Using error_page to a named location - possible?

dubstep · March 6, 2012, 5:38pm

Hi, as part of a larger problem I have a captive portal which grabs all
incoming URLs and after evaluating a bunch of logic redirects the caller
somewhere else. I use proxy_pass to pass the request to my cgi code.

However, I would like to customise the error message in the event that
the upstream daemon is unavailable. I can’t figure out how to do that
affecting part of the incoming URLs. Is it possible to do something
like “error_page @502_error” in conjunction with proxy_pass?

So, I have

server {
listen 8000 default_server;

root /var/www/cp/htdocs/public;

location / {
proxy_read_timeout 5;
proxy_pass http://127.0.0.1:3000;
error_page 502 /502.html;
proxy_set_header Host $http_host;
proxy_set_header X-Server-Addr $server_addr;
proxy_set_header X-Forwarded-For $proxy_add_x_forwarded_for;
proxy_set_header X-Forwarded-HTTPS 0;
proxy_set_header X-Captive-Portal-Redirect 1;
}
location /502.html {
internal;
}
}

However, if I visit “http://somewhere/502.html” I get a 404 response
(presumably matching the location /502.html and the 404 is due to the
“internal”?)

I can’t see that it’s possible to use something like:

location / {
proxy_pass http://127.0.0.1:3000;
error_page 502 @502;
}

location @502 {
index /502.html;
root /var/www;
internal;
}

Does someone have a suggestion on how to customise the error response
without affecting ANY incoming URLs (they must ALL be passed through the
proxy under normal working situation)

(Its occured to me that I could perhaps setup another “server { }” block
listening on another port, then use a different proxy_pass in the
“location @502 {}”? Seems ugly, but would that work?)

Thanks for any thoughts

Ed W

ewildgoose · March 6, 2012, 5:41pm

2012/3/6 Ed W [email protected]:

Does someone have a suggestion on how to customise the error response
without affecting ANY incoming URLs (they must ALL be passed through the
proxy under normal working situation)

Try location = /502.html { } instead.

ewildgoose · March 6, 2012, 6:48pm

Hello!

On Tue, Mar 06, 2012 at 04:37:41PM +0000, Ed W wrote:

Hi, as part of a larger problem I have a captive portal which grabs
all incoming URLs and after evaluating a bunch of logic redirects
the caller somewhere else. I use proxy_pass to pass the request to
my cgi code.

However, I would like to customise the error message in the event
that the upstream daemon is unavailable. I can’t figure out how to
do that affecting part of the incoming URLs. Is it possible to do
something like “error_page @502_error” in conjunction with
proxy_pass?

Yes, see below.

proxy_pass http://127.0.0.1:3000;
}

However, if I visit “http://somewhere/502.html” I get a 404 response
(presumably matching the location /502.html and the 404 is due to
the “internal”?)

Yes.

internal;
}

Try something like:

location @502 {
    internal;
    root /var/www;
    rewrite ^ /502.html break;
}

This should work at least for GET requests.

Does someone have a suggestion on how to customise the error
response without affecting ANY incoming URLs (they must ALL be
passed through the proxy under normal working situation)

(Its occured to me that I could perhaps setup another “server { }”
block listening on another port, then use a different proxy_pass in
the “location @502 {}”? Seems ugly, but would that work?)

This should work too.

Maxim D.

ewildgoose · March 6, 2012, 7:24pm

Hello!

On Tue, Mar 06, 2012 at 10:05:03PM +0400, Max wrote:

proxy_set_header X-Server-Addr $server_addr;
However, if I visit “http://somewhere/502.html” I get a 404 response
}

# all named locations are internal by default, hence no "internal;"
return 502 "Gateway error while servicing $request_uri!\n";

}

The proxy_intercept_errors isn’t needed to handle 502’s generated
by nginx, it’s only needed when backend returns the error in
question and one want to override it.

Maxim D.

ewildgoose · March 6, 2012, 7:39pm

06 марта 2012, 22:24 от Maxim D. [email protected]:

Hi, as part of a larger problem I have a captive portal which grabs
Yes, see below.
proxy_read_timeout 5;
}
error_page 502 @502;
}

location @502 {
# all named locations are internal by default, hence no “internal;”
return 502 “Gateway error while servicing $request_uri!\n”;
}
The proxy_intercept_errors isn’t needed to handle 502’s generated
by nginx, it’s only needed when backend returns the error in
question and one want to override it.

Perhaps I misunderstood the OP, but he mentioned passing the
request to his cgi code, but mentioned no fastcgi_pass directives,
so it seems he wants to intercept 502 errors from the backend
server.

Max

ewildgoose · March 6, 2012, 7:05pm

06 марта 2012, 21:47 от Maxim D. [email protected]:

that the upstream daemon is unavailable. I can’t figure out how to
listen 8000 default_server;
proxy_set_header X-Forwarded-HTTPS 0;
the “internal”?)

