Difference between revisions of "Sunrise.fjfi.cvut.cz"

From NMS
Jump to: navigation, search
(→‎Reinstall)
(→‎Singularity)
 
Line 332: Line 332:
 
  apptainer shell /cvmfs/farm.particle.cz/singularity/fzu_gui-centos7
 
  apptainer shell /cvmfs/farm.particle.cz/singularity/fzu_gui-centos7
 
   
 
   
−
  # Metacentrum containers available in /cvmfs/singularity.metacentrum.cz
+
  # Metacentrum containers available in /cvmfs/singularity.metacentrum.cz, e.g.
 +
apptainer shell /cvmfs/singularity.metacentrum.cz/Metacentrum/ood-meta2.sif
 +
apptainer shell /cvmfs/singularity.metacentrum.cz/Metacentrum/debian11-openpbs.sif
 +
 +
# European Environment for Scientific Software Installations (EESSI)
 +
# doesn't provide apptainers images but rely on `modules` for more details see
 +
# /cvmfs/software.eessi.io/README.eessi
  
 
===ATLAS & Containers===
 
===ATLAS & Containers===

Latest revision as of 08:57, 16 September 2024

Servery / Služby
Přístupné komukoliv
windows
srk
linux / unix
kmlinux
Omezený/individuální účet
linux / unix
bimbo · buon(KF) · km(KM) · lenochod(KJR) · linux · node(KM) · sunrise(KF) · unixlab(KFE) · vkstat(KM)
Služby
backup · DHCP · DNS · doména FJFI · eduroam · fileserver · IdM · forum · gitlab · lists · moodle · indico · mailgw · K4 · mailserver · NMS · openvpn · skolniftp · ssh · videokonference · VoIP · video · VPN · wififjfi · wiki · www
Učebny
e-sklipek · KFE unixlab · KFE pclab · PD1 · KM 105 · KM 115
Ostatní
Network · Blokované porty
[edit] · [view]

Základní informace (sunrise)

Správce 
Michal Broz
HW 
~ 650 core cluster, 200TB storage
OS 
CentOS7 (+ additional CERN packages, e.g. EOS)
Využití 
cluster KF
Konto 
domluvit se správcem

Cluster hardware

year nodes cpumark/node details
2007 1-10 5 serveru s konfiguraci 4x2 jadra, 12x4GB
2010 11-14 6508 2x Intel E5620@2.40GHz (16 HT cores, uveden 2010), 48GB RAM
2011 15,16 6508 2x Intel E5620@2.40GHz (16 HT cores, uveden 2010), 48GB RAM
2012 17,18 6508 2x Intel E5620@2.40GHz (16 HT cores, uveden 2010), 48GB RAM
2012 19,20 6508 2x Intel E5620@2.40GHz (16 HT cores, uveden 2010), 48GB RAM
2013 21,22 5935 2x Intel E5-2609@2.40GHz (8 cores, uveden 2012, avx), 48GB RAM (dva volne sloty)
2014 23,24 5935 2x Intel E5-2609@2.40GHz (8 cores, uveden 2012, avx), 48GB RAM (dva volne sloty)
2018 25-28 42460 2x AMD EPYC 7281@2.1GHZ (64 HT cores, uveden 2017), 256GB RAM
2022 29-34 56562 1x AMD EPYC3 7543@2.8GHz (64 HT cores, uveden 2021), 256GB RAM

Basic info & links

  • monitoring
  • services
    • NAT + DNS (for worker nodes)
    • Squid cache (CVMFS)
    • Apache (kickstart, yum repository, monitoring interfaces)
    • puppet (configuration management)
    • PBSPro server

(Re)Instalace worker nodů

Instalace z lokálního USB

  • použít oficiální boot image SLC6 a nahrát na CD nebo flash
wget http://linuxsoft.cern.ch/cern/slc6X/x86_64/images/boot.iso
livecd-iso-to-disk boot.iso /dev/sd?1
# nahradit "?" písmenem zařízení odpovídající flash
  • nabootovat z CDROM/USB (na sunrise11-24 lze zobrazit boot menu po klávese F11)
  • přidat bootovací parametry pro instalaci, po stisku klávesy TAB doplnit (XX je číslo sunriseXX)
ks=http://192.168.20.1/ks.php?id=XX ksdevice=eth0 ip=192.168.20.1XX gateway=192.168.20.1 netmask=255.255.255.0 dns=147.32.9.4 ssh vnc
  • u worker nodů sunrise01-10 se při bootovaní z flash přehodí pořadí disků a je potřeba instalovat na sdb místo standardního prvního disku sda
ks=http://192.168.20.1/ks.php?id=XX&dev=sdb ...

