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 2.1.0-pre3
=============================
-- Removed sched/gang plugin and moved the logic directly into the slurmctld
daemon so that job preemption and gang scheduling can be used with the
sched/backfill plugin. Added configuration parameter:
PreemptMode=off|suspend|cancel|requeue|checkpoint
to enable/disable the job preemption logic (disabled by default).
(NOTE: There are some problems with memory management which could prevent a
job from starting when memory would be freed by a job being requeued or
otherwise removed, these are being worked on)
-- If the --partition option is used with the sinfo or squeue command then
print information about even hidden partitions.
-- Replaced misc cpu allocation members in job_info_t with select_job_res_t
which will only be populated when requested (show_flags & SHOW_DETAIL)
-- Added a --detail option to "scontrol show job" to display the cpu/mem
allocation info on a node-by-node basis.

Danny Auble
committed
-- Added logic to give correct request uid for individual steps that
were cancelled.
-- Created a spank_get_item() option (S_JOB_ALLOC_MEM) that conveys the memory
that the select/cons_res plugin has allocated to a job.
-- BLUEGENE - blocks in error state are now handled correctly in accounting.
-- Modify squeue to print job step information about a specific job ID using
the following syntax: "squeue -j <job_id> -s".
-- BLUEGENE - scontrol delete block and update block can now remove blocks
on dynamic laid out systems.
-- BLUEGENE - Vastly improve Dynamic layout mode algorithm.
-- Address some issues for SLURM support of Solaris.

Danny Auble
committed
-- Applied patch from Doug Parisek (Doug.Parisek@bull.com) for speeding up
start of sview by delaying to creation of tooltips until requested.
-- Changed GtkToolTips to GtkToolTip for newer versions of GTK.
-- Applied patch from Rod Schultz (Rod.Schultz@Bull.com) that eliminates
ambiguity in the documentation over use of the terms "CPU" and "socket".
-- Modified get_resource_arg_range() to return full min/max values when input
string is null. This fixes the srun -B option to function as documented.
* Changes in SLURM 2.1.0-pre2
=============================
-- Added support for smap to query off node name for display.
-- Slurmdbd modified to set user ID and group ID to SlurmUser if started as
user root.
-- Configuration parameter ControlMachine changed to accept multiple comma-
separated hostnames for support of some high-availability architectures.

Danny Auble
committed
-- ALTERED API CALL slurm_get_job_steps 0 has been changed to NO_VAL for both
job and step id to recieve all jobs/steps. Please make adjustments to
your code.
-- salloc's --wait=<secs> option deprecated by --immediate=<secs> option to
match the srun command.
-- Add new slurmctld list for node features with node bitmaps for simplified
scheduling logic.

Moe Jette
committed
-- Multiple features can be specified when creating a reservation. Use "&"
(AND) or "|" (OR) separators between the feature names.
-- Changed internal node name caching so that front-end mode would work with
multiple lines of node name definitions.
-- Add node state flag for power-up/configuring. Represented by "#" suffix
on the node state name (e.g. "ALLOCATED#") for command output.
-- Add CONFIGURING/CF job state flag for node power-up/configuring.
-- Modify job step cancel logic for scancel and srun (on reciept of SIGTERM
or three SIGINT) to immediately send SIGKILL to spawned tasks. Previous
logic would send SIGCONT, SIGTERM, wait KillWait seconds, SIGKILL.
-- Created a spank_get_item() option (S_JOB_ALLOC_CORES) that conveys the cpus
that the select/cons_res plugin has allocated to a job.
-- Improve sview performance (outrageously) on very large machines.
-- Add support for licenses in resource reservation.
-- BLUEGENE - Jobs waiting for a block to boot will now be in Configuring
state.
-- bit_fmt now does not return brackets surrounding any set of data.
* Changes in SLURM 2.1.0-pre1
=============================
-- Slurmd notifies slurmctld of node boot time to better clean up after node
reboots.
-- Slurmd sends node registration information repeatedly until successful
transmit.
-- Change job_state in job structure to dedicate 8-bits to state flags.
Added macros to get state information (IS_JOB_RUNNING(job_ptr), etc.)
-- Added macros to get node state information (IS_NODE_DOWN(node_ptr), etc).
-- Added support for Solaris. Patch from David Hoppner.
-- Rename "slurm-aix-federation-<version>.rpm" to just
"slurm-aix-<version>.rpm" (federation switch plugin may not be present).

Moe Jette
committed
-- Eliminated the redundant squeue output format and sort options of
"%o" and "%b". Use "%D" and "%S" formats respectively. Also eliminated

Moe Jette
committed
"%X" and "%Y" and "%Z" formats. Use "%z" instead.
-- Added mechanism for SPANK plugins to set environment variables for
Prolog, Epilog, PrologSLurmctld and EpilogSlurmctld programs using
the functions spank_get_job_env, spank_set_job_env, and
spank_unset_job_env. See "man spank" for more information.
-- Completed the work to begun in 2.0.0 to standardize on using '-Q' as the
--quiet flag for all the commands.

