def parse_slurm_conf(slurm_conf_file, part):
cluster = []
r_part = re.compile('PartitionName=(\S+).*?Nodes=(\S+)')
r_node = re.compile("(?<=NodeName=)\S+")
try:
for line in open(slurm_conf_file, 'r'):
if line[0] != '#':
m = r_node.search(line)
if m:
exp_nodelist = hostlist.expand_hostlist(m.group(0))
if not part:
cluster.extend(exp_nodelist)
elif part:
m = r_part.search(line)
if m and m.group(1) in part:
cluster.extend(hostlist.expand_hostlist(m.group(2)))
if len(part) == 1:
break
except IOError:
print "Error: cannot open Slurm conf file at", slurm_conf_file
return cluster
def parse_slurm_conf(slurm_conf_file,part,node_dict):
cluster=[]
r_part=re.compile('PartitionName=(\S+).*?Nodes=(\S+)')
r_node=re.compile("(?<=NodeName=)\S+")
try:
for line in open(slurm_conf_file,'r'):
if line[0]!='#':
m=r_node.search(line)
if m:
exp_nodelist=hostlist.expand_hostlist(m.group(0))
parse_slurm_conf_cores(line,exp_nodelist,node_dict)
if not part:
cluster.extend(exp_nodelist)
elif part:
m=r_part.search(line)
if m and m.group(1) in part:
cluster.extend(hostlist.expand_hostlist(m.group(2)))
if len(part)==1:
break
except IOError:
print "Error: cannot open Slurm conf file at",slurm_conf_file
return cluster
def create_cluster(self, name, names, ips, management):
'''
creates a cluster with the given name specification ip specifications, and the identification of management node.
:param name: name of the cluster
:param names: the names of the cluster servers. 'i[001-003]'. creates the objects with names i001, i002, i003
:param ips: the names of the ips for the servers. 'i[001-003].futuregrid.org' creates the ips for the previously defined names
:param management: the names of the management nodes. 'i[001-002]' sets the nodes i001 and i002 to management nodes. The rest will be set to compute nodes automatically.
'''
name_list = expand_hostlist(names)
ip_list = expand_hostlist(ips)
management_list = expand_hostlist(management)
server_list = zip(name_list, ip_list)
servers = []
for (server_name, server_ip) in server_list:
server = FabricServer(name=server_name, cluster=name, ip=server_ip)
if server_name in management_list:
server.tags = ["manage"]
else:
server.tags = ["compute"]
self.stamp()
server.save(cascade=True)
servers.append(server)
cluster = FabricCluster(name=name, cluster=name, definition=names)
cluster.servers = servers
self.stamp()
cluster.save(cascade=True)
pass
def expand_names(name_spec):
if type(name_spec) != str and isinstance(name_spec, collections.Sequence):
names = []
for name in name_spec:
names += hostlist.expand_hostlist(name)
else:
names = hostlist.expand_hostlist(name_spec)
return names
def validate(self):
cluster = inventory.get("cluster", self.cluster.data)
posibilities = expand_hostlist(cluster.definition)
choice = expand_hostlist(self.nodespec.data)
if choice == []:
ok = False
else:
ok = set(choice).issubset(posibilities)
print "Validate", ok, choice
return ok
def validate(self):
cluster = inventory.get("cluster", self.cluster.data)
posibilities = expand_hostlist(cluster.definition)
choice = expand_hostlist(self.nodespec.data)
if choice == []:
ok = False
else:
ok = set(choice).issubset(posibilities)
print "Validate", ok, choice
return ok
def returnjobinfo(jobid):
global GANGLIA, GANGLIA_PROC
t0 = time.time()
# print jobid
s = hide_usernames(
os.popen("scontrol show -d --oneliner job " + str(jobid)).read()).split()
print("slurm response took ", time.time() - t0)
if s:
j = dict()
for x in s:
y = x.split('=', 1)
if len(y) == 2:
j[y[0]] = y[1]
if not j['JobState'] == 'PENDING':
cpu_mapping = list()
h = ['Nodes', 'CPU_IDs', 'Mem']
nodelist = ""
for i, n in enumerate(s):
if n.startswith("Nodes="):
cpu_mapping.append([s[i].replace('Nodes=', ''),
s[i + 1].replace('CPU_IDs=', ''),
s[i + 2].replace('Mem=', '')])
if n.startswith("NodeList="):
nodelist = n.replace("NodeList=", "")
j['cpu_mapping'] = {'headers': h, 'nodes': cpu_mapping}
j['expanded_nodelist'] = list(
map(str.strip, expand_hostlist(nodelist)))
if 'NodeList' in j:
j['nodeinfo'] = returnnodeinfo(j['NodeList'])['nodeinfo']
elif 'Nodes' in j:
j['nodeinfo'] = returnnodeinfo(j['Nodes'])['nodeinfo']
if GANGLIA == 1 and GANGLIA_PROC == 1:
t_procs0 = time.time()
j['procs'] = get_procs(j['expanded_nodelist'])
print("get procs took", time.time() - t_procs0)
else:
# not an active job, fetch finished job stats
# remark, these stats have a different format, leave it up to the client side
# to fix it.
yesterday = (dt.datetime.today() -
dt.timedelta(1)).strftime("%Y-%m-%d")
sacct = "sacct -X --format=jobid,jobname,user,account,state,elapsed,submit,start,end,nnodes,ncpus,reqnodes,reqcpus,nodelist --parsable2 -S %s --job %s"
s = hide_usernames(run_slurmcommand(sacct % (yesterday, jobid)))
t = io.StringIO(s)
reader = csv.reader(t, delimiter='|')
headers = list(map(convert, next(reader)))
jobinfo = list(map(convert, next(reader)))
j = dict(list(zip(headers, jobinfo)))
j['expanded_nodelist'] = list(
map(str.strip, expand_hostlist(j["NodeList"])))
j['GANGLIA'] = GANGLIA
# print j
print("jobinfo", time.time() - t0)
return j
def main(argv):
global cstring
# Initialize empty cluster list
cluster_list = []
# Default slurm configuration
slurm_conf = "/etc/slurm/slurm.conf"
# Default load threshold to report
load_thr = 8.0
guilty = False
reverse = False
# Default amount of parallelism
workers = 8
try:
opts, args = getopt.getopt(argv, "n:l:c:s:p:Lgh")
except getopt.GetoptError:
usage()
sys.exit()
for opt, arg in opts:
if opt in ("-h"):
usage()
sys.exit()
elif opt in ("-n"): # override default cluster file
cluster_list = hostlist.expand_hostlist(arg)
elif opt in ("-s"): # override default Slurm config file name
slurm_conf = arg
elif opt in ("-l"): # override default load level
load_thr = float(arg)
if load_thr < 0:
print "Error: invalid load level - " + arg + "!"
