def test_slurm_vo_accounts(self):
"""Test that the commands to create accounts are correctly generated."""
vos = [
VO(vsc_id="gvo00001", institute={"site": "gent"}),
VO(vsc_id="gvo00002", institute={"site": "gent"}),
VO(vsc_id="gvo00012", institute={"site": "gent"}),
VO(vsc_id="gvo00016", institute={"site": "gent"}),
VO(vsc_id="gvo00017", institute={"site": "gent"}),
VO(vsc_id="gvo00018", institute={"site": "gent"}),
]
commands = slurm_vo_accounts(vos, [], ["mycluster"])
self.assertEqual([tuple(x) for x in commands], [tuple(x) for x in [
shlex.split("/usr/bin/sacctmgr add account gvo00001 Parent=gent Organization=ugent Cluster=mycluster"),
shlex.split("/usr/bin/sacctmgr add account gvo00002 Parent=gent Organization=ugent Cluster=mycluster")
]])
def test_slurm_vo_accounts(self):
"""Test that the commands to create accounts are correctly generated."""
vos = [
VO(vsc_id="bvo00001", institute={"name": "brussel"}, fairshare=18),
VO(vsc_id="bvo00002", institute={"name": "brussel"}, fairshare=17),
VO(vsc_id="bvo00003", institute={"name": "brussel"}, fairshare=16),
VO(vsc_id="bvo00004", institute={"name": "brussel"}, fairshare=15),
VO(vsc_id="bvo00005", institute={"name": "brussel"}, fairshare=14),
VO(vsc_id="bvo00006", institute={"name": "brussel"}, fairshare=13),
]
commands = slurm_vo_accounts(vos, [], ["mycluster"], 'brussel')
self.assertEqual([tuple(x) for x in commands], [
tuple(x) for x in [
shlex.split(
"/usr/bin/sacctmgr -i add account bvo00005 Parent=brussel Organization=vub Cluster=mycluster Fairshare=14"
),
shlex.split(
"/usr/bin/sacctmgr -i add account bvo00006 Parent=brussel Organization=vub Cluster=mycluster Fairshare=13"
)
]
])
def main():
"""
Main script. The usual.
"""
options = {
"nagios-check-interval-threshold":
NAGIOS_CHECK_INTERVAL_THRESHOLD,
"access_token": ("OAuth2 token to access the account page REST API",
None, "store", None),
"account_page_url": (
"URL of the account page where we can find the REST API",
str,
"store",
"https://apivsc.ugent.be/django",
),
'host_institute':
('Name of the institute where this script is being run', str, 'store',
GENT),
"clusters": (
"Cluster(s) (comma-separated) to sync for. "
"Overrides <host_institute>_SLURM_COMPUTE_CLUSTERS that are in production.",
"strlist",
"store",
[],
),
'start_timestamp':
('Timestamp to start the sync from', str, 'store', None),
'cluster_classes':
('Classes of clusters that should be synced, comma-separated',
"strlist", 'store', [PRODUCTION, PILOT])
}
opts = ExtendedSimpleOption(options)
stats = {}
(last_timestamp, start_time) = retrieve_timestamp_with_default(
SYNC_TIMESTAMP_FILENAME, start_timestamp=opts.options.start_timestamp)
logging.info("Using timestamp %s", last_timestamp)
logging.info("Using startime %s", start_time)
try:
client = AccountpageClient(token=opts.options.access_token,
url=opts.options.account_page_url + "/api/")
host_institute = opts.options.host_institute
slurm_account_info = get_slurm_acct_info(SyncTypes.accounts)
slurm_user_info = get_slurm_acct_info(SyncTypes.users)
logging.debug("%d accounts found", len(slurm_account_info))
logging.debug("%d users found", len(slurm_user_info))
if opts.options.clusters:
clusters = opts.options.clusters
else:
clusters = [
cs for p in opts.options.cluster_classes
for cs in VSC_SLURM_CLUSTERS[host_institute][p]
]
sacctmgr_commands = []
# All users belong to a VO, so fetching the VOs is necessary/
account_page_vos = [
mkVo(v)
for v in client.vo.institute[opts.options.host_institute].get()[1]
]
# make sure the institutes and the default accounts (VOs) are there for each cluster
institute_vos = dict([
(v.vsc_id, v) for v in account_page_vos
if v.vsc_id in INSTITUTE_VOS_BY_INSTITUTE[host_institute].values()
])
sacctmgr_commands += slurm_institute_accounts(slurm_account_info,
clusters, host_institute,
institute_vos)
