331 lines
9.6 KiB
Nix
331 lines
9.6 KiB
Nix
{ pkgs, lib, config, ... }:
|
|
with lib;
|
|
let
|
|
cfg = config.features.host;
|
|
|
|
computeNodes =
|
|
import ./c0/nodes.nix ++
|
|
import ./c1/nodes.nix;
|
|
|
|
mkSANs = host: [
|
|
host.name
|
|
host.address
|
|
"127.0.0.1"
|
|
];
|
|
|
|
configuration = {
|
|
system.autoUpgrade.enable = lib.mkForce false;
|
|
|
|
nixpkgs.overlays = [
|
|
(import ./overlays.nix)
|
|
(import ../modules/overrides/certmgr.nix)
|
|
];
|
|
|
|
boot = {
|
|
loader.systemd-boot.enable = true;
|
|
loader.efi.canTouchEfiVariables = true;
|
|
# kernelPackages = pkgs.linuxKernel.packages.linux_6_9;
|
|
kernelModules = [ "ib_umad" "ib_ipoib" "ceph" ];
|
|
# kernelParams = [
|
|
# "console=ttyS0,115200"
|
|
# "console=tty0"
|
|
# ];
|
|
};
|
|
|
|
console = {
|
|
font = "Lat2-Terminus16";
|
|
keyMap = "us";
|
|
};
|
|
|
|
i18n = {
|
|
defaultLocale = "en_US.UTF-8";
|
|
extraLocaleSettings = {
|
|
LC_CTYPE="en_DK.UTF-8";
|
|
LC_TIME="en_DK.UTF-8";
|
|
LC_PAPER="en_DK.UTF-8";
|
|
LC_NAME="en_DK.UTF-8";
|
|
LC_ADDRESS="en_DK.UTF-8";
|
|
LC_TELEPHONE="en_DK.UTF-8";
|
|
LC_MEASUREMENT="en_DK.UTF-8";
|
|
LC_IDENTIFICATION="en_DK.UTF-8";
|
|
};
|
|
};
|
|
|
|
time.timeZone = "Europe/Oslo";
|
|
|
|
programs.msmtp = {
|
|
enable = true;
|
|
accounts = {
|
|
default = {
|
|
auth = false;
|
|
tls = false;
|
|
tls_starttls = false;
|
|
port = 24;
|
|
from = "ekman@oceanbox.io";
|
|
host = "smtpgw.itpartner.no";
|
|
# user = "utvikling";
|
|
# password = "S0m3rp0m@de#21!";
|
|
};
|
|
};
|
|
defaults = {
|
|
aliases = "/etc/aliases";
|
|
};
|
|
};
|
|
|
|
environment.etc = {
|
|
"aliases" = {
|
|
text = ''
|
|
root: jonas.juselius@oceanbox.io
|
|
'';
|
|
mode = "0644";
|
|
};
|
|
};
|
|
|
|
features = {
|
|
os = {
|
|
# boot.uefi = true;
|
|
adminAuthorizedKeys = [
|
|
"ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIKiAS30ZO+wgfAqDE9Y7VhRunn2QszPHA5voUwo+fGOf jonas"
|
|
"ssh-rsa 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 jonas.juselius@juselius.io"
|
|
"ssh-rsa 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 Simen Kirkvik"
|
|
"ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIHVwcJOtx9YTWy+aD4xGbyPFLOdMN6NqY8wcfDtHczyT Stig Rune Jensen"
|
|
"ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIKfgY468dPNpdXZCkD9jw1p2qA0+z56Wi/c1VYE+riki Stig Rune Jensen"
|
|
"ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAII77Aa2MFZMTha8PdkNg32UR8y6Hwb4R0aR9Ad9qifNq mrtz@wurst"
|
|
];
|
|
docker.enable = false;
|
|
};
|
|
cachix.enable = false;
|
|
monitoring.nodeExporter.enable = false;
|
|
hpc.mft.enable = false; # Mellanox MFT
|
|
};
|
|
|
|
networking = {
|
|
domain = mkDefault "compute.local";
|
|
