CERN server and HTTP/1 clash?
Marc Andreessen (marca@ncsa.uiuc.edu)
Thu, 14 Oct 93 13:29:37 -0700
Frederick G.M. Roeber writes:
> A couple days ago I started playing with NCSA Mosaic 2.0.5 (yes, I
> know it's unsupported). Using it, I receive a lot of errors from
> http servers, particularly the info.cern.ch server (which is
> presumably the CERN httpd). The error is the usual "Error /
> Requested document (URL blah) could not be accessed. The
> information server either is not accessible or is refusing to serve
> the document to you." Bouncing on reload doesn't help.
>
> I've tried telnetting to port 80 of info when I receive this
> message, and sure enough "Connection refused." If I wait a little
> while, then I can telnet or reload.
>
> This can't be just a server problem (like overload) -- it started
> exactly when I started using 2.0.5. But it can't be just Mosaic, or
> telnetting would work fine. So I wonder: might 2.0.5's new HTTP/1
> support annoy old HTTP/0 servers like what's running on info?
Nope, that's not it -- the CERN server is the original HTTP/1.0
server.
Believe it or not, I think the CERN server is dying on a SIGPIPE in
some circumstances (i.e. when the connection is interrupted on the
client end -- the socket breaks, the server tries to keep writing, the
client isn't trapping the signal, and boom). I sent a note around a
while back but never got a response from Tim, and I don't know what he
plans to do about it (if anything), or even if it really is the
problem.
Marc