Compare commits

..

1 Commits

Author SHA1 Message Date
eeac26ccdb Fix missing which in nodes checkPhase
The check still works, but it pollutes the logs with errors
2025-10-23 14:34:37 +02:00
20 changed files with 83 additions and 280 deletions

View File

@@ -5,8 +5,8 @@
vim wget git htop tmux pciutils tcpdump ripgrep nix-index nixos-option
nix-diff ipmitool freeipmi ethtool lm_sensors cmake gnumake file tree
ncdu config.boot.kernelPackages.perf ldns pv
# From jungle overlay
osumb nixgen
# From bsckgs overlay
osumb
];
programs.direnv.enable = true;

View File

@@ -180,19 +180,6 @@
"ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIFmMqKqPg4uocNOr3O41kLbZMOMJn3m2ZdN1JvTR96z3 bsccns@arnau-bsc"
];
};
aaguirre = {
uid = 9655;
isNormalUser = true;
home = "/home/Computational/aaguirre";
description = "Alejandro Aguirre";
group = "Computational";
hosts = [ "apex" "hut" ];
hashedPassword = "$6$TXRXQT6jjBvxkxU6$E.sh5KspAm1qeG5Ct7OPHpo8REmbGDwjFGvqeGgTVz3GASGOAnPL7UMZsMAsAKBoahOw.v8LNno6XGrTEPzZH1";
openssh.authorizedKeys.keys = [
"ssh-ed25519 AAAAC3NzaC1lZDI1NTE5AAAAIOlRX7ZCnqtUJYCxKgWmgSrFCYuA2LHY96rVwqxXPl86 aaguirre@BSC-8488184117"
];
};
};
groups = {

View File

@@ -93,4 +93,20 @@
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;
}

View File

@@ -17,7 +17,6 @@
./postgresql.nix
./nginx.nix
./p.nix
./ompss2-timer.nix
#./pxe.nix
];

View File

@@ -29,9 +29,6 @@
};
};
# Allow gitea user to send mail
users.users.gitea.extraGroups = [ "mail-robot" ];
services.gitea-actions-runner.instances = {
runrun = {
enable = true;

View File

@@ -1,11 +1,8 @@
{ config, lib, ... }:
{
# Robot user that can see the password to send mail from jungle-robot
users.groups.mail-robot = {};
age.secrets.jungleRobotPassword = {
file = ../../secrets/jungle-robot-password.age;
group = "mail-robot";
group = "gitea";
mode = "440";
};

View File

@@ -4,8 +4,8 @@ let
name = "jungle-web";
src = pkgs.fetchgit {
url = "https://jungle.bsc.es/git/rarias/jungle-website.git";
rev = "52abaf4d71652a9ef77a0b098db14ca33bffff4c";
hash = "sha256-/ul9GazbOrOkmlvSgDz/+2W+V+ir5725Y7mVLc3rb0M=";
rev = "739bf0175a7f05380fe7ad7023ff1d60db1710e1";
hash = "sha256-ea5DzhYTzZ9TmqD+x95rdNdLbxPnBluqlYH2NmBYmc4=";
};
buildInputs = [ pkgs.hugo ];
buildPhase = ''

View File

@@ -1,85 +0,0 @@
{ config, pkgs, ... }:
{
systemd.timers = {
"ompss2-closing" = {
wantedBy = [ "timers.target" ];
timerConfig = {
Unit = "ompss2-closing.service";
OnCalendar = [ "*-03-15 07:00:00" "*-09-15 07:00:00"];
};
};
"ompss2-freeze" = {
wantedBy = [ "timers.target" ];
timerConfig = {
Unit = "ompss2-freeze.service";
OnCalendar = [ "*-04-15 07:00:00" "*-10-15 07:00:00" ];
};
};
"ompss2-release" = {
wantedBy = [ "timers.target" ];
timerConfig = {
Unit = "ompss2-release.service";
OnCalendar = [ "*-05-15 07:00:00" "*-11-15 07:00:00" ];
};
};
};
systemd.services =
let
closing = pkgs.writeText "closing.txt"
''
Subject: OmpSs-2 release enters closing period
Hi,
You have one month to merge the remaining features for the next OmpSs-2
release. Please, identify what needs to be merged and discuss it in the next
OmpSs-2 meeting.
Thanks!,
Jungle robot
'';
freeze = pkgs.writeText "freeze.txt"
''
Subject: OmpSs-2 release enters freeze period
Hi,
The period to introduce new features or breaking changes is over, only bug
fixes are allowed now. During this time, please prepare the release notes
to be included in the next OmpSs-2 release.
Thanks!,
Jungle robot
'';
release = pkgs.writeText "release.txt"
''
Subject: OmpSs-2 release now
Hi,
The period to introduce bug fixes is now over. Please, proceed to do the
OmpSs-2 release.
Thanks!,
Jungle robot
'';
mkServ = name: mail: {
"ompss2-${name}" = {
script = ''
set -eu
set -o pipefail
cat ${mail} | ${config.security.wrapperDir}/sendmail star@bsc.es
'';
serviceConfig = {
Type = "oneshot";
DynamicUser = true;
Group = "mail-robot";
};
};
};
in
(mkServ "closing" closing) //
(mkServ "freeze" freeze) //
(mkServ "release" release);
}

