Return 503 when no backends are available

[abh]

We had an outage in our k8s cluster yesterday. After it cleared up the remaining side effect was that haproxy while the backends were unavailable had sent a bunch of 404's to the CDN and browsers that were cached, rather than 503s.

In the nginx ingress they recently changed this to be, I think, more sensible: kubernetes/ingress-nginx#1513

[colinjfw]

I was affected by this recently as well. It's good to see it's been changed upstream in the nginx ingress. I believe the haproxy-ingress should reflect this pattern as well.

For a temporary solution, I built a custom default backend that will return 503 on specific subdomains. This is definitely a workaround though, I think the default backend receiving traffic on no endpoints available is confusing.

[jcmoraisjr]

This should be fixed since the current snapshot version, v0.5-snapshot.2. Closing. Let me know if the problem persists updating this same issue. Thanks for pointing this out.

[abh]

It’d be nice if you included the git commit that made the change... (either by referencing the issue in ththe commit message or by adding a reference to the commit when you close the issue...

[jcmoraisjr]

Sure. This should have been fixed on 011df80 where the upstream controller was updated.