Bug ? - Dual CPU NT4 M/C against same SAMBA share ...

[Ravi Subramaniam]

Hi,

We have recently come across a problem when running jobs on a dual CPU NT
machine where two process have open files on the same SAMBA share and are
writing to these files. These problems have been seen on versions newer
than 1.9.17p1. Versions 1.9.15px and 1.9.16px don't seem to demonstrate
this problem. These process have been spawned by a in-house batch
processing program.

The configuration file is the same across the versions except for the dual
NIC support enabled from version 1.9.17p1. (The problem has been verified
with this version on a single NIC server too !!).

The problem seen is as follows:

Timeline:
           T1     |       T2          |         T3        |       T4
                  |                   |                   |
Process 1 opens   |  Process 2 open   | Process 1 closes  |  Process 2 close
file (A)          |  another file B   |     file A        |     file B

The process and files are independent of each other except sharing the
same SAMBA share.

In such a situation is it noted that file A contents are either empty or
is the desired one. File B invariably has corrupted data. This corruption
either is that the expected data is incomplete OR has file A data instead
of the expected data OR file A data with part of the data expected for
file B.

The clients are running NT4-SP3.

The type and occurrence of problem seems to be a function of the load on
the SAMBA server. Higher the load the greater the likelihood.

It should be noted that this has been primarily seen on jobs run in batch
mode by the batch program we have and not in interactive usage. But... the
load on the SAMBA server is greater in batch mode because of the number of
jobs spawned. Once the batch mode daemon launches the jobs it redirects
the standard output/error to its log file but the files (A and B) above
are opened directly by the launched jobs.

Has anyone seen such problems ? Does anyone have pointers on what could be
a source of these problems ? Best yet, is there a know fix ? 

Thanks in advance for your help.

Ravi

--
=============================================================================
Ravi Subramaniam				
Senior Software Systems Engineer,
Engineering Computing, Design Technology,	
Mailstop: RN4-36,				Ph : (408)-765-3566
Intel Corp., Santa Clara, 			Email: rsubram at scdt.intel.com
California, 95052
=============================================================================