class PutMetricBolt(storm.BasicBolt):
BOLT_NAME = "PutMetricBolt"
def initialize(self, stormconf, context):
self.cass = Cassandra()
self.metrics = {}
def log(self, msg):
storm.log("[%s] %s" % (self.BOLT_NAME, msg))
def tracelog(self, e):
msg = traceback.format_exc(e)
for line in msg.splitlines():
self.log("TRACE: " + line)
def process_put_metric_data_msg(self, metric_key, message):
"""
데이터베이스에 MetricArchive 컬럼패밀리에 입력된 값 추가. 메모리
(self.metrics)에도 입력된 값 추가.
메모리 상의 메트릭을 기반으로 데이터베이스에 StatArchive 컬럼패밀리
업데이트.
"""
# 메시지 값이 없는 경우 종료
if message["value"] is None:
return
# 메트릭 가져오기
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
timestamp = utils.parse_strtime(message["timestamp"])
self.metrics[metric_key].put_metric_data(
metric_key, timestamp=timestamp, value=message["value"], unit=message["unit"]
)
def process_put_metric_alarm_msg(self, metric_key, message):
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
project_id = message["project_id"]
metricalarm = message["metricalarm"]
self.metrics[metric_key].put_alarm(project_id, metricalarm)
def process_delete_metric_alarms_msg(self, metric_key, message):
alarmkey = UUID(message["alarmkey"])
self.log("debug: %s" % self.metrics.keys())
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
self.metrics[metric_key].delete_metric_alarm(alarmkey)
def process_set_alarm_state_msg(self, metric_key, message):
project_id = message.get("project_id")
alarm_name = message.get("alarm_name")
state_reason_data = message.get("state_reason_data")
alarm_key = self.cass.get_metric_alarm_key(project_id, alarm_name)
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
metric = self.metrics[metric_key]
metricalarm = metric.alarms[alarm_key]
metricalarm["state_reason"] = message.get("state_reason")
metricalarm["state_value"] = message.get("state_value")
metricalarm["state_reason_data"] = message.get("state_reason_data")
# write into database
alarm_columns = {"state_reason": message.get("state_reason"), "state_value": message.get("state_value")}
if state_reason_data:
alarm_columns["state_reason_data"] = state_reason_data
self.cass.put_metric_alarm(alarm_key, alarm_columns)
def process_check_metric_alarms_msg(self):
now = datetime.utcnow()
for metric in self.metrics.itervalues():
min_start_period = timedelta(seconds=metric.MIN_START_PERIOD)
if (metric.lastchecked and (now - metric.lastchecked > min_start_period)) or (metric.lastchecked == None):
metric.check_alarms()
def process(self, tup):
try:
metric_key = UUID(tup.values[0]) if tup.values[0] else None
except ValueError:
self.log("badly formed hexadecimal UUID string - %s" % tup.values[0])
return
message = json.loads(tup.values[1])
message_id = message.get("message_id")
if message_id == PUT_METRIC_DATA_MSG_ID:
self.log("process put_metric_data_msg (%s)" % message)
self.process_put_metric_data_msg(metric_key, message)
elif message_id == PUT_METRIC_ALARM_MSG_ID:
self.log("process put_metric_alarm_msg (%s)" % message)
self.process_put_metric_alarm_msg(metric_key, message)
elif message_id == DELETE_ALARMS_MSG_ID:
self.log("process put_metric_alarm_msg (%s)" % message)
self.process_delete_metric_alarms_msg(metric_key, message)
elif message_id == SET_ALARM_STATE_MSG_ID:
self.log("process set_alarm_state_msg (%s)" % message)
self.process_set_alarm_state_msg(metric_key, message)
elif message_id == CHECK_METRIC_ALARM_MSG_ID:
self.log("process check_metric_alarm_msg (%s)" % message)
self.process_check_metric_alarms_msg()
else:
self.log("unknown message")
class PutMetricBolt(storm.BasicBolt):
BOLT_NAME = "PutMetricBolt"
def initialize(self, stormconf, context):
self.pid = os.getpid()
self.cass = Cassandra()
self.metrics = {}
self.mc = memcache.Client(FLAGS.memcached_servers, debug=0)
def process_put_metric_data_msg(self, metric_key, message):
"""
Put metric data into both memory and database
"""
# Load statistics data in memory
if metric_key not in self.metrics:
max_retries = 3
for i in range(max_retries + 1):
try:
self.metrics[metric_key] = MetricMonitor(metric_key,
self.cass)
break
except ResourceNotFound:
if i + 1 < max_retries:
LOG.warn("Metric %s is not in the database. " \
"retry... %d", metric_key, i + 1)
time.sleep(1)
else:
LOG.error("Metric %s is not in the database.",
metric_key)
return
timestamp = utils.parse_strtime(message['timestamp'])
self.metrics[metric_key].put_metric_data(metric_key,
timestamp=timestamp,
value=message['value'],
unit=message['unit'])
def process_put_metric_alarm_msg(self, metric_key, message):
def get_alarm_key(project_id, alarm_name):
key = self.cass.get_metric_alarm_key(project_id, alarm_name)
return key
def metricalarm_for_json(metricalarm):
cut = metricalarm.get('alarm_configuration_updated_timestamp')
alarm_for_json = {
'actionEnabled': metricalarm.get('actions_enabled', False),
'alarmActions': metricalarm.get('alarm_actions', []),
'alarmArn': metricalarm.get('alarm_arn'),
'alarmConfigurationUpdatedTimestamp': utils.strtime(cut),
'alarmDescription': metricalarm.get('alarm_description'),
'alarmName': metricalarm.get('alarm_name'),
'comparisonOperator': metricalarm.get('comparison_operator'),
'dimensions': metricalarm.get('dimensions'),
'evaluationPeriods': metricalarm.get('evaluation_periods'),
'insufficientDataActions':
metricalarm.get('insufficient_data_actions', []),
'metricName':metricalarm.get('metric_name'),
'namespace':metricalarm.get('namespace'),
'okactions':metricalarm.get('ok_actions', []),
'statistic':metricalarm.get('statistic'),
'threshold':metricalarm.get('threshold'),
'unit':metricalarm.get('unit'),
}
return alarm_for_json
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
project_id = message['project_id']
metricalarm = message['metricalarm']
# build metricalarm column, alarmhistory column
alarm_key = get_alarm_key(project_id, metricalarm['alarm_name'])
