burst_buffer.conf - Slurm configuration file for burst buffer management.
burst_buffer.conf is an ASCII file which describes the configuration of
burst buffer resource management. This file is only required on the head
node(s), where the slurmctld daemon executes. The file location can be
modified at system build time using the DEFAULT_SLURM_CONF parameter or at
execution time by setting the SLURM_CONF environment variable. The file will
always be located in the same directory as the
slurm.conf file. In
order to support multiple configuration files for multiple burst buffer
plugins, the configuration file may alternately be given a name containing the
plugin name. For example, if "burst_buffer.conf" is not found, the
burst_buffer/generic configuration could be read from a file named
"burst_buffer_generic.conf".
Parameter names are case insensitive. Any text following a "#" in the
configuration file is treated as a comment through the end of that line.
Changes to the configuration file take effect upon restart of Slurm daemons,
daemon receipt of the SIGHUP signal, or execution of the command
"scontrol reconfigure".
The configuration parameters available include:
- AllowUsers
- Comma separated list of user names and/or IDs permitted to use burst
buffers. The options AllowUsers and DenyUsers can not both
be specified. By default all users are permitted to use burst buffers.
- CreateBuffer
- Fully qualified path name of a program which will create both persistent
and per-job burst buffers. This option is not used by the
burst_buffer/cray plugin.
- DefaultPool
- Name of the pool used by default for resource allocations. The default
value is the first pool reported by the burst buffer infrastructure. This
option is only used by the burst_buffer/cray plugin.
- DenyUsers
- Colon delimited list of user names and/or IDs prevented from using burst
buffers. The options AllowUsers and DenyUsers can not both
be specified. By default all users are permitted to use burst buffers.
- DestroyBuffer
- Fully qualified path name of a program which will destroy both persistent
and per-job burst buffers. This option is not used by the
burst_buffer/cray plugin.
- Flags
- String used to control various functions. Multiple options may be comma
separated. Supported options include:
- DisablePersistent
- Prevents regular users from being able to create and destroy persistent
burst buffers. This is the default behaviour, only privileged users (Slurm
operators and administrators) can create or destroy persistent burst
buffers.
- EmulateCray
- Emulating a Cray DataWarp system using the dw_wlm_cli script in the
burst_buffer/cray plugin.
- EnablePersistent
- Enables regular users to create and destroy persistent burst buffers. By
default, only privileged users (Slurm operators and administrators) can
create or destroy persistent burst buffers.
- PrivateData
- If set, then only Slurm operators and the burst buffer owner can see burst
buffer data.
- TeardownFailure
- If set, then teardown a burst buffer after file staging error. Otherwise
preserve the burst buffer for analysis and manual teardown.
- GetSysState
- Fully qualified path name of a program which will return the current burst
buffer state. See the
src/plugins/burst_buffer/generic/bb_get_state.example in the Slurm
distribution for an example. For the Cray plugin, this should be the path
of the dw_wlm_cli command and it's default value is
/opt/cray/dw_wlm/default/bin/dw_wlm_cli.
- Granularity
- Granularity of job space allocations in units of bytes. The numeric value
may have a suffix of "m" (megabytes), "g" (gigabytes),
"t" (terabytes), "p" (petabytes), or "n"
(nodes). Bytes is assumed if no suffix is supplied. This option is not
used by the burst_buffer/cray plugin.
- OtherTimeout
- If a burst buffer operation (other than job validation, stage in, or stage
out) runs for longer than this number of seconds, the job will be placed
in a held state. A Slurm administrator will be required to release the
job. By default there is a 300 second (5 minute) timeout for these
operations. Also see StageInTimeout, StageOutTimeout, and
ValidateTimeout options.
- PrivateData
- If set to "true" then users will only be able to view burst
buffers they can use. Slurm administrators will still be able to view all
burst buffers. By default, users can view all burst buffers.
- StageInTimeout
- If the stage in of files for a job takes more than this number of seconds,
the burst buffer will be released and the job will be placed in a held
state. A Slurm administrator will be required to release the job. By
default there is a one day timeout for the stage in process.
- StageOutTimeout
- If the stage out of files for a job takes more than this number of
seconds, the burst buffer will be released and the job will be purged. By
default there is a one day timeout for the stage out process.
- StartStageIn
- Fully qualified path name of a program which will stage files in for a
job. See the src/plugins/burst_buffer/generic/bb_start_stage_in.example in
the Slurm distribution for an example. This option is not used by the
burst_buffer/cray plugin.
- StartStageOut
- Fully qualified path name of a program which will stage files out for a
job. See the src/plugins/burst_buffer/generic/bb_start_stage_out.example
in the Slurm distribution for an example. This option is not used by the
burst_buffer/cray plugin.
- StopStageIn
- Fully qualified path name of a program which will stop staging files in
for a job. See the
src/plugins/burst_buffer/generic/bb_stop_stage_out.example in the Slurm
distribution for an example. This option is not used by the
burst_buffer/cray plugin.
- StopStageOut
- Fully qualified path name of a program which will stop staging files in
for a job. See the
src/plugins/burst_buffer/generic/bb_stop_stage_out.example in the Slurm
distribution for an example. This option is not used by the
burst_buffer/cray plugin.
- ValidateTimeout
- If the validation of a job submission request takes more than this number
of seconds, the submission will be rejected. The value of
ValidateTimeout must be less than the value of
MessageTimeout configured in the slurm.conf file or job submission
requests may fail with a response timeout error. By default there is a 5
second timeout for the validation operations. (NOTE: This option was added
after the release of Slurm version 15.08 and will its not be visible to
users with Slurm tools until the version 16.05 release.)
##################################################################
# Slurm's burst buffer configuration file (burst_buffer.conf)
##################################################################
AllowUsers=alan,brenda
PrivateData=true
#
Granularity=1G
#
StageInTimeout=30 # Seconds
StageOutTimeout=30 # Seconds
#
CreateBuffer=/usr/local/slurm/15.08/sbin/CB
DestroyBuffer=/usr/local/slurm/15.08/sbin/DB
GetSysState=/usr/local/slurm/15.08/sbin/GSS
StartStageIn=/usr/local/slurm/15.08/sbin/SSI
StartStageOut=/usr/local/slurm/15.08/sbin/SSO
StopStageIn=/usr/local/slurm/15.08/sbin/PSI
StopStageOut=/usr/local/slurm/15.08/sbin/PSO
Copyright (C) 2014-2015 SchedMD LLC.
This file is part of Slurm, a resource management program. For details, see
<https://slurm.schedmd.com/>.
Slurm is free software; you can redistribute it and/or modify it under the terms
of the GNU General Public License as published by the Free Software
Foundation; either version 2 of the License, or (at your option) any later
version.
Slurm is distributed in the hope that it will be useful, but WITHOUT ANY
WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS FOR
A PARTICULAR PURPOSE. See the GNU General Public License for more details.
slurm.conf(5)