Not sure if you already did or the default config does it but if you only have one backend in an nginx reverse proxy it might also make sense to configure the max_fails and possibly the fail_timeout options so nginx won't consider your backend down for a few seconds every time it receives a TCP error connecting to that (single) backend. max_fails=0 in the named upstream section (see http://nginx.org/en/docs/http/ngx_http_upstream_module.html#upstream ) is what you want here.