history_type = 'Update' if alarm_key else 'Create'
now = utils.utcnow()
if history_type == 'Update':
original_alarm = self.cass.get_metric_alarm(alarm_key)
for dict_key in ['state_updated_timestamp', 'state_reason',
'state_reason_data', 'state_value', 'project_id']:
metricalarm[dict_key] = original_alarm[dict_key]
metricalarm['alarm_configuration_updated_timestamp'] = now
history_data = json.dumps({
'updatedAlarm':metricalarm_for_json(metricalarm),
'type':history_type,
'version': '1.0'
})
summary = "Alarm %s updated" % metricalarm['alarm_name']
else:
alarm_key = uuid.uuid4()
state_reason = "Unchecked: Initial alarm creation"
metricalarm.update({'state_updated_timestamp': now,
'alarm_configuration_updated_timestamp': now,
'state_reason': state_reason,
'state_reason_data': json.dumps({}),
'state_value': "INSUFFICIENT_DATA",
'project_id': project_id})
history_data = json.dumps({
'createdAlarm': metricalarm_for_json(metricalarm),
'type':history_type, 'version': '1.0'
})
summary = "Alarm %s created" % metricalarm['alarm_name']
metricalarm['metric_key'] = metric_key
history_key = uuid.uuid4()
history_column = {
'project_id': project_id,
'alarm_key': alarm_key,
'alarm_name': metricalarm['alarm_name'],
'history_data': history_data,
'history_item_type': 'ConfigurationUpdate',
'history_summary':summary,
'timestamp': utils.utcnow()
}
self.cass.put_metric_alarm(alarm_key, metricalarm)
self.cass.insert_alarm_history(history_key, history_column)
LOG.info("metric alarm inserted: %s %s", alarm_key, metricalarm)
# load metric in memory
self.metrics[metric_key].put_alarm(alarm_key, metricalarm)
def process_delete_metric_alarms_msg(self, metric_key, message):
alarmkey = UUID(message['alarmkey'])
LOG.debug("Metric keys %s", self.metrics.keys())
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
self.metrics[metric_key].delete_metric_alarm(alarmkey)
def process_set_alarm_state_msg(self, metric_key, message):
project_id = message.get('project_id')
alarm_name = message.get('alarm_name')
state_reason_data = message.get('state_reason_data')
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
metric = self.metrics[metric_key]
ret = self.cass.get_metric_alarm_key(project_id, alarm_name)
if ret:
alarm_key = ret
try:
metricalarm = metric.alarms[alarm_key]
except KeyError:
LOG.warn("alarm key [%s] is found, but alarm is not found.",
alarm_key)
return
else:
LOG.warn("alarm key [%s] is not found.", alarm_key)
return
metricalarm['state_reason'] = message.get('state_reason')
metricalarm['state_value'] = message.get('state_value')
metricalarm['state_reason_data'] = message.get('state_reason_data')
# write into database
alarm_columns = {'state_reason':message.get('state_reason'),
'state_value':message.get('state_value')}
if state_reason_data:
alarm_columns['state_reason_data'] = state_reason_data
alarm_columns['project_id'] = project_id
self.cass.put_metric_alarm(alarm_key, alarm_columns)
def process_check_metric_alarms_msg(self, message):
query_time = datetime.utcnow()
stale_metrics = []
ready_to_evaluate = message.get('ready_to_evaluate')
for key, metric in self.metrics.iteritems():
is_stale = metric.is_stale()
if is_stale:
stale_metrics.append(key)
if (not is_stale) and ready_to_evaluate:
metric.check_alarms(query_time)
for key in stale_metrics:
try:
metric = self.metrics.pop(key)
metric.delete()
LOG.audit("Stale metric(%s) is deleted", str(key))
except KeyError:
LOG.error("KeyError occurred when delete stale metric(%s)",
str(key))
def process(self, tup):
message = json.loads(tup.values[1])
message_id = message['message_id']
message_uuid = message.get('message_uuid', None)
LOG.info("start processing msg[%s:%s]", message_id, message_uuid)
try:
metric_key = UUID(tup.values[0]) if tup.values[0] else None
except ValueError:
LOG.error("badly formed hexadecimal UUID string - %s",
tup.values[0])
return
if message_id == PUT_METRIC_DATA_MSG_ID:
# message deduplicate
if message_uuid:
mckey = "%s_message_uuid" % message_uuid
if not self.mc.get(mckey):
# 300 seconds TTL
self.mc.set(mckey, 1, 300)
LOG.info("process put_metric_data_msg (%s)", message)
self.process_put_metric_data_msg(metric_key, message)
else:
LOG.info("Message duplicated. %s", message_uuid)
elif message_id == PUT_METRIC_ALARM_MSG_ID:
LOG.info("process put_metric_alarm_msg (%s)", message)
self.process_put_metric_alarm_msg(metric_key, message)
elif message_id == DELETE_ALARMS_MSG_ID:
LOG.info("process delete_alarms_msg (%s)", message)
self.process_delete_metric_alarms_msg(metric_key, message)
elif message_id == SET_ALARM_STATE_MSG_ID:
LOG.info("process set_alarm_state_msg (%s)", message)
self.process_set_alarm_state_msg(metric_key, message)
elif message_id == CHECK_METRIC_ALARM_MSG_ID:
LOG.info("process check_metric_alarm_msg (%s)", message)
self.process_check_metric_alarms_msg(message)
else:
LOG.error("unknown message")
class API(object):
ROLLING_FUNC_MAP = {
'Average': rolling_mean, 'Minimum': rolling_min,
'Maximum': rolling_max, 'SampleCount': rolling_sum,
'Sum': rolling_sum,
}
def __init__(self):
self.cass = Cassandra()
self.rpc = rpc.RemoteProcedureCall()
def delete_alarms(self, project_id, alarm_names):
alarmkeys = []
for alarm_name in alarm_names:
k = self.cass.get_metric_alarm_key(project_id, alarm_name)
if not k:
raise ResourceNotFound("Alarm %s does not exists." %
alarm_name)
alarmkeys.append(str(k))
body = {'project_id': project_id, 'alarmkeys': alarmkeys} # UUID str
self.rpc.send_msg(rpc.DELETE_ALARMS_MSG_ID, body)
LOG.info("DELETE_ALARMS_MSG sent")
def describe_alarms(self, project_id, action_prefix=None,
alarm_name_prefix=None, alarm_names=None,
max_records=None, next_token=None, state_value=None):
"""
params:
project_id: string
action_prefix: TODO: not implemented yet.