View File

@@ -1,4 +1,4 @@
{ lib, pkgs, ... }:
{ lib, ... }:
{
imports = [
@@ -21,20 +21,4 @@
};
services.slurm.client.enable = true;
# 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;
}

View File

@@ -1,6 +1,31 @@
{ config, pkgs, ... }:
{
let
suspendProgram = pkgs.writeShellScript "suspend.sh" ''
exec 1>>/var/log/power_save.log 2>>/var/log/power_save.log
set -x
export "PATH=/run/current-system/sw/bin:$PATH"
echo "$(date) Suspend invoked $0 $*" >> /var/log/power_save.log
hosts=$(scontrol show hostnames $1)
for host in $hosts; do
echo Shutting down host: $host
ipmitool -I lanplus -H ''${host}-ipmi -P "" -U "" chassis power off
done
'';
resumeProgram = pkgs.writeShellScript "resume.sh" ''
exec 1>>/var/log/power_save.log 2>>/var/log/power_save.log
set -x
export "PATH=/run/current-system/sw/bin:$PATH"
echo "$(date) Suspend invoked $0 $*" >> /var/log/power_save.log
hosts=$(scontrol show hostnames $1)
for host in $hosts; do
echo Starting host: $host
ipmitool -I lanplus -H ''${host}-ipmi -P "" -U "" chassis power on
done
'';
in {
services.slurm = {
controlMachine = "apex";
clusterName = "jungle";
@@ -34,6 +59,16 @@
# the resources. Use the task/cgroup plugin to enable process containment.
TaskPlugin=task/affinity,task/cgroup
# Power off unused nodes until they are requested
SuspendProgram=${suspendProgram}
SuspendTimeout=60
ResumeProgram=${resumeProgram}
ResumeTimeout=300
SuspendExcNodes=fox
# Turn the nodes off after 1 hour of inactivity
SuspendTime=3600
# Reduce port range so we can allow only this range in the firewall
SrunPortRange=60000-61000
@@ -51,7 +86,9 @@
# when a task runs (srun) so we can ssh early.
PrologFlags=Alloc,Contain,X11
LaunchParameters=use_interactive_step
# LaunchParameters=ulimit_pam_adopt will set RLIMIT_RSS in processes
# adopted by the external step, similar to tasks running in regular steps
# LaunchParameters=ulimit_pam_adopt
SlurmdDebug=debug5
#DebugFlags=Protocol,Cgroup
'';

View File

@@ -4,8 +4,8 @@ let
name = "jungle-web";
src = pkgs.fetchgit {
url = "https://jungle.bsc.es/git/rarias/jungle-website.git";
rev = "52abaf4d71652a9ef77a0b098db14ca33bffff4c";
hash = "sha256-/ul9GazbOrOkmlvSgDz/+2W+V+ir5725Y7mVLc3rb0M=";
rev = "739bf0175a7f05380fe7ad7023ff1d60db1710e1";
hash = "sha256-ea5DzhYTzZ9TmqD+x95rdNdLbxPnBluqlYH2NmBYmc4=";
};
buildInputs = [ pkgs.hugo ];
buildPhase = ''

