- 23 Jan, 2019 1 commit
-
-
Danny Auble authored
instead of first as previously done.
-
- 22 Jan, 2019 1 commit
-
-
Morris Jette authored
This fix is needed to address the GRES specification in gres.conf having a Type option, while the GRES specification in slurm.conf does not.
-
- 21 Jan, 2019 5 commits
-
-
Morris Jette authored
If scontrol is used to change a node's GRES and the input string contains socket binding information (e.g. "gres=gpu:4(S:0),...") then ignore the socket binding information input. Use the binding as reported by slurmd in the node registration.
-
Morris Jette authored
If a node's GRES has a count of zero (say after updating GRES counts) then do not print the count. Just don't report anything for that GRES type.
-
Morris Jette authored
If a GRES was defined with Type information in gres.conf, but lacked Type information in slurm.conf, this sets up the data structures in slurmctld for resource allocations.
-
Morris Jette authored
-
Morris Jette authored
Convert node validate logic (on node registration) in gres.c to use the same (new) function as is used for node gres update operation.
-
- 19 Jan, 2019 2 commits
-
-
Morris Jette authored
Add new logic to set node's GRES string on reconfig. Set node GRES sockets based upon real socket/core config info. Validate GRES changes before making them. Specifically we want to make sure no requests to process changes in the count of a GRES associated with File specifications happens. For example, if we have 4 gres/gpu associated with /dev/nvidia[0-3] and "scontrol update NodeName=... Gres=gpu:2" is executed that request will return an error. This is because we have no idea which specific gres/gpu records should be removed. We can issue a request to keep the count unchanged or set the count to 0, but any other count will return an error. This restrictions is not placed on GRES without Files (e.g. gres/craynetwork). Changed some variable names to better reflect their contents.
-
Morris Jette authored
Give job longer time to be scheduled and started (depends on scheduling parameters).
-
- 18 Jan, 2019 4 commits
-
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
Declare as extern; the linker will find this in libc somewhere. Bug 5561.
-
Tim Wickberg authored
A different approach to handling the alias is needed for these systems, this does not begin to cover all of the required function implementations.
-
- 17 Jan, 2019 20 commits
-
-
Felip Moll authored
With the change/refactoring in logics after the new XCC energy plugin we are using the base_watts field as ave_watts. Change all variable names for consistency. Bug 6282
-
Albert Gil authored
Bug 6242
-
Morris Jette authored
This check really is there to test if we have dynamic(inactive) features. Obviously this doesn't work on heterogeneous systems, but there is no good way to check for this as this check is used long before nodes are considered. Bug 6242
-
Broderick Gardner authored
Docs updated with details. 'node' is simply removed from the tres list; the report will continue if other valid tres are specified. Bug 5954
-
Jason Booth authored
This adds the -b option to salloc, srun and sbatch which is equal to --begin. Bug 6356.
-
Danny Auble authored
# Conflicts: # src/sbatch/opt.c
-
Jason Booth authored
Bug 6356.
-
Jason Booth authored
Fixed srun from double printing invalid option msg twice. Also removed the unused -b flag from sbatch. Bug 6356.
-
Jason Booth authored
Corrected extra spacing between function and function arguments in srun and sbatch. Bug 6364.
-
Danny Auble authored
-
Danny Auble authored
It has been shown in a couple of ways these 2 apis are dependent on each other, but both export similar symbols, so when you link to both of them you get duplicate symbols. Bug 1394
-
Morris Jette authored
Note that FastSchedule=0 lets the GRES count used for scheduling be based upon what the node reports (like memory or disk space).
-
Morris Jette authored
This is needed so we can easily determine if we are running in the slurmctld (rather than running in slurmd or in a test program). When running with test7.3 and a configuration of TopologyPlugin=topology/tree the test would generate a bunch of "Invalid GRES configured on node ..." errors and fail. This change means the GRES checking logic is disabled for the test.
-
Morris Jette authored
Just make some formatting changes
-
Morris Jette authored
No change in logic
-
Tim Wickberg authored
-
Tim Wickberg authored
Bug 5670.
-
Tim Wickberg authored
Bug 5670.
-
Tim Wickberg authored
Bug 5670.
-
Tim Wickberg authored
Update the documentation, and split "classic" fair share details off to a separate page. Bug 6036.
-
- 16 Jan, 2019 5 commits
-
-
Felip Moll authored
scontrol and sview won't report Consumed Joules and Lowest Joules anymore. The reason is that these values were collected from slurmd start time which made them not useful for the users. Bug 6283
-
Jacob Jenson authored
SchedMD has moved to Utah, so change the jurisdiction.
-
Tim Wickberg authored
-
Morris Jette authored
-
Morris Jette authored
The test21.30/inc21.30.11 does not work if SelectTypeParameters does not manage memory (not a commonly used configuration), but this adds a check to prevent a failure that is due to the configuration rather than slurm code bug.
-
- 15 Jan, 2019 2 commits
-
-
Morris Jette authored
These tests confirms correct CUDA percentages in an environment where the gres/mps counts are not always 100 and can very from device to device within a node.
-
Morris Jette authored
Slurmd was generating errors of this sort for all GRES: error: Ignoring gres.conf record, invalid name: mps This was due to some GRES setup logic not being invoked on the slurmd. In any case, the relevant checking now only happens on the slurmctld.
-