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 14.11.0pre4
==============================
-- Added job array data structure and removed 64k array size restriction.
-- Added SchedulerParameters options of bf_max_job_array_resv to control how
many tasks of a job array should have resources reserved for them.
-- Added more validity checking of incoming job submit requests.
-- Added srun --export option to set/export specific environment variables.
-- Scontrol modified to print separate error messages for job arrays with
different exit codes on the different tasks of the job array. Applies to
job suspend and resume operations.
-- Fix race condition in CPU frequency set with job preemption.
-- Always call select plugin on step termination, even if the job is also
complete.
-- Srun executable names beginning with "." will be resolved based upon the
working directory and path on the compute node rather than the submit node.
-- Add node state string suffix of "$" to identify nodes in maintenance
reservation or scheduled for reboot. This applies to scontrol, sinfo,
and sview commands.
-- Enable scontrol to clear a nodes's scheduled reboot by setting its state
to "RESUME".
-- As per sbatch and srun documentation when the --signal option is used
signal only the steps and unless, in the case, of a batch job B is
specified in which case signal only the batch script.
-- Modify AuthInfo configuration parameter to accept credential lifetime
option.
-- Modify crypto/munge plugin to use socket and timeout specified in AuthInfo.
-- If we have a state for a step on completion put that in the database
instead of guessing off the exit_code.
-- Added squeue -P/--priority option that can be used to display pending jobs
in the same order as used by the Slurm scheduler even if jobs are submitted
to multiple partitions (job is reported once per usable partition).
-- Improve the pending reason description for various QOS limits. For each
QOS limit that causes a job to be pending print its specific reason.
For example if job pends because of GrpCpus the squeue command will
print QOSGrpCpuLimit as pending reason.
-- sched/backfill - Set expected start time of job submitted to multiple
partitions to the earliest start time on any of the partitions.
-- Introduce a MAX_BATCH_REQUEUE define that indicates how many times a job
can be requeued. When the number is reached the job is put on hold
with reason JobHoldMaxRequeue.
* Changes in Slurm 14.11.0pre3
==============================
-- Move xcpuinfo.[c|h] to the slurmd since it isn't needed anywhere else
and will avoid the need for all the daemons to link to libhwloc.
-- Add memory test to job_submit/partition plugin.
-- Added new internal Slurm functions xmalloc_nz() and xrealloc_nz(), which do
not initialize the allocated memory to zero for improved performance.
-- Modify hostlist function to dynamically allocate buffer space for improved
performance.
-- In the job_submit plugin: Remove all slurmctld locks prior to job_submit()
being called for improved performance. If any slurmctld data structures are
read or modified, add locks directly in the plugin.
-- Added PriorityFlag LEVEL_BASED described in doc/html/level_based.shtml
-- If Fairshare=parent is set on an account, that account's children will be
effectively reparented for fairshare calculations to the first parent of
their parent that is not Fairshare=parent. Limits remain the same,
only it's fairshare value is affected.
* Changes in Slurm 14.11.0pre2
==============================
-- Added AllowSpecResourcesUsage configuration parameter in slurm.conf. This
allows jobs to use specialized resources on nodes allocated to them if the
job designates --core-spec=0.
-- Add new SchedulerParameters option of build_queue_timeout to throttle how
much time can be consumed building the job queue for scheduling.
-- Added HealthCheckNodeState option of "cycle" to cycle through the compute
nodes over the course of HealthCheckInterval rather than running all at
the same time.
-- Add job "reboot" option for Linux clusters. This invokes the configured
RebootProgram to reboot nodes allocated to a job before it begins execution.
-- Added squeue -O/--Format option that makes all job and step fields available
for printing.
-- Improve database slurmctld entry speed dramatically.
-- Add "CPUs" count to output of "scontrol show step".
-- scancel -b signals only the batch step neither any other step nor any
children of the shell script.
-- MySQL - enforce NO_ENGINE_SUBSTITUTION
-- Added CpuFreqDef configuration parameter in slurm.conf to specify the
default CPU frequency and governor to be set at job end.
-- Added support for job email triggers: TIME_LIMIT, TIME_LIMIT_90 (reached
90% of time limit), TIME_LIMIT_80 (reached 80% of time limit), and
TIME_LIMIT_50 (reached 50% of time limit). Applies to salloc, sbatch and
srun commands.
-- In slurm.conf add the parameter SrunPortRange=min-max. If this is configured
then srun will use its dynamic ports only from the configured range.
-- Make debug_flags 64 bit to handle more flags.
* Changes in Slurm 14.11.0pre1
==============================
-- Modify etc/cgroup.release_common.example to set specify full path to the
scontrol command. Also find cgroup mount point by reading cgroup.conf file.
-- Improve qsub wrapper support for passing environment variables.
-- Modify sdiag to report Slurm RPC traffic by user, type, count and time
consumed.
-- In select plugins, stop triggering extra logging based upon the debug flag
-- Added SchedulerParameters options of bf_yield_interval and bf_yield_sleep
to control how frequently and for how long the backfill scheduler will
relinquish its locks.
-- To support larger numbers of jobs when the StateSaveDirectory is on a
file system that supports a limited number of files in a directory, add a
subdirectory called "hash.#" based upon the last digit of the job ID.
-- More gracefully handle missing batch script file. Just kill the job and do
not drain the compute node.
-- Add support for allocation of GRES by model type for heterogenous systems
(e.g. request a Kepler GPU, a Tesla GPU, or a GPU of any type).
-- Record and enable display of nodes anticipated to be used for pending jobs.
-- Modify squeue --start option to print the nodes expected to be used for
pending job (in addition to expected start time, etc.).
-- Add association hash to the assoc_mgr.
-- Better logic to handle resized jobs when the DBD is down.
