: "http://sk.jd.com/unionCreative/?platform=1&device=-1"
2019/05/15 11:01:48 [warn] 198694#0: *14690923 an upstream response is buffered to a temporary file /dev/shm/nginx_temp/proxy/7/24/0000000247 while reading upstream, client: 172.31.127.6, server: sk.jd.com, request: "POST /unionCreative/export HTTP/1.1", upstream: "http://127.0.0.1:1601/unionCreative/export", host: "sk.jd.com", referrer: "http://sk.jd.com/unionCreative/?platform=1&device=-1"
2019/05/15 11:02:22 [crit] 198694#0: *14690923 pwrite() "/dev/shm/nginx_temp/proxy/7/24/0000000247" failed (28: No space left on device) while reading upstream, client: 172.31.127.6, server: sk.jd.com, request: "POST /unionCreative/export HTTP/1.1", upstream: "http://127.0.0.1:1601/unionCreative/export", host: "sk.jd.com", referrer: "http://sk.jd.com/unionCreative/?platform=1&device=-1"
2019/05/15 11:16:13 [warn] 198697#0: *14693421 an upstream response is buffered to a temporary file /dev/shm/nginx_temp/proxy/8/24/0000000248 while reading upstream, client: 172.31.127.6, server: sk.jd.com, request: "POST /unionCreative/export HTTP/1.1", upstream: "http://127.0.0.1:1601/unionCreative/export", host: "sk.jd.com", referrer: "http://sk.jd.com/unionCreative/?platform=1&device=-1"
2019/05/15 11:16:48 [crit] 198697#0: *14693421 pwrite() "/dev/shm/nginx_temp/proxy/8/24/0000000248" failed (28: No space left on device) while reading upstream, client: 172.31.127.6, server: sk.jd.com, request: "POST /unionCreative/export HTTP/1.1", upstream: "http://127.0.0.1:1601/unionCreative/export", host: "sk.jd.com", referrer: "http://sk.jd.com/unionCreative/?platform=1&device=-1"
其中特别注意到
an upstream response is buffered to a temporary file /dev/shm/nginx_temp/proxy/6/24/0000000246 while reading upstream
pwrite() "/dev/shm/nginx_temp/proxy/6/24/0000000246" failed (28: No space left on device) while reading upstream
再结合nginx官网的说明，传输的数据被先写到了nginx的临时buffer中，再向浏览器客户端传输数据。但buffer的空间不足，导致写失败，下载异常。
问题解决
1、针对下载类的请求，文件过大时，针对这类URL设置nginx的location里面的proxy_buffering参数为off。
默认情况下proxy_buffering这个参数是on状态，关闭后，nginx会对服务器端的数据实时传输给客户端，不会先写buffer后再响应数据。
网友评论