Newer
Older
This file describes changes in recent versions of SLURM. It primarily
documents those changes that are of interest to users and admins.
-- Added srun option "--checkpoint=time" for job step to automatically be
checkpointed on a period basis.
-- Change behavior of "scancel -s KILL <jobid>" to send SIGKILL to all job
steps rather than cancelling the job. This now matches the behavior of
all other signals. "scancel <jobid>" still cancels the job and all steps.
-- Add support for new job step options --exclusive and --immediate. Permit
job steps to be queued when resources are not available within an existing
job allocation to dedicate the resources to the job step. Useful for
executing simultaneous job steps. Provides resource management both at
the level of jobs and job steps.
-- Add support for feature count in job contraints, for example
srun --nodes=16 --constraint=graphics*4 ...
Based upon work by Kumar Krishna (HP, India).
-- Add multi-core options to salloc and sbatch commands (sbatch.patch and
cleanup.patch from Chris Holmes, HP).
-- In select/cons_res properly release resources allocated to job being suspended
(rmbreak.patch, from Chris Holmes, HP).
-- Removed database and jobacct plugin replaced with jobacct_storage
and jobacct_gather for easier hooks for further expansion of the
jobacct plugin.
* Changes in SLURM 1.3.0-pre2
=============================
-- Added new srun option --pty to start job with pseudo terminal attached
to task 0 (all other tasks have I/O discarded)
-- Disable user specifying jobid when sched/wiki2 configured (needed for
Moab releases until early 2007).
-- Report command, args and working directory for batch jobs with
"scontrol show job".
* Changes in SLURM 1.3.0-pre1
=============================
Christopher J. Morrone
committed
-- !!! SRUN CHANGES !!!
The srun options -A/--allocate, -b/--batch, and -a/--attach have been
removed! That functionality is now available in the separate commands
salloc, sbatch, and sattach, respectively.
-- Add new node state FAILING plus trigger for when node enters that state.
-- Add new configuration paramter "PrivateData". This can be used to
prevent a user from seeing jobs or job steps belonging to other users.
-- Added configuration parameters for node power save mode: ResumeProgram
ResumeRate, SuspendExcNodes, SuspendExcParts, SuspendProgram and
SuspendRate.
-- Slurmctld maintains the IP address (rather than hostname) for srun
communications. This fixes some possible network routing issues.
-- Added global database plugin. Job accounting and Job completion are the
first to use it. Follow documentation to add more to the plugin.
-- Removed no-longer-needed jobacct/common/common_slurmctld.c since that is
replaced by the database plugin.
-- Added new configuration parameter: CryptoType.
Moved existing digital signature logic into new plugin: crypto/openssl.
Added new support for crypto/munge (available with GPL license).
* Changes in SLURM 1.2.17
=========================
-- In select/cons_res properly release resources allocated to job being
suspended (rmbreak.patch, from Chris Holmes, HP).
-- Fix AIX linking problem for PMI (mpich2) support.
-- Improve PMI logic for greater scalability (up to 16k tasks run).
-- Add srun support for SLURM_THREADS and PMI_FANOUT environment variables.
* Changes in SLURM 1.2.16
=========================
-- Add --overcommit option to the salloc command.
-- Run task epilog from job's working directory rather than directory
where slurmd daemon started from.
-- Log errors running task prolog or task epilog to srun's output.
-- In sched/wiki2, fix bug processing condensed hostlist expressions.
-- Release contribs/mpich1.slurm.patch without GPL license.
-- Fix bug in mvapich plugin for read/write calls that return EAGAIN.
-- Don't start MVAPICH timeout logic until we know that srun is starting
an MVAPICH program.
-- Fix to srun only allocating number of nodes needed for requested task
count when combining allocation and step creation in srun.
-- Execute task-prolog within proctrack container to insure that all
child processes get terminated.
-- Fixed job accounting to work with sgi_job proctrack plugin.
* Changes in SLURM 1.2.15
=========================
-- In sched/wiki2, fix bug processing hostlist expressions where hosts
lack a numeric suffix.
-- Fix bug in srun. When user did not specify time limit, it defaulted to
INFINITE rather than partition's limit.
-- In select/cons_res with SelectTypeParameters=CR_Socket_Memory, fix bug in
memory allocation tracking, mem.patch from Chris Holmes, HP.
-- Add --overcommit option to the sbatch command.
* Changes in SLURM 1.2.14
=========================
-- Fix a couple of bugs in MPICH/MX support (from Asier Roa, BSC).
-- Fix perl api for AIX
-- Add wiki.conf parameter ExcludePartitions for selected partitions to
be directly schedule by Slurm without Moab control
-- Optimize load leveling for shared nodes (alloc.patch, contributed
by Chris Holmes, HP).
-- Added PMI_TIME environment variable for user to control how PMI
communications are spread out in time. See "man srun" for details.
-- Added PMI timing information to srun debug mode to aid in tuning.
Use "srun -vv ..." to see the information.
-- Added checkpoint/ompi (OpenMPI) plugin (still under development).
-- Fix bug in load leveling logic added to v1.2.13 which can cause an
infinite loop and hang slurmctld when sharing nodes between jobs.
-- Added support for sbatch to read in #PBS options from a script
* Changes in SLURM 1.2.13
=========================
-- Add slurm.conf parameter JobFileAppend.
-- Fix for segv in "scontrol listpids" on nodes not in SLURM config.