alarm_name_prefix: string
alarm_names: string list
max_records: integer
next_token: string (uuid type)
state_value: string (OK | ALARM | INSUFFICIENT_DATA)
"""
alarms = self.cass.describe_alarms(project_id, action_prefix,
alarm_name_prefix, alarm_names,
max_records, next_token,
state_value)
return alarms
def describe_alarms_for_metric(self, project_id, namespace, metric_name,
dimensions=None, period=None,
statistic=None, unit=None):
"""
params:
project_id: string
metric_name: string
namespace: string
dimensions: dict
period: integer
statistic: string (SampleCount | Average | Sum | Minimum |
Maximum)
unit: string
"""
alarms = self.cass.describe_alarms_for_metric(project_id, namespace,
metric_name, dimensions=dimensions, period=period,
statistic=statistic, unit=unit)
return alarms
def describe_alarm_history(self, project_id, alarm_name=None,
end_date=None, history_item_type=None,
max_records=None, next_token=None,
start_date=None):
histories = self.cass.describe_alarm_history(
alarm_name=alarm_name, end_date=end_date,
history_item_type=history_item_type,
max_records=max_records, next_token=next_token,
start_date=start_date, project_id=project_id
)
return histories
def set_alarm_actions(self, project_id, alarm_names, enabled):
for alarm_name in alarm_names:
alarm_key = self.cass.get_metric_alarm_key(project_id, alarm_name)
self.cass.put_metric_alarm(alarm_key, {'actions_enabled':enabled})
def set_alarm_state(self, project_id, alarm_name, state_reason,
state_value, state_reason_data=None):
k = self.cass.get_metric_alarm_key(project_id, alarm_name)
if not k:
raise ResourceNotFound("Alarm %s does not exists." % alarm_name)
body = {'project_id': project_id, 'alarm_name': alarm_name,
'state_reason': state_reason, 'state_value': state_value,
'state_reason_data': state_reason_data}
self.rpc.send_msg(rpc.SET_ALARM_STATE_MSG_ID, body)
LOG.info("SET_ALARM_STATE_MSG sent")
def get_metric_statistics(self, project_id, end_time, metric_name,
namespace, period, start_time, statistics,
unit=None, dimensions=None):
"""
입력받은 조건에 일치하는 메트릭의 통계자료 리스트를 반환한다.
"""
def to_datapoint(df, idx):
datapoint = df.ix[idx].dropna()
if len(datapoint):
return idx, datapoint
end_idx = end_time.replace(second=0, microsecond=0)
start_idx = start_time.replace(second=0, microsecond=0)
start_ana_idx = start_idx - datetools.Minute() * (period / 60)
daterange = DateRange(start_idx, end_idx, offset=datetools.Minute())
daterange_ana = DateRange(start_ana_idx, end_idx,
offset=datetools.Minute())
# load default unit for metric from database
if unit == "None" or not unit:
metric_key = self.cass.get_metric_key(
project_id=project_id, namespace=namespace,
metric_name=metric_name, dimensions=dimensions
)
if metric_key:
unit = self.cass.get_metric_unit(metric_key)
else:
unit = "None"
# load statistics data from database
stats = self.cass.get_metric_statistics(
project_id=project_id, namespace=namespace,
metric_name=metric_name, start_time=start_ana_idx,
end_time=end_time, period=period, statistics=statistics,
dimensions=dimensions
)
period = period / 60 # convert sec to min
stat = DataFrame(index=daterange)
for statistic, series in zip(statistics, stats):
func = self.ROLLING_FUNC_MAP[statistic]
if statistic == 'SampleCount':
ts = TimeSeries(series, index=daterange_ana).fillna(0)
else:
ts = TimeSeries(series, index=daterange_ana)
stat[statistic] = func(ts, period, min_periods=0)
ret = filter(None, (to_datapoint(stat, i) for i in stat.index))
return ret
def list_metrics(self, project_id, next_token=None, dimensions=None,
metric_name=None, namespace=None):
"""
입력받은 조건과 일치하는 메트릭의 리스트를 반환한다.
"""
metrics = self.cass.list_metrics(project_id, namespace, metric_name,
dimensions, next_token)
return metrics
def put_metric_alarm(self, project_id, metricalarm):
"""
알람을 DB에 넣고 값이 빈 dictionary 를 반환한다.
메트릭 유무 확인
알람 히스토리 발생.
"""
def metricalarm_for_json(metricalarm):
alarm_for_json = {
'actionEnabled': metricalarm.get('actions_enabled', False),
'alarmActions': metricalarm.get('alarm_actions', []),
'alarmArn': metricalarm.get('alarm_arn'),
'alarmConfigurationUpdatedTimestamp':
metricalarm.get('alarm_configuration_updated_timestamp'),
'alarmDescription': metricalarm.get('alarm_description'),
'alarmName': metricalarm.get('alarm_name'),
'comparisonOperator': metricalarm.get('comparison_operator'),
'dimensions': metricalarm.get('dimensions'),
'evaluationPeriods': metricalarm.get('evaluation_periods'),
'insufficientDataActions':
metricalarm.get('insufficient_data_actions', []),
'metricName':metricalarm.get('metric_name'),
'namespace':metricalarm.get('namespace'),
'okactions':metricalarm.get('ok_actions', []),
'statistic':metricalarm.get('statistic'),
'threshold':metricalarm.get('threshold'),
'unit':metricalarm.get('unit'),
}
return alarm_for_json
now = utils.utcnow()
metricalarm = metricalarm.to_columns()
# 메트릭 유무 확인
metric_key = self.cass.get_metric_key_or_create(
project_id=project_id,
namespace=metricalarm['namespace'],
metric_name=metricalarm['metric_name'],
dimensions=json.loads(metricalarm['dimensions']),
unit=metricalarm['unit'],
)
metricalarm['project_id'] = project_id
metricalarm['metric_key'] = metric_key
metricalarm['alarm_arn'] = "arn:spcs:synaps:%s:alarm:%s" % (
project_id, metricalarm['alarm_name']
)
metricalarm['alarm_configuration_updated_timestamp'] = now
# 알람 유무 확인
alarm_key = self.cass.get_metric_alarm_key(
project_id=project_id, alarm_name=metricalarm['alarm_name']
)
if alarm_key:
history_type = 'Update'
before_alarm = self.cass.get_metric_alarm(alarm_key)
if before_alarm['metric_key'] != metricalarm['metric_key']:
raise InvalidRequest("Metric cannot be changed.")
