def test_sshshowq(self):
"""Test sshshowq"""
clusters = {'delcatty': {'path': '/opt/moab/bin/checkjob', 'master': 'master15.delcatty.gent.vsc'}, 'phanpy': {'path': '/opt/moab/bin/checkjob', 'master': 'master17.phanpy.gent.vsc'}, 'raichu': {'path': '/opt/moab/bin/checkjob', 'master': 'master13.raichu.gent.vsc'}, 'golett': {'path': '/opt/moab/bin/checkjob', 'master': 'master19.golett.gent.vsc'}, 'swalot': {'path': '/opt/moab/bin/checkjob', 'master': 'master21.swalot.gent.vsc'}}
showq = SshShowq(
'master1',
'testuser',
clusters=clusters,
cache_pickle=True,
dry_run=True)
self.assertEqual(showq._command('/opt/moab/bin/checkjob'), ['sudo', 'ssh', 'testuser@master1', '/opt/moab/bin/checkjob'])
self.assertEquals(showq.info, ShowqInfo)
self.assertEquals(showq.info(), {})
def test_sshshowq(self):
"""Test sshshowq"""
clusters = {
'delcatty': {
'path': '/opt/moab/bin/checkjob',
'master': 'master15.delcatty.gent.vsc'
},
'phanpy': {
'path': '/opt/moab/bin/checkjob',
'master': 'master17.phanpy.gent.vsc'
},
'raichu': {
'path': '/opt/moab/bin/checkjob',
'master': 'master13.raichu.gent.vsc'
},
'golett': {
'path': '/opt/moab/bin/checkjob',
'master': 'master19.golett.gent.vsc'
},
'swalot': {
'path': '/opt/moab/bin/checkjob',
'master': 'master21.swalot.gent.vsc'
}
}
showq = SshShowq('master1',
'testuser',
clusters=clusters,
cache_pickle=True,
dry_run=True)
self.assertEqual(
showq._command('/opt/moab/bin/checkjob'),
['sudo', 'ssh', 'testuser@master1', '/opt/moab/bin/checkjob'])
self.assertEquals(showq.info, ShowqInfo)
self.assertEquals(showq.info(), {})
def main():
# Collect all info
# Note: debug option is provided by generaloption
# Note: other settings, e.g., ofr each cluster will be obtained from the configuration file
options = {
'nagios-check-interval-threshold':
NAGIOS_CHECK_INTERVAL_THRESHOLD,
'hosts':
('the hosts/clusters that should be contacted for job information',
None, 'extend', []),
'information': ('the sort of information to store: user, vo, project',
None, 'store', 'user'),
'location': ('the location for storing the pickle file: delcatty, muk',
str, 'store', 'delcatty'),
'account_page_url':
('the URL at which the account page resides', None, 'store', None),
'access_token':
('the token that will allow authentication against the account page',
None, 'store', None),
'target_master':
('the master used to execute showq commands', None, 'store', None),
'target_user': ('the user for ssh to the target master', None, 'store',
None),
}
opts = ExtendedSimpleOption(options)
try:
rest_client = AccountpageClient(token=opts.options.access_token)
gpfs = GpfsOperations()
storage = VscStorage()
storage_name = cluster_user_pickle_store_map[opts.options.location]
login_mount_point = storage[storage_name].login_mount_point
gpfs_mount_point = storage[storage_name].gpfs_mount_point
clusters = {}
for host in opts.options.hosts:
master = opts.configfile_parser.get(host, "master")
showq_path = opts.configfile_parser.get(host, "showq_path")
clusters[host] = {'master': master, 'path': showq_path}
logger.debug("clusters = %s" % (clusters, ))
showq = SshShowq(opts.options.target_master,
opts.options.target_user,
clusters,
cache_pickle=True,
dry_run=opts.options.dry_run)
logger.debug("Getting showq information ...")
(queue_information, _, _) = showq.get_moab_command_information()
timeinfo = time.time()
active_users = queue_information.keys()
logger.debug("Active users: %s" % (active_users))
logger.debug("Queue information: %s" % (queue_information))
# We need to determine which users should get an updated pickle. This depends on
# - the active user set
# - the information we want to provide on the cluster(set) where this script runs
# At the same time, we need to determine the job information each user gets to see
tup = (opts.options.information, active_users, queue_information,
rest_client)
(target_users, target_queue_information,
user_map) = determine_target_information(*tup)
nagios_user_count = 0
nagios_no_store = 0
stats = {}
for user in target_users:
try:
path = get_pickle_path(opts.options.location, user,
rest_client)
user_queue_information = target_queue_information[user]
user_queue_information['timeinfo'] = timeinfo
store_on_gpfs(user, path, "showq",
(user_queue_information, user_map[user]), gpfs,
login_mount_point, gpfs_mount_point,
".showq.json.gz", opts.options.dry_run)
nagios_user_count += 1
except Exception:
logger.error("Could not store pickle file for user %s" %
(user))
nagios_no_store += 1
stats["store_users"] = nagios_user_count
stats["store_fail"] = nagios_no_store
stats["store_fail_critical"] = STORE_LIMIT_CRITICAL
except Exception, err:
logger.exception("critical exception caught: %s" % (err))
opts.critical("Script failed in a horrible way")
sys.exit(NAGIOS_EXIT_CRITICAL)