-- Introduce MemLimitEnforce yes|no in slurm.conf. If set no Slurm will
not terminate jobs if they exceed requested memory.
-- Add support for non-consumable generic resources for resources that are
limited, but can be shared between jobs.
-- Introduce 5 new Slurm errors in slurm_errno.h related to job to better
-- Modify scontrol to print error message for each array task when updating
the entire array.
-- Added gres_drain and gres_used fields to node_info_t.
-- Added PriorityParameters configuration parameter in slurm.conf.
-- Introduce automatic job requeue policy based on exit value. See RequeueExit
and RequeueExitHold descriptions in slurm.conf man page.
-- Modify slurmd to cache launched job IDs for more responsive job suspend and
gang scheduling.
-- Permit jobs steps full control over cpu_bind options if specialized cores
are included in the job allocation.
-- Added ChosLoc configuration parameter to specifiy the pathname of the
Chroot OS tool.
-- Sent SIGCONT/SIGTERM when a job is selected for preemption with GraceTime
configured rather than waiting for GraceTime to be reached before notifying
the job.
-- Do not resume a job with specialized cores on a node running another job
with specialized cores (only one can run at a time).
-- Add specialized core count to job suspend/resume calls.
-- task/affinity and task/cgroup - Correct specialized core task binding with
user supplied invalid CPU mask or map.
-- Add srun --cpu-freq options to set the CPU governor (OnDemand, Performance,
PowerSave or UserSpace).
-- Add support for a job step's CPU governor and/or frequency to be reset on
suspend/resume (or gang scheduling). The default for an idle CPU will now
be "ondemand" rather than "userspace" with the lowest frequency (to recover
from hard slurmd failures and support gang scheduling).
-- Added PriorityFlags option of Calulate_Running to continue recalculating
the priority of running jobs.
-- Replace round-robin front-end node selection with least-loaded algorithm.
-- CRAY - Improve support of XC30 systems when running natively.
-- Add new node configuration parameters CoreSpecCount, CPUSpecList and
MemSpecLimit which support the reservation of resources for system use
with Linux cgroup.
-- Add child_forked() function to the slurm_acct_gather_profile plugin to
close open files, leaving application with no extra open file descriptors.
-- Cray/ALPS system - Enable backup controller to run outside of the Cray to
accept new job submissions and most other operations on the pending jobs.
-- Have sacct print job and task array id's for job arrays.
-- If <sys/prctl.h> is present name major threads in slurmctld, for
example backfill
thread: slurmctld_bckfl, the rpc manager: slurmctld_rpcmg etc.
The name can be seen for example using top -H.
-- Provide more precise error message when job allocation can not be satisfied
(e.g. memory, disk, cpu count, etc. rather than just "node configuration
not available").
-- Create a new DebugFlags named TraceJobs in slurm.conf to print detailed
information about jobs in slurmctld. The information include job ids, state
and node count.
-- When a job dependency can never be satisfied do not cancel the job but keep
pending with reason WAIT_DEP_INVALID (DependencyNeverSatisfied).
* 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 - Do not explicitly unload windows for a job on termination,
only unload its table (which automatically unloads its windows).
-- When HealthCheckNodeState is configured as IDLE don't run the
HealthCheckProgram for nodes in any other states than IDLE.
-- Remove all slurmctld locks prior to job_submit() being called in plugins.
If any slurmctld data structures are read or modified, add locks directly
in the plugin.
-- 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.
* Changes in Slurm 14.03.6
==========================
-- Added examples to demonstrate the use of the sacct -T option to the man
page.
-- Fix for regression in 14.03.5 with sacctmgr load when Parent has "'"
around it.
-- Update comments in sacctmgr dump header.
-- Fix for possible abort on change in GRES configuration.
-- CRAY - fix modules file, (backport from 14.11 commit 78fe86192b.
-- Fix race condition which could result in requeue if batch job exit and node
registration occur at the same time.
-- switch/nrt - Unload job tables (in addition to windows) in user space mode.
-- Differentiate between two identical debug messages about purging vestigial
job scripts.
-- If the socket used by slurmstepd to communicate with slurmd exist when
slurmstepd attempts to create it, for example left over from a previous
requeue or crash, delete it and recreate it.
* Changes in Slurm 14.03.5
==========================
-- If a srun runs in an exclusive allocation and doesn't use the entire
allocation and CR_PACK_NODES is set layout tasks appropriately.
-- Correct Shared field in job state information seen by scontrol, sview, etc.
-- Print Slurm error string in scontrol update job and reset the Slurm errno
before each call to the API.
-- Fix task/cgroup to handle -mblock:fcyclic correctly
-- Fix for core-based advanced reservations where the distribution of cores
across nodes is not even.
-- Fix issue where association maxnodes wouldn't be evaluated correctly if a
QOS had a GrpNodes set.
-- GRES fix with multiple files defined per line in gres.conf.
-- When a job is requeued make sure accounting marks it as such.
-- Print the state of requeued job as REQUEUED.
-- Fix if a job's partition was taken away from it don't allow a requeue.
-- Make sure we lock on the conf when sending slurmd's conf to the slurmstepd.
-- Fix issue with sacctmgr 'load' not able to gracefully handle bad formatted
file.
-- sched/backfill: Correct job start time estimate with advanced reservations.
-- Error message added when in proctrack/cgroup the step freezer path isn't
able to be destroyed for debug.
-- Added extra index's into the database for better performance when
deleting users.
-- Fix issue with wckeys when tracking wckeys, but not enforcing them,
you could get multiple '*' wckeys.
-- Fix bug which could report to squeue the wrong partition for a running job
that is submitted to multiple partitions.
-- Report correct CPU count allocated to job when allocated whole node even if
not using all CPUs.
-- If job's constraints cannot be satisfied put it in pending state with reason
BadConstraints and don't remove it.
