- 30 May, 2014 1 commit
-
-
Morris Jette authored
If shutdown of the slurmctld daemon is in progress, then stop trying to schedule jobs or process reconfigure requests. These are the only operations that take a significant amount of time and only service to slow down the shutdown process. We want the daemon to stop processing incoming RPCs and save state as soon as possible.
-
- 29 May, 2014 5 commits
-
-
Morris Jette authored
-
Morris Jette authored
-
Morris Jette authored
Previous limit was 4 secs, raised to 10.
-
Morris Jette authored
select/cons_res plugin: Fix memory leak related to job preemption. bug 837
-
Danny Auble authored
d75bcaa5
-
- 28 May, 2014 10 commits
-
-
Danny Auble authored
-
Danny Auble authored
-
Morris Jette authored
Show exactly waht the bad ProfileHDF5Default value is.
-
Morris Jette authored
ProfileHDF5Default=Filesystem should be ProfileHDF5Default=Lustre
-
Morris Jette authored
Added double brackets so brackets could be used within test program
-
Morris Jette authored
Fix for commit 3f7bd421
-
Morris Jette authored
Replace AC_TRY_RUN with more modern AC_RUN_IFELSE and assume printf of NULL does not seg fault
-
Morris Jette authored
This give system administrators the option on AMD Opteron 6000 series processors of either considering each NUMA node on a socket as a separate socket (resulting in some incorrect logging of socket count information) or not (resulting in sub-optimal job allocations since each core in the socket will be considered equivalent, even if on different NUMA nodes within the socket). bug 838
-
Danny Auble authored
67fdbce5
-
Morris Jette authored
When a batch job requeue completes, clear it's bitmap of completing nodes. If the bitmap were to persist, and nodes are added/remove in slurm.conf, and "scontrol reconfigure" is executed, and one of the bits in that bitmap now point to a DOWN node, avoid killing the job. bug 805
-
- 27 May, 2014 4 commits
-
-
Morris Jette authored
If a batch job is discovered to be missing from it's head node, set its exit code to 1 rather than leaving it as zero. Bug 833
-
Morris Jette authored
Was printing unsigned as int
-
Morris Jette authored
-
Danny Auble authored
-
- 23 May, 2014 7 commits
-
-
David Bigagli authored
-
David Bigagli authored
-
Yu Watanabe authored
-
Danny Auble authored
compiler would treat 1 as a 32 bit number and wrap.
-
Danny Auble authored
more of a hey, the user is asking for something out of the norm.
-
Danny Auble authored
not able to be separated into multiply patches. If EnforcePartLimits=Yes and QOS job is using can override limits, allow it. Fix issues if partition allows or denys account's or QOS' and either are not set. If a job requests a partition and it doesn't allow a QOS or account the job is requesting pend unless EnforcePartLimits=Yes. Before it would always kill the job at submit.
-
Danny Auble authored
-
- 22 May, 2014 1 commit
-
-
wickberg authored
-
- 21 May, 2014 8 commits
-
-
Morris Jette authored
This reverts commit 859839a7 The ntasks_per_core option was previously really treated like number of CPUs (rather than tasks) to allocate per core, which seems to be what is desired.
-
David Bigagli authored
-
Danny Auble authored
-
Danny Auble authored
wait for.
-
Danny Auble authored
based on the mask given.
-
Danny Auble authored
task/affinity.
-
Danny Auble authored
thread in a core.
-
Danny Auble authored
it can bind cyclically across sockets.
-
- 20 May, 2014 4 commits
-
-
Morris Jette authored
Previous logic assumed cpus_per_task=1, so ntasks_per_core option could spread the job across more cores than desired
-
Morris Jette authored
cpus-per-task support: Try to pack all CPUs of each tasks onto one socket. Previous logic could spread the tasks CPUs across multiple sockets.
-
Morris Jette authored
Previous logic was counting CPUs, but assuming each task would only use one CPU.
-
Dan Weeks authored
-