metricalarm['state_updated_timestamp'] = \
before_alarm['state_updated_timestamp']
metricalarm['state_reason'] = before_alarm['state_reason']
metricalarm['state_reason_data'] = \
before_alarm['state_reason_data']
metricalarm['state_value'] = before_alarm['state_value']
else:
history_type = "Create"
alarm_key = uuid.uuid4()
metricalarm['state_updated_timestamp'] = utils.utcnow()
metricalarm['state_reason'] = "Unchecked: Initial alarm creation"
metricalarm['state_reason_data'] = json.dumps({})
metricalarm['state_value'] = "INSUFFICIENT_DATA"
# insert alarm into database
self.cass.put_metric_alarm(alarm_key, metricalarm)
LOG.debug("metric alarm inserted alarm key: %s" % (alarm_key))
# to make json, convert datetime type into str
metricalarm['state_updated_timestamp'] = utils.strtime(
metricalarm['state_updated_timestamp']
)
metricalarm['alarm_configuration_updated_timestamp'] = utils.strtime(
metricalarm['alarm_configuration_updated_timestamp']
)
metricalarm['metric_key'] = str(metric_key)
if history_type == "Update":
history_data = json.dumps({
'updatedAlarm':metricalarm_for_json(metricalarm),
'type':history_type,
'version': '1.0'
})
summary = "Alarm %s updated" % metricalarm['alarm_name']
else:
history_data = json.dumps({
'createdAlarm': metricalarm_for_json(metricalarm),
'type':history_type, 'version': '1.0'
})
summary = "Alarm %s created" % metricalarm['alarm_name']
history_key = uuid.uuid4()
history_column = {
'project_id': project_id,
'alarm_key': alarm_key,
'alarm_name': metricalarm['alarm_name'],
'history_data': history_data,
'history_item_type': 'ConfigurationUpdate',
'history_summary':summary,
'timestamp': utils.utcnow()
}
self.cass.insert_alarm_history(history_key, history_column)
message = {'project_id': project_id, 'metric_key': str(metric_key),
'metricalarm': metricalarm}
self.rpc.send_msg(rpc.PUT_METRIC_ALARM_MSG_ID, message)
LOG.info("PUT_METRIC_ALARM_MSG sent")
return {}
def put_metric_data(self, project_id, namespace, metric_name, dimensions,
value, unit, timestamp, is_admin=False):
"""
metric data 를 입력받아 MQ 에 넣고 값이 빈 dictionary 를 반환한다.
"""
if namespace.startswith("SPCS/") and not is_admin:
raise AdminRequired()
message = {'project_id': project_id, 'namespace':namespace,
'metric_name': metric_name, 'dimensions': dimensions,
'value':value, 'unit':unit, 'timestamp':timestamp}
self.rpc.send_msg(rpc.PUT_METRIC_DATA_MSG_ID, message)
LOG.info("PUT_METRIC_DATA_MSG sent")
return {}
class API(object):
ROLLING_FUNC_MAP = {
'Average': rolling_mean, 'Minimum': rolling_min,
'Maximum': rolling_max, 'SampleCount': rolling_sum,
'Sum': rolling_sum,
}
def __init__(self):
self.cass = Cassandra()
self.rpc = rpc.RemoteProcedureCall()
def delete_alarms(self, context, project_id, alarm_names):
alarmkeys = []
for alarm_name in alarm_names:
k = self.cass.get_metric_alarm_key(project_id, alarm_name)
if not k:
raise ResourceNotFound("Alarm %s does not exists." %
alarm_name)
alarmkeys.append(str(k))
body = {'project_id': project_id, 'alarmkeys': alarmkeys,
'context': context.to_dict()} # UUID str
self.rpc.send_msg(rpc.DELETE_ALARMS_MSG_ID, body)
LOG.info("DELETE_ALARMS_MSG sent")
def describe_alarms(self, project_id, action_prefix=None,
alarm_name_prefix=None, alarm_names=None,
max_records=None, next_token=None, state_value=None):
"""
params:
project_id: string
action_prefix: TODO: not implemented yet.
