Add support to configure file limits for slurm module

Signed-off-by: Sanskar Bhushan <sanskar@obmondo.com>

Author: sbdtu5498

modules/enableit/profile/manifests/computing/slurm.pp

@@ -36,6 +36,8 @@
   Boolean                        $hwloc_enabled           = $::role::computing::slurm::hwloc_enabled,
   String                         $db_buffer_pool_size     = $::role::computing::slurm::db_buffer_pool_size,
   String                         $db_log_file_size        = $::role::computing::slurm::db_log_file_size,
+  Integer[1]                     $all_users_limit_nofile  = $::role::computing::slurm::all_users_limit_nofile,
+  Integer[1]                     $slurmd_limit_nofile     = $::role::computing::slurm::slurmd_limit_nofile,
 ) inherits ::profile::computing {
 
   # We manually install SLURM and munge packages because we're using packages
@@ -156,11 +158,21 @@
     })
 
     class { '::profile::computing::slurm::slurmd':
-      interface  => $interface,
-      node_cidrs => $node_cidrs,
+      interface           => $interface,
+      node_cidrs          => $node_cidrs,
+      slurmd_limit_nofile => $slurmd_limit_nofile,
     }
   }
 
+  # Configure PAM file descriptor limits for all users on SLURM nodes
+  pam::limits::fragment { 'slurm_nofile':
+    ensure => 'present',
+    list   => [
+      "* soft nofile ${all_users_limit_nofile}",
+      "* hard nofile ${all_users_limit_nofile}",
+    ],
+  }
+
   if !$enable {
     package::remove( ['slurm', 'slurm-slurmdbd', 'slurm-slurmctld', 'slurm-slurmd', 'munge'])
   }

modules/enableit/profile/manifests/computing/slurm/slurmd.pp

@@ -2,6 +2,7 @@
 class profile::computing::slurm::slurmd (
   Eit_types::SimpleString  $interface,
   Array[Eit_types::IPCIDR] $node_cidrs,
+  Integer[1]               $slurmd_limit_nofile,
 ) {
 
   firewall_multi {
@@ -21,5 +22,13 @@
       ;
   }
 
+  # Configure systemd drop-in file to set file descriptor limits for slurmd service
+  systemd::dropin_file { 'slurmd-90-nofile.conf':
+    unit    => 'slurmd.service',
+    filename => '90-nofile.conf',
+    ensure  => 'present',
+    content => "[Service]\nLimitNOFILE=${slurmd_limit_nofile}\n",
+  }
+
   include ::slurm::slurmd
 }

modules/enableit/role/manifests/computing/slurm.pp

@@ -61,6 +61,10 @@
 #
 # @param db_log_file_size Maximum size allocated for database log files.
 #
+# @param all_users_limit_nofile PAM `nofile` value for all users.
+#
+# @param slurmd_limit_nofile Systemd `LimitNOFILE` value for slurmd service.
+#
 # @groups authentication munge_key, jwt_key, encrypt_params
 #
 # @groups daemon_control slurmctld, slurmdbd, slurmd
@@ -77,6 +81,8 @@
 #
 # @groups db db_buffer_pool_size, db_log_file_size
 #
+# @groups limits all_users_limit_nofile, slurmd_limit_nofile
+#
 # @encrypt_params munge_key, jwt_key, slurm_gateway.*.bind_password
 #
 class role::computing::slurm (
@@ -112,6 +118,8 @@
   Optional[Eit_types::Slurm::Policy]  $slurm_policy            = undef,
   String                              $db_buffer_pool_size     = '256M',
   String                              $db_log_file_size        = '24M',
+  Integer[1]                          $all_users_limit_nofile  = 10000,
+  Integer[1]                          $slurmd_limit_nofile     = 10000,
   Eit_types::Encrypt::Params          $encrypt_params          = ['munge_key', 'jwt_key','slurm_gateway.*.bind_password']
 
 ) inherits ::role::computing {