def main():
unis_client = UNISInstance(unis_url=settings.UNIS_URL)
unis_client.register_service_to_unis()
s = Scheduler()
while 1:
start_time = time.time()
s.check_settings()
s.check_probes()
s.check_procs()
sleep_time = (start_time + settings.CHECK_INTERVAL) - time.time()
if sleep_time > 0.001:
time.sleep(sleep_time)
def get_interface_subjects(self):
netdev = self._proc.open('net', 'dev')
faces = []
subjects = {}
type_map = {'ipv4': netifaces.AF_INET,
'ipv6': netifaces.AF_INET6,
'mac': netifaces.AF_LINK}
for line in netdev:
line = line.split()
if line[0].count(":"):
faces.append(line[0][:line[0].index(":")])
unis = UNISInstance(unis_url=self.unis_url)
for face in faces:
post_dict = {}
try:
capacity = ethtool.get_speed(face)
except OSError:
capacity = 0
# assume each port is a layer2 port for the main 'address'
try:
l2_addr = netifaces.ifaddresses(face)[type_map['mac']]
if len(l2_addr):
addr = {"type": "mac", "address": l2_addr[0]['addr']}
post_dict['address'] = addr
except:
pass
# add all the other address info we can find
post_dict['properties'] = {}
for t in type_map:
try:
addrs = netifaces.ifaddresses(face)[type_map[t]]
for a in addrs:
addr = {"type": t, "address": a['addr']}
post_dict['properties'][t] = addr
except:
pass
### some sort of verification here that capacity is right
post_dict['name'] = face
post_dict['capacity'] = capacity
# hack in a 'nodeRef' so we can find port from rspec
post_dict['nodeRef'] = settings.URN_STRING[:-1]
resp = unis.post_port(post_dict)
if isinstance(resp, dict):
subjects[face]=resp['selfRef']
return subjects
def __init__(self, service, measurement):
self.config = measurement["configuration"]
self.service = service
self.unis = UNISInstance(service)
self.pname = None
self.pidfile = None
self.id = None
try:
self.serviceType = self.config["service_type"]
except Exception:
logger.error("__init__", msg="Must specify service_type!")
try:
self.accessPoint = self.config["service_accesspoint"]
except Exception:
logger.error("__init__", msg="Must specify access point!")
try:
self.pidfile = self.config.get("pidfile", None)
except Exception:
logger.warn("__init__", msg="Config does not specify pidfile")
self.pname = self.config.get("process_name", None)
# check for existing service given accessPoint and serviceType
try:
service = self.unis.get("/services?accessPoint=%s&serviceType=%s" %
(self.accessPoint, self.serviceType))
if service and len(service):
self.id = service[0]["id"]
except Exception as e:
logger.error("%s" % e)
def __init__(self, initial_config={}, node_id=None, urn=None):
if not node_id:
node_id = settings.UNIS_ID
self.node_id = node_id
self.urn = urn
self.config = initial_config
self.unis = UNISInstance(self.config)
self.service_setup = False
def __init__(self, initial_config={}, node_id=None, urn=None):
if not node_id:
node_id = settings.UNIS_ID
self.node_id = node_id
self.urn = urn
self.config = initial_config
self.unis = UNISInstance(self.config)
self.node_setup = False
self.service_setup = False
self.exponential_backoff = int(self.config["properties"]["configurations"]["unis_poll_interval"])
def __init__(self, service, measurement):
self.config = measurement["configuration"]
self.service = service
self.measurement = measurement
self.collections_created = False
self.ms = MSInstance(service, measurement)
self.dl = DataLogger(service, measurement)
self.mids = {} # {subject1: {metric1:mid, metric2:mid}, subj2: {...}}
# {mid: [{"ts": ts, "value": val}, {"ts": ts, "value": val}]}
self.mid_to_data = {}
self.mid_to_et = {}
self.unis = UNISInstance(service)
self.num_collected = 0
def __init__(self, service, measurement):
self.service = service
self.measurement = measurement
self.config = measurement["configuration"]
logger.debug('Probe.__init__', config=self.config)
self._proc = Proc(self.config.get("proc_dir", "/proc/"))
self.node_subject=self.config.get(
"subject",
self.config.get("runningOn", {}).get('href', 'not found'))
self.port_match_method=self.config.get(
"port_match_method", "geni_utils.mac_match")
self.port_match_method=blipp_import_method(self.port_match_method)
self.unis = UNISInstance(service)
logger.debug('Probe.__init__ ', subject=self.node_subject)
self.subjects=self.get_interface_subjects()
def __init__(self, service, measurement):
self.service = service
self.measurement = measurement
self.config = measurement["configuration"]
self.unis = UNISInstance(service)
self.command = self._substitute_command(str(self.config.get("command")), self.config)
