commit c30bd1c1316d778ad7eb53973232f1a929f6e878 · pyrox.dev/nixpkgs

+56 -9

nixos/modules/services/computing/slurm/slurm.nix

···

       6
       6
        
       

     

       7
       7
        
         cfg = config.services.slurm;

     

       8
       8
        
         # configuration file can be generated by http://slurm.schedmd.com/configurator.html

     

       9
       9
       -
         configFile = pkgs.writeText "slurm.conf"

     

       9
       9
       +
         configFile = pkgs.writeTextDir "slurm.conf"

     

       10
       10
        
           ''

     

       11
       11
        
             ${optionalString (cfg.controlMachine != null) ''controlMachine=${cfg.controlMachine}''}

     

       12
       12
        
             ${optionalString (cfg.controlAddr != null) ''controlAddr=${cfg.controlAddr}''}

     
···

       17
       17
        
             ${cfg.extraConfig}

     

       18
       18
        
           '';

     

       19
       19
        
       

     

       20
       20
       -
         plugStackConfig = pkgs.writeText "plugstack.conf"

     

       20
       20
       +
         plugStackConfig = pkgs.writeTextDir "plugstack.conf"

     

       21
       21
        
           ''

     

       22
       22
        
             ${optionalString cfg.enableSrunX11 ''optional ${pkgs.slurm-spank-x11}/lib/x11.so''}

     

       23
       23
       +
             ${cfg.extraPlugstackConfig}

     

       23
       24
        
           '';

     

       25
       25
       +
       

     

       26
       26
       +
       

     

       27
       27
       +
         cgroupConfig = pkgs.writeTextDir "cgroup.conf"

     

       28
       28
       +
          ''

     

       29
       29
       +
            ${cfg.extraCgroupConfig}

     

       30
       30
       +
          '';

     

       31
       31
       +
       

     

       32
       32
       +
         # slurm expects some additional config files to be

     

       33
       33
       +
         # in the same directory as slurm.conf

     

       34
       34
       +
         etcSlurm = pkgs.symlinkJoin {

     

       35
       35
       +
           name = "etc-slurm";

     

       36
       36
       +
           paths = [ configFile cgroupConfig plugStackConfig ];

     

       37
       37
       +
         };

     

       38
       38
       +
       

     

       24
       39
        
       in

     

       25
       40
        
       

     

       26
       41
        
       {

     
···

       46
       61
        
       

     

       47
       62
        
             client = {

     

       48
       63
        
               enable = mkEnableOption "slurm client daemon";

     

       64
       64
       +
             };

     

       49
       65
        
       

     

       66
       66
       +
             enableStools = mkOption {

     

       67
       67
       +
               type = types.bool;

     

       68
       68
       +
               default = false;

     

       69
       69
       +
               description = ''

     

       70
       70
       +
                 Wether to provide a slurm.conf file.

     

       71
       71
       +
                 Enable this option if you do not run a slurm daemon on this host

     

       72
       72
       +
                 (i.e. <literal>server.enable</literal> and <literal>client.enable</literal> are <literal>false</literal>)

     

       73
       73
       +
                 but you still want to run slurm commands from this host.

     

       74
       74
       +
               '';

     

       50
       75
        
             };

     

       51
       76
        
       

     

       52
       77
        
             package = mkOption {

     
···

       97
       122
        
               example = "debug Nodes=linux[1-32] Default=YES MaxTime=INFINITE State=UP";

     

       98
       123
        
               description = ''

     

       99
       124
        
                 Name by which the partition may be referenced. Note that now you have

     

       100
       100
       -
                 to write patrition's parameters after the name.

     

       125
       125
       +
                 to write the partition's parameters after the name.

     

       101
       126
        
               '';

     

       102
       127
        
             };

     

       103
       128
        
       

     
···

       107
       132
        
               description = ''

     

       108
       133
        
                 If enabled srun will accept the option "--x11" to allow for X11 forwarding

     

       109
       134
        
                 from within an interactive session or a batch job. This activates the

     

       110
       110
       -
                 slurm-spank-x11 module. Note that this requires 'services.openssh.forwardX11'

     

       111
       111
       -
                 to be enabled on the compute nodes.

     

       135
       135
       +
                 slurm-spank-x11 module. Note that this option also enables

     

       136
       136
       +
                 'services.openssh.forwardX11' on the client.

     

       137
       137
       +
       

     

       138
       138
       +
                 This option requires slurm to be compiled without native X11 support.

     

       112
       139
        
               '';

     

       113
       140
        
             };

     

       114
       141
        
       

     
···

       130
       157
        
                 the end of the slurm configuration file.

     

       131
       158
        
               '';

     

       132
       159
        
             };

     

       160
       160
       +
       

     

       161
       161
       +
             extraPlugstackConfig = mkOption {

     

       162
       162
       +
               default = "";

     

       163
       163
       +
               type = types.lines;

     

       164
       164
       +
               description = ''

     

       165
       165
       +
                 Extra configuration that will be added to the end of <literal>plugstack.conf</literal>.