-- sched/backfill - If job started with infinite time limit, set its end_time
one year in the future.
-- Clear QOS GrpUsedCPUs when resetting raw usage if QOS is not using any cpus.
-- Remove log message left over from debugging.
-- When using CR_PACK_NODES fix make --ntasks-per-node work correctly.
-- Report correct partition associated with a step if the job is submitted to
multiple partitions.
-- Fix to allow removing of preemption from a QOS
-- If the proctrack plugins fail to destroy the job container print an error
message and avoid to loop forever, give up after 120 seconds.
-- Make srun obey POSIX convention and increase the exit code by 128 when the
process terminated by a signal.
-- Sanity check for acct_gather_energy/rapl
-- If the proctrack plugins fail to destroy the job container print an error
message and avoid to loop forever, give up after 120 seconds.
-- If the sbatch command specifies the option --signal=B:signum sent the signal
to the batch script only.
-- If we cancel a task and we have no other exit code send the signal and
exit code.
-- Added note about InnoDB storage engine being used with MySQL.
-- Set the job exit code when the job is signaled and set the log level to
debug2() when processing an already completed job.
-- Reset diagnostics time stamp when "sdiag --reset" is called.
-- squeue and scontrol to report a job's "shared" value based upon partition
options rather than reporting "unknown" if job submission does not use
--exclusive or --shared option.
-- task/cgroup - Fix cpuset binding for batch script.
-- sched/backfill - Fix anomaly that could result in jobs being scheduled out
of order.
-- Expand pseudo-terminal size data structure field sizes from 8 to 16 bits.
-- Set the job exit code when the job is signaled and set the log level to
-- Distinguish between two identical error messages.
-- If using accounting_storage/mysql directly without a DBD fix issue with
start of requeued jobs.
-- If a job fails because of batch node failure and the job is requeued and an
epilog complete message comes from that node do not process the batch step
information since the job has already been requeued because the epilog
script running isn't guaranteed in this situation.
-- Change message to note a NO_VAL for return code could of come from node
failure as well as interactive user.
-- Modify test4.5 to only look at one partition instead of all of them.
-- Fix sh5util -u to accept username different from the user that runs the
command.
-- Corrections to man pages:salloc.1 sbatch.1 srun.1 nonstop.conf.5
slurm.conf.5.
-- Have sacctmgr dump cluster handle situations where users or such have
special characters in their names like ':'
-- Add more debugging for information should the job ran on wrong node
and should there be problems accessing the state files.
-- Fix issue where not enforcing QOS but a partition either allows or denies
them.
-- CRAY - Make switch/cray default when running on a Cray natively.
-- CRAY - Make job_container/cncu default when running on a Cray natively.
-- Disable job time limit change if it's preemption is in progress.
-- Correct logic to properly enforce job preemption GraceTime.
-- Fix sinfo -R to print each down/drained node once, rather than once per
partition.
-- If a job has non-responding node, retry job step create rather than
returning with DOWN node error.
-- Support SLURM_CONF path which does not have "slurm.conf" as the file name.
-- CRAY - make job_container/cncu default when running on a Cray natively
-- Fix issue where batch cpuset wasn't looked at correctly in
jobacct_gather/cgroup.
-- Correct squeue's job node and CPU counts for requeued jobs.
-- Correct SelectTypeParameters=CR_LLN with job selecition of specific nodes.
-- Only if ALL of their partitions are hidden will a job be hidden by default.
-- Run EpilogSlurmctld for a job is killed during slurmctld reconfiguration.
-- Close window with srun if waiting for an allocation and while printing
something you also get a signal which would produce deadlock.
-- Add SelectTypeParameters option of CR_PACK_NODES to pack a job's tasks
tightly on its allocated nodes rather than distributing them evenly across
the allocated nodes.
-- 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.
-- Add new distribution method fcyclic so when a task is using multiple cpus
it can bind cyclically across sockets.
-- task/affinity - When using --hint=nomultithread only bind to the first
thread in a core.
-- Make cgroup task layout (block | cyclic) method mirror that of
task/affinity.
-- If TaskProlog sets SLURM_PROLOG_CPU_MASK reset affinity for that task
based on the mask given.
-- Keep supporting 'srun -N x --pty bash' for historical reasons.
-- If EnforcePartLimits=Yes and QOS job is using can override limits, allow
it.
-- Fix issues if partition allows or denies 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.
-- Fix format output of scontrol command when printing node state.
-- Improve the clean up of cgroup hierarchy when using the
jobacct_gather/cgroup plugin.
-- Added SchedulerParameters value of Ignore_NUMA.
-- Fix issues with code when using automake 1.14.1
-- select/cons_res plugin: Fix memory leak related to job preemption.
-- After reconfig rebuild the job node counters only for jobs that have
not finished yet, otherwise if requeued the job may enter an invalid
COMPLETING state.
-- Do not purge the script and environment files for completed jobs on
slurmctld reconfiguration or restart (they might be later requeued).
-- scontrol now accepts the option job=xxx or jobid=xxx for the requeue,
requeuehold and release operations.
-- task/cgroup - fix to bind batch job in the proper CPUs.
-- Added strigger option of -N, --noheader to not print the header when
displaying a list of triggers.
-- Modify strigger to accept arguments to the program to execute when an
event trigger occurs.
-- Attempt to create duplicate event trigger now generates ESLURM_TRIGGER_DUP
("Duplicate event trigger").
-- Treat special characters like %A, %s etc. literally in the file names
when specified escaped e.g. sbatch -o /home/zebra\\%s will not expand
%s as the stepid of the running job.
-- CRAYALPS - Add better support for CLE 5.2 when running Slurm over ALPS.
-- Test time when job_state file was written to detect multiple primary
slurmctld daemons (e.g. both backup and primary are functioning as
primary and there is a split brain problem).
