def test_sshcheckjob(self):
"""Test sshcheckjob"""
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'}}
checkjob = SshCheckjob(
'master1',
'testuser',
clusters=clusters,
cache_pickle=True,
dry_run=True)
self.assertEqual(checkjob._command('/opt/moab/bin/showq'), ['sudo', 'ssh', 'testuser@master1', '/opt/moab/bin/showq'])
self.assertEquals(checkjob.info, CheckjobInfo)
self.assertEquals(checkjob.info(), {})
def test_sshcheckjob(self):
"""Test sshcheckjob"""
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'
}
}
checkjob = SshCheckjob('master1',
'testuser',
clusters=clusters,
cache_pickle=True,
dry_run=True)
self.assertEqual(
checkjob._command('/opt/moab/bin/showq'),
['sudo', 'ssh', 'testuser@master1', '/opt/moab/bin/showq'])
self.assertEquals(checkjob.info, CheckjobInfo)
self.assertEquals(checkjob.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', []),
'location': ('the location for storing the pickle file: delcatty, muk', str, 'store', 'delcatty'),
'access_token': ('the token that will allow authentication against the account page', None, 'store', None),
'account_page_url': ('', 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")
checkjob_path = opts.configfile_parser.get(host, "checkjob_path")
clusters[host] = {
'master': master,
'path': checkjob_path
}
checkjob = SshCheckjob(
opts.options.target_master,
opts.options.target_user,
clusters,
cache_pickle=True,
dry_run=opts.options.dry_run)
(job_information, _, _) = checkjob.get_moab_command_information()
active_users = job_information.keys()
logger.debug("Active users: %s" % (active_users))
logger.debug("Checkjob information: %s" % (job_information))
nagios_user_count = 0
nagios_no_store = 0
stats = {}
for user in active_users:
path = get_pickle_path(opts.options.location, user, rest_client)
try:
user_queue_information = CheckjobInfo({user: job_information[user]})
store_on_gpfs(user, path, "checkjob", user_queue_information, gpfs, login_mount_point,
gpfs_mount_point, ".checkjob.json.gz", opts.options.dry_run)
nagios_user_count += 1
except Exception:
logger.exception("Could not store cache 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)
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', []),
'location': ('the location for storing the pickle file: delcatty, muk',
str, 'store', 'delcatty'),
'access_token':
('the token that will allow authentication against the account page',
None, 'store', None),
'account_page_url': ('', 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")
checkjob_path = opts.configfile_parser.get(host, "checkjob_path")
clusters[host] = {'master': master, 'path': checkjob_path}
checkjob = SshCheckjob(opts.options.target_master,
opts.options.target_user,
clusters,
cache_pickle=True,
dry_run=opts.options.dry_run)
(job_information, _, _) = checkjob.get_moab_command_information()
active_users = job_information.keys()
logger.debug("Active users: %s" % (active_users))
logger.debug("Checkjob information: %s" % (job_information))
nagios_user_count = 0
nagios_no_store = 0
stats = {}
for user in active_users:
path = get_pickle_path(opts.options.location, user, rest_client)
try:
user_queue_information = CheckjobInfo(
{user: job_information[user]})
store_on_gpfs(user, path, "checkjob", user_queue_information,
gpfs, login_mount_point, gpfs_mount_point,
".checkjob.json.gz", opts.options.dry_run)
nagios_user_count += 1
except Exception:
logger.exception("Could not store cache 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)
