modules.services.slurm: fix

devices/pc/default.nix

@@ -136,7 +136,7 @@ inputs:
             memoryMB = 90112;
             gpus."4060" = 1;
           };
-          partitions.default = [ "pc" ];
+          partitions.localhost = [ "pc" ];
         };
         ollama = {};
         waydroid = {};

devices/srv1/default.nix

@@ -46,20 +46,20 @@ inputs:
             };
             srv1-node2 =
             {
-              name = "n1"; address = "192.168.178.2";
+              name = "n2"; address = "192.168.178.3";
               cpu = { sockets = 4; cores = 8; threads = 2; };
               memoryMB = 30720;
             };
             srv1-node3 =
             {
-              name = "n1"; address = "192.168.178.2";
+              name = "n3"; address = "192.168.178.4";
               cpu = { sockets = 4; cores = 8; threads = 2; };
               memoryMB = 30720;
             };
           };
           partitions =
           {
-            default = [ "srv1-node0" ];
+            localhost = [ "srv1-node0" ];
             old = [ "srv1-node1" "srv1-node2" "srv1-node3" ];
           };
         };

devices/srv1/node0/default.nix

@@ -55,5 +55,8 @@ inputs:
         options = [ "rbind" ];
       };
     };
+    # without this, tproxy does not work
+    # TODO: why?
+    networking.firewall.trustedInterfaces = [ "eno146" ];
   };
 }

devices/srv1/node3/default.nix

@@ -8,7 +8,7 @@ inputs:
       {
         nixpkgs.march = "broadwell";
         networking.networkd.static.eno2 =
-          { ip = "192.168.178.3"; mask = 24; gateway = "192.168.178.1"; dns = "192.168.178.1"; };
+          { ip = "192.168.178.4"; mask = 24; gateway = "192.168.178.1"; dns = "192.168.178.1"; };
         cluster.nodeType = "worker";
         initrd.sshd.enable = true;
         nix.remote.slave.enable = true;

devices/xmupc1/default.nix

@@ -81,7 +81,7 @@ inputs:
             memoryMB = 94208;
             gpus = { "p5000" = 1; "3090" = 1; "4090" = 1; };
           };
-          partitions.default = [ "xmupc1" ];
+          partitions.localhost = [ "xmupc1" ];
         };
         xrdp = { enable = true; hostname = [ "xmupc1.chn.moe" ]; };
         samba =

devices/xmupc2/default.nix

@@ -80,7 +80,7 @@ inputs:
             memoryMB = 253952;
             gpus."4090" = 1;
           };
-          partitions.default = [ "xmupc2" ];
+          partitions.localhost = [ "xmupc2" ];
         };
         xrdp = { enable = true; hostname = [ "xmupc2.chn.moe" ]; };
         samba = { enable = true; hostsAllowed = ""; shares = { home.path = "/home"; root.path = "/"; }; };

modules/services/slurm.nix

@@ -22,7 +22,7 @@ inputs:
       gpus = mkOption { type = types.nullOr (types.attrsOf types.ints.unsigned); default = null; };
     };}));};
     partitions = mkOption { type = types.attrsOf (types.listOf types.nonEmptyStr); default = {}; };
-    defaultPartition = mkOption { type = types.nonEmptyStr; default = "default"; };
+    defaultPartition = mkOption { type = types.nonEmptyStr; default = "localhost"; };
   };
   config = let inherit (inputs.config.nixos.services) slurm; in inputs.lib.mkIf slurm.enable (inputs.lib.mkMerge
   [
@@ -81,6 +81,7 @@ inputs:
             (inputs.localLib.attrsToList slurm.partitions);
           procTrackType = "proctrack/cgroup";
           controlMachine = slurm.master;
+          controlAddr = slurm.node.${slurm.master}.address;
           extraConfig =
           ''
             SelectType=select/cons_tres