-- Fix scontrol to accept update jobid=# numtasks=#
-- If the backup slurmctld assumes primary status, then do NOT purge any
job state files (batch script and environment files) and do not re-use them.
This may indicate that multiple primary slurmctld daemons are active (e.g.
both backup and primary are functioning as primary and there is a split
brain problem).
-- Set correct error code when requeuing a completing/pending job
-- When checking for if dependency of type afterany, afterok and afternotok
don't clear the dependency if the job is completing.
-- Cleanup the JOB_COMPLETING flag and eventually requeue the job when the
last epilog completes, either slurmd epilog or slurmctld epilog, whichever
comes last.
-- When attempting to requeue a job distinguish the case in which the job is
JOB_COMPLETING or already pending.
-- When reconfiguring the controller don't restart the slurmctld epilog if it
is already running.
-- Email messages for job array events print now use the job ID using the
format "#_# (#)" rather than just the internal job ID.
-- Set the number of free licenses to be 0 if the global license count
decreases and total is less than in use.
-- Add DebugFlag of BackfillMap. Previously a DebugFlag value of Backfill
logged information about what it was doing plus a map of expected resouce
use in the future. Now that very verbose resource use map is only logged
with a DebugFlag value of BackfillMap
-- Modify the description of -E and -S option of sacct command as point in time
'before' or 'after' the database records are returned.
-- Correct support for partition with Shared=YES configuration.
-- If job requests --exclusive then do not use nodes which have any cores in an
advanced reservation. Also prevents case where nodes can be shared by other
jobs.
-- For "scontrol --details show job" report the correct CPU_IDs when thre are
multiple threads per core (we are translating a core bitmap to CPU IDs).
-- If DebugFlags=Protocol is configured in slurm.conf print details of the
connection, ip address and port accepted by the controller.
-- Fix minor memory leak when reading in incomplete node data checkpoint file.
-- Enlarge the width specifier when printing partition SHARE to display larger
sharing values.
-- sinfo locks added to prevent possibly duplicate record printing for
resources in multiple partitions.
* Changes in Slurm 14.03.3-2
============================
* Changes in Slurm 14.03.3
==========================
-- Correction to default batch output file name. In version 14.03.2 was using
"slurm_<jobid>_4294967294.out" due to error in job array logic.
-- In slurm.spec file, replace "Requires cray-MySQL-devel-enterprise" with
"Requires mysql-devel".
-- Fix race condition if PrologFlags=Alloc,NoHold is used.
-- Cray - Make NPC only limit running other NPC jobs on shared blades instead
of limited non NPC jobs.
-- Fix for sbatch #PBS -m (mail) option parsing.
-- Fix job dependency bug. Jobs dependent upon multiple other jobs may start
prematurely.
-- Set "Reason" field for all elements of a job array on short-circuited
scheduling for job arrays.
-- Allow -D option of salloc/srun/sbatch to specify relative path.
-- Added SchedulerParameter of batch_sched_delay to permit many batch jobs
to be submitted between each scheduling attempt to reduce overhead of
scheduling logic.
-- Added job reason of "SchedTimeout" if the scheduler was not able to reach
the job to attempt scheduling it.
-- Add job's exit state and exit code to email message.
-- scontrol hold/release accepts job name option (in addition to job ID).
-- Handle when trying to cancel a step that hasn't started yet better.
-- Add --priority option to salloc, sbatch and srun commands.
-- Honor partition priorities over job priorities.
-- Fix sacct -c when using jobcomp/filetxt to read newer variables
-- Fix segfault of sacct -c if spaces are in the variables.
-- Release held job only with "scontrol release <jobid>" and not by resetting
the job's priority. This is needed to support job arrays better.
-- Correct squeue command not to merge jobs with state pending and completing
together.
-- Fix issue where user is requesting --acctg-freq=0 and no memory limits.
-- Fix issue with GrpCPURunMins if a job's timelimit is altered while the job
is running.
-- Temporary fix for handling our typemap for the perl api with newer perl.
-- Fix allowgroup on bad group seg fault with the controller.
-- Handle node ranges better when dealing with accounting max node limits.
-- Update configure to set correct version without having to run autogen.sh
* Changes in Slurm 14.03.1
==========================
-- Add support for job std_in, std_out and std_err fields in Perl API.
-- Add "Scheduling Configuration Guide" web page.
-- BGQ - fix check for jobinfo when it is NULL
-- Do not check cleaning on "pending" steps.
-- task/cgroup plugin - Fix for building on older hwloc (v1.0.2).
-- In the PMI implementation by default don't check for duplicate keys.
Set the SLURM_PMI_KVS_DUP_KEYS if you want the code to check for
duplicate keys.
-- Permit user root to propagate resource limits higher than the hard limit
slurmd has on that compute node has (i.e. raise both current and maximum
limits).
-- Fix issue with license used count when doing an scontrol reconfig.
-- Fix the PMI iterator to not report duplicated keys.
-- Fix issue with sinfo when -o is used without the %P option.
-- Rather than immediately invoking an execution of the scheduling logic on
every event type that can enable the execution of a new job, queue its
execution. This permits faster execution of some operations, such as
modifying large counts of jobs, by executing the scheduling logic less
frequently, but still in a timely fashion.
-- If the environment variable is greater than MAX_ENV_STRLEN don't
set it in the job env otherwise the exec() fails.
-- Optimize scontrol hold/release logic for job arrays.
-- Modify srun to report an exit code of zero rather than nine if some tasks
exit with a return code of zero and others are killed with SIGKILL. Only an
exit code of zero did this.
-- Avoid slurmctld crash getting job info if detail_ptr is NULL.
-- Fix sacctmgr add user where both defaultaccount and accounts are specified.
-- Added SchedulerParameters option of max_sched_time to limit how long the
main scheduling loop can execute for.
