[mapserver-users] mapcache_seed segfault when using threading [SEC=UNCLASSIFIED]

[Matthew Doyle]

Hi Thomas & MapCache users,

A little background: we are developing a weather forecasting tool which provides a visual representation of a quite a large dataset (NetCDF grid, 7 days of data every three hours). As such, we have many diferent layers and a very large number of tiles which require seeding multiple times per day.

Based on very impressive seeding results from tests carried out on on RHEL 5 (a VM box with 8 cores), seeding using 8 threads, and encountering no issues during these tests, the decision was made to purchase the best hardware available in order to seed our tiles as quickly as possible when going operational.


The issue: mapcache_seed is halting with "segmentation fault (core dumped)" on our new operational machine when using any number of threads > 1. The higher the number of threads, the more likely it is to segfault. The issue appears during the initialisation phase just prior to seeding as once the seeding run is initialised it will complete successfully, but if the segfault occurs it will occur and exit almost immediately after execution.

I will try to provide as much information as possible, please let me know what would be useful, perhaps the core dump ? or a simple test case (MapServer mapfile / data file / mapcache xml / seeding parameters) in the hope that you might be able to reproduce?

Hardware:

 *   Power Edge R820, 4-socket, 2U blade server
 *   Populated with four Intel Xeon E5 processors
 *
Processor E5-4650L - 20M Cache, 2.60 GHz, 8.00 GT/s), Clock Speed2.6 GHz. Max
 *
Turbo Frequency 3.1 GHz,
 *
Each processor has 8 (microprocessor) cores,
 *
Supporting 16 threads per core with hyper threading,
 *
A total of 64 threads, and
 *   300GB RAM. (48 DIMMS (upgradable to 1.5 TB of memory))

OS:

 *   Red Hat Enterprise Linux Server release 6.3 (Santiago)

Software:

 *   Apache Server version: Apache/2.2.15 (Unix)
    *   mod_wsgi
    *   mod_mapcache
 *   MapServer 6.0.3
 *   MapCache (current devel 6.2.0beta2)
 *   Gdal 1.8.1
 *   Python 2.6.6
    *   Python Mapscript
 *   Perl 5.10.1
    *   Perl Mapscript
 *   Postgres 8.4.12/PostGIS 1.5.3

Our WMS service is a very basic Python MapScript wrapper via mod_wsgi, but i have also tried standard CGI Python and CGI Perl wrappers with the same results.

We are using non-linear resolution steps to avoid seeding too many tiles. Perhaps that is a problem with depth seeding?
We are using a metatile area of 8 x 8.

Attempting to seed using any number of threads will cause a segfault (usually any number higher than 4, but i have seen it fail with less). It's worth mentioning that if n threads is set to 1, then the seeding process executes and completes without any issues. Any light you can shed on this situation and any suggestions are greatly appreciated.

Best regards,
Matt
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/mapserver-users/attachments/20121010/e5ad6de2/attachment.html>