pyrox.dev
1import ./make-test.nix ({ ... }:
2let mungekey = "mungeverryweakkeybuteasytointegratoinatest";
3 slurmconfig = {
4 controlMachine = "control";
5 nodeName = ''
6 control
7 NodeName=node[1-3] CPUs=1 State=UNKNOWN
8 '';
9 partitionName = "debug Nodes=node[1-3] Default=YES MaxTime=INFINITE State=UP";
10 };
11in {
12 name = "slurm";
13
14 nodes =
15 let
16 computeNode =
17 { ...}:
18 {
19 # TODO slrumd port and slurmctld port should be configurations and
20 # automatically allowed by the firewall.
21 networking.firewall.enable = false;
22 services.slurm = {
23 client.enable = true;
24 } // slurmconfig;
25 };
26 in {
27
28 control =
29 { ...}:
30 {
31 networking.firewall.enable = false;
32 services.slurm = {
33 server.enable = true;
34 } // slurmconfig;
35 };
36
37 submit =
38 { ...}:
39 {
40 networking.firewall.enable = false;
41 services.slurm = {
42 enableStools = true;
43 } // slurmconfig;
44 };
45
46 node1 = computeNode;
47 node2 = computeNode;
48 node3 = computeNode;
49 };
50
51
52 testScript =
53 ''
54 startAll;
55
56 # Set up authentification across the cluster
57 foreach my $node (($submit,$control,$node1,$node2,$node3))
58 {
59 $node->waitForUnit("default.target");
60
61 $node->succeed("mkdir /etc/munge");
62 $node->succeed("echo '${mungekey}' > /etc/munge/munge.key");
63 $node->succeed("chmod 0400 /etc/munge/munge.key");
64 $node->succeed("chown munge:munge /etc/munge/munge.key");
65 $node->succeed("systemctl restart munged");
66 }
67
68 # Restart the services since they have probably failed due to the munge init
69 # failure
70
71 subtest "can_start_slurmctld", sub {
72 $control->succeed("systemctl restart slurmctld");
73 $control->waitForUnit("slurmctld.service");
74 };
75
76 subtest "can_start_slurmd", sub {
77 foreach my $node (($node1,$node2,$node3))
78 {
79 $node->succeed("systemctl restart slurmd.service");
80 $node->waitForUnit("slurmd");
81 }
82 };
83
84 # Test that the cluster work and can distribute jobs;
85
86 subtest "run_distributed_command", sub {
87 # Run `hostname` on 3 nodes of the partition (so on all the 3 nodes).
88 # The output must contain the 3 different names
89 $submit->succeed("srun -N 3 hostname | sort | uniq | wc -l | xargs test 3 -eq");
90 };
91 '';
92})