Hi,

Note: I had to change the hostnames and domain names.

nginx version: nginx/1.4.6 (Ubuntu)

I have the following the configuration:
upstream ssl-app-cluster {
ip_hash;
server app01.example.com:8443 max_fails=1 fail_timeout=60s;
server app02.example.com:8443 max_fails=1 fail_timeout=60s;
}

And during the Application deployment(rolling deployment) to those backend servers(app0{1,2}.example.com) the port 8443 will not be available for 60sec. So, during the deployment I tried hitting the server(api.example.com) continuously to know, how nginx is routing the traffic, when one of the servers in down. As I am using ip_hash, my ip is bound to app01.example.com initially.

Deployment process:
----------------------------
1. I am running the curl command to hit the server api.example.com in a for loop and I am being served by app01(I see this on error log with debug enabled).
2. Deployment process has taken down the app02 host for upgrading the application on it. This doesn't effect anything, as my IP is bound to app01 and I am being served by app01.
3. Now, the deployment has taken down app01 for deployment. So, I see there's a "111 connection refused error" and nginx tried to connect to "http next upstream" which is app02. which is successful.
4. But, nginx has to wait for 60sec(fail_timeout) to establish connection to app01, But I see nginx is trying to connect to app01 immediately and I see there's an error: "111 connection refused error" again and then it connecting to app02.

Can someone please, tell me why this is happening and how can I change this?

Thanks,
Ravi