Strange error_page behavior

[Fabiano Furtado Pessoa Coelho]

Hi...

I have the following setup in my NGINX 1.18.0 server:

http {
  ...
  error_page 400     /my400.html;
  error_page 401     /my401.html;
  error_page 403     /my403.html;
  error_page 404     /my404.html;
  error_page 405     /my405.html;
  error_page 413     /my413.html;
  error_page 500     /my500.html;
  error_page 502 503 /my503.html;
  error_page 504     /my504.html;
  ...
  server {
    ...
    allow 10.0.0.0/8; # my internal network
    deny all;         # all others networks

    error_page 403 =503 /my503.html;
    #error_page 502 503  /my503.html;

    location ~ ^/(my400|my401|my403|my404|my405|my413|my500|my503|my504)\.html$
{
      root /usr/share/nginx/my_custom_error_pages;
      internal;
      allow all;
    }
    ...
    location / {
      ...
      proxy_pass http://...;
    }
  }
}


In this setup, I want to allow my internal network to access my
backend server via proxy_pass AND deny this access to others networks
(error 403 "redirected" to 503 --> defined in "error_page 403 =503
/my503.html;").
This is working great, but when my backend gets down, I receive a
default NGINX "502 Bad Gateway" error page instead of my custom
my503.html page.
Why this is happening?
I can solve this issue uncommenting the "#error_page 502 503
/my503.html;" line in server block, but this configuration is already
defined in http block, which is hierarchically superior in comparison
with server block. Why NGINX isn't regarging the error_page in http
block?

Thanks in advance.