-- Add support for SCANCEL_CTLD env var.
-- In mpi/mvapich plugin, add startup timeout logic. Time based upon
SLURM_MVAPICH_TIMEOUT (value in seconds).
-- Fixed pick_step_node logic to only pick the number of nodes requested
from the user when excluding nodes, to avoid an error message.
-- Disable salloc, sbatch and srun -I/--immediate options with
Moab scheduler.
-- Added "contribs" directory with a Perl API and Torque wrappers for Torque
to SLURM migration. This directory should be used to put anything that
is outside of SLURM proper such as a different API. Perl APIs contributed
by Hongjia Cao (NUDT).
-- In sched/wiki2: add support for tasklist with node name expressions
and task counts (e.g. TASKLIST=tux[1-4]*2:tux[12-14]*4").
-- In select/cons_res with sched/wiki2: fix bug in task layout logic.
-- Removed all curses info from the bluegene plugin putting it into smap
where it belongs.
-- Add support for job time limit specification formats: min, min:sec,
hour:min:sec, and days-hour:min:sec (formerly only supported minutes).
Applies to salloc, sbatch, and srun commands.
-- Improve scheduling support for exclusive constraint list, nodes can
now be in more than one constraint specific exclusively for a job
(e.g. "srun -C [rack1|rack2|rack3|rowB] srun")
-- Create separate MPICH/MX plugin (split out from MPICH/GM plugin)
-- Increase default MessageTimeout (in slurm.conf) from 5 to 10 secs.
-- Fix bug in batch job requeue if node zero of allocation fails to respond
to task launch request.
-- Improve load leveling logic to more evenly distribute the workload
(best_load.patch, contributed by Chris Holmes, HP).
* Changes in SLURM 1.2.12
=========================
-- Increase maximum message size from 1MB to 16MB (from Ernest Artiaga, BSC).
-- In PMI_Abort(), log the event and abort the entire job step.
-- Add support for additional PMI functions: PMI_Get_clique_ranks and
PMI_Get_clique_size (from Chuck Clouston, Bull).
-- Report an error when a hostlist comes in appearing to be a box but not
formatted in XYZxXYZ format.
-- Add support for partition configuration "Shared=exclusive". This is
equivalent to "srun --exclusive" when select/cons_res is configured.
-- In sched/wiki2, report the reason for a node being unavailable for the
GETNODES command using the CAT="<reason>" field.
-- In sched/wiki2 with select/linear, duplicate hostnames in HOSTLIST, one
per allocated processor.
-- Fix bug in scancel with specific signal and job lacks active steps.
-- In sched/wiki2, add support for NOTIFYJOB ARG=<jobid> MSG=<message>.
This sends a message to an active srun command.
-- salloc will now set SLURM_NPROCS to improve srun's behavior under salloc.
-- In sched/wiki2 and select/cons_res: insure that Slurm's CPU allocation
is identical to Moab's (from Ernest Artiaga and Asier Roa, BSC).
-- Added "scontrol show slurmd" command to status local slurmd daemon.
-- Set node DOWN if prolog fails on node zero of batch job launch.
-- Properly handle "srun --cpus-per-task" within a job allocation when
SLURM_TASKS_PER_NODE environment varable is not set.
Christopher J. Morrone
committed
-- Fixed return of slurm_send_rc_msg if msg->conn_fd is < 0 set errno ENOTCONN
and return SLURM_ERROR instead of return ENOTCONN
-- Added read before we send anything down a socket to make sure the socket
is still there.
-- Add slurm.conf variables UnkillableStepProgram and UnkillableStepTimeout.
Christopher J. Morrone
committed
-- Enable nice file propagation from sbatch command.
Christopher J. Morrone
committed
* Changes in SLURM 1.2.11
=========================
-- Updated "etc/mpich1.slurm.patch" for direct srun launch of MPICH1_P4
tasks. See the "README" portion of the patch for details.
-- Added new scontrol command "show hostlist <hostnames>" to translate a list
of hostnames into a hostlist expression (e.g. "tux1,tux2" -> "tux[1-2]")
and "show hostnames <list>", returns a list of of nodes (one node per line)
from SLURM hostlist expression or from SLURM_NODELIST environment variable
if no hostlist specified.
Christopher J. Morrone
committed
-- Add the sbatch option "--wrap".
-- Add the sbatch option "--get-user-env".
-- Added support for mpich-mx (use the mpichgm plugin).
Christopher J. Morrone
committed
-- Make job's stdout and stderr file access rights be based upon user's umask
at job submit time.
-- Add support for additional PMI functions: PMI_Parse_option,
PMI_Args_to_keyval, PMI_Free_keyvals and PMI_Get_options (from Puenlap Lee
and Nancy Kritkausky, Bull).
-- Make default value of SchedulerPort (configuration parameter) be 7321.
-- Use SLURM_UMASK environment variable (if set) at job submit time as umask
for spawned job.
-- Correct some format issues in the man pages (from Gennero Oliva, ICAR).
-- Added support for parallel make across an existing SLURM allocation
based upon GNU make-3.81. Patch is in "etc/make.slurm.patch".
-- Added '-b' option to sbatch for easy MOAB trasition to sbatch instead of
srun. Option does nothing in sbatch.
-- Changed wiki2's handling of a node state in Completing to return 'busy'
Loading
Loading full blame...