>From what I can tell there's nothing (technically) wrong with what Tomcat + ISAPI Redirector is doing here.
What's actually happening here is that Tomcat internally only provides a Content-Length header if it can determine the length of the content easily (e.g. it's a static file) or the Servlet generating dynamic content provides one itself. Any other response content is just written out to whatever connector (HTTP/AJP) is being used. If it's via the HTTP connector, then chunk encoding is automatically provided. Likewise with the AJP connector and mod_jk in Apache - chunk encoding is automatically provided by Apache for all responses that would benefit from it (mod_jk doesn't do anything special to achieve this). IIS being the braindead poor cousin is not so accomodating, as it requires any ISAPI extension to not only tell it that it would like to use persistent HTTP connections, but also provide all of the HTTP level details (including headers and content encoding) to make it work. All IIS does is detect if you've done enough to make the connection persistent and keep open/close the connection if you haven't. Since the current ISAPI redirector doesn't implement chunk encoding, IIS whacks in a Connection: close header on all responses without Content-Length and closes the connection to the client. Closing the connection is actually a valid method of terminating a response message in HTTP 1.1 (as Rainer alluded to, the statement attributed to IBM below about a Content-Length being required in HTTP 1.1 is wrong in a lot of ways - indeed in some responses Content-Length must not be included). http://www.w3.org/Protocols/rfc2616/rfc2616-sec4.html#sec4.4 and http://www.w3.org/Protocols/rfc2616/rfc2616-sec14.html#sec14.10 seem to be pretty clear on how an HTTP application that doesn't support persistent connections should behave - what IIS + ISAPI Redirector is doing is (from what I can tell) valid HTTP 1.1, it's just not polite in this day and age. The fact that your web service call works when accessing Tomcat directly via the HTTP connector implies that the client can handle chunk encoded responses, since the Tomcat HTTP connector provides this for anything that doesn't have a Content-Length set, and your logs indicate your web app isn't setting one. I might have missed some magic Content-Length calculation for small responses in the Tomcat HTTP connector, but I'd imagine that wouldn't work in all cases (e.g if you had a really large response message). You could test this theory by sniffing the network traffic when connecting directly to Tomcat, by installing Apache + mod_jk, or by using my patched IIS connector from http://sourceforge.net/projects/timsjk (the latter two options will provide chunked encoding on all responses coming from Tomcat that don't already provide a Content-Length. (btw I'd be very surprised if my chunked encoding patch attached to the BZ issue worked, as it hasn't been updated to trunk for quite a while. http://www.w3.org/Protocols/rfc2616/rfc2616-sec3.html#sec3.6.1 states that HTTP 1.1 applications must be able to receive chunk encoded responses so if adherence to HTTP 1.1 is important in your environment, you should be able to argue that this is a valid solution. Other more desperate options would involve content buffering Servlet Filters that wrap the response to calculate and set the Content-Length headers (there were a couple floating around the Tomcat world a while back) and hacking your web service toolkit to buffer messages pre sending and set the Content-Length header. I've used the filter approach in the past (pre HTTP 1.1), and it might be workable as long as your web services responses have predictably and reasonably small content sizes. cheers tim -----Original Message----- From: Woytasik Joe [mailto:[EMAIL PROTECTED] Sent: Saturday, 5 January 2008 10:10 a.m. To: Tomcat Users List Subject: RE: Content_Length Problem Rainer, Thanks for the quick response! I am able to repeat this request, and each time I get the same response. The logging level is set to debug, but unfortunately I am unable to send the log file (company policy). I am going to scrub the log file to remove any sensitive information, I will send that your way shortly. I did some network sniffing and CONTENT_LENGTH is not sent. I built a new isapi_redirect.dll using the patch provided in Bugzilla. This patch was supposed to allow chunked encoding, but I am not sure if I applied it right. Is there a registry setting that I need to change to allow chunked encoding with this patch, or does it do it automatically? Thanks- Joe -----Original Message----- From: Rainer Jung [mailto:[EMAIL PROTECTED] Sent: Friday, January 04, 2008 2:06 PM To: Tomcat Users List Subject: Re: Content_Length Problem Hi Joe, are you able to reproduce the behaviour with few, maybe only a single request? If so: you can increase JkLogLevel to "debug" (not recommended for high load production size, because it produces a lot of log lines), reproduce the problem and make the log