Upstream lookup returns IPv6 address on Alpine

Using this (partial) configuration:

    resolver                   8.8.8.8 valid=300s ipv6=off;
    resolver_timeout           10s;
    upstream gs {
        server storage.googleapis.com:443;
        keepalive 128;
    }
    server {
        location / {
            proxy_set_header    Host storage.googleapis.com;
            proxy_pass          https://gs/$bucket_name$uri;
            proxy_http_version  1.1;
            proxy_set_header    Connection "";
        }
    }

on nginx:1.17-alpine getting errors:

*758 connect() to [2a00:1450:4001:824::2010]:443 failed (101: Network unreachable) while connecting to upstream, client: 10.12.0.1, server: , request: "GET / HTTP/1.1", upstream: "https://[2a00:1450:4001:824::2010]:443/example.com/index.html", host: "10.156.0.15"
*758 upstream server temporarily disabled while connecting to upstream, client: 10.12.0.1, server: , request: "GET / HTTP/1.1", upstream: "https://[2a00:1450:4001:824::2010]:443/example.com/index.html", host: "10.156.0.15"

The proxying did still work, apparently it was getting both IPv4 and IPv6 addresses, successfully using the IPv4 ones.

This is while running on Google's Kubernetes Engine (GKE) which does not support IPv6 networking.

Switching to nginx:1.17 made the errors disappear, so I assume the Alpine setup is missing something the Debian setup does.

Start by reproducing the supplied resolver and upstream configuration with nginx:1.17-alpine, then compare it with nginx:1.17, focusing on IPv6 resolution and the image setup. Done means the Alpine image no longer attempts unreachable IPv6 upstream connections when ipv6=off is configured, while IPv4 proxying continues to work.