Slurm QoS 使用
草稿: 本文因某种原因尚未公开。
专栏文章
Slurm QoS 介绍
QoS 优先级
如果在 Partition 的 QOS 和 Job 的 QOS 中都设置了限制,Partition 的 QOS 优先,除非 Job 的 QOS 设置了 OverPartQOS 标志时,Job 的 QOS 优先。优先级如下:
- Partition QOS limit
- Job QOS limit
- User association
- Account association(s), ascending the hierarchy
- Root/Cluster association
- Partition limit
- None
注意:如果在这个层次结构中有多个点定义了限制,那么列表中第一个定义限制将生效。示例:
MaxJobs=20,MaxSubmitJobs在Partition QOS中未定义Job QOS和中没有设置限制MaxJobs=4和MaxSubmitJobs=50在User association中定义
生效的限制为:MaxJobs=20 和 MaxSubmitJobs=50
特例:
- 配置
PartitionTimeLimit如果设置了,使用此QOS的作业将能够覆盖请求分区的 TimeLimit(如MaxWallDurationPerJob)。
示例
account 指定 qos 的限制
# 创建 account
[root@sms1 ~]# sacctmgr add account xiexianbin Description="xiexianbin" Organization=xiexianbin
Adding Account(s)
xiexianbin
Settings
Description = xiexianbin
Organization = xiexianbin
Associations
A = xiexianbin C = poc
Settings
Parent = root
Would you like to commit changes? (You have 30 seconds to decide)
(N/y): y
# 创建 qos
[root@sms1 ~]# sacctmgr add qos xiexianbin_qos
Adding QOS(s)
xiexianbin_qos
Settings
Description = xiexianbin_qos
Would you like to commit changes? (You have 30 seconds to decide)
(N/y): y
# 更新 QoS 配置
[root@sms1 ~]# sacctmgr modify qos xiexianbin_qos set MaxJobsPerUser=256
Modified qos...
xiexianbin_qos
Would you like to commit changes? (You have 30 seconds to decide)
(N/y): y
[root@sms1 ~]# sacctmgr modify qos xiexianbin_qos set MaxTRESPerUser=CPU=128
Modified qos...
xiexianbin_qos
Would you like to commit changes? (You have 30 seconds to decide)
(N/y): y
[root@sms1 ~]# sacctmgr modify qos xiexianbin_qos set MaxJobsPU=1024
Modified qos...
xiexianbin_qos
Would you like to commit changes? (You have 30 seconds to decide)
(N/y): y
[root@sms1 ~]# sacctmgr modify qos xiexianbin_qos set MaxSubmitJobsPerUser=2
Modified qos...
xiexianbin_qos
Would you like to commit changes? (You have 30 seconds to decide)
(N/y): y
# 查看 QoS
[root@sms1 ~]# sacctmgr show qos xiexianbin_qos
Name Priority GraceTime Preempt PreemptExemptTime PreemptMode Flags UsageThres UsageFactor GrpTRES GrpTRESMins GrpTRESRunMin GrpJobs GrpSubmit GrpWall MaxTRES MaxTRESPerNode MaxTRESMins MaxWall MaxTRESPU MaxJobsPU MaxSubmitPU MaxTRESPA MaxJobsPA MaxSubmitPA MinTRES
---------- ---------- ---------- ---------- ------------------- ----------- ---------------------------------------- ---------- ----------- ------------- ------------- ------------- ------- --------- ----------- ------------- -------------- ------------- ----------- ------------- --------- ----------- ------------- --------- ----------- -------------
xiexianbi+ 0 00:00:00 cluster 1.000000 cpu=128 1024 2
# 绑定 user 和 account
[root@sms1 ~]# sacctmgr add user xiexianbin Account=xiexianbin QOS=xiexianbin_qos Cluster=poc
Adding User(s)
xiexianbin
Associations =
U = xiexianbi A = xiexianbin C = poc
Non Default Settings
QOS = xiexianbin_qos
Would you like to commit changes? (You have 30 seconds to decide)
(N/y): y# 更新 user 的限制
sacctmgr modify user xiexianbin set MaxJobs=2
sacctmgr modify user xiexianbin set MaxSubmitJobs=2
sacctmgr modify user xiexianbin set MaxCPUs=1
# 查看信息
[root@sms1 ~]# sacctmgr show user
User Def Acct Admin
---------- ---------- ---------
root root Administ+
xiexianbin xiexianbin None
[root@sms1 ~]# sacctmgr show account
Account Descr Org
---------- -------------------- --------------------
root default root account root
xiexianbin xiexianbin xiexianbin
[root@sms1 ~]# sacctmgr show Association
Cluster Account User Partition Share Priority GrpJobs GrpTRES GrpSubmit GrpWall GrpTRESMins MaxJobs MaxTRES MaxTRESPerNode MaxSubmit MaxWall MaxTRESMins QOS Def QOS GrpTRESRunMin
---------- ---------- ---------- ---------- --------- ---------- ------- ------------- --------- ----------- ------------- ------- ------------- -------------- --------- ----------- ------------- -------------------- --------- -------------
poc root 1 normal
poc root root 1 normal
poc xiexianbin 1 normal
poc xiexianbin xiexianbin 1 xiexianbin_qos参考
- 上一页: Slurm FaQ
- 下一页: Linux procfs 介绍
专栏文章
- Slurm 开发
- Slurm rpmbuild
- Slurm 相关术语
- Slurm QoS 使用(当前)
- Slurm Job
- Slurm Partition 使用
- Slurm REST API 使用
- Slurm scontrol 命令
- Slurm sacct/sacctmgr 命令
- Slurm 命令使用说明
- Slurm sreport 命令
- Slurm FaQ
- Slurm 安装部署
最近更新
最新评论