----------------------------------------------------------------
BEFORE YOU POST, search the faq at <http://java.apache.org/faq/>
WHEN YOU POST, include all relevant version numbers, log files,
and configuration files. Don't make us guess your problem!!!
----------------------------------------------------------------
Jean-Luc:
I was not being clear. We have only changed the platform of the Web Server
piece:
Before (1 box):
Web Server/JServ on one box. Sun e250 Solaris 2.7 w/JDK 1.1.8_09a. JServ
1.1b3
After (2 boxes):
Web Server:
Intel Linux RH6.1 Apache 1.3.9
JServ:
Sun e250 Solaris 2.7 w/JDK 1.1.8_09a JServ 1.1b3
As I said in the previous note, what we are doing is a Multi-Part file
upload. I will try to get you a succint demonstration of the problem. For
now, what we are doing roughly is (I know that we are using a BufferedReader
rather than a BufferedInputStream, but this never caused a problem before). I
will work on the test scenario to let you know if that makes a difference:
Find the content-type:
String contentType = req.getContentType();
String boundaryString = null;
if (contentType!=null) {
int semicolon = contentType.indexOf(";");
int boundary = contentType.indexOf("boundary=");
// save boundary string and content type
if (boundary != -1) {
boundaryString = contentType.substring(boundary+9);
}
if (semicolon != -1){
//trim contentType upto before the first semicolon.
contentType = contentType.substring(0,semicolon);
}
}
Load into Multi-Part:
contentType.equals("multipart/form-data")) {
try {
// #### from multipart request ####
MultiPartForm mpf = new MultiPartForm(req, boundaryString);
// place parameters in context and get fileVect
Vector fileVect = mpf.readFormData(context);
Here's the rough Multi-Part code:
public Vector readFormData (StringHashtable context)
throws IOException, WeaselException
{
int bsLen = bs.length();
char[] c = new char[4096];
int numTries = 0;
FastStringBuffer sb = FastBufferPoolMgr.getNewFastStringBuffer();
InputStream instream = req.getInputStream();
BufferedReader br = new BufferedReader( new InputStreamReader
(instream)
);
Vector fileVect = new Vector();
// read in 4K chunks
/**
* for separate JServ/Apache instances the following code is needed to
handle
* some subtle I/O differences. The same code will bomb if it is
* used for JServ/Apache combo boxes
*/
// for (numTries = 0; !br.ready() && numTries < MAX_NUM_TRIES;
numTries++)
{
// Timer.runTimer();
// }
// if (!br.ready()) {
// throw new WeaselException("MultipartFormData.readFormData: could not
r
ead input stream -> #tries = "+numTries);
// }
int len = br.read(c,0,4096);
while (len >= 0) {
// append to the stringbuffer
if (len > 0) sb.append(c,0,len);
/**
* for separate JServ/Apache instances the following code is needed
to
handle
* some subtle I/O differences. The same code will bomb if it is
* used for JServ/Apache combo boxes
*/
// for (numTries = 0; !br.ready() && numTries < MAX_NUM_TRIES;
numTries++
) {
// Timer.runTimer();
// }
// if (!br.ready()) {
// break;
// }
len = br.read(c,0,4096);
}
// convert string buffer to string
String mpd = sb.toString();
FastBufferPoolMgr.returnFastStringBuffer(sb);
// loop through the form looking for the boundary string
Jean-Luc Rochat wrote:
> ----------------------------------------------------------------
> BEFORE YOU POST, search the faq at <http://java.apache.org/faq/>
> WHEN YOU POST, include all relevant version numbers, log files,
> and configuration files. Don't make us guess your problem!!!
> ----------------------------------------------------------------
>
> Michael Amster wrote:
> >
> > Hi:
> >
> > We have been developing a site using Apache JServ 1.1b3 on Sun Solaris
> > 2.7 with JVM 1.1.8_09 and Apache 1.3.9. During development we had
> > apache and JServ on a single box and all worked well.
> >
> > Now we are going to production with a load balancing. We have Apache
> > 1.3.9 running on 2 Intel Linux boxes running RH6.1.
> >
> > It seems that any input stream behavior does not work reliably, in
> > particular POST operations and multi-part form reading.
> >
> > We see that I/O is acting very differently on the separated setup. It
> > looks like what is good for the goose is not good for the different
> > production architecture. Yesterday I made some changes in the
> > Multipart form implementation to handle the case of the stream not being
> >
> > available. It turns out that there is something truly different,
> > I/O-wise, between having Apache/JServ on one box and having them on
> > separate boxes. In other words one multipart implementation does not
> > satisfy both scenarios. The change I made to fix the production setup
> > was bombing out on the development setup. Here is the gist:
> >
> > On one box ->
> > 1. the input stream always seems to report not ready and no
> > bytes
> > available.
> > 2. calls to read() simply work and and return non-zero lengths
> > until
> > the end of file is reached. this (I believe) is the way things
> > are supposed to work.
> >
> > On 2 boxes ->
> > 1. the input stream correctly reports if it is ready with calls
> > to ready(). the fix I had was to wait until the stream was ready
> >
> > before each read.
> > 2. calls to read when the stream was not ready would hang the vm
> >
> > 3. discovery of EOF was due to a wait for a finite amount of
> > time
> > without the stream being ready. I.E. read() never returns
> > negative number.
> >
> > Any help really appreciated.
> >
> > Thanks,
>
> Hi,
>
> difficult to imagine what's happening, because you changed 4 parameters
> at the same time :
> the OS (from Solaris to Linux), the JVM, separated Apache & JServ on 2
> machines, and used load-balancing.
> The test I'd try : run Apache on 1 linux box & JServ on the "less
> unstable" JVM (Solaris).
> You can use load-balancing here. just don't start JServ ont the Linux
> box.
> And see what's happening. If it works, you've something wrong either
> with Linux (glibc version?), either with the JVM.
> Believe me load-balancing is working on Solaris. Seems to be working on
> Linux too.
> Send me an exemple of a servlet causing problems, and I'll try it here.
>
> Jean-Luc
>
> --
> --------------------------------------------------------------
> Please read the FAQ! <http://java.apache.org/faq/>
> To subscribe: [EMAIL PROTECTED]
> To unsubscribe: [EMAIL PROTECTED]
> Archives and Other: <http://java.apache.org/main/mail.html>
> Problems?: [EMAIL PROTECTED]
--
~-~-~-~-~-~-~-~-~-~-~-~-~-~-WEBEASY-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~-~
Michael Amster [EMAIL PROTECTED]
1148 4th Street, Suite 204 Tel: 310.576.0770
Santa Monica, CA 90403 Fax: 310.576.2011
--
--------------------------------------------------------------
Please read the FAQ! <http://java.apache.org/faq/>
To subscribe: [EMAIL PROTECTED]
To unsubscribe: [EMAIL PROTECTED]
Archives and Other: <http://java.apache.org/main/mail.html>
Problems?: [EMAIL PROTECTED]