# TODO: get the latest ts and store it and use it as query bound next run.
query_url = 'http://*****:*****@r*Buve5'
query_db = "db=collectd"
query_table = self.config.get("table")
query_every = self.config.get("schedule_params").get("every")
query = "q=select * from {0} where time > now() - {1}s;".format(query_table, query_every)
self.command = ['curl', '-GET', query_url,
'--data-urlencode', query_db,
'--data-urlencode', query]
class ServiceConfigure(object):
'''
ServiceConfigure is meant to be a generic class for any service
which registers itself to, and gets configuration from UNIS. It
was originally developed for BLiPP, but BLiPP specific features
should be in the BlippConfigure class which extends
ServiceConfigure.
'''
def __init__(self, initial_config={}, node_id=None, urn=None):
if not node_id:
node_id = settings.UNIS_ID
self.node_id = node_id
self.urn = urn
self.config = initial_config
self.unis = UNISInstance(self.config)
self.service_setup = False
def initialize(self):
self._setup_node(self.node_id)
self._setup_service()
def refresh(self):
r = self.unis.get("/services/" + self.config["id"])
if not r:
logger.warn('refresh', msg="refresh failed")
else:
self.config = r
def _setup_node(self, node_id):
config = self.config
logger.debug('_setup_node', config=pprint.pformat(config))
hostname = settings.HOSTNAME
urn = settings.HOST_URN if not self.urn else self.urn
if node_id:
r = self.unis.get("/nodes/" + str(node_id))
if not r:
logger.warn('_setup_node', msg="node id %s not found" % node_id)
node_id = None
if not node_id:
r = self.unis.get("/nodes?urn=" + urn)
if r and len(r):
r = r[0]
logger.info('_setup_node',
msg="found node with our URN and id %s" % r["id"])
else:
r = self.unis.post("/nodes",
data={
"$schema": settings.SCHEMAS["nodes"],
"name": hostname,
"urn": urn})
if r:
self.node_id = r["id"]
if r:
config["runningOn"] = {
"href": r["selfRef"],
"rel": "full"}
self.node_setup = True
else:
config["runningOn"] = {"href": ""}
logger.warn('_setup_node', msg="Unable to set up node in UNIS")
def _setup_service(self):
config = self.config
logger.debug('_setup_service', config=pprint.pformat(config))
r = None
if config.get("id", None):
r = self.unis.get("/services/" + config["id"])
if not r:
logger.warn('_setup_service',
msg="service id not specified or not found "\
"unis instance ...querying for service")
rlist = self.unis.get("/services?name=" + config.get("name", None) +\
"&runningOn.href=" + config["runningOn"]["href"] + "&limit=2")
# loop over the returned services and find one that
# doesn't return 410 see
# https://uisapp2.iu.edu/jira-prd/browse/GEMINI-98
if rlist:
for i in range(len(rlist)):
r = self.unis.get('/services/' + rlist[i]["id"])
if r:
if isinstance(r, list):
logger.warn('_setup_service',
msg="id not unique... taking first result")
r = r[0]
logger.info('_setup_service',
msg="%s service found with id %s" % (config["name"], r["id"]))
break
else:
logger.warn('_setup_service',
msg="no service found by id or querying "\
"...creating new service")
if r:
merge_dicts(config, r)
# always update UNIS with the merged config
if config.get("id", None):
r = self.unis.put("/services/" + config["id"], data=config)
else:
r = self.unis.post("/services", data=config)
if r:
merge_dicts(config, r)
if r:
self.service_setup = True
else:
logger.warn('_setup_service', msg="unable to set up service in UNIS")
def get(self, key, default=None):
try:
return self.config[key]
except KeyError:
return default
def __getitem__(self, key):
'''
This allows an object which is an instance of this class to behave
like a dictionary when queried with [] syntax
'''
return self.config[key]
class Collector:
"""Collects reported measurements and aggregates them for
sending to MS at appropriate intervals.