sys.exit()
elif opt in ("-c"): # override default community string
cstring = arg
elif opt in ("-g"): # print top user on each overused system
guilty = True
elif opt in ("-L"): # invert load comparison
reverse = True
elif opt in ("-p"): # invert load comparison
workers = int(arg)
if not cluster_list:
cluster_list = parse_slurm_conf(slurm_conf)
if not cluster_list:
print "Error: no cluster nodes specified!"
sys.exit()
pool = Pool(workers)
results = pool.map(query_node, cluster_list)
pool.close()
# sort results by load value and pass to report printer
print_report(sorted(results, key=itemgetter(1), reverse=True),
reduce(max_name, results, 0), load_thr, guilty, reverse)
def listPBSnodes(hostExpression):
""" Create a list of nodes that are free to run jobs
Example: freePBSnodes('node01[05-11],node02[01-80]')
"""
nodeNames = expand_hostlist(hostExpression)
nodeString = "".join(["%s " % n for n in nodeNames])
pbsOut = check_output(["pbsnodes -x %s" % nodeString], shell=True)
nodes = xml2obj(pbsOut)
freenodelist = []
# find free nodes
# state = free
# no jobs are runing
# no message
for node in nodes["Node"]:
status = {}
messages = []
jobs = []
name = node["name"]
state = node["state"]
freenodelist.append(name)
return freenodelist
def setup_servers(self, protocol):
if self.num_nodes > 1:
from hostlist import expand_hostlist
task_index = int(os.environ['SLURM_PROCID'])
hostlist = expand_hostlist(os.environ['SLURM_NODELIST'])
hostlist_w_port = [("%s:2222" % host) for host in hostlist]
cluster = tf.train.ClusterSpec({"localhost":hostlist_w_port}).as_cluster_def()
protocol = f'grpc+{protocol}' if protocol else 'grpc'
server = tf.distribute.Server(cluster, job_name="localhost",
task_index=task_index, protocol=protocol)
session_target = server.target
if task_index != 0:
utils.join_tasks(task_index, hostlist)
quit()
else:
task_index = 0
hostlist = ['localhost']
session_target = ''
[print(f'{arg} : {val}') for arg, val in vars(self.args).items()]
self.session_target = session_target
self.task_index = task_index
self.hostlist = hostlist
def check_nodes_computesets(clusterid, computenodeids):
hosts_param = hostlist.expand_hostlist(computenodeids)
hosts_param_set = set(hosts_param)
nodes_free = True
nodes_allocated = False
nodes_checked = False
# computesetid = -1
computesets = Comet.get_computeset()
# get all active computeset and put nodes into a set
allhosts = set()
for computeset in computesets:
if computeset["cluster"] == clusterid \
and (computeset["state"] in Cluster.ACTIVE_COMPUTESETS):
computesetid = computeset["id"]
# print (computesetid)
for compute in computeset["computes"]:
allhosts.add(compute["name"])
# all nodes allocated
if hosts_param_set <= allhosts:
nodes_allocated = True
nodes_free = False
nodes_checked = True
# at least one specified host not in any Active computeset
else:
for host in hosts_param:
# some specified nodes are in Active computeset
if host in allhosts:
nodes_free = False
nodes_checked = True
break
# print ("nodes_checked: %s" % nodes_checked)
# print ("nodes_free: %s" % nodes_free)
return [nodes_free, nodes_allocated]
def generate(self, nodes, prefix, topology=None, test_name=None):
if topology:
self.logger.info('node: ignoring topology (not used)')
node_set = collections.defaultdict(set)
node_set = set(hostlist.expand_hostlist(bnc.config['nodes']))
node_set &= set(nodes) #Don't include error/excluded nodes
for node in node_set:
test_dir = os.path.join(prefix, "tests", node)
bench.util.mkdir_p(test_dir)
script_file = os.path.join(test_dir, '{0}.job'.format(node))
with open(script_file, 'w') as fp:
fp.write(self.TEMPLATE.render(
job_name = 'bench-node-{0}'.format(node),
modules = " ".join(bnc.config['modules']),
node_name = node,
linpack_path = bnc.config['linpack_path'],
stream_path = bnc.config['stream_path'],
))
node_list_file = os.path.join(test_dir, 'node_list')
bench.util.write_node_list(node_list_file, [node])
self.logger.info('node: add: {0}'.format(len(nodes)))
def get_partition_nodes(partition):
p = scontrol('show partition {0}'.format(partition))
for line in p.stdout:
for match in NODES_P.finditer(line):
for node in hostlist.expand_hostlist(match.group(1)):
yield node
p.wait()
def main(argv):
global cstring
# Initialize empty cluster list
cluster_list=[]
# Default slurm configuration
slurm_conf="/etc/slurm/slurm.conf"
# Default load threshold to report
load_thr=8.0
guilty=False
reverse=False
# Default amount of parallelism
workers=8
try:
opts,args=getopt.getopt(argv,"n:l:c:s:p:Lgh")
except getopt.GetoptError:
usage()
sys.exit()
for opt,arg in opts:
if opt in ("-h"):
usage()
sys.exit()
elif opt in ("-n"): # override default cluster file
cluster_list=hostlist.expand_hostlist(arg)
elif opt in ("-s"): # override default Slurm config file name
slurm_conf=arg
elif opt in ("-l"): # override default load level
load_thr=float(arg)
if load_thr<0:
print "Error: invalid load level - "+arg+"!"
