When using proxy_pass, can /etc/hosts be used to resolve domain names instead of "resolver"?

Question

Can /etc/hosts be used instead of resolver when using proxy_pass?

I need to perform a proxy_pass to the same nginx machine. Is there a way to resolve the domains using the machine's /etc/hosts file instead of specifying a DNS server thru the "resolver" property?

This will save me the additional hops needed to reach the same server. I have tried setting up the internal IP mapped to the DNS in /etc/hosts file but nginx is still reading from the DNS server set in the resolver property. Or is there a way to make the HTTPProxy module to consider the /etc/hosts file settings?

Thanks for any advice you could share..

This is the same question I posted in the nginx forum: http://forum.nginx.org/read.php?11,218997

Cirrostratus

The strange thing , this is working if you use upstream (host resolution) – Kurtz

Upstream worked for me as well, it's a bit of extra code but it seems cleaner than some of the other methods listed below. – Camshaft

It also possible to workaround with systemd-resolved if using a server with systemd. See this answer https://mcmap.net/q/393611/-proxy_pass-does-not-resolve-dns-using-etc-hosts – Announcement

Answer 1

You can get around this by installing dnsmasq and setting your resolver to 127.0.0.1. Basically this uses your local DNS as a resolver, but it only resolves what it knows about (among those things is your /etc/hosts) and forwards the rest to your default DNS.

Dewayne

But sadly dnsmasq does not automatically detect changes in hosts file. You have to send HUP. – Ephemeral

Can you give more details on how we should setup dnsmasq in order to get our nginx to resolve the other container dns name properly ? just installing dnsmasq and adding extra_hosts: - "container_name:127.0.0.1" to my docker-compose.yml config didn't resolve the issue – Triggerfish

If you are using docker I think DNS works a little bit different, especially if you're doing container to container communication. Check out the docker networking configuration for more info. – Dewayne

Will this work with other *nix flavors (Rasbian specifically) that don't have proxy_pass? – Inhaler

Answer 2

A workaround is to use Nginx map, in order to copy the /etc/hosts content.

map $wanted_host $wanted_host_ip
{
    default 127.0.0.1;
    b.dev.local X.X.X.X;
    a.dev.local X.X.X.X;
}

server
{
    listen              80;
    server_name         ~^(?P<wanted_port>[0-9]+?)-(?P<wanted_host>.+?)\.HOSTNAME$;

    location /
    {
        proxy_pass http://$wanted_host_ip:$wanted_port;

    }
}

This will map wanted_hostto wanted_host_ip , like a resolver.

Kurtz

Would this work if has multiple (sub)domains, where just an IP is not enough to know which virtual server to hit? – Se

Sure, you could use domain as X.X.X.X, if it resolves. – Kurtz

Answer 3

In my own case on my Windows development and testing machine, I had to install technitium DNS server. Then created a new zone (example.com) and added an A record for my proxy domain's TLD (@) pointing to 127.0.0.1. Then pointed the nginx resolver in the config file to the default Windows system's domain name (desktop-hmvl5656). Note: In my own default installation the technitium DNS server also used the default Windows system domain name for the backend management webportal

location /example {
   resolver desktop-hmvl5656;
   proxy_pass http://example.com$request_uri;
}

Intent