alarm_name_prefix: string
alarm_names: string list
max_records: integer
next_token: string (uuid type)
state_value: string (OK | ALARM | INSUFFICIENT_DATA)
"""
alarms = self.cass.describe_alarms(project_id, action_prefix,
alarm_name_prefix, alarm_names,
max_records, next_token,
state_value)
return alarms
def describe_alarms_for_metric(self, project_id, namespace, metric_name,
dimensions=None, period=None,
statistic=None, unit=None):
"""
params:
project_id: string
metric_name: string
namespace: string
dimensions: dict
period: integer
statistic: string (SampleCount | Average | Sum | Minimum |
Maximum)
unit: string
"""
alarms = self.cass.describe_alarms_for_metric(project_id, namespace,
metric_name, dimensions=dimensions, period=period,
statistic=statistic, unit=unit)
return alarms
def describe_alarm_history(self, project_id, alarm_name=None,
end_date=None, history_item_type=None,
max_records=None, next_token=None,
start_date=None):
histories = self.cass.describe_alarm_history(
alarm_name=alarm_name, end_date=end_date,
history_item_type=history_item_type,
max_records=max_records, next_token=next_token,
start_date=start_date, project_id=project_id
)
return histories
def set_alarm_actions(self, context, project_id, alarm_names, enabled):
for alarm_name in alarm_names:
alarm_key = self.cass.get_metric_alarm_key(project_id, alarm_name)
if not alarm_key:
raise InvalidParameterValue("Alarm %s does not exist" %
alarm_name)
for alarm_name in alarm_names:
alarm_key = self.cass.get_metric_alarm_key(project_id, alarm_name)
history_data = {'actions_enabled':enabled,
'project_id': project_id}
self.cass.put_metric_alarm(alarm_key, history_data)
if enabled:
summary = "Alarm actions for %s are enabled" % alarm_name
else:
summary = "Alarm actions for %s are disabled" % alarm_name
history_key = uuid.uuid4()
history_column = {
'project_id': project_id,
'alarm_key': alarm_key,
'alarm_name': alarm_name,
'history_data': json.dumps(history_data),
'history_item_type': 'ConfigurationUpdate',
'history_summary':summary,
'timestamp': utils.utcnow()
}
self.cass.insert_alarm_history(history_key, history_column)
def set_alarm_state(self, context, project_id, alarm_name, state_reason,
state_value, state_reason_data=None):
k = self.cass.get_metric_alarm_key(project_id, alarm_name)
if not k:
raise ResourceNotFound("Alarm %s does not exists." % alarm_name)
body = {'project_id': project_id, 'alarm_name': alarm_name,
'state_reason': state_reason, 'state_value': state_value,
'state_reason_data': state_reason_data,
'context': context.to_dict()}
self.rpc.send_msg(rpc.SET_ALARM_STATE_MSG_ID, body)
LOG.info("SET_ALARM_STATE_MSG sent")
def get_metric_statistics(self, project_id, end_time, metric_name,
namespace, period, start_time, statistics,
unit=None, dimensions=None):
"""
입력받은 조건에 일치하는 메트릭의 통계자료 리스트를 반환한다.
"""
def to_datapoint(df, idx):
datapoint = df.ix[idx].dropna()
if len(datapoint):
return idx, datapoint
end_idx = end_time.replace(second=0, microsecond=0)
start_idx = start_time.replace(second=0, microsecond=0)
start_ana_idx = start_idx - datetools.Minute() * (period / 60)
daterange = DateRange(start_idx, end_idx, offset=datetools.Minute())
daterange_ana = DateRange(start_ana_idx, end_idx,
offset=datetools.Minute())
# load default unit for metric from database
if unit == "None" or not unit:
metric_key = self.cass.get_metric_key(
project_id=project_id, namespace=namespace,
metric_name=metric_name, dimensions=dimensions
)
if metric_key:
unit = self.cass.get_metric_unit(metric_key)
else:
unit = "None"
# load statistics data from database
stats = self.cass.get_metric_statistics(
project_id=project_id, namespace=namespace,
metric_name=metric_name, start_time=start_ana_idx,
end_time=end_time, period=period, statistics=statistics,
dimensions=dimensions
)
period = period / 60 # convert sec to min
stat = DataFrame(index=daterange)
for statistic, series in zip(statistics, stats):
func = self.ROLLING_FUNC_MAP[statistic]
ts = TimeSeries(series, index=daterange_ana)
rolled_ts = func(ts, period, min_periods=0)
stat[statistic] = rolled_ts.ix[::period]
LOG.debug("stat %s\n%s" % (statistic, stat[statistic]))
ret = filter(None, (to_datapoint(stat, i) for i in stat.index))
return ret, unit
def list_metrics(self, project_id, next_token=None, dimensions=None,
metric_name=None, namespace=None):
"""
List Metrics
"""
metrics = self.cass.list_metrics(project_id, namespace, metric_name,
dimensions, next_token)
return metrics
def put_metric_alarm(self, context, project_id, metricalarm):
"""
Send put metric alarm message to Storm
"""
def _validate_actions(alarm):
for actions in (alarm.ok_actions, alarm.insufficient_data_actions,
alarm.alarm_actions):
for action in actions:
if utils.validate_groupnotification_action(action):
group = utils.parse_groupnotification_action(action)
if not self.cass.get_notification_group(group):
raise InvalidNotificationGroup()
now = utils.utcnow()
_validate_actions(metricalarm)
metricalarm = metricalarm.to_columns()
alarm_name = metricalarm['alarm_name']
namespace = metricalarm['namespace']
metric_name = metricalarm['metric_name']
dimensions = json.loads(metricalarm['dimensions'])
# check if we have metric in database
metric_key = self.cass.get_metric_key_or_create(project_id=project_id,
namespace=namespace, metric_name=metric_name,
dimensions=dimensions, unit=metricalarm['unit'])
update_data = {
'project_id': project_id,
'metric_key': str(metric_key),
'alarm_arn': "arn:spcs:synaps:%s:alarm:%s" % (project_id,
alarm_name),
'alarm_configuration_updated_timestamp': utils.strtime(now)
}
metricalarm.update(update_data)
# check if metric is changed
alarm_key = self.cass.get_metric_alarm_key(project_id=project_id,
alarm_name=alarm_name)
if alarm_key:
original_alarm = self.cass.get_metric_alarm(alarm_key)
if (str(original_alarm['metric_key']) !=
str(metricalarm['metric_key'])):
raise InvalidRequest("Metric cannot be changed. "
"Delete alarm and retry.")
