nginx reverse proxy causing TCP queuing spikes

[Robert Paprocki]

> I do think this is related to 'proxy_read_timeout 60m;' leaving too
> many connections open.  Can I somehow allow pages to load for up to
> 60m but not bog my server down with too many connections?

Pardon me, but why on earth do you have an environment in which an HTTP request can take an hour? That seems like a serious abuse of the protocol. 

Keeping an HTTP request open means keeping the associated TCP connection open as well. If you have connections open for an hour, you're probably going to run into concurrency issues.