def resmon_on_off(request, action, name):
"""Activate/deactivate a specific resmon, create a new resmon DB if necessary.
:param request: The original HTTP request.
:type request: HttpRequest
:param action: The action to take, must be one of: 'on' or 'off'.
:type action: str
:param name: The name of the monitor against which to apply the action.
:type name: str
:return: The response to this request.
:rtype: HttpResponse
"""
global _task_repository
global _lock
try:
resp = RestCmd('resmon', '')
if action == 'on':
period = request.GET.get('period')
with _lock:
if name in _task_repository:
logger.warning('Monitor {0} already running, restarting.'.format(name))
_task_repository.pop(name).stop()
logger.info('Staring monitor {0} at intervals {1}.'.format(name, period))
_task_repository[name] = start_monitor(name, parse_timespec(period), ResmonListener())
if 'xml' in request.META.get('HTTP_ACCEPT'):
return HttpResponse(ElementTree.tostring(resp.export_xml()), content_type='application/xml')
return HttpResponse(resp.export_json(), content_type='application/json')
with _lock:
if name in _task_repository:
task = _task_repository.pop(name)
logger.info('Stopping {0}'.format(name))
task.stop()
else:
logger.warning('{0} not in resmon repository. Attempt to stop is ignored.'.format(name))
if 'xml' in request.META.get('HTTP_ACCEPT'):
return HttpResponse(ElementTree.tostring(resp.export_xml()), content_type='application/xml')
return HttpResponse(resp.export_json(), content_type='application/json')
except Exception as e:
tb = sys.exc_info()[2]
logger.warning('Encountered unexpected {0} at {1}'.format(e, traceback.format_tb(tb)))
return HttpResponseServerError('Encountered unexpected internal server error')
def main(args):
"""Main entry point of this program.
:param args: The command-line argument list
:return: Exit code
:rtype: int
"""
parser = argparse.ArgumentParser(prog='resmon.py', description='Monitor the system resource utilisation, periodically')
top = parser
group1 = top.add_argument_group()
group1.add_argument('-n', metavar='name', help='Name of the resource monitoring session, artifacts will follow this'
' name', type=str)
group1.add_argument('-p', metavar='period', help='The sampling period of the monitoring process. E.g. 10s, 1m, 2h.'
' Default is 1m', default='1m', type=str)
group2 = top.add_argument_group()
group2.add_argument('-l', help='List all the resmon databases found in the working directory and return.',
action='count')
arg = parser.parse_args(args)
if arg.l:
print(execute_list_dbs())
return 0
if not arg.n:
logger.error('No name specified')
return errno.EINVAL
if arg.p:
period = parse_timespec(arg.p)
else:
period = 60 # Default period is 1m
monitor = start_monitor(arg.n, period)
try:
while 1: # Run until stopped
# Wake up periodically and do something, like kicking a watchdog or something like that, not sure what yet.
sleep(600)
except KeyboardInterrupt:
logger.info('User interrupted')
return errno.EINTR
finally:
monitor.stop()
logger.info('resmon terminated')
return errno.EINVAL # If we are here we were interrupted by something other than KeyboardInterrupt
def main(args):
"""Main entry to program.
:param args: Parsed command-line args.
:return: Exit code.
"""
try:
start_ts = (int(time.time()) / _step) * _step # TS should be aligned with the period
# Step 1: Start the monitoring
if (not os.path.exists(_base_name + 'cpu.rrd')) or (not os.path.exists(_base_name + 'mem.rrd')):
fill_in_the_rrds(_base_name + 'cpu.rrd', _base_name + 'mem.rrd', start_ts - (24 * 3600), start_ts, _step)
else:
fill_in_the_rrds(_base_name + 'cpu.rrd', _base_name + 'mem.rrd',
int(rrdutil.get_last_update(_base_name + 'cpu.rrd')) + _step, start_ts, _step)
monitor = resmon_util.start_monitor(_base_name, _step)
try:
prev_ts = start_ts
for loop in range(0, 360): # Run for an hour (360 loops of 10 seconds).
# Step 2: Wait for some time (10s) for the monitor to gather data
logger.info('Waiting for data, loop #{0}'.format(loop))
time.sleep(10)
# Step 3: Observe the result
last_cpu_data = rrdutil.fetch_last_data(monitor.get_cpu_rrd(), start_ts=prev_ts, resolution=1)
last_mem_data = rrdutil.fetch_last_data(monitor.get_mem_rrd(), start_ts=prev_ts)
# Print out the result
print('\nLAST CPU data {0}, curr time = {1}'.format(last_cpu_data[0], int(time.time())))
print(rrdutil.format_cpu_stats(last_cpu_data))
print('\nLAST Memory data {0}'.format(last_mem_data[0]))
print(rrdutil.format_mem_stats(last_mem_data))
prev_ts = (int(time.time()) / _step) * _step # TS should be aligned with the period
avg_data = rrdutil.fetch_avg_data(monitor.get_cpu_rrd(), start_ts=start_ts, resolution=10)
print('\nAVG CPU data {0}'.format(avg_data[0]))
print(rrdutil.format_cpu_stats(avg_data))
avg_data = rrdutil.fetch_avg_data(monitor.get_mem_rrd(), start_ts=start_ts, resolution=10)
print('\nAVG Memory data {0}'.format(avg_data[0]))
print(rrdutil.format_mem_stats(avg_data))
finally:
# Don't forget to stop the monitor when finish
logger.info('Stopping {0}'.format(monitor))
monitor.stop()
# Generate a graph
end_graf = time.time()
start_graf = end_graf - (2 * SECONDS_IN_HOUR)
gen_graph('cpu.png', rrd_name=monitor.get_cpu_rrd(), title='CPU Utilisation', cf='AVERAGE',
start_ts=start_graf, end_ts=end_graf,
dataset=[('idle', '%Idle'), ('wait', '%Wait'), ('load', 'Load avg')])
gen_graph('mem.png', rrd_name=monitor.get_mem_rrd(), title='Memory Utilisation', cf='AVERAGE',
start_ts=start_graf, end_ts=end_graf,
dataset=[('memtotal', 'Total'), ('memfree', 'Free'), ('buffers', 'Buffers'),
('cached', 'Cached')])
except (TypeError, ValueError) as e:
logger.warning('Encountered invalid monitor period {0}, at {1}'.format(e, format_tb(sys.exc_info()[2])))
except KeyboardInterrupt:
logger.warning('User interruption at {0}'.format(format_tb(sys.exc_info()[2])))
except BaseException as e:
logger.error('Encountered unexpected exception {0} at {1}'.format(repr(e), format_tb(sys.exc_info()[2])))
