Re: [Sqlrelay-discussion] short write

[Firstworks/4access <dav...@fi...>]

Now I remember that code :)

To improve performance, the filedescriptor class supports buffered reads
and writes.  SQL Relay sets up an 8k buffer for both read and write
buffers on both server and client sides.  When one side tries to write()
to the other, the data just gets buffered.  When the buffer is full or
if flushWriteBuffer() is called, the entire buffer is sent over the
network.  A short-write condition (ie. if an attempt to write 8192 bytes
wrote fewer than 8192 bytes) can occur under several circumstances, most
notably if some buffer in the kernel or in the underlying hardware is
full, which can happen if the machine's network interfaces are really
slammed.

The code should just loop back and try to write the remaining data
again.  Whether the file descriptor is in blocking or non-blocking mode
could affect that though.  In blocking mode, when looping back and
writing again, the write() will block until the device can accept data,
but in non-blocking mode, the write() will fall through with another
short-write condition which will loop back and retry again, over and
over.  I'll have to think about that a little.

Dave

On Wed, 2005-08-24 at 11:45 -0400, Matt Flynn wrote:
> typo in original message: 'I assume...writebufferspace=8 are default 
> values...' should be '...writebufferspace=0...'
> 
> Matt Flynn wrote:
> 
> > I am receiving the message 'aaaah, short write!!!!!' from 
> > librudiments.so using rudiments 0.28.2 and sqlrelay 0.36.4.
> >
> > I have enabled DEBUG_WRITE in rudiments-0.28.2/src/filedescriptor.C 
> > and here is a sample of the output during error conditions:
> > 22687: string write(5,sjir,4)
> > bufferedWrite of 4 bytes
> >       writebuffersize=8192
> >       writebufferspace=0
> >       byteswritten=0
> >       bytesunwritten=4
> > just buffering 0 bytes
> > safeWrite of 8192 bytes
> > aaaaah, short write!!!!!
> > 22687: ushort write(5,1)
> > bufferedWrite of 2 bytes
> >       writebuffersize=8192
> >       writebufferspace=0
> >       byteswritten=0
> >       bytesunwritten=2
> > just buffering 0 bytes
> > safeWrite of 8192 bytes
> > aaaaah, short write!!!!!
> >
> > I assume writebuffersize=8192 and writebufferspace=8 are default 
> > values.  Under normal working conditions writebuffersize is usually 
> > less than 8192 and writebufferspace is always greater than 0.   
> > Possibly memory not being initialized properly?  Possibly flaky 
> > network card?
> >
> > Any suggestions?
> >
> > Thanks,
> >
> > Matt
> >
> >
> > -------------------------------------------------------
> > SF.Net email is Sponsored by the Better Software Conference & EXPO
> > September 19-22, 2005 * San Francisco, CA * Development Lifecycle 
> > Practices
> > Agile & Plan-Driven Development * Managing Projects & Teams * Testing 
> > & QA
> > Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf
> > _______________________________________________
> > Sqlrelay-discussion mailing list
> > Sql...@li...
> > https://lists.sourceforge.net/lists/listinfo/sqlrelay-discussion
> >
> 
> 
> 
> -------------------------------------------------------
> SF.Net email is Sponsored by the Better Software Conference & EXPO
> September 19-22, 2005 * San Francisco, CA * Development Lifecycle Practices
> Agile & Plan-Driven Development * Managing Projects & Teams * Testing & QA
> Security * Process Improvement & Measurement * http://www.sqe.com/bsce5sf
> _______________________________________________
> Sqlrelay-discussion mailing list
> Sql...@li...
> https://lists.sourceforge.net/lists/listinfo/sqlrelay-discussion
>