Instalace ze sítě přes PXE

Aktuální konfigurace DHCP pro privátní subnet KF clusteru by při nastavení / výběru bootování ze sítě měla zobrazit PXE bootovací menu s různými možnostmi síťové instalace. Aktuálně zde nejsou žádné specializované volby pro sunset stroje, takže je nutné vybranou konfiguraci doplnit o konfigurační volby uvedené výše (ks, ksdevice, ip, gateway, netmask, dns, ...).

Aby TFTP protokol fungoval i za NATem, tak je nutné zavést moduly nf_conntrack_tftp a nf_nat_tftp 

cat <<EOF > /etc/modules-load.d/tftp.conf
nf_conntrack_tftp
nf_nat_tftp
EOF

a zkonfigurovat TFTP helper v konfiguraci nftables 

cat <<EOF >> /etc/sysconfig/nftables.conf
table ip handletftp
delete table ip handletftp

table ip handletftp {
    ct helper helper-tftp {
        type "tftp" protocol udp
    }

    chain sethelper {
        type filter hook forward priority 0; policy accept;
        ip saddr 192.168.20.0/24 ip daddr 147.32.9.2 udp dport 69 ct helper set "helper-tftp"
    }
}
EOF

Reinstall

  • před instalací (ale po zastavení puppetu) je potřeba odstranit aktuální certifikát z puppet serveru
puppetserver ca clean sunsetXX.kfcluster
  • z boot.iso vykopírovat vmlinuz a initramfs do /boot
  • do konfigurace grubu v /boot/grub/menu.lst přidat další záznam s výše uvedenými soubory
  • přidat parametry odpovídající bootovacím parametrům uvedeným v sekci Instalace
  • v principu by mělo stačit spustit níže uvedený skript (01 je potřeba nahradit číslem sunrise stroje):
XX=01
cd /boot
wget -O vmlinuz http://linuxsoft.cern.ch/cern/slc6X/x86_64/isolinux/vmlinuz
wget -O initrd.img http://linuxsoft.cern.ch/cern/slc6X/x86_64/isolinux/initrd.img
wget -O ks.cfg "http://192.168.20.1/ks.php?id=${XX}"

cat >> /boot/grub/menu.lst <<EOF
title Install
        root (hd0,0)
        kernel /vmlinuz ks=hd:/dev/sda1:/ks.cfg ksdevice=eth0 ip=192.168.20.1${XX} gateway=192.168.20.1 netmask=255.255.255.0 dns=147.32.9.4 ssh vnc
        initrd /initrd.img
EOF
  • pokud na puppet serveru nefunguje autosign, tak je potreba podepsat novy puppet certifikat
# vypsání nepodepsaných certifikátů (parametr --all vypíše všechny)
puppetserver ca list
# podepsání certifikátu
puppetserver ca sign sunsetXX.kfcluster

Konfigurace

Konfigurace worker nodů se provádí pomocí puppetu z ashley.fjfi.cvut.cz.

Creating puppet.git repository

# source scl_source enable git19
cd
git init --bare puppet.git
git clone ssh://root@ashley.fjfi.cvut.cz:/root/puppet.git
cd puppet
cp -a /etc/puppetlabs/code/environments/production/* 
rm -rf .git
git commit -m "Initial commit with basic configuration for puppet 4.10.1"
git branch -m master production
git push origin production
# create puppet.git/hooks/post-receive using code from
# https://puppet.com/blog/git-workflow-and-puppet-environments
# modify code to skip post-receive hook for branches with "tmp" prefix

Puppet configuration workflow

The layout of production directory is inspired by common profiles/roles/nodes abstraction layers described e.g. here.

Our puppet configuration is stored in GIT repository and automatically applied on puppet server after successful GIT push in main puppet.git repository. Don't directly modify files in `puppet config print environmentpath` subdirectories. You can create your own test environment directly on puppet server only if its name starts with "tmp" or "work" prefix. Branches from puppet.git with "tmp" prefix are excluded from automatic post-receive commit hook and they are not cloned in the puppet configuration environment.

  • on RHEL6 it is necessary to use SCL version of git 1.9
source scl_source enable git19
  • checkout current data from remote production branch
git clone -b production ssh://root@ashley.fjfi.cvut.cz:/root/puppet.git work_user
cd work_user
git checkout -b work_user
git config user.name "First Surename"
git config user.email "first.surename@fjfi.cvut.cz"
  • modify files in "work_user" directory and if you cloned these files in puppet environment directory /etc/puppetlabs/code/environments you can test updated configuration
puppet apply --environment=work_user --test --debug
  • when you are happy with updated configuration merge modification in "production" branch and commit to master repository
git commit -m "summary info for modifications" file1 file2 ...
git checkout production
git pull
git checkout work_user
git rebase production
git checkout production
git merge work_user
git push

Monitoring puppet

Monitoring web interfaces listen only on localhost, you have to tunnel local ports from ashley.fjfi.cvut.cz to you machine before you are able to see provided data 

ssh -L 1080:127.0.0.1:1080 -L 8080:127.0.0.1:8080 ashley.fjfi.cvut.cz

Batch

Currently this cluster is using PBSPro as a batch system.

