def prologue(self):
"""Checks the options given for settings and takes appropriate action.
See _merge_options for the format.
- if nagios_report is set, creates a SimpleNagios instance and prints the report.
- if ha is set, checks if running on the correct host, set the appropriate nagios message and bail if not.
- if locking_filename is set, take a lock. If the lock fails, bork and set the nagios exit accordingly.
"""
# bail if nagios report is requested
self.nagios_reporter = SimpleNagios(
_cache=self.options.nagios_check_filename,
_report_and_exit=self.options.nagios_report,
_threshold=self.options.nagios_check_interval_threshold,
_cache_user=self.options.nagios_user,
)
# check for HA host
if self.options.ha and not proceed_on_ha_service(self.options.ha):
self.log.warning(
"Not running on the target host %s in the HA setup. Stopping."
% (self.options.ha, ))
self.nagios_reporter.ok("Not running on the HA master.")
sys.exit(NAGIOS_EXIT_OK)
if not self.options.disable_locking and not self.options.dry_run:
self.lockfile = TimestampedPidLockfile(
self.options.locking_filename,
threshold=self.options.nagios_check_interval_threshold * 2)
lock_or_bork(self.lockfile, self.nagios_reporter)
self.log.info("%s has started" % (_script_name(sys.argv[0])))
def test_cache(self):
"""Test the caching"""
(handle, filename) = tempfile.mkstemp()
os.unlink(filename)
n = SimpleNagios(_cache=filename, _cache_user=self.nagios_user)
message = "mywarning"
n.warning(message)
os.close(handle)
self.buffo.seek(0)
self.buffo.truncate(0)
raised_exception = None
try:
reporter_test = NagiosReporter('test_cache', filename, -1,
self.nagios_user)
reporter_test.report_and_exit()
except SystemExit as err:
raised_exception = err
bo = self.buffo.getvalue().rstrip()
self.assertEqual(bo, "WARNING %s" % message)
self.assertEqual(raised_exception.code, NAGIOS_EXIT_WARNING[0])
statres = os.stat(filename)
self.assertFalse(statres.st_mode & stat.S_IROTH)
def test_eval(self):
"""Test the evaluation of the warning/critical level."""
nagios = SimpleNagios(
foo=100,
foo_critical=90,
bar=20,
)
nagios._eval()
def test_eval(self):
"""Test the evaluation of the warning/critical level."""
nagios = SimpleNagios(
foo=100,
foo_critical=90,
bar=20,
)
nagios._eval()
def prologue(self):
"""Checks the options given for settings and takes appropriate action.
See _merge_options for the format.
- if nagios_report is set, creates a SimpleNagios instance and prints the report.
- if ha is set, checks if running on the correct host, set the appropriate nagios message and bail if not.
- if locking_filename is set, take a lock. If the lock fails, bork and set the nagios exit accordingly.
"""
# bail if nagios report is requested
self.nagios_reporter = SimpleNagios(_cache=self.options.nagios_check_filename,
_report_and_exit=self.options.nagios_report,
_threshold=self.options.nagios_check_interval_threshold,
_cache_user=self.options.nagios_user,
)
# check for HA host
if self.options.ha and not proceed_on_ha_service(self.options.ha):
self.log.warning("Not running on the target host %s in the HA setup. Stopping." % (self.options.ha,))
self.nagios_reporter.ok("Not running on the HA master.")