else:
# If alarm is newly added, check quotas
# check alarm quota per project
project_quota = FLAGS.get('alarm_quota_per_project')
alarms_in_project = self.cass.get_alarm_count(project_id)
if alarms_in_project >= project_quota:
LOG.info("Too many alarms(%d) in the project %s",
alarms_in_project, project_id)
raise ProjectAlarmQuotaExceeded()
# check alarm quota per metric
metric_quota = FLAGS.get('alarm_quota_per_metric')
alarms_per_metric = self.cass.get_alarms_per_metric_count(
project_id, namespace, metric_name, dimensions)
if alarms_per_metric >= metric_quota:
LOG.info("Too many alarms(%d) for this metric",
alarms_per_metric)
raise MetricAlarmQuotaExceeded()
message = {'project_id': project_id, 'metric_key': str(metric_key),
'metricalarm': metricalarm, 'context': context.to_dict()}
self.rpc.send_msg(rpc.PUT_METRIC_ALARM_MSG_ID, message)
LOG.info("PUT_METRIC_ALARM_MSG sent")
return {}
def put_metric_data(self, context, project_id, namespace, metric_name,
dimensions, value, unit, timestamp=None,
is_admin=False):
admin_namespace = FLAGS.get('admin_namespace')
if namespace.startswith(admin_namespace) and not is_admin:
raise AdminRequired()
timestamp = timestamp or utils.strtime(utils.utcnow())
message = {'project_id': project_id, 'namespace':namespace,
'metric_name': metric_name, 'dimensions': dimensions,
'value':value, 'unit':unit, 'timestamp':timestamp,
'context': context.to_dict()}
self.rpc.send_msg(rpc.PUT_METRIC_DATA_MSG_ID, message)
LOG.info("PUT_METRIC_DATA_MSG sent")
return {}
class PutMetricBolt(storm.BasicBolt):
BOLT_NAME = "PutMetricBolt"
def initialize(self, stormconf, context):
self.pid = os.getpid()
self.cass = Cassandra()
self.metrics = {}
self.mc = memcache.Client(FLAGS.memcached_servers, debug=0)
def process_put_metric_data_msg(self, metric_key, message):
"""
Put metric data into both memory and database
"""
# Load statistics data in memory
if metric_key not in self.metrics:
max_retries = 3
for i in range(max_retries + 1):
try:
self.metrics[metric_key] = MetricMonitor(
metric_key, self.cass)
break
except ResourceNotFound:
if i + 1 < max_retries:
LOG.warn("Metric %s is not in the database. " \
"retry... %d", metric_key, i + 1)
time.sleep(1)
else:
LOG.error("Metric %s is not in the database.",
metric_key)
return
timestamp = utils.parse_strtime(message['timestamp'])
self.metrics[metric_key].put_metric_data(metric_key,
timestamp=timestamp,
value=message['value'],
unit=message['unit'])
def process_put_metric_alarm_msg(self, metric_key, message):
def get_alarm_key(project_id, alarm_name):
key = self.cass.get_metric_alarm_key(project_id, alarm_name)
return key
def metricalarm_for_json(metricalarm):
cut = metricalarm.get('alarm_configuration_updated_timestamp')
alarm_for_json = {
'actionEnabled':
metricalarm.get('actions_enabled', False),
'alarmActions':
metricalarm.get('alarm_actions', []),
'alarmArn':
metricalarm.get('alarm_arn'),
'alarmConfigurationUpdatedTimestamp':
utils.strtime(cut),
'alarmDescription':
metricalarm.get('alarm_description'),
'alarmName':
metricalarm.get('alarm_name'),
'comparisonOperator':
metricalarm.get('comparison_operator'),
'dimensions':
metricalarm.get('dimensions'),
'evaluationPeriods':
metricalarm.get('evaluation_periods'),
'insufficientDataActions':
metricalarm.get('insufficient_data_actions', []),
'metricName':
metricalarm.get('metric_name'),
'namespace':
metricalarm.get('namespace'),
'okactions':
metricalarm.get('ok_actions', []),
'statistic':
metricalarm.get('statistic'),
'threshold':
metricalarm.get('threshold'),
'unit':
metricalarm.get('unit'),
}
return alarm_for_json
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
project_id = message['project_id']
metricalarm = message['metricalarm']
# build metricalarm column, alarmhistory column
alarm_key = get_alarm_key(project_id, metricalarm['alarm_name'])
history_type = 'Update' if alarm_key else 'Create'
now = utils.utcnow()
if history_type == 'Update':
original_alarm = self.cass.get_metric_alarm(alarm_key)
for dict_key in [
'state_updated_timestamp', 'state_reason',
'state_reason_data', 'state_value', 'project_id'
]:
metricalarm[dict_key] = original_alarm[dict_key]
metricalarm['alarm_configuration_updated_timestamp'] = now
history_data = json.dumps({
'updatedAlarm':
metricalarm_for_json(metricalarm),
'type':
history_type,
'version':
'1.0'
})
summary = "Alarm %s updated" % metricalarm['alarm_name']
else:
alarm_key = uuid.uuid4()
state_reason = "Unchecked: Initial alarm creation"
metricalarm.update({
'state_updated_timestamp': now,
'alarm_configuration_updated_timestamp': now,
'state_reason': state_reason,
'state_reason_data': json.dumps({}),
'state_value': "INSUFFICIENT_DATA",
'project_id': project_id
})
history_data = json.dumps({
'createdAlarm':
metricalarm_for_json(metricalarm),
'type':
history_type,
'version':
'1.0'
})
summary = "Alarm %s created" % metricalarm['alarm_name']
metricalarm['metric_key'] = metric_key
history_key = uuid.uuid4()
history_column = {
'project_id': project_id,
'alarm_key': alarm_key,
'alarm_name': metricalarm['alarm_name'],
'history_data': history_data,
'history_item_type': 'ConfigurationUpdate',
'history_summary': summary,
'timestamp': utils.utcnow()
}
self.cass.put_metric_alarm(alarm_key, metricalarm)
self.cass.insert_alarm_history(history_key, history_column)
LOG.info("metric alarm inserted: %s %s", alarm_key, metricalarm)
# load metric in memory
self.metrics[metric_key].put_alarm(alarm_key, metricalarm)
def process_delete_metric_alarms_msg(self, metric_key, message):
alarmkey = UUID(message['alarmkey'])
LOG.debug("Metric keys %s", self.metrics.keys())
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
self.metrics[metric_key].delete_metric_alarm(alarmkey)
def process_set_alarm_state_msg(self, metric_key, message):
project_id = message.get('project_id')
alarm_name = message.get('alarm_name')
state_reason_data = message.get('state_reason_data')
if metric_key not in self.metrics:
self.metrics[metric_key] = MetricMonitor(metric_key, self.cass)
metric = self.metrics[metric_key]
ret = self.cass.get_metric_alarm_key(project_id, alarm_name)
if ret:
alarm_key = ret
try:
metricalarm = metric.alarms[alarm_key]
except KeyError:
LOG.warn("alarm key [%s] is found, but alarm is not found.",
alarm_key)
