Job Submission Lua Plugin

The Job Submit Plugin runs on the head node. It executes after a user's job request is submitted to the head node and before the job is actually queued.

Purpose

The Job Submit Plugin can do the following:

• Modify job submission information, such as the job's partition, cores, and memory amount.
• Check job information to determine whether it is allowed to run, and so on.

Installation and Configuration

Add the following to slurm.conf:

JobSubmitPlugins=job_submit/lua

Save the Lua script to /etc/slurm/job_submit.lua.

Supported Functions

-- Executed on job submission
function slurm_job_submit(job_desc, part_list, submit_uid)
end

-- Executed on job modification
function slurm_job_modify(job_desc, job_rec, part_list, modify_uid)
end

Parameters:

• job_desc: job description information
• part_list: all partition information
• submit_uid: UID of the submitting user
• job_rec: current job record (job_modify only)
• modify_uid: UID of the user performing the modification (job_modify only)

Data Structures

Job Descriptor (job_desc)

Field	Type	Description
account	string	Job account name
admin_comment	string	Admin comment
array_task_cnt	number	Number of array tasks
batch_features	string	Job batch features
burst_buffer	string	Job burst buffer
comment	string	Job comment
cpus_per_tres	string	CPUs per TRES (resource manager)
delay_boot	number	Delay boot time
direct_set_prio	number	Directly set priority
features	string	Job features
gres	string	Requested GRES (generic resources)
group_id	number	Job group ID
job_id	number	Job ID
job_state	number	Job state
licenses	string	Job licenses
max_cpus	number	Maximum CPUs available to the job
max_nodes	number	Maximum nodes available to the job
mem_per_tres	string	Memory allocated per TRES
min_cpus	number	Minimum CPUs requested by the job
min_mem_per_node	number	Minimum memory requirement per node
min_mem_per_cpu	number	Minimum memory requirement per CPU
min_nodes	number	Minimum nodes requested by the job
name	string	Job name
nice	number	Job priority offset (scheduling)
pack_job_id	number	Pack job ID
pack_job_id_set	string	Pack job ID set
pack_job_offset	number	Pack job offset
partition	string	Job partition; note this is the requested value and may be empty
pn_min_cpus	number	Minimum CPUs per node
pn_min_memory	number	Minimum memory per node (64-bit integer)
priority	number	Job priority
qos	string	Job Quality of Service (QoS)
reboot	number	Job reboot setting
req_switch	number	Job requested switch information
site_factor	number	Site factor, usually for scheduling policies
spank_job_env	table	SPANK (Slurm plugin) job environment variables
spank_job_env_size	number	Size of SPANK job environment variables
time_limit	number	Job time limit
time_min	number	Minimum job time limit
tres_bind	string	TRES binding information
tres_freq	string	TRES frequency information
tres_per_job	string	TRES per job
tres_per_node	string	TRES per node
tres_per_socket	string	TRES per socket
tres_per_task	string	TRES per task
user_id	number	User ID
user_name	string	User name
wait4switch	number	Job wait-for-switch status
wckey	string	Job workload key (workload key)

Partition (part_list)

Field	Type	Description
allow_accounts	string	Allowed account names
allow_alloc_nodes	string	Allowed node names
allow_groups	string	Allowed group names
allow_qos	string	Allowed QoS settings
alternate	string	Alternate partition name
billing_weights_str	string	Partition billing weight string
default_time	number	Default time limit (minutes)
def_mem_per_cpu	number	Default memory per CPU (if allocated by CPU)
def_mem_per_node	number	Default memory per node
deny_accounts	string	Denied account names
deny_qos	string	Denied QoS settings
flag_default	number	Whether it is the default partition (0: no, 1: yes)
flags	number	Partition flags field (e.g., enabled)
max_cpus_per_node	number	Maximum CPUs allowed per node
max_mem_per_cpu	number	Maximum memory per CPU
max_mem_per_node	number	Maximum memory per node
max_nodes	number	Maximum nodes
max_nodes_orig	number	Original maximum nodes
max_share	number	Maximum sharing ratio
max_time	number	Maximum time limit (minutes)
min_nodes	number	Minimum nodes
min_nodes_orig	number	Original minimum nodes
name	string	Partition name
nodes	string	Node list
priority_job_factor	number	Job priority factor
priority_tier	number	Priority tier
qos	string	QoS settings
state_up	number	Whether the partition is active (0: inactive, 1: active)

Global Object slurm

Log Functions

Field/Function	Description
log_error	Log error messages
log_info	Log informational messages (level 0)
log_verbose	Log verbose messages (level 1)
log_debug	Log debug messages (level 2)
log_debug2	Log more detailed debug messages (level 3)
log_debug3	Log even more detailed debug messages (level 4)
log_debug4	Log the most detailed debug messages (level 5)
log_user	Log user messages

Error Codes

Field/Function	Description
ERROR	Indicates a general error
FAILURE	Indicates operation failure
SUCCESS	Indicates operation success
ESLURM_ACCESS_DENIED	Access denied
ESLURM_ACCOUNTING_POLICY	Accounting policy error
ESLURM_INVALID_ACCOUNT	Invalid account
ESLURM_INVALID_LICENSES	Invalid licenses
ESLURM_INVALID_NODE_COUNT	Invalid node count
ESLURM_INVALID_TIME_LIMIT	Invalid time limit
ESLURM_JOB_MISSING_SIZE_SPECIFICATION	Job missing size specification
ESLURM_MISSING_TIME_LIMIT	Missing time limit

Other Definitions

Field/Function	Description
ALLOC_SID_ADMIN_HOLD	Allocation ID reserved by admin
ALLOC_SID_USER_HOLD	Allocation ID reserved by user
INFINITE	Infinite value
INFINITE64	64-bit infinite value
MAIL_JOB_BEGIN	Send mail when job begins
MAIL_JOB_END	Send mail when job ends
MAIL_JOB_FAIL	Send mail when job fails
MAIL_JOB_REQUEUE	Send mail when job requeues
MAIL_JOB_TIME100	Send mail when job reaches 100% time
MAIL_JOB_TIME90	Send mail when job reaches 90% time
MAIL_JOB_TIME80	Send mail when job reaches 80% time
MAIL_JOB_TIME50	Send mail when job reaches 50% time
MAIL_JOB_STAGE_OUT	Send mail when job stage ends
MEM_PER_CPU	Memory per CPU
NICE_OFFSET	Priority offset
JOB_SHARED_NONE	Job not shared
JOB_SHARED_OK	Job shared
JOB_SHARED_USER	User shared job
JOB_SHARED_MCS	MCS shared job
NO_VAL64	64-bit invalid value
NO_VAL	Invalid value
NO_VAL16	16-bit invalid value
NO_VAL8	8-bit invalid value
SHARED_FORCE	Force sharing

Job Descriptor Bit Flags

Field/Function	Description
GRES_DISABLE_BIND	Disable resource binding
GRES_ENFORCE_BIND	Enforce resource binding
KILL_INV_DEP	Kill invalid dependencies
NO_KILL_INV_DEP	Do not kill invalid dependencies
SPREAD_JOB	Spread job
USE_MIN_NODES	Use minimum nodes