-- Added SchedulerParameters option of sched_interval to control how frequently
the main scheduling loop will execute.
-- Move start time of main scheduling loop timeout after locks are aquired.
-- Add squeue job format option of "%y" to print a job's nice value.
-- Update scontrol update jobID logic to operate on entire job arrays.
-- Fix PrologFlags=Alloc to run the prolog on each of the nodes in the
allocation instead of just the first.
-- Fix race condition if a step is starting while the slurmd is being
restarted.
-- Make sure a job's prolog has ran before starting a step.
-- BGQ - Fix invalid memory read when using DefaultConnType in the
bluegene.conf
-- Make sure we send node state to the DBD on clean start of controller.
-- Fix some sinfo and squeue sorting anomalies due to differences in data
types.
-- Only send message back to slurmctld when PrologFlags=Alloc is used on a
Cray/ALPS system, otherwise use the slurmd to wait on the prolog to gate
the start of the step.
-- Remove need to check PrologFlags=Alloc in slurmd since we can tell if prolog
has ran yet or not.
-- Fix squeue to use a correct macro to check job state.
-- BGQ - Fix incorrect logic issues if MaxBlockInError=0 in the bluegene.conf.
-- priority/basic - Insure job priorities continue to decrease when jobs are
submitted with the --nice option.
-- Make the PrologFlag=Alloc work on batch scripts
-- Make PrologFlag=NoHold (automatically sets PrologFlag=Alloc) not hold in
salloc/srun, instead wait in the slurmd when a step hits a node and the
prolog is still running.
-- Added --cpu-freq=highm1 (high minus one) option.
-- Expand StdIn/Out/Err string length output by "scontrol show job" from 128
to 1024 bytes.
-- squeue %F format will now print the job ID for non-array jobs.
-- Use quicksort for all priority based job sorting, which improves performance
significantly with large job counts.
-- If a job has already been released from a held state ignore successive
release requests.
-- Fix srun/salloc/sbatch man pages for the --no-kill option.
-- Add squeue -L/--licenses option to filter jobs by license names.
-- Handle abort job on node on front end systems without core dumping.
-- Fix dependency support for job arrays.
-- When updating jobs verify the update request is not identical to
the current settings.
-- When sorting jobs and priorities are equal sort by job_id.
-- Do not overwrite existing reason for node being down or drained.
-- Requeue batch job if Munge is down and credential can not be created.
-- Make _slurm_init_msg_engine() tolerate bug in bind() returning a busy
ephemeral port.
-- Don't block scheduling of entire job array if it could run in multiple
partitions.
-- Introduce a new debug flag Protocol to print protocol requests received
together with the remote IP address and port.
-- CRAY - Set up the network even when only using 1 node.
-- CRAY - Greatly reduce the number of error messages produced from the task
plugin and provide more information in the message.
* Changes in Slurm 14.03.0
==========================
-- job_submit/lua: Fix invalid memory reference if script returns error message
for user.
-- Add logic to sleep and retry if slurm.conf can't be read.
-- Reset a node's CpuLoad value at least once each SlurmdTimeout seconds.
-- Scheduler enhancements for reservations: When a job needs to run in
reservation, but can not due to busy resources, then do not block all jobs
in that partition from being scheduled, but only the jobs in that
reservation.
-- Export "SLURM*" environment variables from sbatch even if --export=NONE.
-- When recovering node state if the Slurm version is 2.6 or 2.5 set the
protocol version to be SLURM_2_5_PROTOCOL_VERSION which is the minimum
supported version.
-- Update the scancel man page documenting the -s option.
-- Update sacctmgr man page documenting how to modify account's QOS.
-- Fix for sjstat which currently does not print >1TB memory values correctly.
-- Change xmalloc()/xfree() to malloc()/free() in hostlist.c for better
performance.
-- Update squeue.1 man page describing the SPECIAL_EXIT state.
-- Added scontrol option of errnumstr to return error message given a slurm
error number.
-- If srun invoked with the --multi-prog option, but no task count, then use
the task count provided in the MPMD configuration file.
-- Prevent sview abort on some systems when adding or removing columns to the
display for nodes, jobs, partitions, etc.
-- Add job array hash table for improved performance.
-- Make AccountingStorageEnforce=all not include nojobs or nosteps.
-- Added sacctmgr mod qos set RawUsage=0.
-- Modify hostlist functions to accept more than two numeric ranges (e.g.
"row[1-3]rack[0-8]slot[0-63]")
-- Run job scheduling logic immediately when nodes enter service.
-- Added sbatch '--parsable' option to output only the job id number and the
cluster name separated by a semicolon. Errors will still be displayed.
-- Added failure management "slurmctld/nonstop" plugin.
-- Prevent jobs being killed when a checkpoint plugin is enabled or disabled.
-- Update the documentation about SLURM_PMI_KVS_NO_DUP_KEYS environment
variable.
-- select/cons_res bug fix for range of node counts with --cpus-per-task
option (e.g. "srun -N2-3 -c2 hostname" would allocate 2 CPUs on the first
node and 0 CPUs on the second node).
-- Change reservation flags field from 16 to 32-bits.
-- Add reservation flag value of "FIRST_CORES".
-- Added the idea of Resources to the database. Framework for handling
license servers outside of Slurm.
-- When starting the slurmctld only send past job/node state information to
accounting if running for the first time (should speed up startup
dramatically on systems with lots of nodes or lots of jobs).
-- Make job array expressions more flexible to accept multiple step counts in
the expression (e.g. "--array=1-10:2,50-60:5,123").
-- switch/cray - add state save/restore logic tracking allocated ports.
-- SchedulerParameters - Replace max_job_bf with bf_max_job_start (both will
work for now).
-- Add SchedulerParameters options of preempt_reorder_count and
preempt_strict_order.
