class LogConsumerThread(threading.Thread):
def __init__(self, host, *args, **kwargs):
super(LogConsumerThread, self).__init__(*args, **kwargs)
self.l = logging.getLogger('collector')
self._host = host
def callback(self, channel, method, properties, body):
if not s_queue.full():
#print "adding to queue"
s_queue.put(body)
def run(self):
self.rabbitcom = Comm('10.10.0.134','logs','logs','/','logs','logs','collector')
self.connect = self.rabbitcom.ampq_connect()
#credentials = pika.PlainCredentials("logs", "logs")
#connection = pika.BlockingConnection(pika.ConnectionParameters(host=self._host,credentials=credentials))
#channel = connection.channel()
#channel.queue_bind(exchange='logs',queue='logs')
self.rabbitcom.channel.basic_consume(self.callback,
"logs",
no_ack=True)
self.rabbitcom.channel.start_consuming()
class LogConsumerThread(threading.Thread):
def __init__(self, host, *args, **kwargs):
super(LogConsumerThread, self).__init__(*args, **kwargs)
self.l = logging.getLogger("collector")
self._host = host
def callback(self, channel, method, properties, body):
if not s_queue.full():
# print "adding to queue"
s_queue.put(body)
def run(self):
self.rabbitcom = Comm("10.10.0.134", "logs", "logs", "/", "logs", "logs", "collector")
self.connect = self.rabbitcom.ampq_connect()
# credentials = pika.PlainCredentials("logs", "logs")
# connection = pika.BlockingConnection(pika.ConnectionParameters(host=self._host,credentials=credentials))
# channel = connection.channel()
# channel.queue_bind(exchange='logs',queue='logs')
self.rabbitcom.channel.basic_consume(self.callback, "logs", no_ack=True)
self.rabbitcom.channel.start_consuming()
def run(self):
self.rabbitcom = Comm('10.10.0.134','logs','logs','/','logs','logs','collector')
self.connect = self.rabbitcom.ampq_connect()
#credentials = pika.PlainCredentials("logs", "logs")
#connection = pika.BlockingConnection(pika.ConnectionParameters(host=self._host,credentials=credentials))
#channel = connection.channel()
#channel.queue_bind(exchange='logs',queue='logs')
self.rabbitcom.channel.basic_consume(self.callback,
"logs",
no_ack=True)
self.rabbitcom.channel.start_consuming()
def initialize(self):
self.rabbitcom = Comm('10.10.0.134', 'logs', 'logs', '/', 'jobs', '',
'jobmon')
# Server main loop
while True:
self.getJobs()
time.sleep(4)
def initialize(self):
self.rabbitcom = Comm('10.10.0.134', 'logs', 'logs', '/', 'logs', '',
'sysmon')
self.connect = self.rabbitcom.ampq_connect()
# Pre-Compile our regex for the minimal performance gain
self.dev_re = re.compile(
'([A-Za-z0-9\.]+):\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)'
)
self.meminfo_re = re.compile('([\w]+):\s(\d+)\s')
self.proc_stats_re = re.compile('([A-Za-z0-9]+)\s+(\d+)')
self.diskstatus_re = re.compile(
'\s+(\d+)\s+(\d+)\s(\w+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)'
)
# Server main loop
while True:
self.getProcesses()
time.sleep(1)
def initialize(self):
self.rabbitcom = Comm('10.10.0.134','logs','logs','/','logs','','sysmon')
self.connect = self.rabbitcom.ampq_connect()
# Pre-Compile our regex for the minimal performance gain
self.dev_re = re.compile('([A-Za-z0-9\.]+):\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)')
self.meminfo_re = re.compile('([\w]+):\s(\d+)\s')
self.proc_stats_re = re.compile('([A-Za-z0-9]+)\s+(\d+)')
self.diskstatus_re = re.compile('\s+(\d+)\s+(\d+)\s(\w+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)')
# Server main loop
while True:
self.getProcesses()
time.sleep(1)
def run(self):
self.rabbitcom = Comm("10.10.0.134", "logs", "logs", "/", "logs", "logs", "collector")
self.connect = self.rabbitcom.ampq_connect()
# credentials = pika.PlainCredentials("logs", "logs")
# connection = pika.BlockingConnection(pika.ConnectionParameters(host=self._host,credentials=credentials))
# channel = connection.channel()
# channel.queue_bind(exchange='logs',queue='logs')
self.rabbitcom.channel.basic_consume(self.callback, "logs", no_ack=True)
self.rabbitcom.channel.start_consuming()
class SysmonDaemon(Daemon):
def run(self):
self.l = logging.getLogger('sysmon')
ch = logging.StreamHandler()
ch.setLevel(logging.ERROR)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
ch.setFormatter(formatter)
self.l.addHandler(ch)
self.initialize()
def initialize(self):
self.rabbitcom = Comm('10.10.0.134','logs','logs','/','logs','','sysmon')
self.connect = self.rabbitcom.ampq_connect()
