Hello,

We have an Nginx where we configured http://nginx.org/en/docs/http/ngx_http_proxy_module.html#proxy_next_upstream in Nginx main config file at http level.

We don't want to send traffic to the next upstream on 504:
----
proxy_next_upstream error timeout http_502 http_503 non_idempotent;
----

At vhost level we don't redefine this directive. But periodically we see the following in Nginx logs:
----
141.101.69.85 - 192.168.1.10 - 59.489 - [08/Aug/2020:11:10:41.098 +0000] - POST - /api/ - HTTP/1.1 - 499 - - 0 - 2054 - "Java/1.8.0_242" - api-worker - "192.168.1.11:8080, 192.168.1.12:8080" - "504, -" - "0.000, 0.000" - "31.000, 28.489"
----

api-worker - "192.168.1.11:8080, 192.168.1.12:8080" - "504, -"

From the logs we see that Nginx received 504 status from the first upstream and then for some reason send traffic to the next one, despite the fact that it should not do it on 504 http status.

We did a short test using http://httpstat.us/504
----
upstream test-504 {
server 104.31.86.226:80;
server 104.31.87.226:80;
}

server {
listen 443 ssl http2;
server_name domain.tld

# Test 504
location /test-504 {

# proxy_next_upstream error timeout http_502 http_503 http_504 non_idempotent;

proxy_pass http://test-504/504;

proxy_set_header Host httpstat.us;
}
}
----

If we comment 'proxy_next_upstream' it uses one defined in the main config at http level and don't send traffic to the next upstream. If we uncomment it, we see that Nginx send traffic to the next upstream. All works as expected and described in the documentation.

But the question remained for our production: Why Nginx send traffic to the next upstream on 504 error?
1. It is some misconfiguration on our side, maybe timeouts on any other directives should be enabled/disabled?
2. It is some kind if misunderstanding how 'proxy_next_upstream' directive works?
3. It is some kind of bug?

Thank you!