Danny Auble
committed
-- BLUEGENE - sinfo and sview now display correct cpu counts for partitions
-- Cleaned up the cons_res plugin. It now uses a ptr to a part_record
instead of having to do strcmp's to find the correct one.
-- Pushed most all the plugin specific info in src/common/node_select.c
into the respected plugin.
-- BLUEGENE - closed some corner cases where a block could had been removed
while a job was waiting for it to become ready because an underlying
part of the block was put into an error state.
-- Modify sbcast logic to prevent a user from moving files to nodes they
have not been allocated (this would be possible in previous versions
only by hacking the sbcast code).
-- Add contribs/sjstat script (Perl tool to report job state information).
Put into new RPM: sjstat.
-- Add sched/wiki2 (Moab) JOBMODIFY command support for VARIABLELIST option
to set supplemental environment variables for pending batch jobs.
-- BLUEGENE - add support for scontrol show blocks.
-- Added support for job step time limits.
* Changes in SLURM 2.0.5
========================
-- BLUEGENE - Added support for emulating systems with a X-dimension of 4.
-- BLUEGENE - When a nodecard goes down on a non-Dynamic system SLURM will
now only drain blocks under 1 midplane, if no such block exists then SLURM
will drain the entire midplane and not mark any block in error state.
Previously SLURM would drain every overlapping block of the nodecard
making it possible for a large block to make other blocks not work since
they overlap some other part of the block that really isn't bad.
-- BLUEGENE - Handle L3 errors on boot better.
-- Don't revoke a pending batch launch request from the slurmctld if the
job is immediately suspended (a normal event with gang scheduling).
-- BLUEGENE - Fixed issue with restart of slurmctld would allow error block
nodes to be considered for building new blocks when testing if a job would
run. This is a visual bug only, jobs would never run on new block, but
the block would appear in slurm tools.
-- Better responsiveness when starting new allocations when running with the
slurmdbd.
-- Fixed race condition when reconfiguring the slurmctld and using the
consumable resources plugin which would cause the controller to core.