Also does a bunch of other stuff which should probably be handled by separate classes.
Creates all the metadata objects, and the measurement object in UNIS for all data inserted.
Depends directly on the MS and UNIS... output could be far more modular.
"""
def __init__(self, service, measurement):
self.config = measurement["configuration"]
self.service = service
self.measurement = measurement
self.collections_created = False
self.ms = MSInstance(service, measurement)
self.dl = DataLogger(service, measurement)
self.mids = {} # {subject1: {metric1:mid, metric2:mid}, subj2: {...}}
# {mid: [{"ts": ts, "value": val}, {"ts": ts, "value": val}]}
self.mid_to_data = {}
self.mid_to_et = {}
self.unis = UNISInstance(service)
self.num_collected = 0
def insert(self, data, ts):
'''
Called (by probe_runner) to insert new data into this collector object.
'''
mids = self.mids
for subject, met_val in data.iteritems():
if "ts" in met_val:
ts = met_val["ts"]
del met_val["ts"]
for metric, value in met_val.iteritems():
if metric not in self.measurement["eventTypes"]:
self._add_et(metric)
if not metric in mids.get(subject, {}):
r = self.unis.find_or_create_metadata(subject,
metric,
self.measurement)
mids.setdefault(subject, {})[metric] = r["id"]
self.mid_to_data[r["id"]] = []
self.mid_to_et[mids[subject][metric]] = metric
self._insert_datum(mids[subject][metric], ts, value)
self.num_collected += 1
if self.num_collected >= self.config["reporting_params"]:
ret = self.report()
if ret:
self.num_collected = 0
def _insert_datum(self, mid, ts, val):
item = dict({"ts": ts * 10e5,
"value":val})
self.mid_to_data[mid].append(item)
def _add_et(self, metric):
self.measurement["eventTypes"].append(metric)
r = self.unis.put("/measurements/" +
self.measurement["id"],
data=self.measurement)
if r:
self.measurement = r
def report(self):
'''
Send all data collected so far, then clear stored data.
'''
post_data = []
for mid, data in self.mid_to_data.iteritems():
if len(data):
post_data.append({"mid":mid, "data":data})
ms_ret = self.ms.post_data(post_data)
dl_ret = self.dl.write_data(post_data, self.mid_to_et)
if not ms_ret and not dl_ret and self.num_collected < self.config["reporting_tolerance"] * self.config["reporting_params"]:
return None
self._clear_data()
return True
def _clear_data(self):
for mid in self.mid_to_data:
self.mid_to_data[mid]=[]
class Probe:
TTL_DEFAULT = 600
def __init__(self, service, measurement):
self.config = measurement["configuration"]
self.service = service
self.unis = UNISInstance(service)
self.pname = None
self.pidfile = None
self.id = None
try:
self.serviceType = self.config["service_type"]
except Exception:
logger.error("__init__", msg="Must specify service_type!")
try:
self.accessPoint = self.config["service_accesspoint"]
except Exception:
logger.error("__init__", msg="Must specify access point!")