sys.exit(NAGIOS_EXIT_OK)
if not self.options.disable_locking and not self.options.dry_run:
self.lockfile = TimestampedPidLockfile(self.options.locking_filename,
threshold=self.options.nagios_check_interval_threshold * 2)
lock_or_bork(self.lockfile, self.nagios_reporter)
self.log.info("%s has started" % (_script_name(sys.argv[0])))
def test_cache(self):
"""Test the caching"""
(handle, filename) = tempfile.mkstemp()
os.unlink(filename)
n = SimpleNagios(_cache=filename, _cache_user=self.nagios_user)
message = "mywarning"
n.warning(message)
os.close(handle)
self.buffo.seek(0)
self.buffo.truncate(0)
try:
reporter_test = NagiosReporter('test_cache', filename, -1, self.nagios_user)
reporter_test.report_and_exit()
except SystemExit, e:
pass
def test_world_readable(self):
"""Test world readable cache"""
(handle, filename) = tempfile.mkstemp()
os.unlink(filename)
n = SimpleNagios(_cache=filename, _cache_user=self.nagios_user, _world_readable=True)
n.ok("test")
os.close(handle)
try:
reporter_test = NagiosReporter('test_cache', filename, -1, self.nagios_user)
reporter_test.report_and_exit()
except SystemExit:
pass
statres = os.stat(filename)
self.assertTrue(statres.st_mode & stat.S_IROTH)
def _basic_test_single_instance(self, kwargs, message, nagios_exit):
"""Basic test"""
self.buffo.seek(0)
self.buffo.truncate(0)
try:
SimpleNagios(**kwargs)
except SystemExit, e:
pass
def test_cache(self):
"""Test the caching"""
(handle, filename) = tempfile.mkstemp()
os.unlink(filename)
n = SimpleNagios(_cache=filename, _cache_user=self.nagios_user)
message = "mywarning"
n.warning(message)
os.close(handle)
self.buffo.seek(0)
self.buffo.truncate(0)
try:
reporter_test = NagiosReporter('test_cache', filename, -1,
self.nagios_user)
reporter_test.report_and_exit()
except SystemExit, e:
pass
def test_world_readable(self):
"""Test world readable cache"""
(handle, filename) = tempfile.mkstemp()
os.unlink(filename)
n = SimpleNagios(_cache=filename,
_cache_user=self.nagios_user,
_world_readable=True)
n.ok("test")
os.close(handle)
try:
reporter_test = NagiosReporter('test_cache', filename, -1,
self.nagios_user)
reporter_test.report_and_exit()
except SystemExit:
pass
statres = os.stat(filename)
self.assertTrue(statres.st_mode & stat.S_IROTH)
def main():
"""Main function"""
options = {
'nagios_check_filename': ('filename of where the nagios check data is stored', str, 'store', NAGIOS_CHECK_FILENAME),
'nagios_check_interval_threshold': ('threshold of nagios checks timing out', None, 'store', 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: gengar, muk', str, 'store', 'gengar'),
'ha': ('high-availability master IP address', None, 'store', None),
'dry-run': ('do not make any updates whatsoever', None, 'store_true', False),
}
opts = simple_option(options)
nag = SimpleNagios(_cache=NAGIOS_CHECK_FILENAME)
if opts.options.ha and not proceed_on_ha_service(opts.options.ha):
_log.info("Not running on the target host in the HA setup. Stopping.")
nag.ok("Not running on the HA master.")
else:
# parse config file
clusters = {}
for host in opts.options.hosts:
master = opts.configfile_parser.get(host, "master")
showq_path = opts.configfile_parser.get(host, "showq_path")
mjobctl_path = opts.configfile_parser.get(host, "mjobctl_path")
clusters[host] = {
'master': master,
'spath': showq_path,
'mpath': mjobctl_path,
}
# process the new and previous data
released_jobids, stats = process_hold(clusters, dry_run=opts.options.dry_run)
# nagios state
stats.update(RELEASEJOB_LIMITS)
stats['message'] = "released %s jobs in hold" % len(released_jobids)
nag._eval_and_exit(**stats)
_log.info("Cached nagios state: %s %s" % (nag._final_state[0][1], nag._final_state[1]))
def _basic_test_single_instance_and_exit(self, fn, msg, message,
nagios_exit):
"""Basic test"""
self.buffo.seek(0)
self.buffo.truncate(0)
n = SimpleNagios()
f = getattr(n, fn)
try:
f(msg)
except SystemExit, e:
pass
def _basic_test_single_instance(self, kwargs, message, nagios_exit):
"""Basic test"""
self.buffo.seek(0)
self.buffo.truncate(0)
raised_exception = None
try:
SimpleNagios(**kwargs)
except SystemExit as err:
raised_exception = err
bo = self.buffo.getvalue().rstrip()
self.assertEqual(bo, message)
self.assertEqual(raised_exception.code, nagios_exit[0])
def _basic_test_single_instance_and_exit(self, fn, msg, message,
nagios_exit):
"""Basic test"""
self.buffo.seek(0)
self.buffo.truncate(0)
nagios = SimpleNagios()
func = getattr(nagios, fn)
raised_exception = None
try:
func(msg)
except SystemExit as err:
raised_exception = err
bo = self.buffo.getvalue().rstrip()
self.assertEqual(bo, message)
self.assertEqual(raised_exception.code, nagios_exit[0])
class ExtendedSimpleOption(SimpleOption):
"""
Extends the SimpleOption class to allow other checks to occur at script prologue and epilogue.