return
else:
LOG.warn("alarm key [%s] is not found.", alarm_key)
return
metricalarm['state_reason'] = message.get('state_reason')
metricalarm['state_value'] = message.get('state_value')
metricalarm['state_reason_data'] = message.get('state_reason_data')
# write into database
alarm_columns = {
'state_reason': message.get('state_reason'),
'state_value': message.get('state_value')
}
if state_reason_data:
alarm_columns['state_reason_data'] = state_reason_data
alarm_columns['project_id'] = project_id
self.cass.put_metric_alarm(alarm_key, alarm_columns)
def process_check_metric_alarms_msg(self, message):
query_time = datetime.utcnow()
stale_metrics = []
ready_to_evaluate = message.get('ready_to_evaluate')
for key, metric in self.metrics.iteritems():
is_stale = metric.is_stale()
if is_stale:
stale_metrics.append(key)
if (not is_stale) and ready_to_evaluate:
metric.check_alarms(query_time)
for key in stale_metrics:
try:
metric = self.metrics.pop(key)
metric.delete()
LOG.audit("Stale metric(%s) is deleted", str(key))
except KeyError:
LOG.error("KeyError occurred when delete stale metric(%s)",
str(key))
def process(self, tup):
message = json.loads(tup.values[1])
message_id = message['message_id']
message_uuid = message.get('message_uuid', None)
LOG.info("start processing msg[%s:%s]", message_id, message_uuid)
try:
metric_key = UUID(tup.values[0]) if tup.values[0] else None
except ValueError:
LOG.error("badly formed hexadecimal UUID string - %s",
tup.values[0])
return
if message_id == PUT_METRIC_DATA_MSG_ID:
# message deduplicate
if message_uuid:
mckey = "%s_message_uuid" % message_uuid
if not self.mc.get(mckey):
# 300 seconds TTL
self.mc.set(mckey, 1, 300)
LOG.info("process put_metric_data_msg (%s)", message)
self.process_put_metric_data_msg(metric_key, message)
else:
LOG.info("Message duplicated. %s", message_uuid)
elif message_id == PUT_METRIC_ALARM_MSG_ID:
LOG.info("process put_metric_alarm_msg (%s)", message)
self.process_put_metric_alarm_msg(metric_key, message)
elif message_id == DELETE_ALARMS_MSG_ID:
LOG.info("process delete_alarms_msg (%s)", message)
self.process_delete_metric_alarms_msg(metric_key, message)
elif message_id == SET_ALARM_STATE_MSG_ID:
LOG.info("process set_alarm_state_msg (%s)", message)
self.process_set_alarm_state_msg(metric_key, message)
elif message_id == CHECK_METRIC_ALARM_MSG_ID:
LOG.info("process check_metric_alarm_msg (%s)", message)
self.process_check_metric_alarms_msg(message)
else:
LOG.error("unknown message")
class API(object):
ROLLING_FUNC_MAP = {
'Average': rolling_mean,
'Minimum': rolling_min,
'Maximum': rolling_max,
'SampleCount': rolling_sum,
'Sum': rolling_sum,
}
def __init__(self):
self.cass = Cassandra()
self.rpc = rpc.RemoteProcedureCall()
def delete_alarms(self, context, project_id, alarm_names):
alarmkeys = []
for alarm_name in alarm_names:
k = self.cass.get_metric_alarm_key(project_id, alarm_name)
if not k:
raise ResourceNotFound("Alarm %s does not exists." %
alarm_name)
alarmkeys.append(str(k))
body = {
'project_id': project_id,
'alarmkeys': alarmkeys,
'context': context.to_dict()
} # UUID str
self.rpc.send_msg(rpc.DELETE_ALARMS_MSG_ID, body)
LOG.info("DELETE_ALARMS_MSG sent")
def describe_alarms(self,
project_id,
action_prefix=None,
alarm_name_prefix=None,
alarm_names=None,
max_records=None,
next_token=None,
state_value=None):
"""
params:
project_id: string
action_prefix: TODO: not implemented yet.
alarm_name_prefix: string
alarm_names: string list
max_records: integer
next_token: string (uuid type)
state_value: string (OK | ALARM | INSUFFICIENT_DATA)
"""
alarms = self.cass.describe_alarms(project_id, action_prefix,
alarm_name_prefix, alarm_names,
max_records, next_token,
state_value)
return alarms
def describe_alarms_for_metric(self,
project_id,
namespace,
metric_name,
dimensions=None,
period=None,
statistic=None,
unit=None):
"""
params:
project_id: string
metric_name: string
namespace: string
dimensions: dict
period: integer
statistic: string (SampleCount | Average | Sum | Minimum |
Maximum)
unit: string
"""
alarms = self.cass.describe_alarms_for_metric(project_id,
namespace,
metric_name,
dimensions=dimensions,
period=period,
statistic=statistic,
unit=unit)
return alarms
def describe_alarm_history(self,
project_id,
alarm_name=None,
end_date=None,
history_item_type=None,
max_records=None,
next_token=None,
start_date=None):
histories = self.cass.describe_alarm_history(
alarm_name=alarm_name,
end_date=end_date,
history_item_type=history_item_type,
max_records=max_records,
next_token=next_token,
start_date=start_date,
project_id=project_id)
return histories
def set_alarm_actions(self, context, project_id, alarm_names, enabled):
for alarm_name in alarm_names:
alarm_key = self.cass.get_metric_alarm_key(project_id, alarm_name)
if not alarm_key:
raise InvalidParameterValue("Alarm %s does not exist" %
alarm_name)
for alarm_name in alarm_names:
alarm_key = self.cass.get_metric_alarm_key(project_id, alarm_name)
history_data = {
'actions_enabled': enabled,
'project_id': project_id
}
self.cass.put_metric_alarm(alarm_key, history_data)
if enabled:
summary = "Alarm actions for %s are enabled" % alarm_name
else:
summary = "Alarm actions for %s are disabled" % alarm_name
history_key = uuid.uuid4()
history_column = {
'project_id': project_id,
'alarm_key': alarm_key,
'alarm_name': alarm_name,
'history_data': json.dumps(history_data),
'history_item_type': 'ConfigurationUpdate',
'history_summary': summary,
'timestamp': utils.utcnow()
}
self.cass.insert_alarm_history(history_key, history_column)
def set_alarm_state(self,
context,
project_id,
alarm_name,
state_reason,
state_value,
state_reason_data=None):
k = self.cass.get_metric_alarm_key(project_id, alarm_name)
if not k:
raise ResourceNotFound("Alarm %s does not exists." % alarm_name)
body = {
'project_id': project_id,
'alarm_name': alarm_name,
'state_reason': state_reason,
'state_value': state_value,
'state_reason_data': state_reason_data,
'context': context.to_dict()
}
self.rpc.send_msg(rpc.SET_ALARM_STATE_MSG_ID, body)
LOG.info("SET_ALARM_STATE_MSG sent")
def get_metric_statistics(self,
project_id,
end_time,
metric_name,
namespace,
period,
start_time,
statistics,
unit=None,
dimensions=None):
"""
입력받은 조건에 일치하는 메트릭의 통계자료 리스트를 반환한다.
