SLURM - Simple Linux Utility for Resource Management

Introduction

Slurm is an open source, fault-tolerant, and highly scalable cluster management and job scheduling system for large and small Linux clusters.

It provides three key functions:

  • allocating exclusive and/or non-exclusive access to resources (computer nodes) to users for some duration of time so they can perform work,
  • providing a framework for starting, executing, and monitoring work (typically a parallel job such as MPI) on a set of allocated nodes, and
  • arbitrating contention for resources by managing a queue of pending jobs.

Installation

Controller name: slurm-ctrl

Install slurm-wlm and tools

ssh slurm-ctrl
apt install slurm-wlm slurm-wlm-doc mailutils mariadb-client mariadb-server libmariadb-dev python-dev python-mysqldb

Install Maria DB Server

apt-get install mariadb-server
systemctl start mysql
mysql -u root
create database slurm_acct_db;
create user 'slurm'@'localhost';
set password for 'slurm'@'localhost' = password('slurmdbpass');
grant usage on *.* to 'slurm'@'localhost';
grant all privileges on slurm_acct_db.* to 'slurm'@'localhost';
flush privileges;
exit

In the file /etc/mysql/mariadb.conf.d/50-server.cnf we should have the following setting:

vi /etc/mysql/mariadb.conf.d/50-server.cnf
bind-address = localhost

Node Authentication

First, let us configure the default options for the munge service:

vi /etc/default/munge
OPTIONS="--syslog --key-file /etc/munge/munge.key"

Central Controller

The main configuration file is /etc/slurm-llnl/slurm.conf this file has to be present in the controller and *ALL* of the compute nodes and it also has to be consistent between all of them.

vi /etc/slurm-llnl/slurm.conf
###############################
# /etc/slurm-llnl/slurm.conf
###############################
# slurm.conf file generated by configurator easy.html.
# Put this file on all nodes of your cluster.
# See the slurm.conf man page for more information.
#
ControlMachine=slurm-ctrl
#ControlAddr=10.7.20.97
#
#MailProg=/bin/mail
MpiDefault=none
#MpiParams=ports=#-#
ProctrackType=proctrack/pgid
ReturnToService=1
SlurmctldPidFile=/var/run/slurm-llnl/slurmctld.pid
##SlurmctldPidFile=/var/run/slurmctld.pid
#SlurmctldPort=6817
SlurmdPidFile=/var/run/slurm-llnl/slurmd.pid
##SlurmdPidFile=/var/run/slurmd.pid
#SlurmdPort=6818
SlurmdSpoolDir=/var/spool/slurmd
SlurmUser=slurm
#SlurmdUser=root
StateSaveLocation=/var/spool
SwitchType=switch/none
TaskPlugin=task/none
#
#
# TIMERS
#KillWait=30
#MinJobAge=300
#SlurmctldTimeout=120
#SlurmdTimeout=300
#
#
# SCHEDULING
FastSchedule=1
SchedulerType=sched/backfill
SelectType=select/linear
#SelectTypeParameters=
#
#
# LOGGING AND ACCOUNTING
AccountingStorageType=accounting_storage/none
ClusterName=cluster
#JobAcctGatherFrequency=30
JobAcctGatherType=jobacct_gather/none
#SlurmctldDebug=3
SlurmctldLogFile=/var/log/slurm-llnl/SlurmctldLogFile
#SlurmdDebug=3
SlurmdLogFile=/var/log/slurm-llnl/SlurmLogFile
#
#
# COMPUTE NODES
NodeName=linux1 NodeAddr=10.7.20.98 CPUs=1 State=UNKNOWN

Copy slurm.conf to compute nodes!

root@slurm-ctrl# scp /etc/slurm-llnl/slurm.conf csadmin@10.7.20.109:/tmp/.; scp /etc/slurm-llnl/slurm.conf csadmin@10.7.20.110:/tmp/.
vi /lib/systemd/system/slurmctld.service
[Unit]
Description=Slurm controller daemon
After=network.target munge.service
ConditionPathExists=/etc/slurm-llnl/slurm.conf
Documentation=man:slurmctld(8)

