- 06 Dec, 2016 7 commits
-
-
Morris Jette authored
test12.2 was consistently failing on smd# cluster with tiny differences in the disk read and written. This change permits those tiny discrepancies to exist without failing the test. Here are the numbers, which are consistent: sacct --noheader -p --job=763.0 --format MaxDiskWrite,AveDiskWrite,MaxDiskRead,AveDiskRead 10.00M|10.00M|10.03M|10.03M| (i.e. 0.3% discrepancy, up to 0.5% allowed with current code)
-
Morris Jette authored
Test was failing due to hitting the memory limit (at least on smd1).
-
Morris Jette authored
There were already several configurations that could cause this test to fail. I just added another.
-
Danny Auble authored
-
Danny Auble authored
-
Morris Jette authored
Fix parsing in regression test1.92 for some prompts. bug 2792
-
Morris Jette authored
Recognize a KNL's proper NUMA count (rather than setting it to the value in slurm.conf) when using FastSchedule=0. Previous logic would change the NUMA count on the node to match what was in slurm.conf, which would mess up task layout with respect to the sockets. bug 3306
-
- 05 Dec, 2016 4 commits
-
-
Morris Jette authored
HWLOC is required to properly determine topology
-
Danny Auble authored
from the slurm.conf when using FastSchedule=0.
-
Morris Jette authored
cray/burst_buffer - If slurmctld daemon restarts with pending job and burst buffer having unknown file stage-in status, teardown the buffer, defer the job, and start stage-in over again. bug 3295
-
Morris Jette authored
Add more detail to log message and change from error to debug2 with an explanation of how this happens
-
- 02 Dec, 2016 3 commits
-
-
Danny Auble authored
bug 3314
-
Danny Auble authored
-
Danny Auble authored
-
- 01 Dec, 2016 5 commits
-
-
Dominik Bartkiewicz authored
-
Dominik Bartkiewicz authored
limits after the node selection to make sure it doesn't violate those limits and if it does change the reason for waiting so we don't reserve resources on jobs violating accounting limits. Bug 3029
-
Morris Jette authored
-
Nicolas Joly authored
Bug 3301.
-
Morris Jette authored
node_features/knl_cray - Fix possible race condition when changing node state that could result in old KNL mode as an active features. bug 3235
-
- 30 Nov, 2016 5 commits
-
-
Morris Jette authored
-
Morris Jette authored
No change in logic
-
Morris Jette authored
cray/burst_buffer - Increase time to synchronize operations between threads from 5 to 60 seconds ("setup" operation time observed over 17 seconds). This should fix a race condition between a thread performing a buffer creation (setup) and a thread looking for unexpected buffers. If a buffer is found during the time window allowed for creation, it's space will be counted twice. First by the status checking thread and second by the thread doing the creation. The deallocation only happens once, so the used space information can be left with an invalid value. bug 3295
-
Tim Wickberg authored
Never used and is uninitialized making backtraces more confusing. Fix whitespace in bcast_parameters struct while here. No functional change.
-
Tim Wickberg authored
static variable means multiple active decompression streams will corrupt zlib's internal state, which can lead to a segfault. Bug 3299.
-
- 29 Nov, 2016 4 commits
-
-
Alejandro Sanchez authored
On a reconfig, the exc_node_bitmap is cleared but then it was not built again since last_work_scan was declared as a local static variable in _do_power_work(). The fix is to make it global within the plugin and reinitialize it to 0 on _init_power_config(). Bug 3078.
-
Tim Wickberg authored
-
Tim Wickberg authored
-
Tim Wickberg authored
-
- 28 Nov, 2016 4 commits
-
-
Alejandro Sanchez authored
-
Tim Wickberg authored
-
Dominik Bartkiewicz authored
Bug 3267.
-
Dominik Bartkiewicz authored
Termination can race against step creation if, e.g., ill-behaved SPANK plugins are in use. Bug 3248.
-
- 23 Nov, 2016 1 commit
-
-
Danny Auble authored
-
- 22 Nov, 2016 5 commits
-
-
Morris Jette authored
sched/backfill plugin: Make malloc match data type (defined as uint32_t and allocated as int). No failures observed, if type "int" is smaller than "uint32_t", it could result in an invalid memory reference.
-
Sergey Meirovich authored
Fix API call: slurm_job_cpus_allocated_str_on_node_id() and in turn slurm_job_cpus_allocated_str_on_node() to return correct results for anything but first node. This was caused by missed logic to calculate fist bit belongs to particular node. Lookup was always starting from bit 0. Bug 3266.
-
Morris Jette authored
After one second of wall time, simulate the termination of all remaining running jobs in order to respond in a reasonable time frame. bug 3275
-
Morris Jette authored
Modify backfill algorithm to improve performance with large numbers of running jobs. Group running jobs that end in a "similar" time frame using a time window that grows exponentially rather than linearly. The original window sizes were (in units of minutes): 0, 1, 2, 3, 4, 5, 6, 7, ... minutes The new window sizes are: 0.5, 1, 2, 4, 8, 16, 32, ... minutes This can dramatically reduce the number of instances where the very time consuming "can the pending job run now" operation is executed, especailly if there are 1000+ running jobs. bug 3275
-
Nicolas Joly authored
-
- 20 Nov, 2016 1 commit
-
-
Morris Jette authored
-
- 15 Nov, 2016 1 commit
-
-
Tim Wickberg authored
Prevent a scrollbar from appearing on the SchedMD logo in the top left.
-