# Pre-Compile our regex for the minimal performance gain
self.dev_re = re.compile('([A-Za-z0-9\.]+):\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)')
self.meminfo_re = re.compile('([\w]+):\s(\d+)\s')
self.proc_stats_re = re.compile('([A-Za-z0-9]+)\s+(\d+)')
self.diskstatus_re = re.compile('\s+(\d+)\s+(\d+)\s(\w+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)')
# Server main loop
while True:
self.getProcesses()
time.sleep(1)
def S(self,key ):
return key.replace('.','_')
def getUUID(self):
return re.sub('_|-|=','0',base64.urlsafe_b64encode(uuid.uuid4().bytes))
# Get all the processes that are not system
# processes
def getProcesses(self, ):
""" Get all system processes """
# walk the /proc
data = {}
stats = {}
data['uuid'] = self.getUUID()
d = datetime.datetime.utcnow()
data['hostname'] = socket.gethostname()
data['timestamp_year'] = d.strftime('%y')
data['timestamp_month'] = d.strftime('%m')
data['timestamp_day'] = d.strftime('%d')
data['timestamp_hour'] = d.strftime('%H')
data['timestamp_minute'] = d.strftime('%M')
data['timestamp_second'] = d.strftime('%S')
for item in os.listdir('/proc'):
if re.match('[\d]+',item):
# Start collecting statistics
# We only want statistics for processes that
# are not system processes, i.e. user jobs
stat_info = None
try:
stat_info = os.stat('/proc/' + item + '/status')
except Exception, err:
self.l.error(err)
if stat_info is not None:
# It is assumed that the ownership of the pid directory
# represents the owner of the process and as such we can
# use this to determine if this is a "non-system" or "non-root"
# process
if stat_info.st_uid >= 500:
self.getProcStatistics(item,stats)
# Get memory information
self.getMemInfo(stats)
# Get network statistics
self.getNetworkStats(stats)
# Get processor statistics
self.getProcessorStats(stats)
# Get Disk statistics
self.getDiskStats(stats)
# Get Infiniband statistics
self.getInfinibandStats(stats)
data['data'] = stats
self.l.info("Sending data")
self.rabbitcom.send_message(data)
class SysmonDaemon(Daemon):
def run(self):
self.l = logging.getLogger('sysmon')
ch = logging.StreamHandler()
ch.setLevel(logging.ERROR)
formatter = logging.Formatter(
'%(asctime)s - %(name)s - %(levelname)s - %(message)s')
ch.setFormatter(formatter)
self.l.addHandler(ch)
self.initialize()
def initialize(self):
self.rabbitcom = Comm('10.10.0.134', 'logs', 'logs', '/', 'logs', '',
'sysmon')
self.connect = self.rabbitcom.ampq_connect()
# Pre-Compile our regex for the minimal performance gain
self.dev_re = re.compile(
'([A-Za-z0-9\.]+):\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)\s+(\d+)'
)
self.meminfo_re = re.compile('([\w]+):\s(\d+)\s')
self.proc_stats_re = re.compile('([A-Za-z0-9]+)\s+(\d+)')
self.diskstatus_re = re.compile(
'\s+(\d+)\s+(\d+)\s(\w+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)\s(\d+)'
)
# Server main loop
while True:
self.getProcesses()
time.sleep(1)
def S(self, key):
return key.replace('.', '_')
def getUUID(self):
return re.sub('_|-|=', '0',
base64.urlsafe_b64encode(uuid.uuid4().bytes))
# Get all the processes that are not system
# processes
def getProcesses(self, ):
""" Get all system processes """
# walk the /proc
data = {}
stats = {}
data['uuid'] = self.getUUID()
d = datetime.datetime.utcnow()
data['hostname'] = socket.gethostname()
data['timestamp_year'] = d.strftime('%y')
data['timestamp_month'] = d.strftime('%m')
data['timestamp_day'] = d.strftime('%d')
data['timestamp_hour'] = d.strftime('%H')
data['timestamp_minute'] = d.strftime('%M')
data['timestamp_second'] = d.strftime('%S')
for item in os.listdir('/proc'):
if re.match('[\d]+', item):
# Start collecting statistics
# We only want statistics for processes that
# are not system processes, i.e. user jobs
stat_info = None
try:
stat_info = os.stat('/proc/' + item + '/status')
except Exception, err:
self.l.error(err)
if stat_info is not None:
# It is assumed that the ownership of the pid directory
# represents the owner of the process and as such we can
# use this to determine if this is a "non-system" or "non-root"
# process
if stat_info.st_uid >= 500:
self.getProcStatistics(item, stats)
# Get memory information
self.getMemInfo(stats)
# Get network statistics
self.getNetworkStats(stats)
# Get processor statistics
self.getProcessorStats(stats)
# Get Disk statistics
self.getDiskStats(stats)
# Get Infiniband statistics
self.getInfinibandStats(stats)
data['data'] = stats
self.l.info("Sending data")
self.rabbitcom.send_message(data)