[Service]
Type=forking
EnvironmentFile=-/etc/default/slurmctld
ExecStart=/usr/sbin/slurmctld $SLURMCTLD_OPTIONS
ExecStartPost=/bin/sleep 2
ExecReload=/bin/kill -HUP $MAINPID
PIDFile=/var/run/slurm-llnl/slurmctld.pid

[Install]
WantedBy=multi-user.target
vi /lib/systemd/system/slurmd.service
[Unit]
Description=Slurm node daemon
After=network.target munge.service
ConditionPathExists=/etc/slurm-llnl/slurm.conf
Documentation=man:slurmd(8)

[Service]
Type=forking
EnvironmentFile=-/etc/default/slurmd
ExecStart=/usr/sbin/slurmd $SLURMD_OPTIONS
ExecStartPost=/bin/sleep 2
ExecReload=/bin/kill -HUP $MAINPID
PIDFile=/var/run/slurm-llnl/slurmd.pid
KillMode=process
LimitNOFILE=51200
LimitMEMLOCK=infinity
LimitSTACK=infinity

[Install]
WantedBy=multi-user.target
root@slurm-ctrl# systemctl daemon-reload
root@slurm-ctrl# systemctl enable slurmdbd
root@slurm-ctrl# systemctl start slurmdbd
root@slurm-ctrl# systemctl enable slurmctld
root@slurm-ctrl# systemctl start slurmctld

Accounting Storage

After we have the slurm-llnl-slurmdbd package installed we configure it, by editing the /etc/slurm-llnl/slurmdbd.conf file:

vi /etc/slurm-llnl/slurmdbd.conf
########################################################################
#
# /etc/slurm-llnl/slurmdbd.conf is an ASCII file which describes Slurm
# Database Daemon (SlurmDBD) configuration information.
# The contents of the file are case insensitive except for the names of
# nodes and files. Any text following a "#" in the configuration file is
# treated as a comment through the end of that line. The size of each
# line in the file is limited to 1024 characters. Changes to the
# configuration file take effect upon restart of SlurmDbd or daemon
# receipt of the SIGHUP signal unless otherwise noted.
#
# This file should be only on the computer where SlurmDBD executes and
# should only be readable by the user which executes SlurmDBD (e.g.
# "slurm"). This file should be protected from unauthorized access since
# it contains a database password.
#########################################################################
AuthType=auth/munge
AuthInfo=/var/run/munge/munge.socket.2
StorageHost=localhost
StoragePort=3306
StorageUser=slurm
StoragePass=slurmdbpass
StorageType=accounting_storage/mysql
StorageLoc=slurm_acct_db
LogFile=/var/log/slurm-llnl/slurmdbd.log
PidFile=/var/run/slurm-llnl/slurmdbd.pid
SlurmUser=slurm
root@slurm-ctrl# systemctl start slurmdbd

Authentication

Copy /etc/munge.key to all compute nodes

scp /etc/munge/munge.key csadmin@10.7.20.98:/tmp/.

Allow password-less access to slurm-ctrl

csadmin@slurm-ctrl:~$ ssh-copy-id -i .ssh/id_rsa.pub 10.7.20.102:

Run a job from slurm-ctrl

ssh csadmin@slurm-ctrl
srun -N 1 hostname
linux1

Test munge

munge -n | unmunge | grep STATUS
STATUS:           Success (0)
munge -n | ssh slurm-ctrl unmunge | grep STATUS
STATUS:           Success (0)

Test Slurm

sinfo
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
debug*       up   infinite      1   idle linux1

If computer node is down or drain

sinfo -a
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
debug*       up   infinite      2   down gpu[02-03]

sinfo 
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
gpu*         up   infinite      1  drain gpu02
gpu*         up   infinite      1   down gpu03
scontrol update nodename=gpu02 state=idle
scontrol update nodename=gpu03 state=idle
scontrol update nodename=gpu02 state=resume
sinfo -a
PARTITION AVAIL  TIMELIMIT  NODES  STATE NODELIST
debug*       up   infinite      2   idle gpu[02-03]

Compute Nodes

A compute node is a machine which will receive jobs to execute, sent from the Controller, it runs the slurmd service.