"""
def to_datapoint(df, idx):
datapoint = df.ix[idx].dropna()
if len(datapoint):
return idx, datapoint
end_idx = end_time.replace(second=0, microsecond=0)
start_idx = start_time.replace(second=0, microsecond=0)
start_ana_idx = start_idx - datetools.Minute() * (period / 60)
daterange = DateRange(start_idx, end_idx, offset=datetools.Minute())
daterange_ana = DateRange(start_ana_idx,
end_idx,
offset=datetools.Minute())
# load default unit for metric from database
if unit == "None" or not unit:
metric_key = self.cass.get_metric_key(project_id=project_id,
namespace=namespace,
metric_name=metric_name,
dimensions=dimensions)
if metric_key:
unit = self.cass.get_metric_unit(metric_key)
else:
unit = "None"
# load statistics data from database
stats = self.cass.get_metric_statistics(project_id=project_id,
namespace=namespace,
metric_name=metric_name,
start_time=start_ana_idx,
end_time=end_time,
period=period,
statistics=statistics,
dimensions=dimensions)
period = period / 60 # convert sec to min
stat = DataFrame(index=daterange)
for statistic, series in zip(statistics, stats):
func = self.ROLLING_FUNC_MAP[statistic]
ts = TimeSeries(series, index=daterange_ana)
rolled_ts = func(ts, period, min_periods=0)
stat[statistic] = rolled_ts.ix[::period]
LOG.debug("stat %s\n%s" % (statistic, stat[statistic]))
ret = filter(None, (to_datapoint(stat, i) for i in stat.index))
return ret, unit
def list_metrics(self,
project_id,
next_token=None,
dimensions=None,
metric_name=None,
namespace=None):
"""
List Metrics
"""
metrics = self.cass.list_metrics(project_id, namespace, metric_name,
dimensions, next_token)
return metrics
def put_metric_alarm(self, context, project_id, metricalarm):
"""
Send put metric alarm message to Storm
"""
def _validate_actions(alarm):
for actions in (alarm.ok_actions, alarm.insufficient_data_actions,
alarm.alarm_actions):
for action in actions:
if utils.validate_groupnotification_action(action):
group = utils.parse_groupnotification_action(action)
if not self.cass.get_notification_group(group):
raise InvalidNotificationGroup()
now = utils.utcnow()
_validate_actions(metricalarm)
metricalarm = metricalarm.to_columns()
alarm_name = metricalarm['alarm_name']
namespace = metricalarm['namespace']
metric_name = metricalarm['metric_name']
dimensions = json.loads(metricalarm['dimensions'])
# check if we have metric in database
metric_key = self.cass.get_metric_key_or_create(
project_id=project_id,
namespace=namespace,
metric_name=metric_name,
dimensions=dimensions,
unit=metricalarm['unit'])
update_data = {
'project_id': project_id,
'metric_key': str(metric_key),
'alarm_arn':
"arn:spcs:synaps:%s:alarm:%s" % (project_id, alarm_name),
'alarm_configuration_updated_timestamp': utils.strtime(now)
}
metricalarm.update(update_data)
# check if metric is changed
alarm_key = self.cass.get_metric_alarm_key(project_id=project_id,
alarm_name=alarm_name)
if alarm_key:
original_alarm = self.cass.get_metric_alarm(alarm_key)
if (str(original_alarm['metric_key']) != str(
metricalarm['metric_key'])):
raise InvalidRequest("Metric cannot be changed. "
"Delete alarm and retry.")
else:
# If alarm is newly added, check quotas
# check alarm quota per project
project_quota = FLAGS.get('alarm_quota_per_project')
alarms_in_project = self.cass.get_alarm_count(project_id)
if alarms_in_project >= project_quota:
LOG.info("Too many alarms(%d) in the project %s",
alarms_in_project, project_id)
raise ProjectAlarmQuotaExceeded()
# check alarm quota per metric
metric_quota = FLAGS.get('alarm_quota_per_metric')
alarms_per_metric = self.cass.get_alarms_per_metric_count(
project_id, namespace, metric_name, dimensions)
if alarms_per_metric >= metric_quota:
LOG.info("Too many alarms(%d) for this metric",
alarms_per_metric)
raise MetricAlarmQuotaExceeded()
message = {
'project_id': project_id,
'metric_key': str(metric_key),
'metricalarm': metricalarm,
'context': context.to_dict()
}
self.rpc.send_msg(rpc.PUT_METRIC_ALARM_MSG_ID, message)
LOG.info("PUT_METRIC_ALARM_MSG sent")
return {}
def put_metric_data(self,
context,
project_id,
namespace,
metric_name,
dimensions,
value,
unit,
timestamp=None,
is_admin=False):
admin_namespace = FLAGS.get('admin_namespace')
if namespace.startswith(admin_namespace) and not is_admin:
raise AdminRequired()
timestamp = timestamp or utils.strtime(utils.utcnow())
message = {
'project_id': project_id,
'namespace': namespace,
'metric_name': metric_name,
'dimensions': dimensions,
'value': value,
'unit': unit,
'timestamp': timestamp,
'context': context.to_dict()
}
self.rpc.send_msg(rpc.PUT_METRIC_DATA_MSG_ID, message)
LOG.info("PUT_METRIC_DATA_MSG sent")
return {}