问题
I am trying to build a reverse proxy with nginx to make all Is in my project reachable from single address. For a single service the configuration below works without problem
/etc/nginx/sites-enabled/reverse-proxy.conf
server {
listen 80;
listen [::]:80;
location / {
resolver 127.0.0.1;
allow "x.x.x.x";
deny all;
proxy_pass http://consul:8500;
}
}
So when I call server's ip x.x.x.x in my browser I see the Consul UI and the URL showing x.x.x.x/ui/dc1. Besides that, I see that the UI did requests for asset files successfully.
My question; is it possible two host different services on the same server and just reference to them with different location? For example, if I want to include Vault UI then I would think of doing something like this:
server {
listen 80;
listen [::]:80;
location /consul {
resolver 127.0.0.1;
allow "x.x.x.x";
deny all;
proxy_pass http://consul:8500;
}
location /vault {
resolver 127.0.0.1;
allow "x.x.x.x";
deny all;
proxy_pass http://vault:8200;
}
}
However I am not sure if this could be done this way. The farest I got, is to open the Consul UI with all other sub requests not found (i.e. loading assets).
UPDATE
I think my problem is that I am wrongly using location and proxy_pass
observing the first configuration (which is working)
server {
listen 80;
listen [::]:80;
location / {
resolver 127.0.0.1;
allow "x.x.x.x";
deny all;
proxy_pass http://consul:8500;
}
}
If I look at the curl command curl localhost -L -vvvv
* Trying 127.0.0.1:80...
* TCP_NODELAY set
* Connected to localhost (127.0.0.1) port 80 (#0)
> GET / HTTP/1.1
> Host: localhost
> User-Agent: curl/7.68.0
> Accept: */*
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 301 Moved Permanently
< Server: nginx/1.18.0 (Ubuntu)
< Date: Fri, 10 Jul 2020 16:24:38 GMT
< Content-Type: text/html; charset=utf-8
< Content-Length: 39
< Connection: keep-alive
< Location: /ui/
<
* Ignoring the response-body
* Connection #0 to host localhost left intact
* Issue another request to this URL: 'http://localhost/ui/'
* Found bundle for host localhost: 0x557b754549e0 [serially]
* Can not multiplex, even if we wanted to!
* Re-using existing connection! (#0) with host localhost
* Connected to localhost (127.0.0.1) port 80 (#0)
> GET /ui/ HTTP/1.1
> Host: localhost
> User-Agent: curl/7.68.0
> Accept: */*
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 200 OK
< Server: nginx/1.18.0 (Ubuntu)
< Date: Fri, 10 Jul 2020 16:24:38 GMT
< Content-Type: text/html; charset=utf-8
< Content-Length: 7806
< Connection: keep-alive
< Accept-Ranges: bytes
< Last-Modified: Fri, 10 Jul 2020 07:37:44 GMT
<
<!DOCTYPE html>
<html lang="en" class="ember-loading">
...
and I can see the html already. However, if I changed the conf file to this:
server {
listen 80;
listen [::]:80;
location /consul/ {
resolver 127.0.0.1;
allow "x.x.x.x";
deny all;
proxy_pass http://consul:8500;
}
}
and then try to call it like curl localhost/consul -L -vvvv, I get the following:
* Trying 127.0.0.1:80...
* TCP_NODELAY set
* Connected to localhost (127.0.0.1) port 80 (#0)
> GET /consul HTTP/1.1
> Host: localhost
> User-Agent: curl/7.68.0
> Accept: */*
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 301 Moved Permanently
< Server: nginx/1.18.0 (Ubuntu)
< Date: Fri, 10 Jul 2020 16:32:35 GMT
< Content-Type: text/html
< Content-Length: 178
< Location: http://localhost/consul/
< Connection: keep-alive
<
* Ignoring the response-body
* Connection #0 to host localhost left intact
* Issue another request to this URL: 'http://localhost/consul/'
* Found bundle for host localhost: 0x55ba7959f9e0 [serially]
* Can not multiplex, even if we wanted to!
* Re-using existing connection! (#0) with host localhost
* Connected to localhost (127.0.0.1) port 80 (#0)
> GET /consul/ HTTP/1.1
> Host: localhost
> User-Agent: curl/7.68.0
> Accept: */*
>
* Mark bundle as not supporting multiuse
< HTTP/1.1 404 Not Found
< Server: nginx/1.18.0 (Ubuntu)
< Date: Fri, 10 Jul 2020 16:32:35 GMT
< Content-Length: 0
< Connection: keep-alive
I would appreciate any ideas on this issue
回答1:
You are right, you are using location and proxy_pass a wrong way. When you use the
location /vault {
proxy_pass http://vault:8200;
}
construction, you are passing your URI to the upstream as-is, while most likely you want to strip the /vault prefix from it. To do it, you should use this one:
location /vault/ {
proxy_pass http://vault:8200/;
}
You can read more about the difference of the first and the second one here. However this still can prevent the assets from loading correctly.
This question - how to proxy some webapp under some URI prefix - is being asked again and again on stackoverflow. The only right way to do it is to made your proxied app request its assets via relative URLs only (consider assets/script.js instead of /assets/script.js) or using the right prefix (/vault/assets/script.js). Some well-written apps are able to detect if they are used under such an URI prefix and use it when an asset link is being generated, some apps allows to specify it via some settings, but some are not suited for the such use at all. The reason why the webapp won't work without fulfilling these requirements is quite obvious - any URL not started with /vault won't match your location /vault/ { ... } block and would be served via main location block instead. So the best way to do it is to fix your webapp, however several workarounds can be used if you really cannot.
Some web frameworks already builds their webapps with relative URLs, but uses a
<base href="/">in the head section ofindex.html. For example, React or Angular use this approach. If you have such a line within your webapp rootindex.html, just change it to<base href="/vault/">.Using conditional routing based on HTTP
Refererheader value. This approach works quite well for a single page applications for loading assets, but if a webapp contains several pages this approach won't work, it's logic for the right upstream detection would break after the first jump from one page to another. Here is an example:map $http_referer $prefix { ~https?://[^/]+/vault/ vault; # other webapps prefixes could be defined here # ... default base; } server { # listen port, server name and other global definitions here # ... location / { # "unconditional" jump-to-location idea taken from this answer: # https://serverfault.com/questions/908086/nginx-directly-send-from-location-to-another-named-location/965779#965779 try_files /dev/null @$prefix; } location /vault/ { # proxy request to the vault upstream, remove "/vault" part from the URI proxy_pass http://vault:8200/; } location @vault { # proxy request to the vault upstream, do not change the URI proxy_pass http://vault:8200; } location @base { # default "root" location proxy_pass http://consul:8500; } }Rewriting the links inside the response body using
sub_filterdirective from ngx_http_sub_module. This is the ugliest one, but still can be used as the last available option. This approach has an obvious perfomance impact. Rewrite patterns should be determined from your upstream response body. Usually that type of configuration looked likelocation /vault/ { proxy_pass http://vault:8200/; sub_filter_types text/css application/javascript; sub_filter_once off; sub_filter 'href="/' 'href="/vault/'; sub_filter "href='/" "href='/vault/"; sub_filter 'src="/' 'src="/vault/'; sub_filter "src='/" "src='/vault/"; sub_filter 'url("/' 'url("/vault/'; sub_filter "url('/" "url('/vault/"; sub_filter "url(/" "url(/vault/"; }
来源：https://stackoverflow.com/questions/62836801/nginx-reverse-proxy-how-to-serve-multiple-apps