Yes.

I can’t see that it’s possible to use something like:

location / {
proxy_pass http://127.0.0.1:3000;
error_page 502 @502;
}

It is possible, you should use the proxy_intercept_errors directive:

location / {
proxy_pass http://127.0.0.1:3000;
proxy_intercept_errors on;
error_page 502 @502;
}

location @502 {
# all named locations are internal by default, hence no “internal;”
return 502 “Gateway error while servicing $request_uri!\n”;
}

Max

ewildgoose · March 6, 2012, 11:33pm

On 06/03/2012 17:47, Maxim D. wrote:

internal;
}
Try something like:
 location @502 {
     internal;
     root /var/www;
     rewrite ^ /502.html break;
 }
This should work at least for GET requests.

Perfect - thanks!

Now, I have thought about this a little while and I don’t quite
understand how it works…?

The key seems to be the rewrite clause - near as I can see, without this
the error request gets reprocessed by the “location /” section, but the
rewrite is causing it to terminate that?

Can you teach me to understand this please?

Many thanks!

Ed W

ewildgoose · March 7, 2012, 12:29am

Hello!

On Tue, Mar 06, 2012 at 10:32:56PM +0000, Ed W wrote:

root /var/www;
This should work at least for GET requests.

Can you teach me to understand this please?

Under normal conditions

location @502 {
    root /var/www;
}

will behave much like normal static location, as described here:

http://nginx.org/en/docs/http/request_processing.html

That is, request to a “/file”, if it happens to hit his location,
will be processed as follows:

It doesn’t ends with ‘/’, hence nginx will try to open a file
for it, + . I.e. “/var/www/file” will be opened.
The (1) will likely fail (as there is no such file) and will
generate 404.

More importantly, request to a “/” will be processed as
follows:

It ends with “/”, so nginx will try to use index module to
handle it.
Index module will test + + file, i.e.
“/var/www/502.html” with your config. If it’s found - nginx will
do internal redirect to “/502.html”.
The internal redirect will end up in “location /”. This is
obviously not what you want.

With “rewrite ^ /502.html break;” the following two important
things happen:

a. Every request will be a request to a file (i.e. no internal
redirects). Note that “break” is important, without “break” nginx
will re-do location matching again after rewrite.

b. Every request will be a request to the “/var/www/502.html”
file.

That is, every GET request which hits the @502 location will
return the /var/www/502.html.

Note well:

For POSTs things won’t be that easy, as error_page set to a named
location won’t change request method to GET (in contrast to
error_page set to a uri), and the above will generate 405 Method
Not Allowed. If you have to handle POSTs as well, I see the
following options as of now:

Use proxy_pass to another server{} in location @502 and process
it there. You may also use proxy_method/proxy_pass_request_body
to simplify things.
Use some normally impossible location, like

location / {
proxy_pass http://backend;
error_page 502 //502.html;
}

location = //502.html {
root /var/www;
}

The (2) is more like a hack, but should work as //502.html isn’t
normally reachable via nginx due to merge_slashes being “on” by
default (see Module ngx_http_core_module).

Maxim D.

ewildgoose · March 7, 2012, 10:55am

On 06/03/2012 23:28, Maxim D. wrote:

}
root /var/www;
The key seems to be the rewrite clause - near as I can see, without
will behave much like normal static location, as described here:
generate 404.

The internal redirect will end up in “location /”. This is
obviously not what you want.

With “rewrite ^ /502.html break;” the following two important
things happen:

Hi, Thanks for this REALLY clear post. I wonder if there is somewhere
sensible to record this on the wiki - it’s very incisive on a subtle
subject!

Just to confirm my understanding, the point 2) above seems to the key -
ie the “index” module is doing a “redirect” without a break, hence
adding the explicit redirect WITH a break is what we need to prevent the
cycle. This fits with my observations (of course) because when I added
a location matching “location /502.html” this also prevents the cycle

Thanks for investing quite some time answering this!

Note, the background to this is that I want to bounce any incoming url
to “http://portal/login?redirect_to=$encoded_incoming_url”. I have
tried to achieve this using only core nginx, but I can’t handle the
corner cases such as incoming url having multiple parameters. I have
worked around using my cgi to generate the redirect. IF you believe
that this should be possible using only nginx I would be interested to
re-ask that question - however, I think you are going to confirm that
it’s not currently possible?

Many thanks again!

Ed W

ewildgoose · March 7, 2012, 12:40pm

2012/3/7 Maxim D. [email protected]:

location @502 {
    internal;
    root /var/www;
    rewrite ^ /502.html break;
}

This seems to be working, too:

location @502 {
root /var/www;
try_files /502.html =404;
}