try:
self.pidfile = self.config.get("pidfile", None)
except Exception:
logger.warn("__init__", msg="Config does not specify pidfile")
self.pname = self.config.get("process_name", None)
# check for existing service given accessPoint and serviceType
try:
service = self.unis.get("/services?accessPoint=%s&serviceType=%s" %
(self.accessPoint, self.serviceType))
if service and len(service):
self.id = service[0]["id"]
except Exception as e:
logger.error("%s" % e)
def get_data(self):
stat = "UNKNOWN"
# check pidfile
if self.pidfile:
pid = None
try:
self.pidfile = open(self.config["pidfile"])
pid = self.pidfile.read().rstrip()
except IOError:
logger.warn("__init__", msg="Could not open pidfile: %s" % self.config["pidfile"])
if pid:
try:
os.kill(int(pid), 0)
stat = "ON"
except OSError, err:
if err.errno == errno.ESRCH:
stat = "OFF"
elif err.errno == errno.EPERM:
logger.warn("get_data", msg="No permission to signal this process: %s" % pid)
else:
logger.warn("get_data", msg="Uknown error: %s" % error.errno)
#We could assume if the pidfile exists, the process is running
#if stat is "UNKNOWN" and os.path.exists("/proc/"+pid):
# stat = "ON"
self.pidfile.close()
# check process name, this takes priority
if self.pname:
processes = psutil.process_iter()
for p in processes:
if p.name() == self.pname:
stat = "ON"
self.send_service(status=stat)
return []
class Probe:
"""Get network statistics
"""
UNUSED_METRICS = ["errs_in", "errs_out", "drop_in", "drop_out",
"fifo_in", "fifo_out", "frame_in", "compressed_in",
"compressed_out", "multicast_in", "colls_out", "colls_in",
"carrier_out", "carrier_in", "frame_out", "multicast_out"]
def __init__(self, service, measurement):
self.service = service
self.measurement = measurement
self.config = measurement["configuration"]
logger.debug('Probe.__init__', config=self.config)
self._proc = Proc(self.config.get("proc_dir", "/proc/"))
self.node_subject=self.config.get(
"subject",
self.config.get("runningOn", {}).get('href', 'not found'))
self.port_match_method=self.config.get(
"port_match_method", "geni_utils.mac_match")
self.port_match_method=blipp_import_method(self.port_match_method)
self.unis = UNISInstance(service)
logger.debug('Probe.__init__ ', subject=self.node_subject)
self.subjects=self.get_interface_subjects()
def get_data(self):
netdev = self._proc.open('net', 'dev')
netsnmp = self._proc.open('net', 'snmp')
netdev.readline()
data = self._get_dev_data(netdev.read())
sdata = self._get_snmp_data(netsnmp.read())
data[self.node_subject] = sdata
data = full_event_types(data, EVENT_TYPES)
return data
def _get_dev_data(self, dev_string):
headers_regex = re.compile(
'[^|]*\|(?P<rxheaders>[^|]*)\|(?P<txheaders>.*)')
dev_lines = dev_string.splitlines()
matches = headers_regex.search(dev_lines.pop(0)).groupdict()
txheaders = [ head + "_in" for head in matches['txheaders'].split() ]
rxheaders = [ head + "_out" for head in matches['rxheaders'].split() ]
headers = txheaders + rxheaders
data = {}
for line in dev_lines:
if not line:
continue
line = line.replace(':', ' ')
line = line.split()
iface = line.pop(0)
face_data = dict(zip(headers, line))
self._vals_to_int(face_data)
errors = face_data.pop('errs_in') + face_data.pop('errs_out')
drops = face_data.pop('drop_in') + face_data.pop('drop_out')
face_data['errors'] = errors
face_data['drops'] = drops
