On Mon, Aug 23, 1999 at 04:38:23PM -0600, I could have sworn Jason Gunthorpe wrote: > Almost every proxy out there seems to have problems dealing with APT's > very agressive use of HTTP/1.1 features such as keepalive. AFAIK APT does > not violate the RFC, is it just the only agressive user of HTTP/1.1 out > there.
I think that the keepalive may screw with the concurrent server model of many of these, including rproxy. squid is an iterative server, and I guess that any decent administrator would make sure there were enough fd's around that multiple keepalive requests would not starve the process. Oh, and squid also groks HTTP/1.1. rproxy is simply a proof-of-concept implementation and technically isn't meant to grok anything ;) FWIW, I have my system setup somewhat wonky in that I have Netscape talk to rproxy which talks to the local squid cache, it then talks to rproxy at my ISP which in turn talks to squid at the ISP. This works fine, but if I try it with apt (either apt-get or the usual dselect method) by exporting http_proxy: If http_proxy is the local squid it works fine (even though there is an rproxy at the other end of that) but if I point it at the local rproxy it does not. So I draw the conclusion in the first paragraph from this. > wwoffle and apache-proxy have problems, squid and many others do not, I > suggest you strace the http method process and see what system call is > causing it to print that error.. Unfortunately I am not fast enough to find out the pid of the http method process and strace it before the problem occurs. As you claim apt is working within spec, I'll try the reverse solution and work with the rproxy authors to fix rproxy. Even if that doesn't occur, all I miss out on is saving most of the bandwidth of those large Packages.gz downloads every day ;) (ie, "would be nice" category :) Cheers, -- Matt
