python - Nginx 的工作进程挂起并占用完整的 CPU

Question

我已经安装了 Nginx、uWSGI 和 Python Flask 应用程序。您可以在下面找到来自 Nginx 配置的服务器指令:

location /api {
    try_files $uri @api;
}

location @api {
    include uwsgi_params;
    uwsgi_pass 127.0.0.1:3031;
}

我用uwsgi --ini/etc/uwsgi.ini启动uWSGI。该文件如下所示:

[uwsgi]
uid = root
gid = root
socket = 127.0.0.1:3031
module = iris.api
callable = app

对 / 的请求工作正常，Nginx 返回“欢迎使用 Nginx”页面。但是对 /api 的请求失败了。 /api/analog_output/1 中的第一个请求通过 uWSGI 传递给 Python 应用程序。 Python 应用返回 HTTP 200 响应，但 Nginx 没有通过将此响应发送回客户端来完成请求。

--- Operational MODE: single process --- WSGI app 0 (mountpoint='') ready in 11 seconds on interpreter 0x1567e8 pid: 957 (default app)
--- uWSGI is running in multiple interpreter mode --- spawned uWSGI worker 1 (and the only) (pid: 957, cores: 1) 
[pid: 957|app: 0|req: 1/1] 10.0.0.125 () {42 vars in 712 bytes} [Sun Jan 14 17:22:49 2007] GET /api/analog_output/1 => generated 135 bytes in 66 msecs (HTTP/1.1 200) 2 headers in 72 bytes (1 switches on core 0)

您可以在下面找到绑定(bind)到 Nginx worker 的 strace 的输出。

17:27:39.127453 gettimeofday({1168795659, 128279}, NULL) = 0
17:27:39.129180 write(4, "2007/01/14 17:27:39 [info] 970#0"..., 83) = 83
17:27:39.130169 epoll_wait(8, {{EPOLLIN, {u32=651592, u64=2410198208213320}}}, 512, -1) = 1
17:27:44.680001 gettimeofday({1168795664, 680353}, NULL) = 0
17:27:44.682734 accept4(6, {sa_family=AF_INET, sin_port=htons(53845), sin_addr=inet_addr("10.0.0.125")}, [16], SOCK_NONBLOCK) = 3
17:27:44.685625 epoll_ctl(8, EPOLL_CTL_ADD, 3, {EPOLLIN|EPOLLRDHUP|EPOLLET, {u32=651816, u64=651816}}) = 0
17:27:44.688045 epoll_wait(8, {{EPOLLIN, {u32=651816, u64=651816}}}, 512, 60000) = 1
17:27:44.690552 gettimeofday({1168795664, 691682}, NULL) = 0
17:27:44.693043 recv(3, "GET /api/analog_output/1 HTTP/1."..., 1024, 0) = 426
17:27:44.695848 stat64("/usr/html/api/analog_output/1", 0xbeb8f730) = -1 ENOENT (No such file or directory)
17:27:44.698599 epoll_ctl(8, EPOLL_CTL_MOD, 3, {EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLET, {u32=651816, u64=13170497834292212264}}) = 0
17:27:44.701146 getsockname(3, {sa_family=AF_INET, sin_port=htons(80), sin_addr=inet_addr("10.0.0.195")}, [16]) = 0
17:27:44.703848 socket(PF_INET, SOCK_STREAM, IPPROTO_IP) = 10
17:27:44.706386 ioctl(10, FIONBIO, [1]) = 0
17:27:44.708823 epoll_ctl(8, EPOLL_CTL_ADD, 10, {EPOLLIN|EPOLLOUT|EPOLLRDHUP|EPOLLET, {u32=651928, u64=2525337691484824}}) = 0
17:27:44.711468 connect(10, {sa_family=AF_INET, sin_port=htons(3031), sin_addr=inet_addr("127.0.0.1")}, 16) = -1 EINPROGRESS (Operation now in progress)
17:27:44.714574 epoll_wait(8, {{EPOLLOUT, {u32=651816, u64=13170497834292212264}}, {EPOLLOUT, {u32=651928, u64=2525337691484824}}}, 512, 60000) = 2
17:27:44.717109 gettimeofday({1168795664, 718064}, NULL) = 0
17:27:44.719461 recv(3, 0xbeb8f89c, 1, MSG_PEEK) = -1 EAGAIN (Resource temporarily unavailable)
17:27:44.721999 getsockopt(10, SOL_SOCKET, SO_ERROR, [0], [4]) = 0
17:27:44.724476 writev(10, [{"\0\310\2\0\f\0QUERY_STRING\0\0\16\0REQUEST_ME"..., 716}], 1) = 716
17:27:44.729618 epoll_wait(8, {{EPOLLIN|EPOLLOUT, {u32=651928, u64=2525337691484824}}}, 512, 60000) = 1
17:27:44.793473 gettimeofday({1168795664, 794585}, NULL) = 0
17:27:44.796026 recv(10, "HTTP/1.1 200 OK\r\nContent-Type: a"..., 4096, 0) = 207

但是现在 Nginx 的 worker 因 CPU 满载而挂起。不处理进一步的请求。

这是怎么回事？我该如何解决？

Answer 1

您的位置 block 看起来不完整。目前不在机器前，但尝试添加:

proxy_redirect 关闭；

编辑:@app 位置 block