sys.exit()
elif opt in ("-c"): # override default community string
cstring=arg
elif opt in ("-g"): # print top user on each overused system
guilty=True
elif opt in ("-L"): # invert load comparison
reverse=True
elif opt in ("-p"): # invert load comparison
workers=int(arg)
if not cluster_list:
cluster_list=parse_slurm_conf(slurm_conf)
if not cluster_list:
print "Error: no cluster nodes specified!"
sys.exit()
pool=Pool(workers)
results=pool.map(query_node,cluster_list)
pool.close()
# sort results by load value and pass to report printer
print_report(sorted(results,key=itemgetter(1),reverse=True),
reduce(max_name,results,0),load_thr,guilty,reverse)
def add(self, **kwargs):
if "host" not in kwargs:
Console.error("no id specified")
sys.exit(1)
hosts = hostlist.expand_hostlist(kwargs['host'])
if 'ip' in kwargs:
ips = Parameter.expand(kwargs['ip'])
else:
ips = [None for i in hosts]
if ips is None:
ips = [None for i in hosts]
for host, ip in zip(hosts, ips):
if host in self.data:
entry = self.data[host]
else:
entry = dict(self.entry)
self.data[host] = entry
for key, value in kwargs.items():
entry[key] = value
entry['ip'] = ip
entry['host'] = host
for attribute in entry:
self.data[host][attribute] = entry[attribute]
def set(self, spec, attribute):
"""add an attribute for the specified hosts in the format
i[1-20]. which would set the attribute for all hosts in i1 to
i20"""
hosts = expand_hostlist(spec)
for host in hosts:
self._set(host, attribute)
def set(self, spec, attribute):
"""add an attribute for the specified hosts in the format
i[1-20]. which would set the attribute for all hosts in i1 to
i20"""
hosts = expand_hostlist(spec)
for host in hosts:
self._set(host, attribute)
def acct_reader(self, filename):
ftr = [3600,60,1]
acct = []
with open(filename) as fd:
for job in csv.DictReader(fd, delimiter = '|'):
if self.jobids and job['JobID'] not in self.jobids: continue
if job['NodeList'] == "None assigned": continue
jent = {}
jent['id'] = job['JobID']
jent['user'] = job['User']
jent['project'] = job['Account']
jent['start_time'] = int(parse(job['Start']).strftime('%s'))
jent['end_time'] = int(parse(job['End']).strftime('%s'))
jent['queue_time'] = int(parse(job['Submit']).strftime('%s'))
jent['queue'] = job['Partition']
jent['name'] = job['JobName']
jent['status'] = job['State'].split()[0]
jent['nodes'] = int(job['NNodes'])
jent['cores'] = int(job['ReqCPUS'])
jent['host_list'] = hostlist.expand_hostlist(job['NodeList'])
if '-' in job['Timelimit']:
days, time = job['Timelimit'].split('-')
else:
time = job['Timelimit']
days = 0
jent['requested_time'] = (int(days) * 86400 +
sum([a*b for a,b in zip(ftr, [int(i) for i in time.split(":")])]))/60
acct += [jent]
return acct
def test_alltoall_switch_tests(self, _):
self.nodes = NODES
bench.util.write_node_list(os.path.join(self.directory, 'node_list'),
self.nodes)
bench.add.execute(self.directory,
TOPOLOGY_FILE,
alltoall_switch_tests=True)
prefix = os.path.join(self.directory, 'alltoall-switch', 'tests')
switches = set((
'hpcf-ib-rack1-u43',
'hpcf-ib-rack1-u45',
'hpcf-ib-rack1-u42',
'hpcf-ib-rack1-u44',
'hpcf-ib-rack1-u46',
))
self.assertEqual(
set(os.listdir(prefix)),
switches,
)
nodes = set()
for switch in switches:
script = os.path.join(prefix, switch, '{0}.job'.format(switch))
with open(script) as fp:
match = NODELIST_P.search(fp.read())
nodes |= set(hostlist.expand_hostlist(match.group(3)))
self.assertEqual(nodes, self.nodes)
def _detect_nodes(self):
# see if we have a PBS_NODEFILE
pbs_nodefile = os.environ.get('PBS_NODEFILE')
slurm_nodelist = os.environ.get('SLURM_NODELIST')
if pbs_nodefile is not None:
# parse PBS the nodefile
self._raw_nodes = [line.strip() for line in open(pbs_nodefile)]
if self.log is not None:
self.log.info(
message="Found PBS_NODEFILE %s: %s" % (pbs_nodefile, self._raw_nodes),
suffix=LOG_SUFFIX)
elif slurm_nodelist is not None:
# parse SLURM nodefile
self._raw_nodes = hostlist.expand_hostlist(slurm_nodelist)
if self.log is not None:
self.log.info(
message="Found SLURM_NODELIST %s. Expanded to: %s" % (slurm_nodelist, self._raw_nodes),
suffix=LOG_SUFFIX)
else:
self._raw_nodes = ['localhost']
if self.log is not None:
self.log.info(
message="No PBS_NODEFILE or SLURM_NODELIST found. Using hosts: %s" % (self._raw_nodes),
suffix=LOG_SUFFIX)
def test_exclude_file(self, _):
fail_nodes = os.path.join(self.directory, 'fail_nodes')
with open(fail_nodes, 'w') as fp:
for node in ['tnode0101', 'tnode0102', 'tnode0208', 'nonode0101']:
fp.write('{0}\n'.format(node))
bench.add.execute(
self.directory,
topology_file=TOPOLOGY_FILE,
node_tests=True,
exclude_files=[fail_nodes],
)
tests_dir = os.path.join(self.directory, 'node', 'tests')
expected_tests = set(self.nodes) - set(
['tnode0101', 'tnode0102', 'tnode0208', 'nonode0101'])
self.assertEqual(
set(os.listdir(tests_dir)),
expected_tests,
)
for node in expected_tests:
script = os.path.join(tests_dir, node, '{0}.job'.format(node))
self.assertNotEqual(os.stat(script).st_size, 0)
with open(script) as fp:
match = NODELIST_P.search(fp.read())
