Update slurm sinfo filter to discard empty cluster

This commit is contained in:
Benoît Seignovert 2024-02-14 16:31:52 +01:00
parent 867e217d44
commit 52b96548c2
Signed by: Benoît Seignovert
GPG key ID: F5D8895227D18A0B
2 changed files with 64 additions and 29 deletions

View file

@ -59,7 +59,7 @@ class SlurmNode:
self.gpu = SlurmGpu(*re.findall(r'gpu:(\w+):(\d+)', gres)[0] if 'gpu:' in gres else []) self.gpu = SlurmGpu(*re.findall(r'gpu:(\w+):(\d+)', gres)[0] if 'gpu:' in gres else [])
def _sinfo_run(username: str = None) -> str: def sinfo_run(username: str = None) -> str:
"""SLURM SINFO run command.""" """SLURM SINFO run command."""
flags = '--federation --noheader --responding' flags = '--federation --noheader --responding'
fmt = 'Cluster,PartitionName,NodeHost,StateLong,CPUsState,Memory,Gres' fmt = 'Cluster,PartitionName,NodeHost,StateLong,CPUsState,Memory,Gres'
@ -71,20 +71,18 @@ def _sinfo_run(username: str = None) -> str:
return subprocess.check_output(shlex.split(cmd, posix=False)).decode('utf-8') return subprocess.check_output(shlex.split(cmd, posix=False)).decode('utf-8')
def _sinfo_reader(result) -> list: def sinfo_reader(result: str) -> list:
"""SLURM SINFO reader.""" """SLURM SINFO reader."""
return [SlurmNode(*re.findall('.{20}', node)) for node in result.splitlines()] return [SlurmNode(*re.findall('.{20}', node)) for node in result.splitlines()]
def sinfo(username: str = None, with_states=('idle', 'mixed')) -> dict: def sinfo_filter(resources: list, with_states=('idle', 'mixed')) -> dict:
"""SLURM SINFO resources available with a given state(s). """SLURM SINFO filtered resources available with a given state(s).
Grouped by cluster and partition names. Grouped by cluster and partition names.
""" """
resources = _sinfo_reader(_sinfo_run(username=username)) resources = {
return {
cluster: { cluster: {
partition: available partition: available
for partition, nodes in groupby(partitions, key=attrgetter('partition')) for partition, nodes in groupby(partitions, key=attrgetter('partition'))
@ -92,3 +90,10 @@ def sinfo(username: str = None, with_states=('idle', 'mixed')) -> dict:
} }
for cluster, partitions in groupby(resources, key=attrgetter('cluster')) for cluster, partitions in groupby(resources, key=attrgetter('cluster'))
} }
return {key: values for key, values in resources.items() if values}
def sinfo(username: str = None, with_states=('idle', 'mixed')) -> dict:
"""SLURM SINFO resources available for a given user."""
return sinfo_filter(sinfo_reader(sinfo_run(username=username)), with_states=with_states)

