Make core files from failed transfer processes visible (*)

What is it?

Occasionally transfer processes die badly. The most common occurrence of this is when the Globus library helpfully calls the system function abort inside the library when it suspects that the response from on of the gridFTP severs is malformed. The core filesize is usually set to zero on SLC3 by default (at least at CERN).

Procedure

Follow the procedure below:

1. Edit the transfer agent daemon's sysconfig file, /etc/sysconfig/glite-data-transfer-agents, and set the core ulimit to something higher than zero, ideally unlimited if you want to actually use it.
2. Check /etc/security/limits.conf and make sure the hard limit for core files for edguser is not set to zero.
3. Restart the transfer agent daemon(s). New processes will be set to write core files if they fail.
4. The transfer processes will dump core files in their CWD which is set to /tmp/.
5. The originating binary for the transfer processes is /opt/glite/libexec/glite-url-copy-exec. See FtsServerAdmin20.

Maintainer: GavinMcCance

(*) Who noticed the extra The in the title of this page?