- nagios reporting
- checking if running on the designated HA host
- locking on a file
The prologue should be called at the start of the script; the epilogue at the end.
"""
def __init__(self, options, run_prologue=True, excepthook=None, **kwargs):
"""Initialise.
If run_prologue is True (default), we immediately execute the prologue.
Note that if taking a lock is requested (default), and the lock cannot be
acquire for some reason, the program will exit,
"""
options_ = _merge_options(options)
super(ExtendedSimpleOption, self).__init__(options_, **kwargs)
self.nagios_reporter = None
self.lockfile = None
if run_prologue:
self.prologue()
if not excepthook:
sys.excepthook = self.critical_exception_handler
else:
sys.excepthook = excepthook
self.log = fancylogger.getLogger()
def prologue(self):
"""Checks the options given for settings and takes appropriate action.
See _merge_options for the format.
- if nagios_report is set, creates a SimpleNagios instance and prints the report.
- if ha is set, checks if running on the correct host, set the appropriate nagios message and bail if not.
- if locking_filename is set, take a lock. If the lock fails, bork and set the nagios exit accordingly.
"""
# bail if nagios report is requested
self.nagios_reporter = SimpleNagios(_cache=self.options.nagios_check_filename,
_report_and_exit=self.options.nagios_report,
_threshold=self.options.nagios_check_interval_threshold,
_cache_user=self.options.nagios_user,
)
# check for HA host
if self.options.ha and not proceed_on_ha_service(self.options.ha):
self.log.warning("Not running on the target host %s in the HA setup. Stopping." % (self.options.ha,))
self.nagios_reporter.ok("Not running on the HA master.")
sys.exit(NAGIOS_EXIT_OK)
if not self.options.disable_locking and not self.options.dry_run:
self.lockfile = TimestampedPidLockfile(self.options.locking_filename,
threshold=self.options.nagios_check_interval_threshold * 2)
lock_or_bork(self.lockfile, self.nagios_reporter)
self.log.info("%s has started" % (_script_name(sys.argv[0])))
def _epilogue(self):
if not self.options.disable_locking and not self.options.dry_run:
release_or_bork(self.lockfile, self.nagios_reporter)
def epilogue(self, nagios_message, nagios_thresholds=None):
"""Run at the end of a script, quitting gracefully if possible."""
if nagios_thresholds is None:
nagios_thresholds = {}
self._epilogue()
nagios_thresholds['message'] = nagios_message
self.nagios_reporter._eval_and_exit(**nagios_thresholds)
self.log.info("%s has finished" % (_script_name(sys.argv[0]))) # may not be reached
def ok(self, nagios_message):
"""Run at the end of a script and force an OK exit."""
self._epilogue()
self.nagios_reporter.ok(nagios_message)
def warning(self, nagios_message):
"""Run at the end of a script and force a Warning exit."""
self._epilogue()
self.nagios_reporter.warning(nagios_message)
def critical(self, nagios_message):
"""Run at the end of a script and force a Critical exit."""
self._epilogue()
self.nagios_reporter.critical(nagios_message)
def critical_exception_handler(self, tp, value, traceback):
"""
Run at the end of a script and force a Critical exit.
