Using Preemption with the External Process Method (Network Computer)

AlanB_22262
AlanB_22262 New Altair Community Member
edited February 2023 in Altair HPCWorks

When using the EXT (external) method for preemption in NetworkComputer, only selected programs in the process tree are signaled.  This is necessary, because some programs do not respond well when the entire process tree is signaled.

The programs signaled are determined by include and exclude patterns for the program name.

The actual work is done by vovjobctrl, and this script reads several properties attached to the job to determine which processes to signal.

Properties JOBCTRL_INCLUDE and JOBCTRL_EXCLUDE hold the names of the programs to signal and avoid, respectively.

The Tcl's string match subcommand determines this and accepts a simplified regular-expression syntax. For more information, see http://www.tcl.tk/man/tcl8.6/TclCmd/string.htm.

Usually, the names of the programs will be enough to filter through the process tree.