for metric in self.UNUSED_METRICS:
if metric in face_data:
del face_data[metric]
data[self.subjects[iface]] = face_data
return data
def _vals_to_int(self, adict):
for k,v in adict.items():
adict[k] = int(v)
return adict
def _get_snmp_data(self, snmp_string):
data = {}
lines = snmp_string.splitlines()
i = 0
while i < len(lines):
line = lines[i].split()
if line[0].lower()=="tcp:":
i += 1
in_index = line.index("InSegs")
out_index = line.index("OutSegs")
retrans_index = line.index("RetransSegs")
dataline = lines[i].split()
data.update({"tcp_segments_in":dataline[in_index],
"tcp_segments_out":dataline[out_index],
"tcp_retrans":dataline[retrans_index]})
elif line[0].lower()=="udp:":
i += 1
in_index = line.index("InDatagrams")
out_index = line.index("OutDatagrams")
dataline = lines[i].split()
data.update({"datagrams_in":dataline[in_index],
"datagrams_out":dataline[out_index]})
i += 1
return self._vals_to_int(data)
def get_interface_subjects(self):
subjects = {}
unis_ports = self.get_interfaces_in_unis()
for face in netifaces.interfaces():
local_port_dict = self._build_port_dict(face)
portRef = self._find_or_post_port(unis_ports, local_port_dict, self.port_match_method)
if isinstance(portRef, str) or isinstance(portRef, unicode):
subjects[face]=portRef
else:
logger.warn('get_interface_subjects',
msg="subject for face %s is of an unexpected type %s, portRef=%s"%(face,
type(portRef),
portRef))
subjects[face]="unexpected type"
return subjects
def _build_port_dict(self, port_name):
type_map = {'ipv4': netifaces.AF_INET,
'ipv6': netifaces.AF_INET6,
'mac': netifaces.AF_LINK}
post_dict = {}
try:
capacity = ethtool.get_speed(port_name)
except OSError:
capacity = 0
# assume each port is a layer2 port for the main 'address'
try:
l2_addr = netifaces.ifaddresses(port_name)[type_map['mac']]
if len(l2_addr):
addr = {"type": "mac", "address": l2_addr[0]['addr']}
post_dict['address'] = addr.strip().replace(':', '').lower()
except:
pass
# add all the other address info we can find
post_dict['properties'] = {}
for t in type_map:
try:
addrs = netifaces.ifaddresses(port_name)[type_map[t]]
for a in addrs:
addr = {"type": t, "address": a['addr']}
post_dict['properties'][t] = addr
except Exception as e:
logger.exc('get_interface_subjects', e)
# TODO some sort of verification here that capacity is right
post_dict['name'] = port_name
post_dict['capacity'] = capacity
# hack in a 'nodeRef' so we can find port from rspec
post_dict['nodeRef'] = settings.HOST_URN[:-1]
return post_dict
def get_interfaces_in_unis(self):
node = self.unis.get(self.service["runningOn"]["href"])
port_list = node.get('ports', [])
ports = []
for port in port_list:
ports.append(self.unis.get(port['href']))
return ports
def _find_or_post_port(self, ports, local_port, matching_method):
for port in ports:
if matching_method(port, local_port):
return port["selfRef"]
post = self.unis.post_port(local_port)
if post:
return post["selfRef"]
else:
logger.warn('_find_or_post_port',
msg="post seems to have failed... subject for %s will be wrong" % local_port['name'])
return "failed"
class Probe:
def __init__(self, service, measurement):
self.service = service
self.measurement = measurement
self.config = measurement["configuration"]
self.unis = UNISInstance(service)
self.command = self._substitute_command(str(self.config.get("command")), self.config)