View file

@ -6,13 +6,15 @@ from glicid_spawner.slurm import (
SlurmCpu, SlurmCpu,
SlurmGpu, SlurmGpu,
SlurmNode, SlurmNode,
_sinfo_reader,
_sinfo_run,
sinfo, sinfo,
sinfo_filter,
sinfo_reader,
sinfo_run,
subprocess, subprocess,
) )
DATA = Path(__file__).parent / 'data' DATA = Path(__file__).parent / 'data'
SINFO = (DATA / 'sinfo.txt').read_text()
def test_slurm_dataclasses(): def test_slurm_dataclasses():
@ -48,7 +50,7 @@ def test_slurm_sinfo_run(monkeypatch):
"""Test SLURM SINFO run command.""" """Test SLURM SINFO run command."""
monkeypatch.setattr(subprocess, 'check_output', lambda cmd: ' '.join(cmd).encode()) monkeypatch.setattr(subprocess, 'check_output', lambda cmd: ' '.join(cmd).encode())
assert _sinfo_run() == ( assert sinfo_run() == (
'sinfo ' 'sinfo '
'--federation ' '--federation '
'--noheader ' '--noheader '
@ -56,7 +58,7 @@ def test_slurm_sinfo_run(monkeypatch):
'--Format=Cluster,PartitionName,NodeHost,StateLong,CPUsState,Memory,Gres' '--Format=Cluster,PartitionName,NodeHost,StateLong,CPUsState,Memory,Gres'
) )
assert _sinfo_run(username='john-doe') == ( assert sinfo_run(username='john-doe') == (
'su - john-doe -c "' 'su - john-doe -c "'
'sinfo ' 'sinfo '
'--federation ' '--federation '
@ -69,7 +71,7 @@ def test_slurm_sinfo_run(monkeypatch):
def test_slurm_sinfo_reader(): def test_slurm_sinfo_reader():
"""Test SLURM SINFO reader.""" """Test SLURM SINFO reader."""
nodes = _sinfo_reader((DATA / 'sinfo.txt').read_text()) nodes = sinfo_reader(SINFO)
for node in nodes: for node in nodes:
assert isinstance(node, SlurmNode) assert isinstance(node, SlurmNode)
@ -113,11 +115,11 @@ def test_slurm_sinfo_reader():
assert [node.gpu.nb for node in nodes if node.gpu] == [2, 2, 2, 4, 1, 1, 2, 4, 2] assert [node.gpu.nb for node in nodes if node.gpu] == [2, 2, 2, 4, 1, 1, 2, 4, 2]
def test_slurm_sinfo_resources(monkeypatch): def test_slurm_sinfo_filter(monkeypatch):
"""Test SLURM SINFO resources.""" """Test SLURM SINFO filtered resources."""
monkeypatch.setattr(subprocess, 'check_output', lambda _: (DATA / 'sinfo.txt').read_bytes()) resources = sinfo_reader(SINFO)
clusters = sinfo() clusters = sinfo_filter(resources)
assert isinstance(clusters, dict) assert isinstance(clusters, dict)
assert len(clusters) == 3 assert len(clusters) == 3
@ -125,21 +127,49 @@ def test_slurm_sinfo_resources(monkeypatch):
assert [len(partitions) for partitions in clusters.values()] == [2, 2, 2] assert [len(partitions) for partitions in clusters.values()] == [2, 2, 2]
nautilus = clusters['nautilus'] partitions = clusters['nautilus']
assert isinstance(nautilus, dict) assert isinstance(partitions, dict)
assert len(nautilus) == 2 assert len(partitions) == 2
assert list(nautilus) == ['gpu', 'all'] assert list(partitions) == ['gpu', 'all']
gpus = nautilus['gpu'] gpu_nodes = partitions['gpu']
assert len(gpus) == 2 assert len(gpu_nodes) == 2
assert [partition.hostname for partition in gpus] == ['gnode1', 'gnode2'] assert [node.hostname for node in gpu_nodes] == ['gnode1', 'gnode2']
assert [partition.cpu.allocated for partition in gpus] == [4, 0] assert [node.cpu.allocated for node in gpu_nodes] == [4, 0]
assert [partition.cpu.idle for partition in gpus] == [92, 96] assert [node.cpu.idle for node in gpu_nodes] == [92, 96]
assert [partition.mem for partition in gpus] == [768, 256] assert [node.mem for node in gpu_nodes] == [768, 256]
assert [partition.gpu.name for partition in gpus] == ['A100', 'A100'] assert [node.gpu.name for node in gpu_nodes] == ['A100', 'A100']
assert [partition.gpu.nb for partition in gpus] == [1, 2] assert [node.gpu.nb for node in gpu_nodes] == [1, 2]
# Get only `idle` nodes # Get only `idle` nodes
assert [len(partitions) for partitions in sinfo(with_states=('idle')).values()] == [1, 2, 1] clusters = sinfo_filter(resources, with_states=('idle'))
assert list(clusters) == [None, 'nautilus', 'waves']
assert [len(partitions) for partitions in clusters.values()] == [1, 2, 1]
# Discard clusters without partition available
clusters = sinfo_filter(resources, with_states=('completing'))
assert list(clusters) == ['nautilus']
assert [len(partitions) for partitions in clusters.values()] == [1]
def test_slurm_sinfo_resources(monkeypatch):
"""Test SLURM SINFO resources."""
monkeypatch.setattr(subprocess, 'check_output', lambda _: SINFO.encode())
clusters = sinfo(username='john-doe', with_states=('completing'))
assert list(clusters) == ['nautilus']
partitions = clusters['nautilus']
assert list(partitions) == ['standard']
std_nodes = partitions['standard']
assert std_nodes == [
SlurmNode(*'nautilus standard cnode001 completing 0/96/0/96 384000 (null)'.split())
]