  • user commands to submit/check/delete jobs
qsub script.sh
qstat
qdel job_id
  • show queue configuration
qstat -Q -f
qmgr -c 'p s'
  • set worker node online/offline
pbsnodes -o sunriseXX-0
pbsnodes -r sunriseXX-0

Squid (CVMFS)

Machines located at FNSPE should used for CVMFS our local squid proxy and as a backup it is also possible (allowed) to utilize FZU proxy. CVMFS configuration in /etc/cvmfs/default.local should contain: 

CVMFS_HTTP_PROXY="http://squid.fjfi.cvut.cz:3128;http://squid.farm.particle.cz:3128;DIRECT"

For KF cluster worker nodes it is now better to use Squid cache directly from headnode 

CVMFS_HTTP_PROXY="http://ashley.fjfi.cvut.cz:3128;http://squid.farm.particle.cz:3128;DIRECT"

User software

Intel compiler

# setup environment for Intel compiler
source /fjfi/apps/intel/Compiler/11.0/074/bin/ifortvars.sh intel64
# Hello World příklad kompilovaný pomocí Intel Fortran Compileru
cat > hello.f <<EOF
        program hello
           print *, "Hello World!"
        end program hello
EOF
ifortbin hello.f
./a.out

Maple software

You can use Maple (and other softwares) on sunrise. The first step is to connect via ssh to the headnode Ashley. (If you are on Windows, you may need to install an ssh client.) For Mac/Linux, you can open the Terminal and type 

ssh USERNAME@sunrise.fjfi.cvut.cz
# USERNAME should be replaced with your username.

The connection will require the password linked with the username. (If you don’t have an account, you shall write the administrator.) From there, you should connect to one of the working nodes (sunset 01 to 28). You can monitor which nodes are currently not being used here. To connect to a node, you can type 

ssh 192.168.20.1XX
# The XX should be changed by the node that you want to connect (from 01 to 28).

One should note that the new nodes on sunrise7 (sunset 25 to 28) may require additional authorizations. Once connected to the working node, you can launch Maple with the command 

source /fjfi/apps/maple/maple/bin/maple

which will launch a command-based maple of the latest version installed on sunrise. You can enter Maple code line by line or copy and paste many lines and Maple will compile them in order. To save or export results on Maple, you can use the commands 

currentdir(path);
save result1_to_save, result2_to_save, “filename.m” 
# By leaving the path empty, the file will be saved in the current folder (/home/USERNAME/).

The list of other softwares that can be used on sunrise is in the folder /fjfi/apps/ and it includes e.g. Maple, Mathematica, Matlab, Python.

LCG Software Elements

One can also use software provided by LCG. To list availabe software you can use: 

export LCGENV_PATH=/cvmfs/sft.cern.ch/lcg/releases
export PATH=/cvmfs/sft.cern.ch/lcg/releases/lcgenv/latest:${PATH}
lcgenv
lcgenv x86_64-slc6-gcc62-opt
lcgenv -p LCG_latest
lcgenv -p LCG_latest x86_64-slc6-gcc62-opt

E.g. if you need ROOT 6.08.06 with all dependencies (Boost, python, ...) use 

export LCGENV_PATH=/cvmfs/sft.cern.ch/lcg/releases
export PATH=/cvmfs/sft.cern.ch/lcg/releases/lcgenv/latest:${PATH}
eval "`lcgenv -p LCG_88 x86_64-slc6-gcc49-opt ROOT`"

For most recent version use 

export LCGENV_PATH=/cvmfs/sft.cern.ch/lcg/releases
export PATH=/cvmfs/sft.cern.ch/lcg/releases/lcgenv/latest:${PATH}
eval "`lcgenv x86_64-slc6-gcc62-opt all`"
eval "`lcgenv -p LCG_latest x86_64-slc6-gcc62-opt ROOT`"

Setup full software stack from LCG Software Elements 

source /cvmfs/sft.cern.ch/lcg/views/setupViews.sh LCG_latest x86_64-slc6-gcc62-opt

ATLAS Software

Setup basic ATLAS environment 

export ATLAS_LOCAL_ROOT_BASE=/cvmfs/atlas.cern.ch/repo/ATLASLocalRootBase
alias setupATLAS='source ${ATLAS_LOCAL_ROOT_BASE}/user/atlasLocalSetup.sh'
setupATLAS

