sacctmgr

Overview

sacctmgr is the command tool in Slurm for managing accounting data:

• Manage Accounts, Users, resource limits (QOS), etc.
• Configure associations and permissions
• Requires administrator privileges

Common Options

Option	Description	Example
add	Add entity	sacctmgr add account
modify	Modify entity	sacctmgr modify user
delete	Delete entity	sacctmgr delete qos
show	Show info	sacctmgr show cluster
-i, --immediate	Take effect immediately	sacctmgr -i add ...
-p, --parsable	Parsable output	sacctmgr -p show ...
-Q, --quiet	Quiet mode	sacctmgr -Q delete ...

Examples

Manage accounts

# Add account
sacctmgr add account myaccount Description="test account" Organization=CS

# Modify account
sacctmgr modify account myaccount set GrpTRES=cpu=100

# Delete account
sacctmgr delete account myaccount

Manage users

# Add user
sacctmgr add user user1 Account=myaccount

# Set default account for user
sacctmgr modify user user1 set DefaultAccount=myaccount

# Delete user
sacctmgr delete user user1

Manage QOS

# Add QoS
sacctmgr add qos high Priority=100

# Add qos: limit high to 10 CPUs, 10 RTX2080 GPUs, and 4 V100 GPUs
sacctmgr add qos high set GrpTRES=cpu=10,gres/gpu:rtx2080=10,gres/gpu:v100=4 

# Limit high to 10 CPUs, 10 RTX2080 GPUs, and 4 V100 GPUs
sacctmgr modify qos high set maxtresperuser=cpu=1,gres/gpu:rtx2080=1,gres/gpu:v100=1

# Modify QoS limits
sacctmgr modify qos high set MaxTRESPerJob=cpu=32

# Clear limits; -1 means unlimited
sacctmgr modify qos high set GrpTRES=cpu=-1,gres/gpu:rtx2080=-1,gres/gpu:v100=-1

# Modify group limits
sacctmgr modify qos high set GrpTRES=cpu=-1,gres/gpu:rtx2080=-1,gres/gpu:v100=-1

# Delete QoS
sacctmgr delete qos high

# Add user to qos
sacctmgr modify user <user_name> set qos=<qos_name>

View associations

sacctmgr show Association  format=Cluster,Account,User,Partition,QOS,GrpSubmit,GrpWall,GrpTRESMins,MaxJobs,MaxTRES format=GrpTRES%60

View user details

sacctmgr show user user1 format=User,DefaultAccount,AdminLevel

View QOS configuration

sacctmgr show qos format=Name,Priority,MaxTRESPerJob

Key Parameter Details

Resource Types (TRES)

Type	Description	Example
cpu	CPU cores	cpu=100
mem	Memory (MB)	mem=100G
gres/gpu	GPU count	gres/gpu=2
billing	Billing weight	billing=30

Common limits

Limit parameter	Description
GrpTRES	Group total resource limit
MaxTRESPerUser	Per-user resource limit
MaxTRESPerJob	Per-job resource limit
MaxJobs	Max jobs limit

Notes

• All modifications take effect immediately and are irreversible

• Deleting a parent account will delete all child accounts

• Resource limits are based on real-time calculation and may affect queued jobs

• For complex configurations, run simulation tests first

• Back up configuration before production operations:

   sacctmgr -p dump <cluster_name> file=./<file_name>.cfg

📌 Tip: For more advanced features (such as federated cluster configuration), see man sacctmgr or visit Slurm official documentation