self.assertEqual(
set(hostlist.expand_hostlist(match.group(3))),
set((node, )),
)
def check_nodes_computesets(clusterid, computenodeids):
hosts_param = hostlist.expand_hostlist(computenodeids)
hosts_param_set = set(hosts_param)
nodes_free = True
nodes_allocated = False
nodes_checked = False
# computesetid = -1
computesets = Comet.get_computeset()
# get all active computeset and put nodes into a set
allhosts = set()
for computeset in computesets:
if computeset["cluster"] == clusterid \
and (computeset["state"] in Cluster.ACTIVE_COMPUTESETS):
computesetid = computeset["id"]
# print (computesetid)
for compute in computeset["computes"]:
allhosts.add(compute["name"])
# all nodes allocated
if hosts_param_set <= allhosts:
nodes_allocated = True
nodes_free = False
nodes_checked = True
# at least one specified host not in any Active computeset
else:
for host in hosts_param:
# some specified nodes are in Active computeset
if host in allhosts:
nodes_free = False
nodes_checked = True
break
# print ("nodes_checked: %s" % nodes_checked)
# print ("nodes_free: %s" % nodes_free)
return [nodes_free, nodes_allocated]
def test_exclude_file (self, _):
fail_nodes = os.path.join(self.directory, 'fail_nodes')
with open(fail_nodes, 'w') as fp:
for node in ['tnode0101', 'tnode0102', 'tnode0208', 'nonode0101']:
fp.write('{0}\n'.format(node))
bench.add.execute(
self.directory,
topology_file=TOPOLOGY_FILE,
node_tests=True,
exclude_files=[fail_nodes],
)
tests_dir = os.path.join(self.directory, 'node', 'tests')
expected_tests = set(self.nodes) - set(['tnode0101', 'tnode0102', 'tnode0208', 'nonode0101'])
self.assertEqual(
set(os.listdir(tests_dir)),
expected_tests,
)
for node in expected_tests:
script = os.path.join(tests_dir, node, '{0}.job'.format(node))
self.assertNotEqual(os.stat(script).st_size, 0)
with open(script) as fp:
match = NODELIST_P.search(fp.read())
self.assertEqual(
set(hostlist.expand_hostlist(match.group(3))),
set((node, )),
)
def read_data_from_yaml(self):
"""
read mac address and bmc configuration information from **mac.yaml** file.
"""
data = read_yaml_config(self.yaml_file)
result = None
if data:
result = {}
data = data["inventory"]
for cluster in data:
cluster_data = data[cluster]
if "bmc" in cluster_data and "common" in cluster_data["bmc"]:
# process the common bmc data in cluster
common_bmc_data = cluster_data["bmc"]["common"]
host_range = common_bmc_data.pop("range", None)
hosts = expand_hostlist(host_range)
mac_data = cluster_data["macaddr"]
for host in mac_data:
if host in hosts:
temp_common_bmc_data = deepcopy(common_bmc_data)
if "bmc" in mac_data[host]:
# bmc config in individual host have a high
# priority than common config
temp_common_bmc_data.update(mac_data[host]["bmc"])
mac_data[host]["bmc"] = temp_common_bmc_data
result[cluster] = mac_data
return result
def dict_key_list_table_printer(d, indexed=False):
'''
accept a dict in the form:
{key1: [list1],
key2: [list2],
.......
=>
| key1 | key2 |
| l
| i
| s
| t
'''
x = PrettyTable()
temp = d.values()
l = 0
for item in temp:
l0 = len(item)
if l0 > l:
l = l0
if indexed:
if l == 0:
index_list = []
else:
index_list = hostlist.expand_hostlist("[1-{0}]".format(str(l)))
x.add_column("index", index_list)
for k, v in d.iteritems():
v0 = v + [" "] * (l - len(v))
x.add_column(k, v0)
x.align = "l"
return x
def display_provision_form():
clusters = cm_config_server().get("cloudmesh.server.provisioner.clusters")
# clusters = ['india','bravo','sierra']
# servers = n_inventory.hostlist(cluster)
# server = n_inventory.host(name,auth=False)
form = ProvisionForm(csrf=False)
if form.validate_on_submit():
flash("Success")
print "FORM"
pprint(form.__dict__)
print "CLUSTER", form.cluster.data
print "Service", form.service.data
hosts = expand_hostlist(form.nodespec.data)
print "Nodespec", hosts
for host in hosts:
print "PROVISION HOST", host
provision.delay(host, form.service.data)
return redirect("provision/tasks/{0}/{1}/{2}".format(
form.cluster.data, form.nodespec.data, form.service.data))
# return redirect("/provision/summary/")
else:
flash("Wrong submission")
inventory.refresh()
return render_template("mesh/provision/provision.html",
clusters=clusters,
form=form)
def init_process(backend="nccl"):
print(f"Starting process with rank {ptu.dist_rank}...", flush=True)
if "SLURM_STEPS_GPUS" in os.environ:
gpu_ids = os.environ["SLURM_STEP_GPUS"].split(",")
os.environ["MASTER_PORT"] = str(12345 + int(min(gpu_ids)))
else:
os.environ["MASTER_PORT"] = str(12345)
if "SLURM_JOB_NODELIST" in os.environ:
hostnames = hostlist.expand_hostlist(os.environ["SLURM_JOB_NODELIST"])
os.environ["MASTER_ADDR"] = hostnames[0]
else:
os.environ["MASTER_ADDR"] = "127.0.0.1"
dist.init_process_group(
backend,
rank=ptu.dist_rank,
world_size=ptu.world_size,
)
print(f"Process {ptu.dist_rank} is connected.", flush=True)
dist.barrier()
silence_print(ptu.dist_rank == 0)
if ptu.dist_rank == 0:
print(f"All processes are connected.", flush=True)
def GET(self, id, metric):
res = db.select(settings["SlurmClusterName"]+"_job_table",
what="nodelist, time_start, time_end",
