Mamba Server Setup

This guide's purpose is to give a quick overview of how to install everything required for the Mamba Server.

Installation

1. Install Fedora Server

2. Install NVidia drivers

   # https://www.reddit.com/r/Fedora/comments/12ju2sg/i_need_help_with_installing_nvidia_drivers_to/
   sudo dnf install https://mirrors.rpmfusion.org/free/fedora/rpmfusion-free-release-$(rpm -E %fedora).noarch.rpm https://mirrors.rpmfusion.org/nonfree/fedora/rpmfusion-nonfree-release-$(rpm -E %fedora).noarch.rpm
   sudo dnf update -y
   sudo dnf install akmod-nvidia -y
   sudo dnf install xorg-x11-drv-nvidia-cuda -y
   sudo reboot now
   

3. Alternative Install NVidia drivers via dnf module

   Links to official install procedure supported by NVidia:
   https://docs.nvidia.com/cuda/cuda-installation-guide-linux/index.html#fedora

   sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/$distro/x86_64/cuda-$distro.repo
   

   Replace $distro with the latest available version matching the server, currently fedora39.

   Remark: often NVidia can be late by updating the version number of their repository. For example the current version of Fedora is 40 and the latest repo version is fedora39. Although the repository version is anterior, there will be no issue installing that version until a new one is made available available.

   sudo dnf config-manager --add-repo https://developer.download.nvidia.com/compute/cuda/repos/fedora39/x86_64/cuda-fedora39.repo
   sudo dnf makecache
   

   cuda-fedora39-x86_64 should appear in the list of enabled repositories, along with an entry in /etc/yum.repos.d/cuda-fedora39.repo.

   Next we install the nvidia driver and cuda toolkit via its now available module. Choose the appropriate version and between preferred closed or opensource module version. That will install a dkms driver that will be automatically updated with every kernel update.
   Choose between the latest version of the driver latest-dkms that will be updated with dnf update, of pin a specific version for example 555-dkms. Do the same for cuda toolkit via the meta-package cuda-toolkit or target a specific verison of cuda.

   If a previously installed kernel via rpmfusion is installed, remove everything first.

   sudo dnf autoremove akmod-nvidia xorg-x11-drv-nvidia-*
   

   Then install the new drivers via its module.

   sudo dnf module list
   sudo dnf module install nvidia-driver:latest-dkms
   

   Check that the dkms module is built successfuly for all installed kernels.

   $ sudo dkms status 
   nvidia/555.42.02, 6.8.10-300.fc40.x86_64, x86_64: installed
   

   Finally, proceed with the installation of the cuda toolkit

   sudo dnf install cuda-toolkit
   

   Select the default cuda version.

   sudo update-alternatives --display cuda
   sudo update-alternatives --config cuda
   

   Set nvcc and other utilities's PATH in
   bashrc or bash_profile per user
   /etc/environment or /etc/profile.d/cuda.sh for all users

   export CUDACXX=/usr/local/cuda/bin/nvcc
   export PATH=/usr/local/cuda/bin${PATH:+:${PATH}}
   

   Make the driver persistent

   sudo systemctl enable nvidia-persistenced.service
   sudo systemctl start nvidia-persistenced.service
   

   Before rebooting update initial ramdisk.

   sudo dracut -f
   sudo reboot now
   

4. Install Munge

   export MUNGEUSER=1111
   sudo groupadd -g $MUNGEUSER munge
   sudo useradd  -m -c "MUNGE Uid 'N' Gid Emporium" -d /var/lib/munge -u $MUNGEUSER -g munge  -s /sbin/nologin munge
   export SLURMUSER=1121
   sudo groupadd -g $SLURMUSER slurm
   sudo useradd  -m -c "SLURM workload manager" -d /var/lib/slurm -u $SLURMUSER -g slurm  -s /bin/bash slurm
   sudo dnf install munge munge-devel munge-libs -y
   
   sudo dnf install rng-tools -y
   rngd -r /dev/urandom
   
   sudo mungekey
   sudo chown munge:munge /etc/munge/munge.key
   sudo chmod 400 /etc/munge/munge.key
   sudo chown -R munge: /etc/munge/ /var/log/munge/
   sudo systemctl enable munge.service
   sudo systemctl start munge.service
   

5. Install Slurm

   sudo dnf openssl openssl-devel pam-devel numactl numactl-devel hwloc hwloc-devel lua lua-devel readline-devel rrdtool-devel ncurses-devel man2html libibmad libibumad -y
   sudo dnf install libcgroup libcgroup-tools libcgroup-devel mariadb mariadb-devel mariadb-server -y
   sudo dnf install autoconf automake perl -y
   dnf install dbus-devel -y
   