David J. Bremer
committed
-- Fixed race condition that sometimes caused jobs to stay in completing
state longer than necessary after being terminated.
-- Fixed issue where if a parent account has a qos added and then a child
account has the qos removed the users still get the qos.
-- BLUEGENE - New blocks in dynamic mode will only be made in the system
when the block is actually needed for a job, not when testing.
-- BLUEGENE - Don't remove larger block used for small block until job starts.
-- Add new squeue output format and sort option of "%L" to print a job's time
left (time limit minus time used).
* Changes in SLURM 2.0.4
========================
-- Permit node suspend/resume logic to be enabled through "scontrol reconfig"
given appropriate changes to slurm configuration file.
-- Check for return codes on functions with warn_unused_result set.
-- Fix memory leak in getting step information (as used by squeue -s).
-- Better logging for when job's request bad output file locations.
-- Fix issue where if user specified non-existant file to write to slurmstepd
will regain privileges before sending batch script ended to the controller.
-- Fix bug when using the priority_multifactor plugin with no associations
yet.
-- BLUEGENE - we no longer check for the images to sync state. This was
needed long ago when rebooting blocks wasn't a possibility and should
had been removed when that functionality was available.
-- Added message about no connection with the database for sacctmgr.
-- On BlueGene, let srun or salloc exit on SIGINT if slurmctld dies while
booting its block.
-- In select/cons_res fix bug that could result in invalid memory pointer
if node configurations in slurm.conf contains 8 or more distinct
socket/core/thread counts.
-- Modify select/cons_res to recognize updated memory size upon node startup
if FastSchedule=0.
-- Fixed bug if not enforcing associations, but running with them and the
priority/multifactor, the slurmctld will not core dump on processing usage.
-- QOS will not be reset to the default when added back a previously deleted
association.
-- Do not set a job's virtual memory limit based upon the job's specified
memory limit (which should be a real memory limit, not virtual).
-- In sched/wiki2 - Strip single and double quotes out of a node's reason
string to avoid confusing Moab's parser.
-- BLUEGENE - fix for sinfo/sview for displaying proper node count for nodes
in draining state.
-- Fix for sview when viewing a certain part of a group (like 1 job) so it
doesn't core when the part is gone.
-- BLUEGENE - Changed order of SYNC's to be on the front of the list to
avoid having a job terminated with a TERM before the SYNC of the
job happens.
-- Validate configured PluginDir value is a valid directory before trying to
use it.
-- Fix to resolve agent_queue_request symbol from some checkpoint plugins.
-- Fix possible execve error for sbatch script read from stdin.
-- Modify handling of user ID/name and group ID/name in the slurm.conf file
to properly handle user names that contain all digits. Return error code
from uid_from_string() and gid_from_string() functions rather than a uid of
-1, which might be a valid uid or gid on some systems.
-- Fix in re-calcuation of job priorities due to DOWN or DRAINED nodes.
* Changes in SLURM 2.0.3
========================
-- Add reservation creation/update flag of Ignore_Jobs to enable the creation
of a reservation that overlaps jobs expected to still be running when
the reservation starts. This would be especially useful to reserve all
nodes for system maintenence without adjusting time limits of running
jobs before creating the reservation. Without this flag, nodes allocated
jobs expected to running when the reservation begins can not be placed
into a reservation.
-- In task/affinity plugin, add layer of abstraction to logic translating
block masks to physical machine masks. Patch from Matthieu Hautreux, CEA.
-- Fix for setting the node_bitmap in a job to NULL if the job does not
start correctly when expected to start.
-- Fixed bug in srun --pty logic. Output from the task was split up
arbitrarily into stdout and stderr streams, and sometimes was printed
out of order.
-- If job requests minimum and maximum node count range with select/cons_res,
try to satisfy the higher value (formerly only allocated the minimum).
-- Fix for checking for a non-existant job when querying steps.
-- For job steps with the --exclusive option, base initial wait time in
partly upon the process ID for better performance with many job steps
started at the same time. Maintain exponential back-off as needed.
-- Fix for correct step ordering in sview.
-- Support optional argument to srun and salloc --immediate option. Specify
timeout value in seconds for job or step to be allocated resources.
* Changes in SLURM 2.0.2
========================
-- Fix, don't remove job details when a job is cancelled while pending.
-- Do correct type for mktime so garbage isn't returned on 64bit systems
for accounting archival.
-- Better checking in sacctmgr to avoid infinite loops.
-- Fix minor memory leak in fake_slurm_step_layout_create()
-- Fix node weight (scheduling priority) calculation for powered down
nodes. Patch from Hongjia Cao, NUDT.
-- Fix node suspend/resume rate calculations. Patch from Hongjia Cao, NUDT.
-- Change calculations using ResumeRate and SuspendRate to provide higher
resolution.
-- Log the IP address for incoming messages having an invalid protocol
version number.
-- Fix for sacct to show jobs that start the same second as the sacct
command is issued.
-- BLUEGENE - Fix for -n option to work on correct cpu counts for each
midplane instead of treating -n as a c-node count.
-- salloc now sets SLURM_NTASKS_PER_NODE if --ntasks-per-node option is set.
-- Fix select/linear to properly set a job's count of allocated processors
(all processors on the allocated nodes).
-- Fix select/cons_res to allocate proper CPU count when --ntasks-per-node
option is used without a task count in the job request.
-- Insure that no node is allocated to a job for which the CPU count is less
than --ntasks-per-node * --cpus-per-task.
-- Correct AllocProcs reported by "scontrol show node" when ThreadsPerCore
is greater than 1 and select/cons_res is used.
-- Fix scontrol show config for accounting information when values are
not set in the slurm.conf.
-- Added a set of SBATCH_CPU_BIND* and SBATCH_MEM_BIND* env variables to keep
jobsteps launched from within a batch script from inheriting the CPU and
memory affinity that was applied to the batch script. Patch from Matthieu
Hautreux, CEA.
-- Ignore the extra processors on a node above configured size if either
sched/gang or select/cons_res is configured.
-- Fix bug in tracking memory allocated on a node for select/cons_res plugin.
-- Fixed a race condition when writing labelled output with a file per task
or per node, which potentially closed a file before all data was written.
-- BLUEGENE - Fix, for if a job comes in spanning both less than and
over 1 midplane in size we check the connection type appropriately.
-- Make sched/backfill properly schedule jobs with constraints having node
counts. NOTE: Backfill of jobs with constraings having exclusive OR
operators are not fully supported.
-- If srun is cancelled by SIGINT, set the job state to cancelled, not
failed.
-- BLUEGENE - Fix, for if you are setting an subbp into an error mode
where the subbp stated isn't the first ionode in a nodecard.
-- Fix for backfill to not core when checking shared nodes.
-- Fix for scontrol to not core when hitting just return in interactive mode.
-- Improve sched/backfill logic with respect to shared nodes (multiple jobs
per node).
-- In sched/wiki (Maui interface) add job info fields QOS, RCLASS, DMEM and
TASKSPERNODE. Patch from Bjorn-Helge Mevik, University of Oslo.
* Changes in SLURM 2.0.1
========================
-- Fix, truncate time of start and end for job steps in sacct.
-- Initialize all messages to slurmdbd. Previously uninitialized string could
cause slurmctld to fail with invalid memory reference.
-- BLUEGENE - Fix, for when trying to finish a torus on a block already
visited. Even though this may be possible electrically this isn't valid
in the under lying infrastructure.
-- Fix, in mysql plugi
Loading
Loading full blame...