View File

@@ -37,7 +37,6 @@ let
nanos6 = callPackage ./pkgs/nanos6/default.nix { };
nanos6Debug = final.nanos6.override { enableDebug = true; };
nixtools = callPackage ./pkgs/nixtools/default.nix { };
nixgen = callPackage ./pkgs/nixgen/default.nix { };
# Broken because of pkgsStatic.libcap
# See: https://github.com/NixOS/nixpkgs/pull/268791
#nix-wrap = callPackage ./pkgs/nix-wrap/default.nix { };
@@ -52,7 +51,6 @@ let
prometheus-slurm-exporter = prev.callPackage ./pkgs/slurm-exporter/default.nix { };
#pscom = callPackage ./pkgs/parastation/pscom.nix { }; # Unmaintaned
#psmpi = callPackage ./pkgs/parastation/psmpi.nix { }; # Unmaintaned
slurm = import ./pkgs/slurm/default.nix { slurm = prev.slurm; };
sonar = callPackage ./pkgs/sonar/default.nix { };
stdenvClangOmpss2 = final.stdenv.override { cc = final.clangOmpss2; allowedRequisites = null; };
stdenvClangOmpss2Nanos6 = final.stdenv.override { cc = final.clangOmpss2Nanos6; allowedRequisites = null; };
@@ -101,16 +99,14 @@ let
pkgsTopLevel = filterAttrs (_: isDerivation) bscPkgs;
# Native build in that platform doesn't imply cross build works
canCrossCompile = platform: default: pkg:
canCrossCompile = platform: pkg:
(isDerivation pkg) &&
# If meta.cross is undefined, use default
(pkg.meta.cross or default) &&
(meta.availableOn final.pkgsCross.${platform}.stdenv.hostPlatform pkg);
# Must be defined explicitly
(pkg.meta.cross or false) &&
(meta.availableOn platform pkg);
# For now only RISC-V
crossSet = genAttrs [ "riscv64" ] (platform:
filterAttrs (_: canCrossCompile platform true)
final.pkgsCross.${platform}.bsc.pkgsTopLevel);
crossSet = { riscv64 = final.pkgsCross.riscv64.bsc.pkgsTopLevel; };
buildList = name: paths:
final.runCommandLocal name { } ''
@@ -130,7 +126,7 @@ let
# For now only RISC-V
crossList = buildList "ci-cross"
(filter
(canCrossCompile "riscv64" false) # opt-in (pkgs with: meta.cross = true)
(canCrossCompile final.pkgsCross.riscv64.stdenv.hostPlatform)
(builtins.attrValues crossSet.riscv64));
in bscPkgs // {

View File

@@ -90,7 +90,7 @@ in
meta = {
description = "Performance analysis tool-suite for x86 based applications";
homepage = "https://www.amd.com/es/developer/uprof.html";
platforms = [ "x86_64-linux" ];
platforms = lib.platforms.linux;
license = lib.licenses.unfree;
maintainers = with lib.maintainers.bsc; [ rarias varcila ];
};

View File

@@ -1,6 +1,5 @@
{
stdenv
, lib
, cudatoolkit
, cudaPackages
, autoAddDriverRunpath
@@ -41,9 +40,4 @@ stdenv.mkDerivation (finalAttrs: {
'';
installPhase = "touch $out";
};
meta = {
platforms = [ "x86_64-linux" ];
maintainers = with lib.maintainers.bsc; [ rarias ];
};
})

View File

@@ -16,19 +16,19 @@
, useGit ? false
, gitUrl ? "ssh://git@bscpm04.bsc.es/llvm-ompss/llvm-mono.git"
, gitBranch ? "master"
, gitCommit ? "872ba63f86edaefc9787984ef3fae9f2f94e0124" # github-release-2025.11
, gitCommit ? "880e2341c56bad1dc14e8c369fb3356bec19018e"
}:
let
stdenv = llvmPackages_latest.stdenv;
release = rec {
version = "2025.11";
version = "2025.06";
src = fetchFromGitHub {
owner = "bsc-pm";
repo = "llvm";
rev = "refs/tags/github-release-${version}";
hash = "sha256-UgwMTUkM9Z87dDH205swZFBeFhrcbLAxginViG40pBM=";
hash = "sha256-ww9PpRmtz/M9IyLiZ8rAehx2UW4VpQt+svf4XfKBzKo=";
};
};

View File

@@ -1,22 +0,0 @@
{
stdenv
, lib
}:
stdenv.mkDerivation {
pname = "nixgen";
version = "0.0.1";
src = ./nixgen;
dontUnpack = true;
phases = [ "installPhase" ];
installPhase = ''
mkdir -p $out/bin
cp -a $src $out/bin/nixgen
'';
meta = {
description = "Quickly generate flake.nix from command line";
maintainers = with lib.maintainers.bsc; [ rarias ];
platforms = lib.platforms.linux;
license = lib.licenses.gpl3Plus;
};
}

