Community
Participate
Working Groups
The LML monitoring fails on some systems because the paths are not necessarily set up correctly, so commands such as qstat may not be available. The RM configuration XML should be able to be used to specify the paths of these remote commands. A suggested XML syntax is: <monitor-data schedulerType="torque"> <driver> <cmd name="jobinfo" exec="/opt/torque/2.5.7/bin/qstat"/> <cmd name="nodeinfo" exec="/opt/torque/2.5.7/bin/pbsnodes"/> </driver> </monitor-data> The DA driver would match the schedulerType "torque" to its know schedulers, then use the commands when obtaining job/node information.
The XML schema has been updated to support this syntax.
Created attachment 202438 [details] Client side implementation
Created attachment 202446 [details] Client side implementation In the 2. and all after wards created requests (LML files) still the request tag is missing. Still work on it.
I've tested this with existing RMs and they appear to function correctly.
Created attachment 202452 [details] Client side implementation - Part 2 Second part, now there is always a request tag generated when the reuquest LML file is generated.
Is the driver tag unique or could it occur often than one in the monitor tag? If this is the case, how could LML_da_driver decide which one of these driver should be taken? And the driver tag needs also a name or id to distinguish its instances.
There is not currently an attribute (name) on the driver. The job scheduler is specified by the schedulerType attribute and applies to all the elements in the monitor-data section. I'm not sure there is any need to provide commands (drivers) for more than one job scheduler, since each configuration file is really specific to a particular system/job scheduler configuration, the user can always create another configuration if necessary.
Created attachment 202458 [details] patch-file for org.eclipse.ptp.rm.lml.da patch-file for org.eclipse.ptp.rm.lml.da version 1.11 -> 1.12: - supports now new LML request structure (driver, command)
After changing the default job scheduler to TORQUE, this works for current RMs. I'll try using the new driver section next.
This is working on keenland which requires a modified path for the jobs and nodes commands.
Applied to ptp_5_0 and HEAD