Nginx reverse proxy causing 504 Gateway Timeout

Question

I am using Nginx as a reverse proxy that takes requests then does a proxy_pass to get the actual web application from the upstream server running on port 8001.

If I go to mywebsite.com or do a wget, I get a 504 Gateway Timeout after 60 seconds... However, if I load mywebsite.com:8001, the application loads as expected!

So something is preventing Nginx from communicating with the upstream server.

All this started after my hosting company reset the machine my stuff was running on, prior to that no issues whatsoever.

Here's my vhosts server block:

server {
    listen   80;
    server_name mywebsite.com;

    root /home/user/public_html/mywebsite.com/public;

    access_log /home/user/public_html/mywebsite.com/log/access.log upstreamlog;
    error_log /home/user/public_html/mywebsite.com/log/error.log;

    location / {
        proxy_pass http://xxx.xxx.xxx.xxx:8001;
        proxy_redirect off;
        proxy_set_header Host $host;
        proxy_set_header X-Real-IP $remote_addr;
        proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
    }
} 

And the output from my Nginx error log:

2014/06/27 13:10:58 [error] 31406#0: *1 upstream timed out (110: Connection timed out) while connecting to upstream, client: xxx.xx.xxx.xxx, server: mywebsite.com, request: "GET / HTTP/1.1", upstream: "http://xxx.xxx.xxx.xxx:8001/", host: "mywebsite.com"

Answer 1

If you want to increase or add time limit to all sites then you can add below lines to the nginx.conf file.

Add below lines to the http section of /usr/local/etc/nginx/nginx.conf or /etc/nginx/nginx.conf file.

fastcgi_read_timeout 600;
proxy_read_timeout 600;

If the above lines doesn't exist in conf file then add them, otherwise increase fastcgi_read_timeout and proxy_read_timeout to make sure that nginx and php-fpm did not timeout.

To increase time limit for only one site then you can edit in vim /etc/nginx/sites-available/example.com

location ~ \.php$ {
    include /etc/nginx/fastcgi_params;
        fastcgi_pass  unix:/var/run/php5-fpm.sock;
    fastcgi_read_timeout 300; 
}

and after adding these lines in nginx.conf, then don't forget to restart nginx.

service php7-fpm reload 
service nginx reload

or, if you're using valet then simply type valet restart.

Answer 2

You can also face this situation if your upstream server uses a domain name, and its IP address changes (e.g.: your upstream points to an AWS Elastic Load Balancer)

The problem is that nginx will resolve the IP address once, and keep it cached for subsequent requests until the configuration is reloaded.

You can tell nginx to use a name server to re-resolve the domain once the cached entry expires:

location /mylocation {
    # use google dns to resolve host after IP cached expires
    resolver 8.8.8.8;
    set $upstream_endpoint http://your.backend.server/;
    proxy_pass $upstream_endpoint;
}

The docs on proxy_pass explain why this trick works:

Parameter value can contain variables. In this case, if an address is specified as a domain name, the name is searched among the described server groups, and, if not found, is determined using a resolver.

Kudos to "Nginx with dynamic upstreams" (tenzer.dk) for the detailed explanation, which also contains some relevant information on a caveat of this approach regarding forwarded URIs.

Answer 3

Had the same problem. Turned out it was caused by iptables connection tracking on the upstream server. After removing --state NEW,ESTABLISHED,RELATED from the firewall script and flushing with conntrack -F the problem was gone.

Answer 4

Probably can add a few more line to increase the timeout period to upstream. The examples below sets the timeout to 300 seconds :

proxy_connect_timeout       300;
proxy_send_timeout          300;
proxy_read_timeout          300;
send_timeout                300;

Answer 5

Increasing the timeout will not likely solve your issue since, as you say, the actual target web server is responding just fine.

I had this same issue and I found it had to do with not using a keep-alive on the connection. I can't actually answer why this is but, in clearing the connection header I solved this issue and the request was proxied just fine:

server {
    location / {
        proxy_set_header   X-Real-IP $remote_addr;
        proxy_set_header   Host      $http_host;
        proxy_http_version 1.1;
        proxy_set_header Connection "";
        proxy_pass http://localhost:5000;
    }
}

Have a look at this posts which explains it in more detail: nginx close upstream connection after request Keep-alive header clarification http://nginx.org/en/docs/http/ngx_http_upstream_module.html#keepalive