Support for HTTP/1.0 Healthchecks by Loadbalancers

Please describe your use case / problem. Some loadbalancer vendors are using http/1.0 healthchecks (haproxy default, digitalocean exclusively) When deploying ambassador behind such a loadbalancer, Ambassador/envoy is replying with 426 UPGRADE REQUIRED answers:

ACCESS [2018-11-21T10:12:10.359Z] "GET /ambassador/v0/check_alive HTTP/1.0" 426 - 0 0 0 - "-" "curl/7.58.0" "-" "localhost:8080" "-"
ACCESS [2018-11-21T10:12:18.139Z] "GET /ambassador/v0/check_ready HTTP/1.0" 426 - 0 0 0 - "-" "curl/7.58.0" "-" "localhost:8080" "-"

Therefore, the upstream Loadbalancers are marking the service as DOWN (they expect a 200 OK answer).

Describe the solution you’d like I think we should be able to answer simple http/1.0 calls. envoy does support it accept_http_10 https://www.envoyproxy.io/docs/envoy/latest/api-v2/api/v2/core/protocol.proto

Describe alternatives you’ve considered I have now resolved it by manually setting the upstream loadbalancer to TCP only check. So the Loadbalancer only checks if the port ist open. Thats not optimal, but better than no service

[phenoxim]

Same issue with me. Could anyone give a hint to resolve this propblem?

FYI

Update http_config here: https://github.com/datawire/ambassador/blob/7ccc4945f7b50af35c77ffddca2cc7f22979882a/ambassador/ambassador/envoy/v2/v2listener.py#L353

http_config['http_protocol_options'] = {
        'accept_http_10': True
}

and build a private ambassador image.

Solved this issue.

[richarddli]

Fixed in 0.61.