[beau]

I tried this a long time ago. At the time, certain queries were hard for a server to answer. Other clients connected to the zombie server stalled until something timed out. With HTTPS, a load balancer can direct new queries to servers that are responsive.

[thinkloop]

I was also curious about this (and thanks for the article, very informative!)

Are you saying that at one point the servers would crash relatively often, which would leave sockets clients hanging, unless some complicated client-side code was written - whereas without sockets, a load-balancer could automatically switch clients to functional servers, without extra coding, and mitigate the issue? Isn't the problem the crashy servers?

[beau]

Sure, but why keep state when you don't need to? Now that HTTP/2 is ubiquitous, sending new GET requests is no more expensive than sending bytes through a WebSocket.