Compare commits

..

7 Commits

5 changed files with 55 additions and 23 deletions

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

@@ -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

View File

@@ -43,7 +43,7 @@ stdenv.mkDerivation rec {
configureFlags = [
"--with-infiniband=${rdma-core-all}"
"--with-mpi=yes" # fixes mpi detection when cross-compiling
"--with-mpi=yes"
"--with-slurm"
"CFLAGS=-fPIC"
"CXXFLAGS=-fPIC"
@@ -70,6 +70,6 @@ stdenv.mkDerivation rec {
maintainers = with lib.maintainers.bsc; [ rarias ];
platforms = lib.platforms.linux;
license = lib.licenses.gpl3Plus;
cross = false; # infiniband detection does not work
cross = false;
};
}

View File

@@ -17,7 +17,6 @@
, gitUrl ? "ssh://git@bscpm04.bsc.es/llvm-ompss/llvm-mono.git"
, gitBranch ? "master"
, gitCommit ? "872ba63f86edaefc9787984ef3fae9f2f94e0124" # github-release-2025.11
, buildLlvmPackages
}:
let
@@ -120,8 +119,6 @@ in stdenv.mkDerivation {
"-DCMAKE_INSTALL_RPATH=${zlib}/lib:${gcc.cc.lib}/lib"
"-DLLVM_APPEND_VC_REV=ON"
"-DLLVM_FORCE_VC_REVISION=${source.version}"
"-DLLVM_TABLEGEN_EXE=${buildLlvmPackages.tblgen}/bin/llvm-tblgen"
"-DCLANG_TABLEGEN=${buildLlvmPackages.tblgen}/bin/clang-tblgen"
)
'';