-- Make memory types in acct_gather uint64_t to handle systems with more than
4TB of memory on them.
-- BGQ - --export=NONE option for srun to make it so only the SLURM_JOB_ID
and SLURM_STEP_ID env vars are set.
-- Munge plugins - Add sleep between retries if can't connect to socket.
-- Added DebugFlags value of "License".
-- Added --enable-developer which will give you -Werror when compiling.
-- Fix for job request with GRES count of zero.
-- Job array dependency logic: Cache results for major performance improvement.
-- Modify squeue to support filter on job states Special_Exit and Resizing.
-- Defer purging job record until after EpilogSlurmctld completes.
-- Fix handling RPCs from a 14.03 slurmctld to a 2.6 slurmd
* Changes in Slurm 14.03.0pre6
==============================
-- Modify slurmstepd to log messages according to the LogTimeFormat
parameter in slurm.conf.
-- Insure that overlapping reservations do not oversubscribe available
licenses.
-- Added core specialization logic to select/cons_res plugin.
-- Added whole_node field to job_resources structure and enable gang scheduling
for jobs with core specialization.
-- When using FastSchedule = 1 the nodes with less than configured resources
are not longer set DOWN, they are set to DRAIN instead.
-- Modified 'sacctmgr show associations' command to show GrpCPURunMins
by default.
-- Replace the hostlist_push() function with a more efficient
hostlist_push_host().
-- Modify the reading of lustre file system statistics to print more
information when debug and when io error occur.
-- Add specialized core count field to job credential data.
NOTE: This changes the communications protocol from other pre-releases of
version 14.03. All programs must be cancelled and daemons upgraded from
previous pre-releases of version 14.03. Upgrades from version 2.6 or earlier
can take place without loss of jobs
-- Add version number to node and front-end configuration information visible
using the scontrol tool.
-- Add idea of a RESERVED flag for node state so idle resources are marked
not "idle" when in a reservation.
-- Added core specialization plugin infrastructure.
-- Added new job_submit/trottle plugin to control the rate at which a user
can submit jobs.
-- CRAY - added network performance counters option.
-- Allow scontrol suspend/resume to accept jobid in the format jobid_taskid
to suspend/resume array elements.
-- In the slurmctld job record, split "shared" variable into "share_res" (share
resource) and "whole_node" fields.
-- Fix the format of SLURM_STEP_RESV_PORTS. It was generated incorrectly
when using the hostlist_push_host function and input surrounded by [].
-- Modify the srun --slurmd-debug option to accept debug string tags
(quiet, fatal, error, info verbose) beside the numerical values.
-- Fix the bug where --cpu_bind=map_cpu is interpreted as mask_cpu.
-- Update the documentation egarding the state of cpu frequencies after
a step using --cpu-freq completes.
-- CRAY - Fix issue when a job is requeued and nhc is still running as it is
being scheduled to run again. This would erase the previous job info
that was still needed to clean up the nodes from the previous job run.
(Bug 526).
-- Set SLURM_JOB_PARTITION environment variable set for all job allocations.
-- Set SLURM_JOB_PARTITION environment variable for Prolog program.
-- Added SchedulerParameters option of partition_job_depth to limit scheduling
logic depth by partition.
-- Handle the case in which errno is not reset to 0 after calling
getgrent_r(), which causes the controller to core dump.
-- Added squeue format option of "%X" (core specialization count).
-- Added core specialization web page (just a start for now).
-- Added the SLURM_ARRAY_JOB_ID and SLURM_ARRAY_TASK_ID
-- Fix bug in job step allocation failing due to memory limit.
-- Modify the pbsnodes script to reflect its output on a TORQUE system.
-- Add ability to clear a node's DRAIN flag using scontrol or sview by setting
it's state to "UNDRAIN". The node's base state (e.g. "DOWN" or "IDLE") will
not be changed.
-- Modify the output of 'scontrol show partition' by displaying
DefMemPerCPU=UNLIMITED and MaxMemPerCPU=UNLIMITED when these limits are
configured as 0.
-- mpirun-mic - Major re-write of the command wrapper for Xeon Phi use.
-- Add new configuration parameter of AuthInfo to specify port used by
authentication plugin.
-- Fixed conditional RPM compiling.
-- Corrected slurmstepd ident name when logging to syslog.
-- Fixed sh5util loop when there are no node-step files.
-- Add SLURM_CLUSTER_NAME to environment variables passed to PrologSlurmctld,
Prolog, EpilogSlurmctld, and Epilog
-- Add the idea of running a prolog right when an allocation happens
instead of when running on the node for the first time.
-- If user runs 'scontrol reconfig' but hostnames or the host count changes
the slurmctld throws a fatal error.
-- gres.conf - Add "NodeName" specification so that a single gres.conf file
can be used for a heterogeneous cluster.
-- Add flag to accounting RPC to indicate if job data is packed or not.
-- After all srun tasks have terminated on a node close the stdout/stderr
channel with the slurmstepd on that node.
-- In case of i/o error with slurmstepd log an error message and abort the
job.
-- Add --test-only option to sbatch command to validate the script and options.
The response includes expected start time and resources to be allocated.
==============================
-- Remove the ThreadID documentation from slurm.conf. This functionality has
been obsoleted by the LogTimeFormat.
-- Sched plugins - rename global and plugin functions names for consistency
with other plugin types.
-- BGQ - Added RebootQOSList option to bluegene.conf to allow an implicate
reboot of a block if only jobs in the list are running on it when cnodes
go into a failure state.
-- Correct task count of pending job steps.
-- Improve limit enforcement for jobs, set RLIMIT_RSS, RLIMIT_AS and/or
RLIMIT_DATA to enforce memory limit.
-- Pending job steps will have step_id of INFINITE rather than NO_VAL and
will be reported as "TBD" by scontrol and squeue commands.