   # Build Slurm RPM
   sudo su
   cd
   wget https://download.schedmd.com/slurm/slurm-24.05.0-0rc1.tar.bz2
   rpmbuild -ta slurm-24.05.0-0rc1.tar.bz2
   cd rpmbuild/RPMS/x86_64/
   dnf --nogpgcheck localinstall *.rpm -y
   # To reinstall if you recompile
   dnf --nogpgcheck reinstall *.rpm -y
   

6. Configure Slurm: Copy the configs from norlab-ulaval/dotfiles-mamba-server to /etc/slurm

7. Ensure the permissions are correct

   mkdir /var/spool/slurmctld
   chown slurm: /var/spool/slurmctld
   chmod 755 /var/spool/slurmctld
   touch /var/log/slurmctld.log
   chown slurm: /var/log/slurmctld.log
   touch /var/log/slurm_jobacct.log /var/log/slurm_jobcomp.log
   chown slurm: /var/log/slurm_jobacct.log /var/log/slurm_jobcomp.log
   

8. Check that slurm is correctly configured: slurmd -C

9. Start the services

   systemctl enable slurmd.service
   systemctl start slurmd.service
   systemctl status slurmd.service
   
   systemctl enable slurmctld.service
   systemctl start slurmctld.service
   systemctl status slurmctld.service
   

10. Setup accounting

    systemctl enable mariadb.service
    systemctl start mariadb.service
    
    # Inspired by: https://github.com/Artlands/Install-Slurm/blob/master/README.md#setting-up-mariadb-database-master
    mysql
    # Change password in the following line
    > create user 'slurm'@'localhost' identified by '${DB_USER_PASSWORD}'; grant all on slurm_acct_db.* TO 'slurm'@'localhost'; create database slurm_acct_db;
    > GRANT ALL ON slurm_acct_db.* TO 'slurm'@'localhost' IDENTIFIED BY '1234' with grant option;
    > SHOW VARIABLES LIKE 'have_innodb';
    > FLUSH PRIVILEGES;
    > CREATE DATABASE slurm_acct_db;
    > quit;
    
    # Verify you can login
    mysql -p -u slurm
    

11. Copy /etc/my.cnf.d/innodb.cnf from norlab-ulaval/dotfiles-mamba-server

12. Restart mariadb

    systemctl stop mariadb
    mv /var/lib/mysql/ib_logfile? /tmp/
    mv /var/lib/mysql/* /tmp/
    systemctl start mariadb
    

13. Check the ownership of some files

    chown slurm slurmdbd.conf
    touch /var/log/slurmctld.log
    chown slurm /var/log/slurmctld.log
    chown slurm slurm*
    

14. Check if slurmdb can start correctly using slurmdbd -D -vvv

15. Start the services

    systemctl enable slurmdbd
    systemctl start slurmdbd
    systemctl status slurmdbd
    
    systemctl enable slurmctld.service
    systemctl start slurmctld.service
    systemctl status slurmctld.service
    

16. Add accounts

    sudo sacctmgr add account norlab Description="Norlab mamba-server" Organization=norlab
    sacctmgr add user wigum Account=norlab
    

17. If the Slurm services crash at startup, add the following lines to each slurm service (slurmctl, slurmdbd and slurmd) using systemctl edit slurmXXX.service

    [Service]
    Restart=always
    RestartSec=5s
    

18. Setup the LVM: Resize LVM.

    lvextend -l +100%FREE fedora
    xfs_growfs /dev/fedora/root
    # Verify the fs took all the place
    lsblk -f
    

19. Install nvidia-container-toolkit: nvidia-container-toolkit and setup for container use: cdi-support

    sudo dnf install dkms
    curl -s -L https://nvidia.github.io/libnvidia-container/stable/rpm/nvidia-container-toolkit.repo | \
         sudo tee /etc/yum.repos.d/nvidia-container-toolkit.repo
    sudo dnf config-manager --enable nvidia-container-toolkit-experimental
    sudo dnf install nvidia-container-toolkit -y
    sudo sed -i 's/^#no-cgroups = false/no-cgroups = true/;' /etc/nvidia-container-runtime/config.toml
    
    # Create a systemd service to run the following line at startup
    sudo nvidia-ctk cdi generate --output=/etc/cdi/nvidia.yaml
    
    sudo reboot now
    

20. Verify that the network interface is correctly configured

    sudo dnf install speedtest-cli -y
    speedtest-cli --secure
    # Should print approximately 1000Mb/s
    

    If not, change the ethernet port and enable the other interface

    # /etc/NetworkManager/system-connections/enp68s0.nmconnection
    # Enable it with
    autoconnect=true
    

21. Install a docker version that supports buildx plugin: Install docker on fedora

Add a new user

sudo useradd -c 'Full name' -m <username> -G docker
sudo passwd <username>

We also recommend setting the following env variable in their .bashrc:

export SQUEUE_FORMAT="%.18i %.9P %.25j %.8u %.2t %.10M %.6D %.20e %b %.8c"

21. Cronjob to clean podman cache

Add the following cronjob to sudo crontab -u root -e:

cat /etc/passwd | grep /bin/bash | awk -F: '{ print $1}' | while read user; do echo "Processing user $user..." && sudo -u $user -H bash -c "cd && podman system prune -af"; done

Running jobs

As users do not have root access on the Mamba Server, every project should be ran in a container. We recommend using podman.

