[PVFS2-developers] Re: smallio server death with old kernel

[Phil Carns]

Pete Wyckoff wrote:
> slang at mcs.anl.gov wrote on Fri, 06 Jan 2006 10:13 -0600:
> 
>>Thanks for looking into this Murali.  I'm back from vacation today so  
>>I'll look at these patches and try to get them committed ASAP.
> 
> 
> Just so you know, the fixes work fine here.  I've restarted all our
> production machines with CVS head.  Thanks both,
> 
> 		-- Pete


I am seeing an error with the current CVS head when running the "read01" 
test from the LTP test suite.  I don't know what is causing the initial 
problem, but there is a state machine issue that is contributing.  In 
small-io.sm we have this transition:

     state start_job
     {
         run small_io_start_job;
         success => send_response;
         default => cleanup;
     }

Should there just be one transition to send_response (default => 
send_response) rather than doing something different for success or 
failure?  Right now the client hangs for a while if something goes wrong 
in small_io_start_job() because no negative ack is sent back.

The initial error that I am seeing that leads to this issue shows up 
like this in the server logs:

[E 19:56:37.776476] ../pvfs2_src/src/io/description/pint-request.c line 
89: PINT_process_request: NULL segmax or bytemax!

<some backtrace stuff>

[E 19:56:37.776654] small_io: Failed to process file request

This previously triggered the assertion on line 183 of sys-small-io.sm 
before it was removed.

-Phil