Strange upstream behaviour

[Maxim Dounin]

Hello!

On Fri, Feb 05, 2016 at 06:26:28AM -0500, abauer wrote:

> We are using nginx as a loadbalancer in front of docker containers. Most of
> the time this works without problems. But sometimes (~0.1% of the requests)
> the requests are sent to the server group name instead of one of the members
> of the servergroup.

[...]

> But randomly this happens:
> 
> 03/Feb/2016:04:00:25 +0100 "/api/v1/login" 502 52 "Jersey/2.7
> (HttpUrlConnection 1.8.0_51)" "time=0.192" "<public dns>"
> "upstream=gateway"
> 
> As you can see, it uses the server group name as the upstream target (which
> fails of course since this is not a valid host).
> 
> What could be the cause of this behaviour?

Upstream name can be seen in the $upstream_addr variable if nginx 
wasn't able to select an upstream server to connect to because all 
servers were down as per max_fails/fail_timeout.  At the same time 
the "no live upstreams" error is logged to the error log.

-- 
Maxim Dounin
http://nginx.org/