First, on your host machine, write a Dockerfile to run your project inside a container. Then, build and test that everything works on your machine before testing it on the server.

We recommend putting your data in a directory and to symlink it to the data folder of your project. We describe here how to add volumes to avoid copying the data in the container.

# Build the image
buildah build --layers -t myproject .

# Run docker image
export CONFIG=path/to/config> # for example `config/segsdet.yaml`
export CUDA_VISIBLE_DEVICES=0 # or `0,1` for specific GPUs, will be automatically set by SLURM

podman run --gpus all --rm -it --ipc host \
  -v .:/app/ \
  -v /app/data \
  -v ./data/coco/:/app/data/coco \
  -v /dev/shm:/dev/shm \
  myproject bash -c "python3 tools/train.py $CONFIG --gpu $CUDA_VISIBLE_DEVICES"

After you verified everything works on your machine, copy the code on the server and write a Slurm job script.

#!/bin/bash
#SBATCH --gres=gpu:1
#SBATCH --cpus-per-task=16
#SBATCH --time=10-00:00
#SBATCH --job-name=$NAME
#SBATCH --output=%x-%j.out

cd ~/myproject || exit
buildah build --layers -t myproject .

export CONFIG=path/to/config> # for example `config/segsdet.yaml`

# Notice there is no -it option
podman run --gpus all --rm --ipc host \
  -v .:/app/ \
  -v /app/data \
  -v ./data/coco/:/app/data/coco \
  -v /dev/shm:/dev/shm \
  myproject bash -c "python3 tools/train.py $CONFIG --gpu $CUDA_VISIBLE_DEVICES"

Then, you can queue the job using sbatch job.sh and see the queued jobs using squeue. For an easier experience, you can use willGuimont/sjm. After you've verified this works, use the following code to kill the container when the slurm job stops.

# Notice the -d option to detach the process, and no -it option
container_id=$(
  podman run --gpus all --rm -d --ipc host \
  -v .:/app/ \
  -v /app/data \
  -v ./data/coco/:/app/data/coco \
  -v /dev/shm:/dev/shm \
  myproject bash -c "python3 tools/train.py $CONFIG --gpu $CUDA_VISIBLE_DEVICES"
)

stop_container() {
  podman logs $container_id
  podman container stop $container_id
}

trap stop_container EXIT
echo "Container ID: $container_id"
podman wait $container_id

You can then run the job using:

sbatch job.sh

And see the running jobs using:

squeue

Remote connection

1. SSH and X11 forwarding with GLX support

   First make sure X11 forwarding is enabled server side. Check these two lines in /etc/ssh/sshd_config

   X11Forwarding yes
   X11DisplayOffset 10
   

   If they were not, enable them and restart sshd

   sudo systemctl restart sshd
   

   Make sure xauth is available on the server or install it

   sudo dnf install xorg-x11-xauth
   

   Next install basic utilities to test GLX and Vulkan capabilities on the server. We'll need them to benchmark the remote connection's performance.

   sudo dnf install glx-utils vulkan-tools
   

   If you encounter problems, make sure on the client-side the server is allowed to display.
   Make sure the ip of the server is valid. + to add - to remove from the trusted list.

   xhost + 132.203.26.231
   

   Connect to the server from your client using ssh.
   Use the options -X or -Y to redirect X11 via the ssh tunnel. The redirection works despite the fact that the server is headless. But Xorg must be installed.
   The -X option will automatically update the DSIPLAY env variable. Note: the ip of the server is subject to change, make sure to have the last updated.

   ssh -X user@132.203.26.231
   

   Test that X redirection is working by executing a simple X graphical application.

   $ xterm
   

   Test GLX support with glxinfo

   glxinfo
   

   Test what GLX implementation is used by default

   $ glxinfo | grep -i vendor
   server glx vendor string: SGI
   client glx vendor string: Mesa Project and SGI
       Vendor: Mesa (0xffffffff)
   OpenGL vendor string: Mesa
   

   Check both NVidia and Mesa implementations work for GLX passthrough.

   __GLX_VENDOR_LIBRARY_NAME=nvidia glxinfo | grep -i vendor
   __GLX_VENDOR_LIBRARY_NAME=mesa glxinfo | grep -i vendor
   

   Choose the best implementation between Nvidia and Mesa.
   On Nvidia GPUs NVidia's implementation gives the best results.

   export __GLX_VENDOR_LIBRARY_NAME=nvidia
   glxgears
   

   For Vulkan aplications the process is similar

   vulkaninfo
   VK_DRIVER_FILES="/usr/share/vulkan/icd.d/nvidia_icd.x86_64.json" vkcube