View File

@@ -1,97 +0,0 @@
#!/bin/sh
#
# Copyright (c) 2025, Barcelona Supercomputing Center (BSC)
# SPDX-License-Identifier: GPL-3.0+
# Author: Rodrigo Arias Mallo <rodrigo.arias@bsc.es>
function usage() {
echo "USAGE: nixgen [-f] [package [...]] [-b package [...]]" >&2
echo " Generates a flake.nix file with the given packages." >&2
echo " After flake.nix is created, use 'nix develop' to enter the shell." >&2
echo "OPTIONS" >&2
echo " -f Overwrite existing flake.nix (default: no)." >&2
echo " packages... Add these packages to the shell." >&2
echo " -b packages... Add the dependencies needed to build these packages." >&2
echo "EXAMPLE" >&2
echo " $ nixgen ovni bigotes -b nosv tampi" >&2
echo " Adds the packages ovni and bigotes as well as all required dependencies" >&2
echo " to build nosv and tampi." >&2
echo "AUTHOR" >&2
echo " Rodrigo Arias Mallo <rodrigo.arias@bsc.es>" >&2
exit 1
}
mode=package
packages=
inputsFrom=
force=
if [[ $# -eq 0 ]]; then
usage
fi
while [[ $# -gt 0 ]]; do
case $1 in -b)
mode=build
shift
;;
-f)
force=1
shift
;;
-h)
usage
;;
-*|--*)
echo "error: unknown option $1" >&2
exit 1
;;
*)
if [ "$mode" == "package" ]; then
packages+="${packages:+ }$1"
else
inputsFrom+="${inputsFrom:+ }$1"
fi
shift
;;
esac
done
if [ ! "$force" -a -e flake.nix ]; then
echo "error: flake.nix exists, force overwrite with -f" >&2
exit 1
fi
cat > flake.nix <<EOF
{
inputs.jungle.url = "git+https://jungle.bsc.es/git/rarias/jungle";
outputs = { self, jungle }:
let
nixpkgs = jungle.inputs.nixpkgs;
customOverlay = (final: prev: {
# Example overlay, for now empty
});
pkgs = import nixpkgs {
system = "x86_64-linux";
overlays = [
# Apply jungle overlay to get our BSC custom packages
jungle.outputs.bscOverlay
# And on top apply our local changes to customize for cluster
customOverlay
];
};
in {
devShells.x86_64-linux.default = pkgs.mkShell {
pname = "devshell";
# Include these packages in the shell
packages = with pkgs; [
$packages
];
# The dependencies needed to build these packages will be also included
inputsFrom = with pkgs; [
$inputsFrom
];
};
};
}
EOF

View File

@@ -14,19 +14,19 @@
, useGit ? false
, gitUrl ? "ssh://git@gitlab-internal.bsc.es/nos-v/nodes.git"
, gitBranch ? "master"
, gitCommit ? "511489e71504a44381e0930562e7ac80ac69a848" # version-1.4
, gitCommit ? "6002ec9ae6eb876d962cc34366952a3b26599ba6"
}:
with lib;
let
release = rec {
version = "1.4";
version = "1.3";
src = fetchFromGitHub {
owner = "bsc-pm";
repo = "nodes";
rev = "version-${version}";
hash = "sha256-+lR/R0l3fGZO3XG7whMorFW2y2YZ0ZFnLeOHyQYrAsQ=";
hash = "sha256-cFb9pxcjtkMmH0CsGgUO9LTdXDNh7MCqicgGWawLrsU=";
};
};

View File

@@ -13,19 +13,19 @@
, useGit ? false
, gitUrl ? "git@gitlab-internal.bsc.es:nos-v/nos-v.git"
, gitBranch ? "master"
, gitCommit ? "1108e4786b58e0feb9a16fa093010b763eb2f8e8" # version 4.0.0
, gitCommit ? "9f47063873c3aa9d6a47482a82c5000a8c813dd8"
}:
with lib;
let
release = rec {
version = "4.0.0";
version = "3.2.0";
src = fetchFromGitHub {
owner = "bsc-pm";
repo = "nos-v";
rev = "${version}";
hash = "sha256-llaq73bd/YxLVKNlMebnUHKa4z3sdcsuDUoVwUxNuw8=";
hash = "sha256-yaz92426EM8trdkBJlISmAoG9KJCDTvoAW/HKrasvOw=";
};
};

View File

@@ -7,7 +7,7 @@
, useGit ? false
, gitBranch ? "master"
, gitUrl ? "ssh://git@bscpm04.bsc.es/rarias/ovni.git"
, gitCommit ? "06432668f346c8bdc1006fabc23e94ccb81b0d8b" # version 1.13.0
, gitCommit ? "e4f62382076f0cf0b1d08175cf57cc0bc51abc61"
, enableDebug ? false
# Only enable MPI if the build is native (fails on cross-compilation)
, useMpi ? (stdenv.buildPlatform.canExecute stdenv.hostPlatform)
@@ -15,13 +15,13 @@
let
release = rec {
version = "1.13.0";
version = "1.12.0";
src = fetchFromGitHub {
owner = "bsc-pm";
repo = "ovni";
rev = "${version}";
hash = "sha256-0l2ryIyWNiZqeYdVlnj/WnQGS3xFCY4ICG8JedX424w=";
} // { shortRev = "0643266"; };
hash = "sha256-H04JvsVKrdqr3ON7JhU0g17jjlg/jzQ7eTfx9vUNd3E=";
} // { shortRev = "a73afcf"; };
};
git = rec {