Re: [5] IndexOutOfBoundsException from CharChunk
I believe 23805 is unrelated to my problem, and may even be invalid. Mine comes from CharChunk. To refresh the memory, the stack trace is Servlet.service() for servlet jsp threw exception java.lang.IndexOutOfBoundsException at sun.nio.cs.StreamEncoder.write(StreamEncoder.java:132) at java.io.OutputStreamWriter.write(OutputStreamWriter.java:191) at org.apache.tomcat.util.buf.WriteConvertor.write(C2BConverter.java:228 ) at org.apache.tomcat.util.buf.C2BConverter.convert(C2BConverter.java:120 ) at org.apache.coyote.tomcat5.OutputBuffer.realWriteChars(OutputBuffer.ja va:606) at org.apache.tomcat.util.buf.CharChunk.flushBuffer(CharChunk.java:463) at org.apache.coyote.tomcat5.OutputBuffer.flush(OutputBuffer.java:357) at org.apache.coyote.tomcat5.CoyoteWriter.flush(CoyoteWriter.java:117) at org.apache.jasper.runtime.JspWriterImpl.flush(JspWriterImpl.java:203) at org.apache.jasper.runtime.JspRuntimeLibrary.include(JspRuntimeLibrary .java:990) ... In my case, it's not the offset, but "end" that is too big. It got the value of 16384 when "limit" is only 8192. I have looked into this more without getting more insight. I use jdb to catch IndexOutOfBoundsException but couldn't figure out what caused "end" to get this big. Since this happens randomly, I cannot stop a thread and follow its progress. Inserting prints in CharChunk was also not useful, since values for "end" in threads appear normal until the exception occurs, as if its value were clobbered by a strayed pointer (like in C). The print did show that in some cases, the value for "end" is 16384 after write(char[]) was invoked, though that didn't cause any exception for some reason. This turns out to be more elusive that it appears. I might have to give this up as another mystery of the universe. :-) -Kin-man > Date: Wed, 15 Oct 2003 10:36:01 +0200 > From: Remy Maucherat <[EMAIL PROTECTED]> > Subject: Re: [5] IndexOutOfBoundsException from CharChunk > To: Tomcat Developers List <[EMAIL PROTECTED]> > > Kin-Man Chung wrote: > > I don't know much about the test; it is one of stress test we have, but it > > should not affect CharChunk this way, right, even if it has something > > weird? Like I said, this happens only in this test, and not always > > reproducible, so I am not surprised that nobody noticed it in 4.1. I > > thought it might be thread related, but each thread should have its own > > CharChunk instance, so I can't see how. I look into this more and report > > back. > > Bug 23805 could be related to this, and I'm awaiting more details. At > least, the offset numbers do match, which is too big a coincidence ;-) > > Remy > > > > - > To unsubscribe, e-mail: [EMAIL PROTECTED] > For additional commands, e-mail: [EMAIL PROTECTED] > - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: [5] IndexOutOfBoundsException from CharChunk
Kin-Man Chung wrote: I don't know much about the test; it is one of stress test we have, but it should not affect CharChunk this way, right, even if it has something weird? Like I said, this happens only in this test, and not always reproducible, so I am not surprised that nobody noticed it in 4.1. I thought it might be thread related, but each thread should have its own CharChunk instance, so I can't see how. I look into this more and report back. Bug 23805 could be related to this, and I'm awaiting more details. At least, the offset numbers do match, which is too big a coincidence ;-) Remy - To unsubscribe, e-mail: [EMAIL PROTECTED] For additional commands, e-mail: [EMAIL PROTECTED]
Re: [5] IndexOutOfBoundsException from CharChunk
> Date: Wed, 08 Oct 2003 21:22:07 +0200
> From: Remy Maucherat <[EMAIL PROTECTED]>
> Subject: Re: [5] IndexOutOfBoundsException from CharChunk
> To: Tomcat Developers List <[EMAIL PROTECTED]>
>
> Kin-Man Chung wrote:
>
> > I am getting a sporadic IndexOutOfBoundsException and it seems to come from
> > org.apache.tomcat.util.buf.CharChunk.flushBuffer. The trace is
> >
> > Servlet.service() for servlet jsp threw exception
> > java.lang.IndexOutOfBoundsException
> > at sun.nio.cs.StreamEncoder.write(StreamEncoder.java:132)
> > at java.io.OutputStreamWriter.write(OutputStreamWriter.java:191)
> > at
org.apache.tomcat.util.buf.WriteConvertor.write(C2BConverter.java:228
> > )
> > at
org.apache.tomcat.util.buf.C2BConverter.convert(C2BConverter.java:120
> > )
> > at
org.apache.coyote.tomcat5.OutputBuffer.realWriteChars(OutputBuffer.ja
> > va:606)
> > at
org.apache.tomcat.util.buf.CharChunk.flushBuffer(CharChunk.java:463)
> > at
org.apache.coyote.tomcat5.OutputBuffer.flush(OutputBuffer.java:357)
> > at
org.apache.coyote.tomcat5.CoyoteWriter.flush(CoyoteWriter.java:117)
> > at
org.apache.jasper.runtime.JspWriterImpl.flush(JspWriterImpl.java:203)
> > at
org.apache.jasper.runtime.JspRuntimeLibrary.include(JspRuntimeLibrary
> > .java:990)
> > ...
