forked from rarias/bscpkgs
113 lines
3.6 KiB
Nix
113 lines
3.6 KiB
Nix
{ lib, config, pkgs, ... }:
|
|
|
|
{
|
|
imports = [
|
|
../common/base.nix
|
|
../common/xeon/console.nix
|
|
../module/amd-uprof.nix
|
|
../module/emulation.nix
|
|
../module/nvidia.nix
|
|
../module/slurm-client.nix
|
|
../module/hut-substituter.nix
|
|
./wireguard.nix
|
|
];
|
|
|
|
# Don't turn off on August as UPC has different dates.
|
|
# Fox works fine on power cuts.
|
|
systemd.timers.august-shutdown.enable = false;
|
|
|
|
# Select the this using the ID to avoid mismatches
|
|
boot.loader.grub.device = "/dev/disk/by-id/wwn-0x500a07514b0c1103";
|
|
|
|
# No swap, there is plenty of RAM
|
|
swapDevices = lib.mkForce [];
|
|
|
|
boot.initrd.availableKernelModules = [ "xhci_pci" "ahci" "nvme" "usbhid" "usb_storage" "sd_mod" ];
|
|
boot.kernelModules = [ "kvm-amd" "amd_uncore" "amd_hsmp" ];
|
|
|
|
hardware.cpu.amd.updateMicrocode = lib.mkDefault config.hardware.enableRedistributableFirmware;
|
|
hardware.cpu.intel.updateMicrocode = lib.mkForce false;
|
|
|
|
# Use performance for benchmarks
|
|
powerManagement.cpuFreqGovernor = "performance";
|
|
|
|
services.amd-uprof.enable = true;
|
|
|
|
# Disable NUMA balancing
|
|
boot.kernel.sysctl."kernel.numa_balancing" = 0;
|
|
|
|
# Expose kernel addresses
|
|
boot.kernel.sysctl."kernel.kptr_restrict" = 0;
|
|
|
|
# Disable NMI watchdog to save one hw counter (for AMD uProf)
|
|
boot.kernel.sysctl."kernel.nmi_watchdog" = 0;
|
|
|
|
services.openssh.settings.X11Forwarding = true;
|
|
|
|
services.fail2ban.enable = true;
|
|
|
|
networking = {
|
|
timeServers = [ "ntp1.upc.edu" "ntp2.upc.edu" ];
|
|
hostName = "fox";
|
|
# UPC network (may change over time, use DHCP)
|
|
# Public IP configuration:
|
|
# - Hostname: fox.ac.upc.edu
|
|
# - IP: 147.83.30.141
|
|
# - Gateway: 147.83.30.130
|
|
# - NetMask: 255.255.255.192
|
|
# Private IP configuration for BMC:
|
|
# - Hostname: fox-ipmi.ac.upc.edu
|
|
# - IP: 147.83.35.27
|
|
# - Gateway: 147.83.35.2
|
|
# - NetMask: 255.255.255.0
|
|
interfaces.enp1s0f0np0.useDHCP = true;
|
|
};
|
|
|
|
# Recommended for new graphics cards
|
|
hardware.nvidia.open = true;
|
|
|
|
# Mount NVME disks
|
|
fileSystems."/nvme0" = { device = "/dev/disk/by-label/nvme0"; fsType = "ext4"; };
|
|
fileSystems."/nvme1" = { device = "/dev/disk/by-label/nvme1"; fsType = "ext4"; };
|
|
|
|
# Mount the NFS home
|
|
fileSystems."/nfs/home" = {
|
|
device = "10.106.0.30:/home";
|
|
fsType = "nfs";
|
|
options = [ "nfsvers=3" "rsize=1024" "wsize=1024" "cto" "nofail" ];
|
|
};
|
|
|
|
# Make a /nvme{0,1}/$USER directory for each user.
|
|
systemd.services.create-nvme-dirs = let
|
|
# Take only normal users in fox
|
|
users = lib.filterAttrs (_: v: v.isNormalUser) config.users.users;
|
|
commands = lib.concatLists (lib.mapAttrsToList
|
|
(_: user: [
|
|
"install -d -o ${user.name} -g ${user.group} -m 0755 /nvme{0,1}/${user.name}"
|
|
]) users);
|
|
script = pkgs.writeShellScript "create-nvme-dirs.sh" (lib.concatLines commands);
|
|
in {
|
|
enable = true;
|
|
wants = [ "local-fs.target" ];
|
|
after = [ "local-fs.target" ];
|
|
wantedBy = [ "multi-user.target" ];
|
|
serviceConfig.ExecStart = script;
|
|
};
|
|
|
|
# Only allow SSH connections from users who have a SLURM allocation
|
|
# See: https://slurm.schedmd.com/pam_slurm_adopt.html
|
|
security.pam.services.sshd.rules.account.slurm = {
|
|
control = "required";
|
|
enable = true;
|
|
modulePath = "${pkgs.slurm}/lib/security/pam_slurm_adopt.so";
|
|
args = [ "log_level=debug5" ];
|
|
order = 999999; # Make it last one
|
|
};
|
|
|
|
# Disable systemd session (pam_systemd.so) as it will conflict with the
|
|
# pam_slurm_adopt.so module. What happens is that the shell is first adopted
|
|
# into the slurmstepd task and then into the systemd session, which is not
|
|
# what we want, otherwise it will linger even if all jobs are gone.
|
|
security.pam.services.sshd.startSession = lib.mkForce false;
|
|
}
|