-- Add logic so PMI_Abort or PMI2_Abort can propagate an exit code.
-- Added SlurmdPlugstack configuration parameter.
-- Added PriorityFlag DEPTH_OBLIVIOUS to have the depth of an association
not effect it's priorty.
-- Multi-thread the sinfo command (one thread per partition).
-- Added sgather tool to gather files from a job's compute nodes into a
central location.
-- Added configuration parameter FairShareDampeningFactor to offer a greater
priority range based upon utilization.
-- Change MaxArraySize and job's array_task_id from 16-bit to 32-bit field.
Additional Slurm enhancements are be required to support larger job arrays.
-- Added -S/--core-spec option to salloc, sbatch and srun commands to reserve
specialized cores for system use. Modify scontrol and sview to get/set
the new field. No enforcement exists yet for these new options.
struct job_info / slurm_job_info_t: Added core_spec
struct job_descriptorjob_desc_msg_t: Added core_spec
-- Do not set SLURM_NODEID environment variable on front-end systems.
-- Convert bitmap functions to use int32_t instead of int in data structures
and function arguments. This is to reliably enable use of bitmaps containing
up to 4 billion elements. Several data structures containing index values
were also changed from data type int to int32_t:
- Struct job_info / slurm_job_info_t: Changed exc_node_inx, node_inx, and
req_node_inx from type int to type int32_t
- job_step_info_t: Changed node_inx from type int to type int32_t
- Struct partition_info / partition_info_t: Changed node_inx from type int
to type int32_t
- block_job_info_t: Changed cnode_inx from type int to type int32_t
- block_info_t: Changed ionode_inx and mp_inx from type int to type int32_t
- Struct reserve_info / reserve_info_t: Changed node_inx from type int to
type int32_t
-- Modify qsub wrapper output to match torque command output, just print the
job ID rather than "Submitted batch job #"
-- Change Slurm error string for ESLURM_MISSING_TIME_LIMIT from
"Missing time limit" to
"Time limit specification required, but not provided"
-- Change salloc job_allocate error message header from
"Failed to allocate resources" to
"Job submit/allocate failed"
-- Modify slurmctld message retry logic to support Cray cold-standby SDB.
-- Added "JobAcctGatherParams" configuration parameter. Value of "NoShare"
disables accounting for shared memory.
-- Added fields to "scontrol show job" output: boards_per_node,
sockets_per_board, ntasks_per_node, ntasks_per_board, ntasks_per_socket,
ntasks_per_core, and nice.
-- Add squeue output format options for job command and working directory
(%o and %Z respectively).
-- Add stdin/out/err to sview job output.
-- Add new job_state of JOB_BOOT_FAIL for job terminations due to failure to
boot it's allocated nodes or BlueGene block.
-- CRAY - Add SelectTypeParameters NHC_NO_STEPS and NHC_NO which will disable
the node health check script for steps and allocations respectfully.
-- Reservation with CoreCnt: Avoid possible invalid memory reference.
-- Add new error code for attempt to create a reservation with duplicate name.
-- Validate that a hostlist file contains text (i.e. not a binary).
-- switch/generic - propagate switch information from srun down to slurmd and
slurmstepd.
-- CRAY - Do not package Slurm's libpmi or libpmi2 libraries. The Cray version
of those libraries must be used.
-- Added a new option to the scontrol command to view licenses that are
configured in use and avalable. 'scontrol show licenses'.
-- MySQL - Made Slurm compatible with 5.6
==============================
-- Add task pointer to the task_post_term() function in task plugins. The
terminating task's PID is available in task->pid.
-- Defer sending SIGKILL signal to processes while core dump in progress.
-- Added JobContainerPlugin configuration parameter and plugin infrastructure.
-- Added partition configuration parameters AllowAccounts, AllowQOS,
DenyAccounts and DenyQOS.
-- The rpmbuild option for a cray system with ALPS has changed from
%_with_cray to %_with_cray_alps.
-- The log file timestamp format can now be selected at runtime via the
LogTimeFormat configuration option. See the slurm.conf and slurmdbd.conf
man pages for details.
-- Added switch/generic plugin to a job's convey network topology.
-- BLUEGENE - If block is in 'D' state or has more cnodes in error than
MaxBlockInError set the job wait reason appropriately.
-- API use: Generate an error return rather than fatal error and exit if the
configuraiton file is absent or invalid. This will permit Slurm APIs to be
more reliably used by other programs.
-- Add support for load-based scheduling, allocate jobs to nodes with the
largest number of available CPUs. Added SchedulingParameters paramter of
"CR_LLN" and partition parameter of "LLN=yes|no".
-- Added job_info() and step_info() functions to the gres plugins to extract
plugin specific fields from the job's or step's GRES data structure.
-- Added sbatch --signal option of "B:" to signal the batch shell rather than
only the spawned job steps.
-- Added sinfo and squeue format option of "%all" to print all fields available
for the data type with a vertical bar separating each field.
-- Add mechanism for job_submit plugin to generate error message for srun,
salloc or sbatch to stderr. New argument added to job_submit function in
the plugin.
-- Add StdIn, StdOut, and StdErr paths to job information dumped with
"scontrol show job".
-- Permit Slurm administrator to submit a batch job as any user.
-- Set a job's RLIMIT_AS limit based upon it's memory limit and VsizeFactor
configuration value.
-- Remove Postgres plugins
-- Make jobacct_gather/cgroup work correctly and also make all jobacct_gather
plugins more maintainable.
-- Proctrack/pgid - Add support for proctrack_p_plugin_get_pids() function.
-- Sched/backfill - Change default max_job_bf parameter from 50 to 100.
-- Added -I|--item-extract option to sh5util to extract data item from series.
* Changes in Slurm 2.6.10
=========================
-- Switch/nrt - On switch resource allocation failure, free partial allocation.