defaultGateway = mkDefault "10.255.241.1";
|
|
nameservers = mkDefault [ "8.8.8.8" ];
|
|
search = mkDefault [];
|
|
extraHosts = import ./hosts.nix;
|
|
firewall.extraCommands = ''
|
|
iptables -I INPUT -s 10.255.241.0/24 -j ACCEPT
|
|
iptables -I INPUT -s 10.255.243.0/24 -j ACCEPT
|
|
'';
|
|
};
|
|
|
|
environment.variables = {};
|
|
|
|
# systemd.services."serial-getty@ttyS0".enable = true;
|
|
|
|
# environment.etc."beegfs/connauthfile" = {
|
|
# source = ./connauthfile;
|
|
# mode = "0400";
|
|
# uid = 0;
|
|
# gid = 0;
|
|
# };
|
|
|
|
services.certmgr.validMin = "120h";
|
|
services.certmgr.renewInterval = "30m";
|
|
|
|
nix.settings = {
|
|
max-jobs = 32;
|
|
trusted-users = [ "@wheel" ];
|
|
substituters = [
|
|
];
|
|
};
|
|
|
|
system.activationScripts = {
|
|
kraken-permissions.text = ''
|
|
chmod 755 /work/kraken
|
|
'';
|
|
};
|
|
};
|
|
|
|
slurm = {
|
|
features.hpc.slurm = {
|
|
enable = true;
|
|
client = true;
|
|
mungeKey = ./munge.key;
|
|
mungeUid = mkDefault 996; # hack
|
|
# pkey = "0x7666";
|
|
controlMachine = "frontend";
|
|
mailDomain = "oceanbox.io";
|
|
nodeName = [
|
|
"c0-[1-18] Sockets=2 CoresPerSocket=64 ThreadsPerCore=1 RealMemory=256000 TmpDisk=500000 State=UNKNOWN"
|
|
"c1-[1-8] Sockets=1 CoresPerSocket=64 ThreadsPerCore=1 RealMemory=256000 TmpDisk=100000 State=UNKNOWN"
|
|
"ekman Sockets=2 CoresPerSocket=64 ThreadsPerCore=2 RealMemory=256000 TmpDisk=500000 State=UNKNOWN"
|
|
"frontend Sockets=2 CoresPerSocket=16 ThreadsPerCore=2 RealMemory=92000 TmpDisk=200000 State=UNKNOWN"
|
|
];
|
|
partitionName = [
|
|
"batch Nodes=c0-[1-18] Default=YES MaxTime=INFINITE State=UP"
|
|
"ekman Nodes=ekman MaxTime=1:00:00 State=UP"
|
|
"short Nodes=c1-[1-8] MaxTime=INFINITE State=UP"
|
|
"long Nodes=c1-[3-8] MaxTime=INFINITE State=UP"
|
|
"stats Nodes=c1-[7-8] MaxTime=INFINITE State=UP"
|
|
];
|
|
};
|
|
};
|
|
|
|
compute = {
|
|
system.activationScripts = {
|
|
mkWorkDir.text = "mkdir -p /work";
|
|
};
|
|
|
|
cluster.slurm = true;
|
|
|
|
features = {
|
|
hpc = {
|
|
enable = true;
|
|
beegfs = {
|
|
enable = false;
|
|
# beegfs = {
|
|
# work = {
|
|
# mgmtdHost = "ibbeegfs0";
|
|
# connAuthFile = "/etc/beegfs/connauthfile";
|
|
# client = {
|
|
# enable = false;
|
|
# mountPoint = "/work";
|
|
# };
|
|
# };
|
|
# };
|
|
};
|
|
};
|
|
};
|
|
};
|
|
|
|
k8s-node = {
|
|
features = {
|
|
k8s = {
|
|
enable = true;
|
|
node.enable = true;
|
|
clusterName = "ekman";
|
|
initca = ./ca;
|
|
cidr = "10.100.0.0/16";
|
|
master = {
|
|
name = "frontend";
|
|
address = "10.255.241.99";
|
|
# extraSANs = [
|
|
# "frontend.oceanbox.io"
|
|
# ];
|
|
};
|
|
ingressNodes = [
|
|
"ekman.oceanbox.io"
|
|
];
|
|
fileserver = "fs-work";