# The VOs do not track active state of users, so we need to fetch all accounts as well
active_accounts = set(
[a["vsc_id"] for a in client.account.get()[1] if a["isactive"]])
# dictionary mapping the VO vsc_id on a tuple with the VO members and the VO itself
account_page_members = dict([(vo.vsc_id, (set(vo.members), vo))
for vo in account_page_vos])
# process all regular VOs
sacctmgr_commands += slurm_vo_accounts(account_page_vos,
slurm_account_info, clusters,
host_institute)
# process VO members
sacctmgr_commands += slurm_user_accounts(account_page_members,
active_accounts,
slurm_user_info, clusters,
opts.options.dry_run)
logging.info("Executing %d commands", len(sacctmgr_commands))
if opts.options.dry_run:
print("Commands to be executed:\n")
print("\n".join([" ".join(c) for c in sacctmgr_commands]))
else:
execute_commands(sacctmgr_commands)
if not opts.options.dry_run:
(_, ldap_timestamp) = convert_timestamp(start_time)
write_timestamp(SYNC_TIMESTAMP_FILENAME, ldap_timestamp)
opts.epilogue("Accounts synced to slurm", stats)
else:
logging.info("Dry run done")
except Exception as err:
logging.exception("critical exception caught: %s", err)
opts.critical("Script failed in a horrible way")
sys.exit(NAGIOS_EXIT_CRITICAL)
def main():
"""
Main script. The usual.
"""
options = {
"nagios-check-interval-threshold":
NAGIOS_CHECK_INTERVAL_THRESHOLD,
"access_token": ("OAuth2 token to access the account page REST API",
None, "store", None),
"account_page_url": (
"URL of the account page where we can find the REST API",
str,
"store",
"https://apivsc.ugent.be/django",
),
"clusters": (
"Cluster(s) (comma-separated) to sync for. "
"Overrides GENT_SLURM_COMPUTE_CLUSTERS that are in production.",
str,
"store",
None,
),
}
opts = ExtendedSimpleOption(options)
stats = {}
try:
client = AccountpageClient(token=opts.options.access_token,
url=opts.options.account_page_url + "/api/")
last_timestamp = "201804010000Z" # the beginning of time
logging.info("Last recorded timestamp was %s" % (last_timestamp))
slurm_account_info = get_slurm_acct_info(SyncTypes.accounts)
slurm_user_info = get_slurm_acct_info(SyncTypes.users)
logging.debug("%d accounts found", len(slurm_account_info))
logging.debug("%d users found", len(slurm_user_info))
if opts.options.clusters is not None:
clusters = opts.options.clusters.split(",")
else:
clusters = [
c for c in GENT_SLURM_COMPUTE_CLUSTERS
if c in GENT_PRODUCTION_COMPUTE_CLUSTERS
]
sacctmgr_commands = []
# make sure the institutes and the default accounts (VOs) are there for each cluster
sacctmgr_commands += slurm_institute_accounts(slurm_account_info,
clusters)
# All users belong to a VO, so fetching the VOs is necessary/
account_page_vos = [mkVo(v) for v in client.vo.get()[1]]
# The VOs do not track active state of users, so we need to fetch all accounts as well
active_accounts = set(
[a["vsc_id"] for a in client.account.get()[1] if a["isactive"]])
# dictionary mapping the VO vsc_id on a tuple with the VO members and the VO itself
account_page_members = dict([(vo.vsc_id, (set(vo.members), vo))
for vo in account_page_vos])
# process all regular VOs
sacctmgr_commands += slurm_vo_accounts(account_page_vos,
slurm_account_info, clusters)
# process VO members
sacctmgr_commands += slurm_user_accounts(account_page_members,
active_accounts,
slurm_user_info, clusters,
opts.options.dry_run)
logging.info("Executing %d commands", len(sacctmgr_commands))
if opts.options.dry_run:
print("Commands to be executed:\n")
print("\n".join([" ".join(c) for c in sacctmgr_commands]))
else:
execute_commands(sacctmgr_commands)
except Exception as err:
logger.exception("critical exception caught: %s" % (err))
opts.critical("Script failed in a horrible way")
sys.exit(NAGIOS_EXIT_CRITICAL)
if not opts.options.dry_run:
opts.epilogue("Accounts synced to slurm", stats)
else:
logger.info("Dry run done")