> >
> > In CharChunk,
> >
> > public void flushBuffer()
> > throws IOException
> > {
> > //assert out!=null
> > if( out==null ) {
> > throw new IOException( "Buffer overflow, no sink " + limit + " "
+
> >buff.length );
> > }
> > out.realWriteChars( buff, start, end - start );
> > end=start;
> > }
> >
> > when realWriteChars is called, the value for start is 0, end is 16384, and
> > buff.length is 8192, hence the exception.
> >
> > Does anyone knows how end can go beyond 8192? This only hapeens very
> > rarely, under heavy load.
>
> I don't know. I suppose limit is 8192 also. Are you sure your test is
> clean, and without anything weird ? (somehow, I doubt it)
> I've changed a bit the algorithm in ByteChunk, and it seems better, but
> the CharChunk algorithm is identical in 4.1's ByteChunk and CharChunk.
> We would have noticed anything bad, I think.
>
I don't know much about the test; it is one of stress test we have, but it
should not affect CharChunk this way, right, even if it has something
weird? Like I said, this happens only in this test, and not always
reproducible, so I am not surprised that nobody noticed it in 4.1. I
thought it might be thread related, but each thread should have its own
CharChunk instance, so I can't see how. I look into this more and report
back.
-Kin-man
> Remy
>
>
>
> -
> To unsubscribe, e-mail: [EMAIL PROTECTED]
> For additional commands, e-mail: [EMAIL PROTECTED]
>
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
Re: [5] IndexOutOfBoundsException from CharChunk
Kin-Man Chung wrote:
I am getting a sporadic IndexOutOfBoundsException and it seems to come from
org.apache.tomcat.util.buf.CharChunk.flushBuffer. The trace is
Servlet.service() for servlet jsp threw exception
java.lang.IndexOutOfBoundsException
at sun.nio.cs.StreamEncoder.write(StreamEncoder.java:132)
at java.io.OutputStreamWriter.write(OutputStreamWriter.java:191)
at org.apache.tomcat.util.buf.WriteConvertor.write(C2BConverter.java:228
)
at org.apache.tomcat.util.buf.C2BConverter.convert(C2BConverter.java:120
)
at org.apache.coyote.tomcat5.OutputBuffer.realWriteChars(OutputBuffer.ja
va:606)
at org.apache.tomcat.util.buf.CharChunk.flushBuffer(CharChunk.java:463)
at org.apache.coyote.tomcat5.OutputBuffer.flush(OutputBuffer.java:357)
at org.apache.coyote.tomcat5.CoyoteWriter.flush(CoyoteWriter.java:117)
at org.apache.jasper.runtime.JspWriterImpl.flush(JspWriterImpl.java:203)
at org.apache.jasper.runtime.JspRuntimeLibrary.include(JspRuntimeLibrary
.java:990)
...
In CharChunk,
public void flushBuffer()
throws IOException
{
//assert out!=null
if( out==null ) {
throw new IOException( "Buffer overflow, no sink " + limit + " " +
buff.length );
}
out.realWriteChars( buff, start, end - start );
end=start;
}
when realWriteChars is called, the value for start is 0, end is 16384, and
buff.length is 8192, hence the exception.
Does anyone knows how end can go beyond 8192? This only hapeens very
rarely, under heavy load.
I don't know. I suppose limit is 8192 also. Are you sure your test is
clean, and without anything weird ? (somehow, I doubt it)
I've changed a bit the algorithm in ByteChunk, and it seems better, but
the CharChunk algorithm is identical in 4.1's ByteChunk and CharChunk.
We would have noticed anything bad, I think.
Remy
-
To unsubscribe, e-mail: [EMAIL PROTECTED]
For additional commands, e-mail: [EMAIL PROTECTED]