Installation slurm and munge

ssh -l csadmin <compute-nodes> 10.7.20.109 10.7.20.110
sudo apt install slurm-wlm libmunge-dev libmunge2 munge
sudo vi /lib/systemd/system/slurmd.service
[Unit]
Description=Slurm node daemon
After=network.target munge.service
ConditionPathExists=/etc/slurm-llnl/slurm.conf
Documentation=man:slurmd(8)

[Service]
Type=forking
EnvironmentFile=-/etc/default/slurmd
ExecStart=/usr/sbin/slurmd $SLURMD_OPTIONS
ExecStartPost=/bin/sleep 2
ExecReload=/bin/kill -HUP $MAINPID
PIDFile=/var/run/slurm-llnl/slurmd.pid
KillMode=process
LimitNOFILE=51200
LimitMEMLOCK=infinity
LimitSTACK=infinity

[Install]
WantedBy=multi-user.target
sudo systemctl enable slurmd
sudo systemctl enable munge
sudo systemctl start slurmd
sudo systemctl start munge

Generate ssh keys

ssh-keygen

Copy ssh-keys to slurm-ctrl

ssh-copy-id -i ~/.ssh/id_rsa.pub csadmin@slurm-ctrl.inf.unibz.it:

Become root to do important things:

sudo -i
vi /etc/hosts

Add those lines below to the /etc/hosts file

10.7.20.97      slurm-ctrl.inf.unibz.it slurm-ctrl
10.7.20.98      linux1.inf.unibz.it     linux1

First copy the munge keys from the slurm-ctrl to all compute nodes, now fix location, owner and permission.

mv /tmp/munge.key /etc/munge/.
chown munge:munge /etc/munge/munge.key
chmod 400 /etc/munge/munge.key

Place /etc/slurm-llnl/slurm.conf in right place,

mv /tmp/slurm.conf /etc/slurm-llnl/
chown root: /etc/slurm-llnl/slurm.conf

Modify user accounts

Add user

sacctmgr add user <usernme> Account=gpu-users Partition=gpu

Modify user, give 12000 minutes/200 hours for usage

sacctmgr modify user misegata set GrpTRESMin=cpu=12000,gres/gpu=12000

Restart the services:

systemctl restart slurmctld.service
systemctl restart slurmdbd.service

Check status:

systemctl status slurmctld.service
systemctl status slurmdbd.service

Modules

Python

Python 3.7.7

cd /opt/packages
mkdir /opt/packages/python/3.7.7
wget https://www.python.org/ftp/python/3.7.7/Python-3.7.7.tar.xz
tar xfJ Python-3.7.7.tar.xz
cd Python-3.7.7/
./configure --prefix=/opt/packages/python/3.7.7/ --enable-optimizations
make
make install

Python 2.7.18

cd /opt/packages
mkdir /opt/packages/python/2.7.18
wget https://www.python.org/ftp/python/2.7.18/Python-2.7.18.tar.xz
cd Python-2.7.18
./configure --prefix=/opt/packages/python/2.7.18/ --enable-optimizations
make
make install

Create modules file

PYTHON

cd /opt/modules/modulefiles/
vi python-2.7.18
#%Module1.0
proc ModulesHelp { } {
global dotversion
 
puts stderr "\tPython 2.7.18"
}
 
module-whatis "Python 2.7.18"
prepend-path PATH /opt/packages/python/2.7.18/bin

CUDA

vi /opt/modules/modulefiles/cuda-10.2
#%Module1.0
proc ModulesHelp { } {
global dotversion

puts stderr "\tcuda-10.2"
}

module-whatis "cuda-10.2"

set     prefix  /usr/local/cuda-10.2

setenv          CUDA_HOME       $prefix
prepend-path    PATH            $prefix/bin
prepend-path    LD_LIBRARY_PATH $prefix/lib64

GCC

This takes a long time!