where="id_job = $id_job",
vars={ "id_job" : id })
job = res[0]
url = "%s/render" % ( settings["GraphiteURL"] )
target = "%s{%s}.%s" % ( settings["GraphitePrefix"],
",".join(hostlist.expand_hostlist(job["nodelist"])),
metric
)
if 'infiniband' not in metric:
if 'cpu-' in metric or 'swap_io' in metric or metric.endswith('tx') or metric.endswith('rx') or metric.startswith('llite-') or metric.endswith('read') or metric.endswith('write'):
target = "scaleToSeconds(derivative(%s),1)" % target
data = urllib.urlencode({
"target": target,
"from": datetime.datetime.fromtimestamp(job["time_start"]).strftime("%H:%M_%Y%m%d"),
"until": datetime.datetime.fromtimestamp(job["time_end"]).strftime("%H:%M_%Y%m%d"),
"format": "json",
"maxDataPoints": 200})
req = urllib2.Request(url, data)
ret = {}
returned = urllib2.urlopen(req).read()
graphiteData = json.loads(returned)
for metric in graphiteData:
node = metric["target"].split(".")[len(settings["GraphitePrefix"].split("."))-1]
ret[node] = [[i[1] * 1000, i[0]] for i in metric["datapoints"]]
return json.dumps(ret)
def create(self, kind, subkind, nameregex):
#"india[9-11].futuregrid.org,india[01-02].futuregrid.org"
names = expand_hostlist(nameregex)
for name in names:
if kind == "server":
object = FabricServer(
name=name,
kind=kind,
subkind=subkind,
)
elif kind == "service":
object = FabricService(
name=name,
kind=kind,
subkind=subkind)
log.info("creating {0} {1} {2}".format(name, kind, subkind))
elif kind == "cluster":
object = FabricCluster(
name=name,
kind=kind,
subkind=subkind)
log.info("creating {0} {1} {2}".format(name, kind, subkind))
else:
log.error(
"kind is not defined, creation of objects failed, kind, nameregex")
return
object.save(cascade=True)
def expand(cls, parameter, allow_duplicates=False, sort=False):
if parameter is None:
return parameter
else:
return expand_hostlist(parameter,
allow_duplicates=False,
sort=False)
def display_provision_form():
clusters = cm_config_server().get("cloudmesh.server.provisioner.clusters")
# clusters = ['india','bravo','sierra']
# servers = n_inventory.hostlist(cluster)
# server = n_inventory.host(name,auth=False)
form = ProvisionForm(csrf=False)
if form.validate_on_submit():
flash("Success")
print "FORM"
pprint(form.__dict__)
print "CLUSTER", form.cluster.data
print "Service", form.service.data
hosts = expand_hostlist(form.nodespec.data)
print "Nodespec", hosts
for host in hosts:
print "PROVISION HOST", host
provision.delay(host, form.service.data)
return redirect("provision/tasks/{0}/{1}/{2}"
.format(form.cluster.data,
form.nodespec.data,
form.service.data))
# return redirect("/provision/summary/")
else:
flash("Wrong submission")
inventory.refresh()
return render_template("mesh/provision/provision.html", clusters=clusters, form=form)
def acct_reader(self, filename):
ftr = [3600,60,1]
acct = []
with open(filename) as fd:
for job in csv.DictReader(fd, delimiter = '|'):
if self.jobids and job['JobID'] not in self.jobids: continue
if job['NodeList'] == "None assigned": continue
if len(job) != 13:
print(job['JobID'] + " is not parsed correctly")
continue
jent = {}
jent['id'] = job['JobID']
jent['user'] = job['User']
jent['project'] = job['Account']
jent['start_time'] = int(parse(job['Start']).strftime('%s'))
jent['end_time'] = int(parse(job['End']).strftime('%s'))
jent['queue_time'] = int(parse(job['Submit']).strftime('%s'))
jent['queue'] = job['Partition']
jent['name'] = job['JobName']
jent['status'] = job['State'].split()[0]
jent['nodes'] = int(job['NNodes'])
jent['cores'] = int(job['ReqCPUS'])
jent['host_list'] = hostlist.expand_hostlist(job['NodeList'])
if '-' in job['Timelimit']:
days, time = job['Timelimit'].split('-')
else:
time = job['Timelimit']
days = 0
jent['requested_time'] = (int(days) * 86400 +
sum([a*b for a,b in zip(ftr, [int(i) for i in time.split(":")])]))/60
acct += [jent]
return acct
def main(argv):
# Default slurm configuration
slurm_conf = "/etc/slurm/slurm.conf"
argc = len(argv)
if argc < 2 or argc > 3:
usage()
else:
if argc == 2:
nodes = parse_slurm_conf(slurm_conf)
else:
nodes = hostlist.expand_hostlist(argv[0])
if len(nodes) > 0:
for node in nodes:
cmdline = "scp " + argv[argc -
2] + " " + node + ":" + argv[argc - 1]
print "Running", cmdline
results = commands.getstatusoutput(cmdline)
if results[0] == 0:
print "OK", results[1]
else:
print "Error:", node + ")", results[1]
else:
print "Error: empty or invalid list of cluster nodes!"
def main():
global logger, ARGS
signal.signal(signal.SIGINT, sig_handler)
ARGS = parse_args()
if ARGS.verbose:
setup_logging("debug")
else:
setup_logging("info")
logger.debug(ARGS)
G.config = ConfigParser.SafeConfigParser()
try:
G.config.read(ARGS.cfgfile)
G.hosts = hostlist.expand_hostlist(G.config.get("global", "pub_hosts"))
G.url = G.config.get("DB", "url")
G.port = G.config.get("global", "pub_port")
except:
logger.error("Can't read configuration file")
sys.exit(1)
ARGS.func()
def dict_key_list_table_printer(d, indexed=False):
"""
accept a dict in the form:
{key1: [list1],
key2: [list2],
.......