     

       166
       166
       +
               '';

     

       167
       167
       +
             };

     

       168
       168
       +
       

     

       169
       169
       +
             extraCgroupConfig = mkOption {

     

       170
       170
       +
               default = "";

     

       171
       171
       +
               type = types.lines;

     

       172
       172
       +
               description = ''

     

       173
       173
       +
                 Extra configuration for <literal>cgroup.conf</literal>. This file is

     

       174
       174
       +
                 used when <literal>procTrackType=proctrack/cgroup</literal>.

     

       175
       175
       +
               '';

     

       176
       176
       +
             };

     

       133
       177
        
           };

     

       134
       178
        
       

     

       135
       179
        
         };

     
···

       142
       186
        
             wrappedSlurm = pkgs.stdenv.mkDerivation {

     

       143
       187
        
               name = "wrappedSlurm";

     

       144
       188
        
       

     

       145
       145
       -
               propagatedBuildInputs = [ cfg.package configFile ];

     

       146
       146
       -
       

     

       147
       189
        
               builder = pkgs.writeText "builder.sh" ''

     

       148
       190
        
                 source $stdenv/setup

     

       149
       191
        
                 mkdir -p $out/bin

     
···

       155
       197
        
                 #!/bin/sh

     

       156
       198
        
                 if [ -z "$SLURM_CONF" ]

     

       157
       199
        
                 then

     

       158
       158
       -
                   SLURM_CONF="${configFile}" "$EXE" "\$@"

     

       200
       200
       +
                   SLURM_CONF="${etcSlurm}/slurm.conf" "$EXE" "\$@"

     

       159
       201
        
                 else

     

       160
       202
        
                   "$EXE" "\$0"

     

       161
       203
        
                 fi

     

       162
       204
        
                 EOT

     

       163
       205
        
                   chmod +x "$wrappername"

     

       164
       206
        
                 done

     

       207
       207
       +
       

     

       208
       208
       +
                 mkdir -p $out/share

     

       209
       209
       +
                 ln -s ${getBin cfg.package}/share/man $out/share/man

     

       165
       210
        
               '';

     

       166
       211
        
             };

     

       167
       212
        
       

     

       168
       168
       -
         in mkIf (cfg.client.enable || cfg.server.enable) {

     

       213
       213
       +
         in mkIf (cfg.enableStools || cfg.client.enable || cfg.server.enable) {

     

       169
       214
        
       

     

       170
       215
        
           environment.systemPackages = [ wrappedSlurm ];

     

       171
       216
        
       

     
···

       189
       234
        
               mkdir -p /var/spool

     

       190
       235
        
             '';

     

       191
       236
        
           };

     

       237
       237
       +
       

     

       238
       238
       +
           services.openssh.forwardX11 = mkIf cfg.client.enable (mkDefault true);

     

       192
       239
        
       

     

       193
       240
        
           systemd.services.slurmctld = mkIf (cfg.server.enable) {

     

       194
       241
        
             path = with pkgs; [ wrappedSlurm munge coreutils ]

+18 -5

nixos/tests/slurm.nix

···

       1
       1
        
       import ./make-test.nix ({ pkgs, ... }:

     

       2
       2
        
       let mungekey = "mungeverryweakkeybuteasytointegratoinatest";

     

       3
       3
        
           slurmconfig = {

     

       4
       4
       -
             client.enable = true;

     

       5
       4
        
             controlMachine = "control";

     

       6
       5
        
             nodeName = ''

     

       7
       6
        
               control

     
···

       20
       19
        
               # TODO slrumd port and slurmctld port should be configurations and

     

       21
       20
        
               # automatically allowed by the  firewall.

     

       22
       21
        
               networking.firewall.enable = false;

     

       23
       23
       -
               services.slurm = slurmconfig;

     

       22
       22
       +
               services.slurm = {

     

       23
       23
       +
                 client.enable = true;

     

       24
       24
       +
               } // slurmconfig;

     

       24
       25
        
             };

     

       25
       26
        
           in {

     

       27
       27
       +
       

     

       26
       28
        
           control =

     

       27
       29
        
             { config, pkgs, ...}:

     

       28
       30
        
             {

     
···

       31
       33
        
                 server.enable = true;

     

       32
       34
        
               } // slurmconfig;

     

       33
       35
        
             };

     

       36
       36
       +
       

     

       37
       37
       +
           submit =

     

       38
       38
       +
             { config, pkgs, ...}:

     

       39
       39
       +
             {

     

       40
       40
       +
               networking.firewall.enable = false;

     

       41
       41
       +
               services.slurm = {

     

       42
       42
       +
                 enableStools = true;

     

       43
       43
       +
               } // slurmconfig;

     

       44
       44
       +
             };

     

       45
       45
       +
       

     

       34
       46
        
           node1 = computeNode;

     

       35
       47
        
           node2 = computeNode;

     

       36
       48
        
           node3 = computeNode;

     

       37
       49
        
         };

     

       50
       50
       +
       

     

       38
       51
        
       

     

       39
       52
        
         testScript =

     

       40
       53
        
         ''

     

       41
       54
        
         startAll;

     

       42
       55
        
       

     

       43
       56
        
         # Set up authentification across the cluster

     