Commands to run to compile gcc-6.1.0

wget https://ftp.gnu.org/gnu/gcc/gcc-6.1.0/gcc-6.1.0.tar.bz2
tar xfj gcc-6.1.0.tar.bz2
cd gcc-6.1.0
./contrib/download_prerequisites
./configure --prefix=/opt/package/gcc/6.1.0 --disable-multilib
make

After some time an error occurs, and the make process stops!

...
In file included from ../.././libgcc/unwind-dw2.c:401:0:
./md-unwind-support.h: In function ‘x86_64_fallback_frame_state’:
./md-unwind-support.h:65:47: error: dereferencing pointer to incomplete type ‘struct ucontext’
       sc = (struct sigcontext *) (void *) &uc_->uc_mcontext;
                                               ^~
../.././libgcc/shared-object.mk:14: recipe for target 'unwind-dw2.o' failed

To fix do: solution

vi /opt/packages/gcc-6.1.0/x86_64-pc-linux-gnu/libgcc/md-unwind-support.h

and replace/comment out line 61 with this:

struct ucontext_t *uc_ = context->cfa;

old line: /* struct ucontext *uc_ = context→cfa; */

make

Next error:

../../.././libsanitizer/sanitizer_common/sanitizer_stoptheworld_linux_libcdep.cc:270:22: error: aggregate ‘sigaltstack handler_stack’ has incomplete type and cannot be defined
   struct sigaltstack handler_stack;

To fix see: solution or https://gcc.gnu.org/bugzilla/show_bug.cgi?id=81066

Amend the files according to solution above!

Next error:

...
checking for unzip... unzip
configure: error: cannot find neither zip nor jar, cannot continue
Makefile:23048: recipe for target 'configure-target-libjava' failed
...
...
apt install unzip zip

and run make again!

make

Next error:

...
In file included from ../.././libjava/prims.cc:26:0:
../.././libjava/prims.cc: In function ‘void _Jv_catch_fpe(int, siginfo_t*, void*)’:
./include/java-signal.h:32:26: error: invalid use of incomplete type ‘struct _Jv_catch_fpe(int, siginfo_t*, void*)::ucontext’
   gregset_t &_gregs = _uc->uc_mcontext.gregs;    \
...

Edit the file: /opt/packages/gcc-6.1.0/x86_64-pc-linux-gnu/libjava/include/java-signal.h

vi /opt/packages/gcc-6.1.0/x86_64-pc-linux-gnu/libjava/include/java-signal.h
Not enough more errors!
// kh
  ucontext_t *_uc = (ucontext_t *);                             \
  //struct ucontext *_uc = (struct ucontext *)_p;                               \
  // kh

Next error:

...
In file included from ../.././libjava/prims.cc:26:0:          
./include/java-signal.h:32:3: warning: multi-line comment [-Wcomment]
   //struct ucontext *_uc = (struct ucontext *)_p;   \                                                
   ^                                                      
../.././libjava/prims.cc: In function ‘void _Jv_catch_fpe(int, siginfo_t*, void*):
./include/java-signal.h:31:15: warning: unused variable ‘_uc’ [-Wunused-variable]               
   ucontext_t *_uc = (ucontext_t *)_p;     \   
               ^         
../.././libjava/prims.cc:192:3: note: in expansion of macro ‘HANDLE_DIVIDE_OVERFLOW’            
   HANDLE_DIVIDE_OVERFLOW;       
   ^~~~~~~~~~~~~~~~~~~~~~
../.././libjava/prims.cc:203:1: error: expected ‘while’ before ‘jboolean’                    
 jboolean                                       
 ^~~~~~~~                                      
../.././libjava/prims.cc:203:1: error: expected ‘(’ before ‘jboolean’
../.././libjava/prims.cc:204:1: error: expected primary-expression before ‘_Jv_equalUtf8Consts’
 _Jv_equalUtf8Consts (const Utf8Const* a, const Utf8Const *b)                   
 ^~~~~~~~~~~~~~~~~~~                                    
../.././libjava/prims.cc:204:1: error: expected ‘)’ before ‘_Jv_equalUtf8Consts’
../.././libjava/prims.cc:204:1: error: expected ‘;’ before ‘_Jv_equalUtf8Consts’
../.././libjava/prims.cc:204:22: error: expected primary-expression before ‘const’
 _Jv_equalUtf8Consts (const Utf8Const* a, const Utf8Const *b)
...

Examples

Example mnist

An simple example to use nvidia GPU!

The example consists of the following files:

  • README.md
  • requirements.txt
  • main.job
  • main.py

Create a folder mnist and place the 4 files in there.

mkdir mnist

cat README.md

# Basic MNIST Example

```bash
pip install -r requirements.txt
python main.py
# CUDA_VISIBLE_DEVICES=2 python main.py  # to specify GPU id to ex. 2
```
cat requirements.txt
torch
torchvision
cat main.job
#!/bin/bash

#SBATCH --job-name=mnist
#SBATCH --output=mnist.out
#SBATCH --error=mnist.err

#SBATCH --partition gpu
#SBATCH --gres=gpu
#SBATCH --mem-per-cpu=4gb
#SBATCH --nodes 2
#SBATCH --time=00:08:00

#SBATCH --ntasks=10

#SBATCH --mail-type=ALL
#SBATCH --mail-user=<your-email@address.com>

ml load miniconda3
python3 main.py

Remove your-email@address.com and add your e-mail address.

main.py

Once you have all files launch this command on slurm-ctrl:

sbatch main.job

Check your job with

squeue

CUDA NVIDIA TESLA Infos

nvidia-smi

root@gpu02:~# watch nvidia-smi
Every 2.0s: nvidia-smi                                           gpu02: Mon Jun 22 17:49:14 2020

Mon Jun 22 17:49:14 2020
+-----------------------------------------------------------------------------+
| NVIDIA-SMI 440.64.00    Driver Version: 440.64.00    CUDA Version: 10.2     |
|-------------------------------+----------------------+----------------------+
| GPU  Name        Persistence-M| Bus-Id        Disp.A | Volatile Uncorr. ECC |
| Fan  Temp  Perf  Pwr:Usage/Cap|         Memory-Usage | GPU-Util  Compute M. |
|===============================+======================+======================|
|   0  Tesla V100-PCIE...  On   | 00000000:3B:00.0 Off |                    0 |
| N/A   53C    P0   139W / 250W |  31385MiB / 32510MiB |     69%      Default |
+-------------------------------+----------------------+----------------------+
|   1  Tesla V100-PCIE...  On   | 00000000:AF:00.0 Off |                    0 |
| N/A   35C    P0    26W / 250W |      0MiB / 32510MiB |      0%      Default |
+-------------------------------+----------------------+----------------------+

+-----------------------------------------------------------------------------+
| Processes:                                                       GPU Memory |
|  GPU       PID   Type   Process name                             Usage      |
|=============================================================================|
|    0      8627      C   /opt/anaconda3/bin/python3                 31373MiB |
+-----------------------------------------------------------------------------+

deviceQuery

To run the deviceQuery it is necessary to make it first!

root@gpu03:~# cd /usr/local/cuda/samples/1_Utilities/deviceQuery
make

Add PATH to the system wide environment

vi /etc/environment

Add this to the end

/usr/local/cuda/samples/bin/x86_64/linux/release

Next enable/source it:

source /etc/environment
root@gpu03:~# deviceQuery 
deviceQuery Starting...