Validate machine configuration for ATLAS

setupATLAS
diagnostics
checkOS

List installed software in ATLAS repository

setupATLAS
showVersions
showVersions root

Use ROOT from ATLAS and LCG software repository

# ROOT directly from ATLAS repository for SLC6 with GCC 6.2
lsetup "root 6.12.04-x86_64-slc6-gcc62-opt"
# ROOT from general LCG repository for SLC6 with GCC 6.2
# gcc doesn't comes as a LCG ROOT dependency and second lsetup for gcc is necessary
lsetup "lcgenv -p LCG_92 x86_64-slc6-gcc62-opt ROOT"

Containers

Singularity

To run application in a completely different environment you can use apptainer (singularity) containers. Use official documentation for information how to use apptainer containers. Running shell from different distribution can be as simple as 

# official ATLAS images
apptainer shell /cvmfs/atlas.cern.ch/repo/containers/fs/singularity/x86_64-slc5
apptainer shell /cvmfs/atlas.cern.ch/repo/containers/fs/singularity/x86_64-centos6
apptainer shell /cvmfs/atlas.cern.ch/repo/containers/fs/singularity/x86_64-centos7
apptainer shell /cvmfs/atlas.cern.ch/repo/containers/fs/singularity/x86_64-almalinux8
apptainer shell /cvmfs/atlas.cern.ch/repo/containers/fs/singularity/x86_64-almalinux9

# CERN test images
apptainer shell /cvmfs/unpacked.cern.ch/registry.hub.docker.com/library/centos:centos6
apptainer shell /cvmfs/unpacked.cern.ch/registry.hub.docker.com/library/centos:centos7
apptainer shell /cvmfs/unpacked.cern.ch/registry.hub.docker.com/library/debian:stable
apptainer shell /cvmfs/unpacked.cern.ch/registry.hub.docker.com/library/fedora:latest

# Fermilab worker node images
apptainer shell /cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-el8:latest
apptainer shell /cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-el9:latest
apptainer shell /cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-sl6:latest
apptainer shell /cvmfs/singularity.opensciencegrid.org/fermilab/fnal-wn-sl7:latest

# OSG images
apptainer shell /cvmfs/singularity.opensciencegrid.org/opensciencegrid/osgvo-el6:latest
apptainer shell /cvmfs/singularity.opensciencegrid.org/opensciencegrid/osgvo-el7:latest
apptainer shell /cvmfs/singularity.opensciencegrid.org/opensciencegrid/osgvo-el8:latest
apptainer shell /cvmfs/singularity.opensciencegrid.org/opensciencegrid/osgvo-el9:latest

# FZU containers available in /cvmfs/farm.particle.cz/singularity
apptainer shell /cvmfs/farm.particle.cz/singularity/fzu_wn-slc6
apptainer shell /cvmfs/farm.particle.cz/singularity/fzu_wn-centos7
apptainer shell /cvmfs/farm.particle.cz/singularity/fzu_wn-centos8
apptainer shell /cvmfs/farm.particle.cz/singularity/fzu_gui-centos7

# Metacentrum containers available in /cvmfs/singularity.metacentrum.cz, e.g.
apptainer shell /cvmfs/singularity.metacentrum.cz/Metacentrum/ood-meta2.sif
apptainer shell /cvmfs/singularity.metacentrum.cz/Metacentrum/debian11-openpbs.sif

# European Environment for Scientific Software Installations (EESSI)
# doesn't provide apptainers images but rely on `modules` for more details see
# /cvmfs/software.eessi.io/README.eessi

ATLAS & Containers

Since ~ 2020 ATLAS runs basically all grid jobs in the containers which provides same environment regardless of installed OS. This is also very useful in case you would like to run old code compiled in the older OS version (e.g. use CentOS7 for SLC6 ATLAS development). ATLAS provides setupATLAS which can automatically bring you selected OS with righ ATLAS environment 

export ATLAS_LOCAL_ROOT_BASE=/cvmfs/atlas.cern.ch/repo/ATLASLocalRootBase
#export ALRB_CONT_CMDOPTS="$ALRB_CONT_CMDOPTS -B /mnt:/mnt"
alias setupATLAS='source ${ATLAS_LOCAL_ROOT_BASE}/user/atlasLocalSetup.sh'
setupATLAS -c sl6

Run `setupATLAS -h` to get more details, available OS environment can be listed with `setupATLAS -c -h`

Retrieved from "https://nms.fjfi.cvut.cz/wiki/index.php?title=Sunrise.fjfi.cvut.cz&oldid=4077"