|
|
charts = {
|
|
acme_email = "acme@oceanbox.io";
|
|
# grafana_smtp_user = "utvikling";
|
|
# grafana_smtp_password = "S0m3rp0m@de#21!";
|
|
};
|
|
};
|
|
};
|
|
|
|
system.activationScripts = {
|
|
copyCaKey.text = "cp ${./ca}/ca-key.pem /var/lib/kubernetes/secrets";
|
|
};
|
|
|
|
# services.kubernetes.kubelet.extraSANs = mkSANs {
|
|
# name = cfg.name;
|
|
# address = cfg.address;
|
|
# };
|
|
};
|
|
|
|
shosts = {
|
|
environment.etc."ssh/shosts.equiv" = {
|
|
mode = "0644";
|
|
uid = 0;
|
|
gid = 0;
|
|
text = ''
|
|
10.255.241.80
|
|
10.255.241.90
|
|
'' + builtins.foldl' (a: x: a + "${x.address}\n") "" computeNodes;
|
|
};
|
|
|
|
programs.ssh.knownHosts = {
|
|
frontend = {
|
|
hostNames = [
|
|
"frontend" "frontend.compute.local" "frontend.oceanbox.io" "10.255.241.99" "10.255.243.99"
|
|
];
|
|
publicKeyFile = ../frontend.pub;
|
|
};
|
|
ekman = {
|
|
hostNames = [
|
|
"ekman" "ekman.compute.local" "ekman.oceanbox.io" "10.255.241.100" "10.255.243.100"
|
|
];
|
|
publicKeyFile = ./ekman/ekman.pub;
|
|
};
|
|
fs-work = {
|
|
hostNames = [
|
|
"fs-work" "fs-work.compute.local" "10.255.241.90" "10.255.243.90"
|
|
];
|
|
publicKeyFile = ./fs-work/fs-work.pub;
|
|
};
|
|
fs-backup = {
|
|
hostNames = [
|
|
"fs-backup" "fs-backup.compute.local" "10.255.241.80" "10.255.243.80"
|
|
];
|
|
publicKeyFile = ./fs-backup/fs-backup.pub;
|
|
};
|
|
} // builtins.foldl' (a: x:
|
|
let n = toString x.idx;
|
|
in a // {
|
|
"${x.name}" = {
|
|
hostNames = [
|
|
"${x.name}"
|
|
"${x.name}.compute.local"
|
|
"10.255.241.${n}"
|
|
"10.255.243.${n}"
|
|
];
|
|
publicKeyFile = x.pubkey;
|
|
};
|
|
}) {} computeNodes;
|
|
|
|
environment.systemPackages = [
|
|
openssh-shosts
|
|
pkgs.inotify-tools
|
|
pkgs.ceph
|
|
pkgs.ceph-client
|
|
];
|
|
|
|
security.wrappers = {
|
|
ssh-keysign = {
|
|
source = "${openssh-shosts}/libexec/ssh-keysign";
|
|
owner = "root";
|
|
group = "root";
|
|
permissions = "u+rs,g+rx,o+rx";
|
|
};
|
|
};
|
|
};
|
|
|
|
openssh-shosts = pkgs.openssh.overrideAttrs (attrs: {
|
|
buildFlags = [ "SSH_KEYSIGN=/run/wrappers/bin/ssh-keysign" ];
|
|
doCheck = false; # the tests take hours
|
|
});
|
|
|
|
in {
|
|
options.cluster = {
|
|
compute = mkEnableOption "Enable compute node configs";
|
|
};
|
|
|
|
options.cluster = {
|
|
k8sNode = mkEnableOption "Enable k8s node";
|
|
};
|
|
|
|
options.cluster = {
|
|
slurm = mkEnableOption "Enable slurm";
|
|
};
|
|
|
|
config = mkMerge [
|
|
configuration
|
|
shosts
|
|
(mkIf config.cluster.compute compute)
|
|
(mkIf config.cluster.k8sNode k8s-node)
|
|
(mkIf config.cluster.slurm slurm)
|
|
];
|
|
|
|
imports = [
|
|
../modules
|
|
../nixos
|
|
./users.nix
|
|
];
|
|
}
|
|
|