=>
| key1 | key2 |
| l
| i
| s
| t
"""
x = PrettyTable()
temp = d.values()
l = 0
for item in temp:
l0 = len(item)
if l0 > l:
l = l0
if indexed:
if l == 0:
index_list = []
else:
index_list = hostlist.expand_hostlist("[1-{0}]".format(str(l)))
x.add_column("index", index_list)
for k, v in d.iteritems():
v0 = v + [" "] * (l - len(v))
x.add_column(k, v0)
x.align = "l"
return x
def create(self, kind, namespec):
'''
creates fabric objects of the specified kind and matching the name specification
:param kind: the kind . see FABRIC_TYPES
:param namespec: the specifacation for a name list. 'i[001-003]'. creates the objects with names i001, i002, i003
'''
elements = []
names = expand_hostlist(namespec)
for name in names:
if kind == "server":
element = FabricServer(name=name, kind=kind)
elif kind == "service":
element = FabricService(name=name, kind=kind)
log.info("creating {0} {1}".format(name, kind))
elif kind == "cluster":
element = FabricCluster(name=name, kind=kind)
log.info("creating {0} {1}".format(name, kind))
elif kind == "image":
element = FabricImage(name=name, kind=kind)
log.info("creating {0} {1}".format(name, kind))
else:
log.error(
"kind is not defined, creation of objects failed, kind, nameregex"
)
return
self.stamp()
element.save(cascade=True)
elements.append(object)
return elements
def test_submit_jobs_3(self, arg1):
'''Test that submit doesn't submit jobs with nodes in --nodelist'''
node_test = bench.tests.node_test.NodeTest("node")
node_test.Submit.execute(self.directory, nodelist='tnode01[01-06]')
self.assertTrue(bench.slurm.sbatch.called)
directories = set()
expected_directories = set()
job_scripts = set()
expected_job_scripts = set()
for ii, call in enumerate(arg1.call_args_list):
script_dir = os.path.join(self.node_test_dir, self.nodes[ii])
args, kwargs = call #call object is two things: args=tuple, kwargs=dict
directories.add(script_dir)
expected_directories.add(kwargs['chdir'])
job_scripts.add(args[0])
expected_job_scripts.add(script_dir + '/' + self.nodes[ii] +
'.job')
# Check that --nodelist nodes not submitted
for node in hostlist.expand_hostlist('tnode01[07-10]'):
self.assertNotIn(node, kwargs['chdir'])
self.assertEqual(directories, expected_directories)
self.assertEqual(job_scripts, expected_job_scripts)
def read_data_from_yaml(self):
"""
read mac address and bmc configuration information from **mac.yaml** file.
"""
data = read_yaml_config(self.yaml_file)
result = None
if data:
result = {}
data = data["inventory"]
for cluster in data:
cluster_data = data[cluster]
if "bmc" in cluster_data and "common" in cluster_data["bmc"]:
# process the common bmc data in cluster
common_bmc_data = cluster_data["bmc"]["common"]
host_range = common_bmc_data.pop("range", None)
hosts = expand_hostlist(host_range)
mac_data = cluster_data["macaddr"]
for host in mac_data:
if host in hosts:
temp_common_bmc_data = deepcopy(common_bmc_data)
if "bmc" in mac_data[host]:
# bmc config in individual host have a high
# priority than common config
temp_common_bmc_data.update(mac_data[host]["bmc"])
mac_data[host]["bmc"] = temp_common_bmc_data
result[cluster] = mac_data
return result
def get_resource() -> Tuple[str, List[str], Optional[List[int]]]:
"""Get SLURM resources: nodename, nodelist, and gpus.
Returns
-------
Tuple[str, List[str], Optional[List[int]]]
nodename, nodelist, and gpus
Raises
------
RuntimeError
if number of nodes could not be retrieved
ValueError
list of nodes is not of the same length sa number of nodes
ValueError
if current nodename is not found in node list
"""
nodelist = hostlist.expand_hostlist(os.environ["SLURM_JOB_NODELIST"])
nodename = os.environ["SLURMD_NODENAME"]
num_nodes_env = os.getenv("SLURM_JOB_NUM_NODES")
if num_nodes_env:
num_nodes = int(num_nodes_env)
else:
raise RuntimeError("Could not get SLURM number of nodes")
if len(nodelist) != num_nodes:
raise ValueError(
f"Number of slurm nodes {len(nodelist)} not equal to {num_nodes}")
if nodename not in nodelist:
raise ValueError(
f"Nodename({nodename}) not in nodelist({nodelist}). This should not happen!"
)
gpus = local.get_gpus()
return nodename, nodelist, gpus
def create (self, kind, namespec):
'''
creates fabric objects of the specified kind and matching the name specification
:param kind: the kind . see FABRIC_TYPES
:param namespec: the specifacation for a name list. 'i[001-003]'. creates the objects with names i001, i002, i003
'''
elements = []
names = expand_hostlist(namespec)
for name in names:
if kind == "server":
element = FabricServer(name=name, kind=kind)
elif kind == "service":
element = FabricService(name=name, kind=kind)
log.info("creating {0} {1}".format(name, kind))
elif kind == "cluster":
element = FabricCluster(name=name, kind=kind)
log.info("creating {0} {1}".format(name, kind))
elif kind == "image":
element = FabricImage(name=name, kind=kind)
log.info("creating {0} {1}".format(name, kind))
else:
log.error(
"kind is not defined, creation of objects failed, kind, nameregex")
return
self.stamp()
element.save(cascade=True)
elements.append(object)
return elements
def filter_node_list(
nodes,
include_nodes=None,
exclude_nodes=None,
include_reservations=None,
exclude_reservations=None,
include_states=None,
exclude_states=None,
include_files=None,
exclude_files=None,
):
nodes = set(nodes)
if include_states or exclude_states:
nodes &= set(
get_nodes(
include_states=include_states,
exclude_states=exclude_states,
))
if include_nodes or include_reservations or include_files:
include_nodes_ = set()
if include_nodes:
for hostlist_ in include_nodes:
include_nodes_ |= set(hostlist.expand_hostlist(hostlist_))
if include_reservations:
for reservation in include_reservations:
include_nodes_ |= get_reserved_nodes(reservation)
if include_files:
for include_file in include_files:
include_nodes_ |= set(read_node_list(include_file))
nodes &= include_nodes_
if exclude_nodes or exclude_reservations or exclude_files:
exclude_nodes_ = set()
if exclude_nodes:
for hostlist_ in exclude_nodes:
exclude_nodes_ |= set(hostlist.expand_hostlist(hostlist_))
if exclude_reservations:
for reservation in exclude_reservations:
exclude_nodes_ |= get_reserved_nodes(reservation)
if exclude_files:
for exclude_file in exclude_files:
exclude_nodes_ |= set(read_node_list(exclude_file))
nodes -= exclude_nodes_
return nodes
def get_status_short(self, raw_hosts=None):
"""get status of baremetal computer
provided for **rain status --short [HOSTS]**
:param string raw_hosts: one or more hosts with the valid formation of hostlist
:return: a dict of the formation {"host1": "deployed", "host2": "deploying", "host3": "failed"}
"""
hosts = expand_hostlist(raw_hosts) if raw_hosts else None
return self.status.get_status_short(hosts)
def get_worker_host_list(base_port, workers_per_host):
hosts = expand_hostlist(os.environ['SLURM_NODELIST'])
ports = [base_port + i for i in range(workers_per_host)]
worker_hlist = []
for h in hosts:
for p in ports:
worker_hlist.append('{}:{}'.format(h, p))
return worker_hlist
def get_slurm_allocated_nodes():
hosts = os.environ.get("SLURM_NODELIST")
hosts = hostlist.expand_hostlist(hosts)
freenodes = []
for h in hosts:
freenodes.append((h + "\n"))
return list(set(freenodes))
def ip_name_pair(self, nameregex, format_string, start=1):
ips = expand_hostlist(nameregex)
i = start
names = []
for ip in ips:
names.append(format_string.format(i))
i += 1
return zip(names, ips)
def get_worker_host_list(base_port,workers_per_host):
hosts = expand_hostlist( os.environ['SLURM_NODELIST'])
ports = [base_port + i for i in range(workers_per_host)]
l = []
for h in hosts:
for p in ports:
l.append('{}:{}'.format(h,p))
return l
def get_status_summary(self, raw_hosts=None):
"""get status summary of baremetal computer
provided for **rain status --summary [HOSTS]**
:param string raw_hosts: one or more hosts with the valid formation of hostlist
:return: a dict of the formation {"deployed": 1, "deploying":2, "failed":2, "total": 5}
"""
hosts = expand_hostlist(raw_hosts) if raw_hosts else None
return self.status.get_status_summary(hosts)
def baremetal_computer_host_off(self, raw_hosts):
"""Disable/OFF computers for baremetal provisioning
provided for **rain admin off HOSTS**
:param string raw_hosts: ne or more hosts with the valid formation of hostlist
:return: True means successfully, otherwise False
"""
hosts = expand_hostlist(raw_hosts) if raw_hosts else None
return self.baremetal.disable_baremetal_computers(hosts)
def filtered_hosts_based_policy(user, projects, hosts):
"""filtered hosts based on policy of the user and his/her projects
"""
# wrapper
wrapper = RainCobblerWrapper()
policy = wrapper.get_policy_based_user_or_its_projects(user, projects)
policy_hosts = hostlist.expand_hostlist(policy) if policy else None
return [h for h in hosts if h in policy_hosts] if policy_hosts else []
def filtered_hosts_based_baremetal(raw_hosts):
"""filtered hosts based on baremetal computers
"""
# wrapper
wrapper = RainCobblerWrapper()
input_hosts = hostlist.expand_hostlist(raw_hosts)
bm_hosts = wrapper.baremetal_computer_host_list()
return [h for h in input_hosts if h in bm_hosts] if bm_hosts else []
def set_compute_nodelist(self, nodelist):
nodelist_expanded = hostlist.expand_hostlist(nodelist)
self._inventory_content['all']['children']['compute_nodes'][
'hosts'] = {}
for node in nodelist_expanded:
self._inventory_content['all']['children']['compute_nodes'][
'hosts'][node] = None
def filtered_hosts_based_policy(user, projects, hosts):
"""filtered hosts based on policy of the user and his/her projects
"""
# wrapper
wrapper = RainCobblerWrapper()
policy = wrapper.get_policy_based_user_or_its_projects(user, projects)
policy_hosts = hostlist.expand_hostlist(policy) if policy else None
return [h for h in hosts if h in policy_hosts] if policy_hosts else []
def filtered_hosts_based_baremetal(raw_hosts):
"""filtered hosts based on baremetal computers
"""
# wrapper
wrapper = RainCobblerWrapper()
input_hosts = hostlist.expand_hostlist(raw_hosts)
bm_hosts = wrapper.baremetal_computer_host_list()
return [h for h in input_hosts if h in bm_hosts] if bm_hosts else []
def generate(self):
self.generate_bootspec()
self._generate_globals()
clusters = self.config.get("cloudmesh.inventory")
for cluster_name in clusters:
cluster = clusters[cluster_name]
names = expand_hostlist(cluster["id"])
net_id = 0
for network in cluster["network"]:
n_index = expand_hostlist(network["id"])
n_label = expand_hostlist(network["label"])
n_range = expand_hostlist(network["range"])
n_name = network["name"]
for i in range(0, len(names)):
name = n_index[i]
element = dict(network)
del element['range']
element.update({'cm_type': "inventory",
'cm_key': 'server',
'cm_kind': 'server',
'cm_id': name,
'cm_cluster': cluster_name,
'id': name,
'label' : n_label[i],
'network_name': n_name,
'cm_network_id': net_id,
'ipaddr': n_range[i],
'cm_attribute': 'network'}
)
self.insert(element)
net_id += 1
# added by HC
# init rack status
self.generate_rack_status()
# init baremetal computer managemnt, maybe will be deprecated later
bdb = BaremetalDB()
bdb.init_base_document_structure()
# insert necessary mac info of baremetal computers to inventory
bmc = BaremetalComputer()
bmc.insert_mac_data_to_inventory()
def _configure(self):
slurm_nodelist = os.environ.get("SLURM_NODELIST")
if slurm_nodelist is None:
msg = "$SLURM_NODELIST not set!"
self._log.error(msg)
raise RuntimeError(msg)
# Parse SLURM nodefile environment variable
slurm_nodes = hostlist.expand_hostlist(slurm_nodelist)
self._log.info("Found SLURM_NODELIST %s. Expanded to: %s", slurm_nodelist, slurm_nodes)
# $SLURM_NPROCS = Total number of cores allocated for the current job
slurm_nprocs_str = os.environ.get("SLURM_NPROCS")
if slurm_nprocs_str is None:
msg = "$SLURM_NPROCS not set!"
