HTTP/1.0 shouldn't be used

[Junien F]

Hi,

It would appear that nginx uses HTTP/1.0 when connecting to haproxy.
It would also appear that haproxy uses HTTP/1.0 when connecting to its non-cached backends.

HTTP/1.0 doesn't have keep-alive support, so this could be causing a lot of churns.
When I tcpdump'ed traffic from a content-cache unit to its backend, filtering only new TCP connections [1], I'm seeing a constant stream of packets, which shouldn't happen.

Thanks

[1] sudo tcpdump -n -i $int 'host $backend and tcp port 443 and tcp[tcpflags] == tcp-syn'

[Junien F (axino)]

tcpdump showing a constant stream of new connections could actually be explained by haproxy healthchecks, which we configure to use HTTP/1.0 for some reason. Not sure if we can/want to make them using a single, kept-alive session or not.

[Haw Loeung (hloeung)]

Landed various things making the switch to either HTTP/2 or HTTP/1.1. What's left is to update the Nginx layer to use upstream module and activate keepalive as documented[1] with the example below:

| upstream http_backend {
| server 127.0.0.1:8080;
|
| keepalive 16;
| }
|
| server {
| ...
|
| location /http/ {
| proxy_pass http://http_backend;
| proxy_http_version 1.1;
| proxy_set_header Connection "";
| ...
| }
| }

The upstream name, in this case "http_backend", needs to be unique across all sites.

[1]https://nginx.org/en/docs/http/ngx_http_upstream_module.html#keepalive