# TODO: get the latest ts and store it and use it as query bound next run.
query_url = 'http://*****:*****@r*Buve5'
query_db = "db=collectd"
query_table = self.config.get("table")
query_every = self.config.get("schedule_params").get("every")
query = "q=select * from {0} where time > now() - {1}s;".format(query_table, query_every)
self.command = ['curl', '-GET', query_url,
'--data-urlencode', query_db,
'--data-urlencode', query]
def get_data(self):
proc = subprocess.Popen(self.command,
stdout = subprocess.PIPE,
stderr = subprocess.PIPE)
output = proc.communicate()
if not output[0]:
raise CmdError(output[1])
try:
data = self._extract_data(output[0])
except ValueError as e:
logger.exc("get_data", e)
return {}
# sample data [{'wsu.stor1-href': {ts: <ts>, et: [instance, type, type_instance, value]}},
# {'um.stor2-href': {ts: <ts>, et: [instance, type, type_instance, value]}}]
return data
def _extract_data(self, stdout):
def normalize(values, time_index, host_index, subject_index, event_index, table_name):
'''
carve the host and turn it into the subject of the data
change name 'time' to 'ts' and remove it from the data
'''
if '.'.join([table_name, values[event_index]]) not in EVENT_MAPS:
# not event type that we are interested
return {}
host_name = values[host_index]
try:
host_obj = self.unis.get("/nodes?name=" + host_name)
except ConnectionError:
host_obj = None
if not host_obj:
# host not registered in this domain OR ConnectionError
return {}
if table_name == ETHTABLE:
if 'ports' in host_obj[0]:
ports = host_obj[0]['ports']
subject = None
for port in ports:
port_id = port['href'].split('/')[-1]
try:
port_obj = self.unis.get("/ports/" + port_id)
except ConnectionError:
port_obj = None
if port_obj and port_obj['name'] == values[subject_index]:
subject = port_obj['selfRef']
break
if not subject:
# none of the eth matches
return {}
else:
# this host has no eth uploaded
return {}
else :
# other tables (disk stat) will aggregate all instances
subject = host_obj[0]['selfRef']
event_type = EVENT_MAPS['.'.join([table_name, values[event_index]])]
dt = datetime.datetime.strptime(values[time_index], "%Y-%m-%dT%H:%M:%S.%fZ")
time = calendar.timegm(dt.timetuple()) + (dt.microsecond / 1000000.0)
values = values[-1]
return {subject: {'ts': time, event_type: values}}
json_output = json.loads(stdout)
if not json_output['results'][0]:
return []
table_name = json_output['results'][0]['series'][0]['name']
event_column = EVENT_COLUMNS[table_name]
i = json_output['results'][0]['series'][0]['columns'].index('time')
j = json_output['results'][0]['series'][0]['columns'].index('host')
k = json_output['results'][0]['series'][0]['columns'].index('instance')
e = json_output['results'][0]['series'][0]['columns'].index(event_column)
ret = map(lambda x: normalize(x, i, j, k, e, table_name), json_output['results'][0]['series'][0]['values'])
ret = [x for x in ret if x]
if table_name == ETHTABLE:
pass
else:
# aggregate all drive on one host
# ts is chosen randomly from the same subject-et items
tmp = dict()
for data in ret:
et = filter(lambda x: x != 'ts', data.itervalues().next().keys())[0]
if (data.iterkeys().next(), et) in tmp:
tmp[(data.iterkeys().next(), et)][et] += data.itervalues().next()[et]
else:
tmp[(data.iterkeys().next(), et)] = data.itervalues().next()
ret = []
for k, v in tmp.iteritems():
ret.append({k[0]: v})
return ret
def _substitute_command(self, command, config):
''' command in form "ping $ADDRESS"
config should have substitutions like "address": "example.com"
Note; now more complex
'''
command = shlex.split(command)
ret = []
for item in command:
if item[0] == '$':
if item[1:] in config:
val = config[item[1:]]
if isinstance(val, bool):
if val:
ret.append(item[1:])
elif item[1]=="-":
ret.append(item[1:])
ret.append(str(val))
else:
ret.append(str(val))
elif item:
ret.append(item)
logger.info('substitute_command', cmd=ret, name=self.config['name'])
return ret
