I would be interested to hear if anyone has encountered the
situation I am going to describe.  We have put in a working patch
but I would like to pin down the nature of the problem more
precisely.  I will try to provide as brief a description as possible, but
it will be more lengthy than the typical question.

Background: For over a year, we have been running a servlet in
production acting as an intermediary between our host system and
client browsers in order to duplicate most of the features of our PC-
based client software that communicates directly with our host
system (e.g. maintain a session with the host, provide some
temporary storage of files).  Nothing unusual in what I have
described, of course.  The system has run well.

Although the operation is more complex than what I am about to
say, one can essentially think of each display in the client browser
as a frameset targeting left and right panels. Typically, the left
panels are small documents and the servlet saves them in a
temporary client area on the server's disk, and points the frameset
(sent directly from the servlet) to the left panel on disk. Often, the
right panels are quite large documents, so the servlet sets the
frameset to point the right panel back at the servlet with appropriate
information to permit the servlet to pass the large document
directly to the client browser as it is received by the servlet from
the host system.

Admittedly, this explanation leaves out much detail, but the basic
notion is that the right panel request from the frameset is for a
document sent directly from the servlet.

All this works fine, usually.  Recently, we have encountered four
users who have experienced problems viewing the right panel,
generated directly from the servlet.  Our patch has remedied the
problem for two of the users, and we await testing by the other two.

We traced one of the users experiencing the problem and found
that the servlet was receiving two requests for the right panel of the
frameset, very close together in time (of the order of 200
milliseconds).  The first request showed no Referer (relatively
unusual, although MAC MSIE browsers seem to leave the Referer
field null) and a User-Agent simply of Mozilla/3.01 compatible.
That is, an older level than we would accept normally, and, more
oddly, no other information.  The second request looked normal,
with a Referer page, and User-Agent info of Mozilla/4.0, browser
type, windows type.

I might note that for other "items" referred to in our HTML (images,
stylesheets, etc.) we saw some accessed by the Mozilla/3.01
"guy" and some accessed by the browser.  Who did the accessing
seemed to depend on what "level" referred to a given item, but that
all worked as expected, with no double requests for any of these
items.

What we feel was happening was that some items were accessed
by the proxy server on behalf of the browser and some directly by
the browser, and that, normally, the coordination was perfect.

The exception seemed to be for the dynamic request for the right
panel of the frameset.  It appeared that there was some indecision
between the proxy server and the browser, resulting in two
requests, one right after the other.  Based on our testing, the
second request, the one directly from the browser, seemed to be
the one that would be displayed in the browser.  So, we patched
our code to ignore the first request, and that has worked well so far.

I might note that one our people spoke to a major provider and they
told him that they had what appeared to be similar complaints from
other customers about other web services than ours. We may learn
more as time goes on.

Well, as you see, although we are pleased to have at least a
temporary fix for the situation, we would like to get a better handle
on it.  We have spoken to the supplier of server software to one of
the customers in question and to ISPs but the most we have heard
is something to the effect that, yes, that might be what is
happening.

If anyone has something to add, I would be pleased to hear it.  I do
apologize for the length and the rather specific nature of this post.

Dave Cottee

___________________________________________________________________________
To unsubscribe, send email to [EMAIL PROTECTED] and include in the body
of the message "signoff SERVLET-INTEREST".

Archives: http://archives.java.sun.com/archives/servlet-interest.html
Resources: http://java.sun.com/products/servlet/external-resources.html
LISTSERV Help: http://www.lsoft.com/manuals/user/user.html

Reply via email to