file available. What I didn't really understand from your post: do you know, if the Content-Length header gets send or not? How do you know? Did you sniff the network traffic or do you only know from the CICS behaviour? Lastly: HTTP/1.1 responses without Content-Length headers are valid if they are using chunked encoding (Transfer-Encoding: chunked). I think at the moment the isapi redirector does not use chunked encoding (didn't yet test, but there's an open RFE to implement chunked encoding in the isapi redirecotr), but I want to clarify the absolute statement concerning the http protocol. Chunked encoding replaces the content-length header with sending the number of bytes available in front of every chunk, s.t. the receiving node knows, how much data to expect, without the sending node needing to know the full size before sending. Dynamically generated content often uses chunked encoding to prevent the need of buffering the whole reposne before sending. Regards, Rainer Woytasik Joe schrieb: > I have a custom webservice hosted on IIS 6.0 and Tomcat 6, and I am > using the latest version of the isapi_redirect.dll. The problem > occurs when a CICS mainframe application tries to call this webservice. > Everything appears to work fine, but the CICS application receives a > response indicating a zero length message. I can view the message > being sent from the webservice and this is definitely not the case > (have also taken several packet traces to confirm this). We sent our > problem to the folks over at IBM and they say that the CONTENT_LENGTH > is not being set. > > Here is their response: > > The problem is that there isn't a Content-Length header sent by the > IIS/Tomcat Server. CICS receives the headers and finds it is an > HTTP/1.1 response for a Connection: Close. There isn't a > Content-Length header so there can't be any user data (HTTP/1.1 has to > supply Content-Length) so DFHWBCL just closes the session. PI domain > then indicates that it failed to receive a response. The customer > needs to investigate why their IIS server didn't return a > Content-Length header. > . > The Content-Length header is mandatory for CICS' HTTP/1.1 > conversations. This is documented in the CICS/TS 3.1 Internet Guide, > section 1.3.11.1 ("CICS Web support behavior in compliance with > HTTP/1.1"); this chapter documents the requirement in a section titled > "New Behavior for CICS TS Version 3", under the first item "CICS > checks inbound messages for compliance with HTTP/1.1, and handles or > rejects non-compliant messages": > Note: CICS requires the Content-Length header on all inbound > HTTP/1.1 messages that have a message body. If a message body is > present but the header is not provided, or its value is inaccurate, > the socket receive for the faulty message or for a subsequent message > can produce unpredictable results. For HTTP/1.0 messages that have a > message body, the Content-Length header is optional. > . > The reason this is mandatory under CICS/TS 3.1, is due to our > adherance to HTTP/1.1 specifications -- in other words, your HTTP/1.1 > Web Service PROVIDER platform must provide this header, to be considered compliant. > . > Please ensure the IIS/Tomcat server sends a proper header. > > If we make the same request directly to Tomcat using the port number > it works fine. The problem either lies in the isapi_redirect.dll or > the IIS configuration. Does anyone have any ideas on what I can try > to resolve this? Is there a know bug with the isapi_redirect.dll and > CONTENT_LENGTH? > > Thanks- > Joe > > This e-mail is confidential. If you are not the intended recipient, you must not disclose or use the information contained in it. If you have received this e-mail in error, please tell us immediately by return e-mail to [EMAIL PROTECTED] and delete the document. > > E-mails containing unprofessional, discourteous or offensive remarks violate Sentry policy. You may report employee violations by forwarding the message to [EMAIL PROTECTED] > > No recipient may use the information in this e-mail in violation of any civil or criminal statute. Sentry disclaims all liability for any unauthorized uses of this e-mail or its contents. > > This e-mail constitutes neither an offer nor an acceptance of any offer. No contract may be entered into by a Sentry employee without express approval from an authorized Sentry manager. > > Warning: Computer viruses can be transmitted via e-mail. Sentry accepts no liability or responsibility for any damage caused by any virus transmitted with this e-mail. --------------------------------------------------------------------- To start a new topic, e-mail: users@tomcat.apache.org To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To start a new topic, e-mail: users@tomcat.apache.org To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED] --------------------------------------------------------------------- To start a new topic, e-mail: users@tomcat.apache.org To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