-- Switch/nrt - Properly track usage of CAU and RDMA resources with multiple
tasks per compute node.
-- Fix issue where user is requesting --acctg-freq=0 and no memory limits.
-- BGQ - Temp fix issue where job could be left on job_list after it finished.
-- BGQ - Fix issue where limits were checked on midplane counts instead of
cnode counts.
-- BGQ - Move code to only start job on a block after limits are checked.
-- Handle node ranges better when dealing with accounting max node limits.
-- Fix perlapi to compile correctly with perl 5.18
-- BGQ - Fix issue with uninitialized variable.
-- Correct sinfo --sort fields to match documentation: E => Reason,
H -> Reason Time (new), R -> Partition Name, u/U -> Reason user (new)
-- If an invalid assoc_ptr comes in don't use the id to verify it.
-- Sched/backfill modified to avoid using nodes in completing state.
-- Correct support for job --profile=none option and related documentation.
-- Properly enforce job --requeue and --norequeue options.
-- If a job --mem-per-cpu limit exceeds the partition or system limit, then
scale the job's memory limit and CPUs per task to satisfy the limit.
-- Correct logic to support Power7 processor with 1 or 2 threads per core
(CPU IDs are not consecutive).
* Changes in Slurm 2.6.9
========================
-- Fix sinfo to work correctly with draining/mixed nodes as well as filtering
on Mixed state.
-- Fix sacctmgr update user with no "where" condition.
-- Fix logic bugs for SchedulerParameters option of max_rpc_cnt.
* Changes in Slurm 2.6.8
========================
-- Add support for Torque/PBS job array options and environment variables.
-- Fix issue where jobs still pending after a reservation would remain
in waiting reason ReqNodeNotAvail.
-- Update last_job_update when a job's state_reason was modified.
-- Free job_ptr->state_desc where ever state_reason is set.
-- Fixed sacct.1 and srun.1 manual pages which contains a hyphen where
a minus sign for options was intended.
-- sinfo - Make sure if partition name is long and the default the last char
doesn't get chopped off.
-- task/affinity - Protect against zero divide when simulating more hardware
than you really have.
-- NRT - Fix issue with 1 node jobs. It turns out the network does need to
be setup for 1 node jobs.
-- Fix recovery of job dependency on task of job array when slurmctld restarts.
-- mysql - Fix invalid memory reference.
-- Lock the /cgroup/freezer subsystem when creating files for tracking processes.
-- Fix preempt/partition_prio to avoid preempting jobs in partitions with
PreemptMode=OFF
-- launch/poe - Implicitly set --network in job step create request as needed.
-- Permit multiple batch job submissions to be made for each run of the
scheduler logic if the job submissions occur at the nearly same time.
-- Fix issue where associations weren't correct if backup takes control and
new associations were added since it was started.
-- Fix race condition is corner case with backup slurmctld.
-- With the backup slurmctld make sure we reinit beginning values in the
slurmdbd plugin.
-- Fix sinfo to work correctly with draining/mixed nodes.
-- MySQL - Fix it so a lock isn't held unnecessarily.
-- Added new SchedulerParameters option of max_rpc_cnt when too many RPCs
are active.
-- BGQ - Fix deny_pass to work correctly.
-- BGQ - Fix sub block steps using a block when the block has passthrough's
in it.
* Changes in Slurm 2.6.7
========================
-- Properly enforce a job's cpus-per-task option when a job's allocation is
constrained on some nodes by the mem-per-cpu option.
-- Correct the slurm.conf man pages and checkpoint_blcr.html page
describing that jobs must be drained from cluster before deploying
-- Fix issue where if using munge and munge wasn't running and a slurmd
needed to forward a message, the slurmd would core dump.
-- Update srun.1 man page documenting the PMI2 support.
-- Fix slurmctld core dump when a jobs gets its QOS updated but there
is not a corresponding association.
-- If a job requires specific nodes and can not run due to those nodes being
busy, the main scheduling loop will block those specific nodes rather than
the entire queue/partition.
-- Fix minor memory leak when updating a job's name.
-- Fix minor memory leak when updating a reservation on a partition using "ALL"
nodes.
-- Fix minor memory leak when adding a reservation with a nodelist and core
count.
-- Update sacct man page description of job states.
-- BGQ - Fix minor memory leak when selecting blocks that can't immediately be
placed.
-- Fixed minor memory leak in backfill scheduler.
-- MYSQL - Fixed memory leak when querying clusters.
-- MYSQL - Fix when updating QOS on an association.
-- NRT - Fix to supply correct error messages to poe/pmd when a launch fails.
-- Add SLURM_STEP_ID to Prolog environment.
-- Add support for SchedulerParameters value of bf_max_job_start that limits
the total number of jobs that can be started in a single iteration of the
backfill scheduler.
-- Don't print negative number when dealing with large memory sizes with
sacct.
-- Fix sinfo output so that host in state allocated and mixed will not be
merged together.
-- GRES: Avoid crash if GRES configurations is inconstent.
-- Make S_SLURM_RESTART_COUNT item available to SPANK.
-- Munge plugins - Add sleep between retries if can't connect to socket.
-- Fix the database query to return all pending jobs in a given time interval.
-- switch/nrt - Correct logic to get dynamic window count.
-- Remove need to use job->ctx_params in the launch plugin, just to simplify
code.
-- NRT - Fix possible memory leak if using multiple adapters.
-- NRT - Fix issue where there are more than NRT_MAXADAPTERS on a system.
-- NRT - Increase Max number of adapters from 8 -> 9
-- NRT - Initialize missing variables when the PMD is starting a job.
-- NRT - Fix issue where we are launching hosts out of numerical order,
this would cause pmd's to hang.
-- NRT - Change xmalloc's to malloc just to be safe.