[torqueusers] memory leak in 2.3.9 pbs_server

Jan Svec svecj at fzu.cz
Thu Apr 15 05:36:42 MDT 2010


Jan Svec wrote:
> Hi all,
> 
> it seems, that there is a memory leak in 2.3.9 torque. After week or so
> (with several thousands jobs runnig every day) the pbs_server process
> dies with following messages  in syslog:
> 
> Apr 13 06:25:02 torque pbs_server: LOG_CRITICAL::socket_to_handle,
> internal socket table is half-full at 578 (expected num_connections is 584)!
> Apr 13 06:25:02 torque pbs_server: LOG_CRITICAL::socket_to_handle,
> internal socket table is half-full at 579 (expected num_connections is 585)!
> Apr 13 06:25:02 torque pbs_server: LOG_ERROR::Cannot allocate memory
> (12) in DIS_tcp_setup, malloc failure
> 
> The server is running on 32bit Scientific Linux 5.2 (RHEL clone) with
> maui 3.2.6p20 as scheduler.
> Anyone observed this behaviour too? Is 2.3.10 correcting any memory
> leaks? (I don't see such changes in changelog). Or is it safe to run 2.4
> branch?
> 
> Thank You for any help
> Jan
> 
Hi

nobody else? It is strange, because I tried to compile 2.3.9. at several
different systems and the problem occurs everywhere, the memory usage
continually grows until the memory is full and the server cannot
allocate more (it should be easily reproducible).
And does somebody know, what the "internal socket table is half-full"
error message means? Can I somehow tune size of this table? Does this
somehow affect performance of the torque server?

Thank You
Jan

-- 
Jan Svec
Institute of Physics AVCR


More information about the torqueusers mailing list