       44
       44
       -
         foreach my $node (($control,$node1,$node2,$node3))

     

       57
       57
       +
         foreach my $node (($submit,$control,$node1,$node2,$node3))

     

       45
       58
        
         {

     

       46
       59
        
           $node->waitForUnit("default.target");

     

       47
       60
        
       

     
···

       60
       73
        
         };

     

       61
       74
        
       

     

       62
       75
        
         subtest "can_start_slurmd", sub {

     

       63
       63
       -
           foreach my $node (($control,$node1,$node2,$node3))

     

       76
       76
       +
           foreach my $node (($node1,$node2,$node3))

     

       64
       77
        
           {

     

       65
       78
        
             $node->succeed("systemctl restart slurmd.service");

     

       66
       79
        
             $node->waitForUnit("slurmd");

     
···

       72
       85
        
         subtest "run_distributed_command", sub {

     

       73
       86
        
           # Run `hostname` on 3 nodes of the partition (so on all the 3 nodes).

     

       74
       87
        
           # The output must contain the 3 different names

     

       75
       75
       -
           $control->succeed("srun -N 3 hostname | sort | uniq | wc -l | xargs test 3 -eq");

     

       88
       88
       +
           $submit->succeed("srun -N 3 hostname | sort | uniq | wc -l | xargs test 3 -eq");

     

       76
       89
        
         };

     

       77
       90
        
         '';

     

       78
       91
        
       })

+21 -6

pkgs/servers/computing/slurm/default.nix

···

       1
       1
       -
       { stdenv, fetchurl, pkgconfig, libtool, curl, python, munge, perl, pam, openssl

     

       1
       1
       +
       { stdenv, fetchurl, pkgconfig, libtool, curl

     

       2
       2
       +
       , python, munge, perl, pam, openssl

     

       2
       3
        
       , ncurses, mysql, gtk2, lua, hwloc, numactl

     

       4
       4
       +
       , readline, freeipmi, libssh2, xorg

     

       5
       5
       +
       # enable internal X11 support via libssh2

     

       6
       6
       +
       , enableX11 ? true

     

       3
       7
        
       }:

     

       4
       8
        
       

     

       5
       9
        
       stdenv.mkDerivation rec {

     
···

       13
       17
        
       

     

       14
       18
        
         outputs = [ "out" "dev" ];

     

       15
       19
        
       

     

       20
       20
       +
         prePatch = stdenv.lib.optional enableX11 ''

     

       21
       21
       +
           substituteInPlace src/common/x11_util.c \

     

       22
       22
       +
               --replace '"/usr/bin/xauth"' '"${xorg.xauth}/bin/xauth"'

     

       23
       23
       +
         '';

     

       24
       24
       +
       

     

       16
       25
        
         # nixos test fails to start slurmd with 'undefined symbol: slurm_job_preempt_mode'

     

       17
       26
        
         # https://groups.google.com/forum/#!topic/slurm-devel/QHOajQ84_Es

     

       18
       27
        
         # this doesn't fix tests completely at least makes slurmd to launch

     
···

       20
       29
        
       

     

       21
       30
        
         nativeBuildInputs = [ pkgconfig libtool ];

     

       22
       31
        
         buildInputs = [

     

       23
       23
       -
           curl python munge perl pam openssl mysql.connector-c ncurses gtk2 lua hwloc numactl

     

       24
       24
       -
         ];

     

       32
       32
       +
           curl python munge perl pam openssl

     

       33
       33
       +
             mysql.connector-c ncurses gtk2

     

       34
       34
       +
             lua hwloc numactl readline freeipmi

     

       35
       35
       +
         ] ++ stdenv.lib.optionals enableX11 [ libssh2 xorg.xauth ];

     

       25
       36
        
       

     

       26
       26
       -
         configureFlags =

     

       37
       37
       +
         configureFlags = with stdenv.lib;

     

       27
       38
        
           [ "--with-munge=${munge}"

     

       28
       39
        
             "--with-ssl=${openssl.dev}"

     

       40
       40
       +
             "--with-hwloc=${hwloc.dev}"

     

       41
       41
       +
             "--with-freeipmi=${freeipmi}"

     

       29
       42
        
             "--sysconfdir=/etc/slurm"

     

       30
       30
       -
           ] ++ stdenv.lib.optional (gtk2 == null)  "--disable-gtktest";

     

       43
       43
       +
           ] ++ (optional (gtk2 == null)  "--disable-gtktest")

     

       44
       44
       +
             ++ (optional enableX11 "--with-libssh2=${libssh2.dev}");

     

       45
       45
       +
       

     

       31
       46
        
       

     

       32
       47
        
         preConfigure = ''

     

       33
       48
        
           patchShebangs ./doc/html/shtml2html.py

     
···

       45
       60
        
           description = "Simple Linux Utility for Resource Management";

     

       46
       61
        
           platforms = platforms.linux;

     

       47
       62
        
           license = licenses.gpl2;

     

       48
       48
       -
           maintainers = [ maintainers.jagajaga ];

     

       63
       63
       +
           maintainers = with maintainers; [ jagajaga markuskowa ];

     

       49
       64
        
         };

     

       50
       65
        
       }