This function is meant to be used as sys.excepthook
"""
self.log.exception("unhandled exception detected: %s - %s", tp, value)
self.log.debug("traceback %s", traceback)
message = "Script failure: %s - %s" % (tp, value)
sys.exc_clear()
self.critical(message)
class ExtendedSimpleOption(SimpleOption):
"""
Extends the SimpleOption class to allow other checks to occur at script prologue and epilogue.
- nagios reporting
- checking if running on the designated HA host
- locking on a file
The prologue should be called at the start of the script; the epilogue at the end.
"""
def __init__(self, options, run_prologue=True, excepthook=None, **kwargs):
"""Initialise.
If run_prologue is True (default), we immediately execute the prologue.
Note that if taking a lock is requested (default), and the lock cannot be
acquire for some reason, the program will exit,
"""
options_ = _merge_options(options)
super(ExtendedSimpleOption, self).__init__(options_, **kwargs)
self.nagios_reporter = None
self.lockfile = None
if run_prologue:
self.prologue()
if not excepthook:
sys.excepthook = self.critical_exception_handler
else:
sys.excepthook = excepthook
self.log = fancylogger.getLogger()
def prologue(self):
"""Checks the options given for settings and takes appropriate action.
See _merge_options for the format.
- if nagios_report is set, creates a SimpleNagios instance and prints the report.
- if ha is set, checks if running on the correct host, set the appropriate nagios message and bail if not.
- if locking_filename is set, take a lock. If the lock fails, bork and set the nagios exit accordingly.
"""
# bail if nagios report is requested
self.nagios_reporter = SimpleNagios(
_cache=self.options.nagios_check_filename,
_report_and_exit=self.options.nagios_report,
_threshold=self.options.nagios_check_interval_threshold,
_cache_user=self.options.nagios_user,
)
# check for HA host
if self.options.ha and not proceed_on_ha_service(self.options.ha):
self.log.warning(
"Not running on the target host %s in the HA setup. Stopping."
% (self.options.ha, ))
self.nagios_reporter.ok("Not running on the HA master.")
sys.exit(NAGIOS_EXIT_OK)
if not self.options.disable_locking and not self.options.dry_run:
self.lockfile = TimestampedPidLockfile(
self.options.locking_filename,
threshold=self.options.nagios_check_interval_threshold * 2)
lock_or_bork(self.lockfile, self.nagios_reporter)
self.log.info("%s has started" % (_script_name(sys.argv[0])))
def _epilogue(self):
if not self.options.disable_locking and not self.options.dry_run:
release_or_bork(self.lockfile, self.nagios_reporter)
def epilogue(self, nagios_message, nagios_thresholds=None):
"""Run at the end of a script, quitting gracefully if possible."""
if nagios_thresholds is None:
nagios_thresholds = {}
self._epilogue()
nagios_thresholds['message'] = nagios_message
self.nagios_reporter._eval_and_exit(**nagios_thresholds)
self.log.info("%s has finished" %
(_script_name(sys.argv[0]))) # may not be reached
def ok(self, nagios_message):
"""Run at the end of a script and force an OK exit."""
self._epilogue()
self.nagios_reporter.ok(nagios_message)
def warning(self, nagios_message):
"""Run at the end of a script and force a Warning exit."""
self._epilogue()
self.nagios_reporter.warning(nagios_message)
def critical(self, nagios_message):
"""Run at the end of a script and force a Critical exit."""
self._epilogue()
self.nagios_reporter.critical(nagios_message)
def critical_exception_handler(self, tp, value, traceback):
"""
Run at the end of a script and force a Critical exit.
This function is meant to be used as sys.excepthook
"""
self.log.exception("unhandled exception detected: %s - %s", tp, value)
self.log.debug("traceback %s", traceback)
message = "Script failure: %s - %s" % (tp, value)
sys.exc_clear()
self.critical(message)