self._log.error(msg)
raise RuntimeError(msg)
else:
slurm_nprocs = int(slurm_nprocs_str)
# $SLURM_NNODES = Total number of (partial) nodes in the job's resource allocation
slurm_nnodes_str = os.environ.get("SLURM_NNODES")
if slurm_nnodes_str is None:
msg = "$SLURM_NNODES not set!"
self._log.error(msg)
raise RuntimeError(msg)
else:
slurm_nnodes = int(slurm_nnodes_str)
# $SLURM_CPUS_ON_NODE = Number of cores per node (physically)
slurm_cpus_on_node_str = os.environ.get("SLURM_CPUS_ON_NODE")
if slurm_cpus_on_node_str is None:
msg = "$SLURM_CPUS_ON_NODE not set!"
self._log.error(msg)
raise RuntimeError(msg)
else:
slurm_cpus_on_node = int(slurm_cpus_on_node_str)
# Verify that $SLURM_NPROCS <= $SLURM_NNODES * $SLURM_CPUS_ON_NODE
if not slurm_nprocs <= slurm_nnodes * slurm_cpus_on_node:
self._log.warning(
"$SLURM_NPROCS(%d) <= $SLURM_NNODES(%d) * $SLURM_CPUS_ON_NODE(%d)",
slurm_nprocs,
slurm_nnodes,
slurm_cpus_on_node,
)
# Verify that $SLURM_NNODES == len($SLURM_NODELIST)
if slurm_nnodes != len(slurm_nodes):
self._log.error("$SLURM_NNODES(%d) != len($SLURM_NODELIST)(%d)", slurm_nnodes, len(slurm_nodes))
# Report the physical number of cores or the total number of cores
# in case of a single partial node allocation.
self.cores_per_node = min(slurm_cpus_on_node, slurm_nprocs)
self.node_list = slurm_nodes
def parse_slurm_env(env_var_name):
cluster=[]
env=os.getenv(env_var_name)
if env is not None:
cluster.extend(hostlist.expand_hostlist(env))
else:
print "Error: cannot get "+env_var_name+" from environment"
return cluster
def lookup(list):
debug("Forward Lookup:")
debug(list)
if "*" in list or ":" in list:
# MOAB events on Cray XE6 use node list format [1-7]*16:[10-99]*7
list = re.sub("\*\d+","", list) # omit *nprocs
list = re.sub(":",",", list) # change : to ,
new_list = hostlist.expand_hostlist(list)
# this takes the list we get back and turns it into a string for splunk
return ",".join(str(n) for n in new_list)
def main(argv):
# Default slurm configuration
slurm_conf="/etc/slurm/slurm.conf"
num_workers=8
try:
opts,args=getopt.getopt(argv,"p:nh")
except getopt.GetoptError:
usage()
sys.exit()
nodes=[]
for opt,arg in opts:
if opt in ("-h"):
usage()
sys.exit()
elif opt in ("-p"): # partition
nodes=parse_slurm_conf(slurm_conf,arg)
if len(nodes)==0:
print "Error: invalid partition -",arg
sys.exit()
elif opt in ("-n"): # dump nodelist
if not nodes:
nodes=parse_slurm_conf(slurm_conf)
for node in nodes:
print node
sys.exit()
elif opt in ('-N'): # set workers
num_workers=int(arg)
argc=len(args)
if argc<2 or argc>3:
usage()
else:
if argc==3:
nodes=hostlist.expand_hostlist(args[0])
elif not nodes:
nodes=parse_slurm_conf(slurm_conf)
if len(nodes)>0:
cmdline=[]
for node in nodes:
cmdline.append([node,"scp "+args[argc-2]+" "+node+":"+args[argc-1]])
pool=Pool(num_workers)
results=pool.map(copy_to_node,cmdline)
pool.close()
for result in results:
if result!="OK":
print result
else:
print "Error: empty or invalid list of cluster nodes!"
def get_ps_host_list(base_port,num_ps):
assert(num_ps < 10000000)
port = base_port
l = []
hosts = expand_hostlist( os.environ['SLURM_NODELIST'])
while True:
for host in hosts:
if len(l) >= num_ps:
return l
l.append('{}:{}'.format(host,port))
port += 1
def stateChangeLogic(record, nodes, start_state, end_state):
global output_results, options, known_states, node_states
if "-" in nodes or "[" in nodes:
try:
# MOAB JOBSTART events on Cray XE6 use node list format [1-7]*16:[10-999]*7
nodes = re.sub("\*\d+","", nodes) # omit *nprocs
nodes = re.sub(":",",", nodes) # change : to ,
node_list = hostlist.expand_hostlist(nodes)
except: # try to deal with BadHostlist exceptions
debug2("---- Bad hostlist: " + str(record['_time']) +" nodes="+nodes)
# guess it is missing a left bracket, and truncate at last comma
m=re.match("(^.*)\[(.*),(.*)", nodes)
if not m == None:
nodes = m.group(1) +"["+ m.group(2) +"]"
debug2("---- Changing to: nodes="+nodes)
node_list = hostlist.expand_hostlist(nodes) # do or die
else:
node_list = nodes.split(",")
if len(nodes) > 500: # omit long node lists
record['_raw'] = record['_raw'].replace(nodes,'(LONG_NODE_LIST)')
# the below is a hack for cielo, because RSV lines don't include some service nodes.
# if RSV lines list at least 98% of all known hosts, apply it to all hosts.
raw = str(record['_raw'])
if (raw.find("RSVSTART")!=-1 or raw.find("RSVEND")!=-1):
these = len(node_list)
all = len(node_states)
# the hardcoded .95 may cause problems later and should be improved.
# eg it could be an argument and set via local/savedsearches.conf for each machine,
# but really it should be a percentage of the hosts in each hpc_system index...
if (all>these and these>0.95*all):
debug2("Applying RSV to all known hosts ("+ str(these) +">0.95*"+ str(all)+"): "+raw)
node_list = node_states.keys()
# debug2("---- in stateChangeLogic:" + str(record['_time']) + " start="+ start_state + " end="+ end_state + " nodes="+nodes)
for node in node_list:
current_state = node_states.get(node)
if current_state == None or current_state=="Unknown" or start_state=="*" or current_state==start_state:
nodeStateChange(record, node, current_state, end_state)
