fix(nginx): set service-upstream to "true" (#532)

Instead of having NGINX maintain it's own list of upstreams, we want it
instead to use the service ClusterIP such that k8s can handle e.g. cases
where pods are evicted.

Without this we end up with 5xx errors e.g. on deploys. See
kubernetes/ingress-nginx#257 for more details.

charts/posthog/values.yaml

@@ -513,6 +513,12 @@ ingress-nginx:
       # -- [ingress-nginx documentation](https://kubernetes.github.io/ingress-nginx/user-guide/nginx-configuration/configmap/#use-forwarded-headers)
       use-forwarded-headers: "true"
 
+      # Ensure that we use the service rather than maintaining a list of
+      # upstreams. Without this NGINX will maintain a list which can cause
+      # issues when pods are terminated. See
+      # https://github.com/kubernetes/ingress-nginx/issues/257 for details
+      service-upstream: "true"
+
       # Use JSON format for logs, such that we can easily parse them in e.g. Promtail
       #
       # We also add in: