Newer
Older
This file describes changes in recent versions of Slurm. It primarily
documents those changes that are of interest to users and admins.
* Changes in Slurm 14.03.9
==========================
-- If slurmd fails to stat(2) the configuration print the string describing
the error code.
-- Fix for mixing core base reservations with whole node based reservations
to avoid overlapping erroneously.
-- BLUEGENE - Remove references to Base Partition.
-- sview - If compiled on a non-bluegene system then used to view a BGQ fix
to allow sview to display blocks correctly.
-- Fix bug in update reservation. When modifying the reservation the end time
was set incorrectly.
-- The start time of a reservation that is in ACTIVE state cannot be modified.
-- Update the cgroup documentation about release agent for devices.
-- MYSQL - fix for setting up preempt list on a QOS for multiple QOS.
-- Correct a minor error in the scancel.1 man page related to the
--signal option.
-- Enhance the scancel.1 man page to document the sequence of signals sent
-- Fix slurmstepd core dump if the cgroup hierarchy is not completed
when terminating the job.
-- Fix hostlist_shift to be able to give correct node names on names with a
different number of dimensions than the cluster.
-- BLUEGENE - Fix invalid pointer in corner case in the plugin.
-- Make sure on a reconfigure the select information for a node is preserved.
-- Correct logic to support job GRES specification over 31 bits (problem
in logic converting int to uint32_t).
-- Remove logic that was creating GRES bitmap for node when not needed (only
needed when GRES mapped to specific files).
-- BLUEGENE - Fix sinfo -tr before it would only print idle nodes correctly.
-- BLUEGENE - Fix for licenses_only reservation on bluegene systems.
-- sview - Verify pointer before using strchr.
-- -M option on tools talking to a Cray from a non-Cray fixed.
-- CRAY - Fix rpmbuild issue for missing file slurm.conf.template.
-- Fix race condition when dealing with removing many associations at
different times when reservations are using the associations that are
being deleted.
-- When a node's state is set to power_down/power_up, then execute
SuspendProgram/ResumeProgram even if previously executed for that node.
-- Fix logic determining when job configuration (i.e. running node power up
logic) is complete.
-- Setting the state of a node in powered down state node to "resume" will
no longer cause it to reboot, but only clear the "drain" state flag.
Brian Christiansen
committed
-- Fix srun documentation to remove SLURM_NODELIST being equivalent as the -w
option (since it isn't).
-- Fix issue with --hint=nomultithread and allocations with steps running
arbitrary layouts (test1.59).
-- PrivateData=reservation modified to permit users to view the reservations
which they have access to (rather then preventing them from seeing ANY
reservation). Backport from 14.11 commit 77c2bd25c.
-- Fix PrivateData=reservation when using associations to give privileges to
a reservation.
* Changes in Slurm 14.03.8
==========================
-- Fix minor memory leak when Job doesn't have nodes on it (Meaning the job
has finished)
-- Fix sinfo/sview to be able to query against nodes in reserved and other
states.
-- Make sbatch/salloc read in (SLURM|(SBATCH|SALLOC))_HINT in order to
handle sruns in the script that will use it.
-- srun properly interprets a leading "." in the executable name based upon
the working directory of the compute node rather than the submit host.
-- Fix Lustre misspellings in hdf5 guide
Kilian Cavalotti
committed
-- Fix wrong reference in slurm.conf man page to what --profile option should
be used for AcctGatherFilesystemType.
-- Update HDF5 document to point out the SlurmdUser is who creates the
ProfileHDF5Dir directory as well as all it's sub-directories and files.
-- CRAY NATIVE - Remove error message for srun's ran inside an salloc that
had --network= specified.
-- Defer job step initiation of required GRES are in use by other steps rather
than immediately returning an error.
-- Deprecate --cpu_bind from sbatch and salloc. These never worked correctly
and only caused confusion since the cpu_bind options mostly refer to a
step we opted to only allow srun to set them in future versions.
-- Modify sgather to work if Nodename and NodeHostname differ.
-- Changed use of JobContainerPlugin where it should be JobContainerType.
-- Fix for possible error if job has GRES, but the step explicitly requests a
GRES count of zero.
-- Make "srun --gres=none ..." work when executed without a job allocation.
-- Change the global eio_shutdown_time to a field in eio handle.
-- Advanced reservation fixes for heterogeneous systems, especially when
reserving cores.
-- If --hint=nomultithread is used in a job allocation make sure any srun's
ran inside the allocation can read the environment correctly.
-- If batchdir can't be made set errno correctly so the slurmctld is notified
correctly.
-- Remove repeated batch complete if batch directory isn't able to be made
since the slurmd will send the same message.
-- sacctmgr fix default format for list transactions.
-- BLUEGENE - Fix backfill issue with backfilling jobs on blocks already
reserved for higher priority jobs.
-- When creating job arrays the job specification files for each elements
are hard links to the first element specification files. If the controller
fails to make the links the files are copied instead.
-- Fix error handling for job array create failure due to inability to copy
job files (script and environment).
-- Added patch in the contribs directory for integrating make version 4.0 with
Slurm and renamed the previous patch "make-3.81.slurm.patch".
-- Don't wait for an update message from the DBD to finish before sending rc
message back. In slow systems with many associations this could speed
responsiveness in sacctmgr after adding associations.
-- Eliminate race condition in enforcement of MaxJobCount limit for job arrays.
-- Fix anomaly allocating cores for GRES with specific device/CPU mapping.
-- cons_res - When requesting exclusive access make sure we set the number
of cpus in the job_resources_t structure so as nodes finish the correct
cpu count is displayed in the user tools.
-- If the job_submit plugin calls take longer than 1 second to run, print a
warning.
-- Make sure transfer_s_p_options transfers all the portions of the
s_p_options_t struct.
-- Correct the srun man page, the SLURM_CPU_BIND_VERBOSE, SLURM_CPU_BIND_TYPE
SLURM_CPU_BIND_LIST environment variable are set only when task/affinity
plugin is configured.
-- sacct - Initialize variables correctly to avoid incorrect structure
reference.
-- Performance adjustment to avoid calling a function multiple times when it
only needs to be called once.
-- Give more correct waiting reason if job is waiting on association/QOS
MaxNode limit.
-- DB - When sending lft updates to the slurmctld only send non-deleted lfts.
-- BLUEGENE - Fix documentation on how to build a reservation less than
a midplane.
-- If Slurmctld fails to read the job environment consider it an error
and abort the job.
-- Add the name of the node a job is running on to the message printed by
slurmstepd when terminating a job.
-- Remove unsupported options from sacctmgr help and the dump function.
-- Update sacctmgr man page removing reference to obsolete parameter
MaxProcSecondsPerJob.
-- Added more validity checking of incoming job submit requests.
* Changes in Slurm 14.03.7
==========================
-- Add note to MaxNodesPerUser and multiple jobs running on the same node
counting as multiple nodes.
-- PerlAPI - fix renamed call from slurm_api_set_conf_file to
slurm_conf_reinit.
-- Fix gres race condition that could result in job deallocation error message.
-- Correct NumCPUs count for jobs with --exclusive option.
-- When creating reservation with CoreCnt, check that Slurm uses
SelectType=select/cons_res, otherwise don't send the request to slurmctld
and return an error.
-- Save the state of scheduled node reboots so they will not be lost should the
slurmctld restart.
-- In select/cons_res plugin - Insure the node count does not exceed the task
count.
-- switch/nrt - Unload tables rather than windows at job end, to release CAU.
-- When HealthCheckNodeState is configured as IDLE don't run the
HealthCheckProgram for nodes in any other states than IDLE.
-- Minor sanity check to verify the string sent in isn't NULL when using
bit_unfmt.
-- CRAY NATIVE - Fix issue on heavy systems to only run the NHC once per
job/step completion.
-- Remove unneeded step cleanup for pending steps.
-- Fix issue where if a batch job was manually requeued the batch step
information wasn't stored in accounting.
-- When job is release from a requeue hold state clean up its previous
exit code.
-- Correct the srun man page about how the output from the user application
is sent to srun.
-- Increase the timeout of the main thread while waiting for the i/o thread.
Allow up to 180 seconds for the i/o thread to complete.
-- When using sacct -c to read the job completion data compute the correct
job elapsed time.
-- Perl package: Define some missing node states.
-- When using AccountingStorageType=accounting_storage/mysql zero out the
database index for the array elements avoiding duplicate database values.
-- Reword the explanation of cputime and cputimeraw in the sacct man page.
-- JobCompType allows "jobcomp/mysql" as valid name but the code used
"job_comp/mysql" setting an incorrect default database.
-- Try to load libslurm.so only when necessary.
-- When nodes scheduled for reboot, set state to DOWN rather than FUTURE so
they are still visible to sinfo. State set to IDLE after reboot completes.
-- Apply BatchStartTimeout configuration to task launch and avoid aborting
srun commands due to long running Prolog scripts.
-- Fix minor memory leaks when freeing node_info_t structure.
-- If a batch script is requeued and running steps get correct exit code/signal
previous it was always -2.
-- If step exitcode hasn't been set display with sacct the -2 instead
of acting like it is a signal and exitcode.
-- Send calculated step_rc for batch step instead of raw status as
done for normal steps.
-- If a job times out, set the exit code in accounting to 1 instead of the
signal 1.
-- Update the acct_gather.conf.5 man page removing the reference to
InfinibandOFEDFrequency.
-- Fix gang scheduling for jobs submitted to multiple partitions.
-- Enable srun to submit job to multiple partitions.
-- Update slurm.conf man page. When Epilog or Prolog fail the node state
is set ro DRAIN.
-- Start a job in the highest priority partition possible, even if it requires
preempting other jobs and delaying initiation, rather than using a lower
priority partition. Previous logic would preempt lower priority jobs, but
then might start the job in a lower priority partition and not use the
resources released by the preempted jobs.
-- Fix SelectTypeParameters=CR_PACK_NODES for srun making both job and step
Loading
Loading full blame...