nginx proxy_pass DNS funkiness

06 Mar 2015

This was noticed with Sugarcane. Sugarcane uses an AWS ELB to direct traffic to its servers. The issue at hand is that the domain used to point to the ELB does not have static IPs and this is something Amazon has warned about.

How this affects nginx is that nginx resolves the domains you pass to it once at startup and holds onto them. So if nginx has been running for two months straight, the IPs it had resolved for the ELB are probably not current.

This nginx ML thread talks about a work-around. If you use a variable in proxy_pass, they will be dynamically resolved and cached for about 5 minutes.

--- old/nginx.conf  2015-02-17 11:22:12.720729211 -0800
+++ new/nginx.conf  2015-03-06 10:57:22.829696006 -0800
@@ -36,7 +36,9 @@
         # proxy everything to a sugarcane test site
         #
         location / {
-            proxy_pass  http://test.sugarcane.rb;
+            resolver 192.168.21.84
+            set $scelb  "test.sugarcane.rb";
+            proxy_pass http://$scelb;
             proxy_set_header  X-Real-IP  $remote_addr;
             proxy_read_timeout 700;
             proxy_set_header Host $http_host;