- 20 Jul, 2006 15 commits
-
-
Christopher J. Morrone authored
Add -D, --dependency option Man page updates.
-
Moe Jette authored
-
Danny Auble authored
removed task dist from srun put into slurmctld. need to remove the addresses from job allocation since that isn't needed anymore. need to fix some issues with the way sbcast does things now which should be easy. The switch plugins can change also since we have the tasks laid out already now before we set up the switch. Some more clean up also will be nice in srun.
-
Moe Jette authored
-
Moe Jette authored
-
Moe Jette authored
-
Christopher J. Morrone authored
-
Moe Jette authored
-
Moe Jette authored
-
Moe Jette authored
-
Moe Jette authored
-
Moe Jette authored
-
Moe Jette authored
-
Christopher J. Morrone authored
- returns the return code of its child command - add timeout for limiting how long it waits for a PENDING job allocation (slurm_allocate_resources_blocking was updated to timeout correctly) - cleans up job allocation on Ctrl-C (Actually, it ignores SIGINT, SIGTERM, SIGQUIT, but unblocks them for the child, so ctrl-c from the shell hits the whole process group, and the child exits, and salloc cleans up like normal. Might to be more proactive with signals, but I am not yet certain.) - trim more unnecessary command line options - man page additions
-
Christopher J. Morrone authored
-
- 19 Jul, 2006 8 commits
-
-
Christopher J. Morrone authored
-
Moe Jette authored
-
Moe Jette authored
-
-
Moe Jette authored
-
Moe Jette authored
-
Moe Jette authored
This should decrease the number of realloc calls and improve performance.
-
Christopher J. Morrone authored
Update slaunch man page. Add beginnings of salloc man page.
-
- 18 Jul, 2006 6 commits
-
-
Christopher J. Morrone authored
Handles SRUN_PING most importantly, but also has stubs for SRUN_TIMEOUT and SRUN_NODE_FAIL. The latter two print messages, but do not take any action.
-
Danny Auble authored
-
Christopher J. Morrone authored
-
Danny Auble authored
-
Moe Jette authored
-
-
- 17 Jul, 2006 1 commit
-
-
Christopher J. Morrone authored
-
- 15 Jul, 2006 1 commit
-
-
Christopher J. Morrone authored
-
- 14 Jul, 2006 3 commits
-
-
Christopher J. Morrone authored
"resp_port" and "io_port". Now they just list the available listening ports in the client (srun/slaunch), instead of listing a port per node (which contains many duplicate ports). The slurmd picks a port to use based on number of available ports modulo its own rank in the job step.
-
Christopher J. Morrone authored
-
Christopher J. Morrone authored
-
- 13 Jul, 2006 3 commits
-
-
Christopher J. Morrone authored
- SlurmdPidFile, SlurmdLogFile, and SlurmdSpoolDir may each contain a "%h" and/or a "%n", which will be replaced by the hostname or NodeName, respectively. - Add new xstring.c xstrsubstitute() function. - launch_tasks_response_msg_t structure now contains a task_ids array - Remove SLURM_DIST_UNKNOWN from enum task_dist_states - job_desc_msg_t now lists a seperate addresses for the RESPONSE_RESOURCE_ALLOCATION message. - Expand the job step context concept used previously for task "spawn" for poe. The step context is now also used for nomal step launch. New SLURM API function slurm_step_launch() and associated support functions. - Fix spaghetti header #includes relating to global_srun.[ch] - Fix slurmctld segfault when slaunch --relative is used
-
-
-
- 11 Jul, 2006 2 commits
-
-
-
Christopher J. Morrone authored
close on the file descriptor.
-
- 10 Jul, 2006 1 commit
-
-