alioth/nixpkgs
3
0
Fork 0
forked from mirrors/nixpkgs
Code Releases Activity

nixos/slurm: convert test from perl to python

Browse source
This commit is contained in:
Markus Kowalewski 2019-11-10 23:07:08 +01:00
parent 472e165b56
commit 9b28dbd36a
No known key found for this signature in database
GPG key ID: D865C8A91D7025EB
1 changed files with 37 additions and 43 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

80
nixos/tests/slurm.nix
View file

@ -1,4 +1,4 @@
import ./make-test.nix ({ lib, ... }:
import ./make-test-python.nix ({ lib, ... }:
let
mungekey = "mungeverryweakkeybuteasytointegratoinatest";
@ -85,63 +85,57 @@ in {
testScript =
''
startAll;
start_all()
# Set up authentification across the cluster
foreach my $node (($submit,$control,$dbd,$node1,$node2,$node3))
{
$node->waitForUnit("default.target");
for node in [submit, control, dbd, node1, node2, node3]:
$node->succeed("mkdir /etc/munge");
$node->succeed("echo '${mungekey}' > /etc/munge/munge.key");
$node->succeed("chmod 0400 /etc/munge/munge.key");
$node->succeed("chown munge:munge /etc/munge/munge.key");
$node->succeed("systemctl restart munged");
node.wait_for_unit("default.target")
node.succeed("mkdir /etc/munge")
node.succeed(
"echo '${mungekey}' > /etc/munge/munge.key"
)
node.succeed("chmod 0400 /etc/munge/munge.key")
node.succeed("chown munge:munge /etc/munge/munge.key")
node.succeed("systemctl restart munged")
node.wait_for_unit("munged")
$node->waitForUnit("munged");
};
# Restart the services since they have probably failed due to the munge init
# failure
subtest "can_start_slurmdbd", sub {
$dbd->succeed("systemctl restart slurmdbd");
$dbd->waitForUnit("slurmdbd.service");
$dbd->waitForOpenPort(6819);
};
with subtest("can_start_slurmdbd"):
dbd.succeed("systemctl restart slurmdbd")
dbd.wait_for_unit("slurmdbd.service")
dbd.wait_for_open_port(6819)
# there needs to be an entry for the current
# cluster in the database before slurmctld is restarted
subtest "add_account", sub {
$control->succeed("sacctmgr -i add cluster default");
# check for cluster entry
$control->succeed("sacctmgr list cluster | awk '{ print \$1 }' | grep default");
};
with subtest("add_account"):
control.succeed("sacctmgr -i add cluster default")
# check for cluster entry
control.succeed("sacctmgr list cluster | awk '{ print $1 }' | grep default")
subtest "can_start_slurmctld", sub {
$control->succeed("systemctl restart slurmctld");
$control->waitForUnit("slurmctld.service");
};
with subtest("can_start_slurmctld"):
control.succeed("systemctl restart slurmctld")
control.waitForUnit("slurmctld.service")
subtest "can_start_slurmd", sub {
foreach my $node (($node1,$node2,$node3))
{
$node->succeed("systemctl restart slurmd.service");
$node->waitForUnit("slurmd");
}
};
with subtest("can_start_slurmd"):
for node in [node1, node2, node3]:
node.succeed("systemctl restart slurmd.service")
node.wait_for_unit("slurmd")
# Test that the cluster works and can distribute jobs;
subtest "run_distributed_command", sub {
# Run `hostname` on 3 nodes of the partition (so on all the 3 nodes).
# The output must contain the 3 different names
$submit->succeed("srun -N 3 hostname | sort | uniq | wc -l | xargs test 3 -eq");
};
with subtest("run_distributed_command"):
# Run `hostname` on 3 nodes of the partition (so on all the 3 nodes).
# The output must contain the 3 different names
submit.succeed("srun -N 3 hostname | sort | uniq | wc -l | xargs test 3 -eq")
subtest "check_slurm_dbd", sub {
# find the srun job from above in the database
sleep 5;
$control->succeed("sacct | grep hostname");
};
with subtest("check_slurm_dbd"):
# find the srun job from above in the database
control.succeed("sleep 5")
control.succeed("sacct | grep hostname")
'';
})