 CUDA Device Query (Runtime API) version (CUDART static linking)

Detected 2 CUDA Capable device(s)

Device 0: "Tesla V100-PCIE-32GB"
  CUDA Driver Version / Runtime Version          10.2 / 10.2
  CUDA Capability Major/Minor version number:    7.0
  Total amount of global memory:                 32510 MBytes (34089730048 bytes)
  (80) Multiprocessors, ( 64) CUDA Cores/MP:     5120 CUDA Cores
  GPU Max Clock rate:                            1380 MHz (1.38 GHz)
  Memory Clock rate:                             877 Mhz
  Memory Bus Width:                              4096-bit
  L2 Cache Size:                                 6291456 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 7 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 59 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >

Device 1: "Tesla V100-PCIE-32GB"
  CUDA Driver Version / Runtime Version          10.2 / 10.2
  CUDA Capability Major/Minor version number:    7.0
  Total amount of global memory:                 32510 MBytes (34089730048 bytes)
  (80) Multiprocessors, ( 64) CUDA Cores/MP:     5120 CUDA Cores
  GPU Max Clock rate:                            1380 MHz (1.38 GHz)
  Memory Clock rate:                             877 Mhz
  Memory Bus Width:                              4096-bit
  L2 Cache Size:                                 6291456 bytes
  Maximum Texture Dimension Size (x,y,z)         1D=(131072), 2D=(131072, 65536), 3D=(16384, 16384, 16384)
  Maximum Layered 1D Texture Size, (num) layers  1D=(32768), 2048 layers
  Maximum Layered 2D Texture Size, (num) layers  2D=(32768, 32768), 2048 layers
  Total amount of constant memory:               65536 bytes
  Total amount of shared memory per block:       49152 bytes
  Total number of registers available per block: 65536
  Warp size:                                     32
  Maximum number of threads per multiprocessor:  2048
  Maximum number of threads per block:           1024
  Max dimension size of a thread block (x,y,z): (1024, 1024, 64)
  Max dimension size of a grid size    (x,y,z): (2147483647, 65535, 65535)
  Maximum memory pitch:                          2147483647 bytes
  Texture alignment:                             512 bytes
  Concurrent copy and kernel execution:          Yes with 7 copy engine(s)
  Run time limit on kernels:                     No
  Integrated GPU sharing Host Memory:            No
  Support host page-locked memory mapping:       Yes
  Alignment requirement for Surfaces:            Yes
  Device has ECC support:                        Enabled
  Device supports Unified Addressing (UVA):      Yes
  Device supports Compute Preemption:            Yes
  Supports Cooperative Kernel Launch:            Yes
  Supports MultiDevice Co-op Kernel Launch:      Yes
  Device PCI Domain ID / Bus ID / location ID:   0 / 175 / 0
  Compute Mode:
     < Default (multiple host threads can use ::cudaSetDevice() with device simultaneously) >
> Peer access from Tesla V100-PCIE-32GB (GPU0) -> Tesla V100-PCIE-32GB (GPU1) : Yes
> Peer access from Tesla V100-PCIE-32GB (GPU1) -> Tesla V100-PCIE-32GB (GPU0) : Yes

deviceQuery, CUDA Driver = CUDART, CUDA Driver Version = 10.2, CUDA Runtime Version = 10.2, NumDevs = 2
Result = PASS

Links

/var/www/wiki.inf.unibz.it/data/pages/tech/slurm.txt · Last modified: 2021/02/25 10:25 by kohofer
CC Attribution-Share Alike 4.0 International
Driven by DokuWiki Recent changes RSS feed Valid CSS Valid XHTML 1.0