Nginx – Amazon Load Balancer gives out Error 503

I managed to accomplish this using an approach like the one described here: http://scottwb.com/blog/2013/10/28/always-on-https-with-nginx-behind-an-elb/.

In NGINX config, I added:

1. A specific location directive for the healthcheck URL ("/healthcheck") that does not do any HTTPS enforcement.
2. A redirect if the "X-Forwarded-Proto: https" header does not exist.

I finally figured out the issue, hope this saves somebody else a few hours. The problem turned out to be a combination of my load balancer Probe settings and the IIS bindings for my websites.

In IIS on the VMs I have 2 websites configured. Since the 2 websites share 1 public IP, I'm using a Host Name when setting up the bindings in IIS. So basically my bindings are configured to send www.site.com traffic to site #1 and test.site.com traffic to site #2. I didn't set up a default binding to handle port 80 traffic with a DNS name other than those 2. So if you go to http://[internal_ip_or_machine_name] no page gets loaded because there isn't a binding to handle it.

My probe was an HTTP probe set to hit /probe.html on port 80. The problem is that the probe likely hits http://[internal_ip_or_machine_name]/probe.html since the probe knows nothing about the DNS names I use, and that URL doesn't have a binding to handle it in IIS. So the probe was failing and assuming all of my VMs were offline. This resulted in the 502 error.

Not sure what fix I'll do yet. I'll either add a default binding for port 80 in IIS that points to the production website, or I'll use a TCP probe.

What really threw me for a loop is that HTTP worked fine as I mentioned above. I must have happened to have a default IIS binding in place while testing HTTP and then removed it before testing HTTPS. Now without a valid probe, both HTTP and HTTPS fail. With a valid probe, both work.