def graph(self, type="png"):
graphConfig = {}
graphConfig["title"] = DataOperations.getTitle(
self.timelyMetric.metric, self.analyticConfig)
return TimelyMetric.graph(self.analyticConfig,
self.dataFrame,
self.timelyMetric.metric,
seriesConfig=self.seriesConfig,
graphConfig=graphConfig,
notebook=self.notebook,
type=type)
'how': 'mean',
'rolling_average_period': '12 hours',
'min_threshold': None,
'average_min_threshold': None,
'max_threshold': None,
'average_max_threshold': None,
'min_threshold_percentage': -50,
'max_threshold_percentage': 50,
'boolean': 'and',
'min_alert_period': '5 minutes',
'last_alert': '1 hour',
'display': 'all',
'output_dir': '/path/to/output'
})
alert = TimelyAnalytic.find_alerts(timelyMetric, analyticConfig)
if alert is not None:
# write graph to file
oldmask = os.umask(022)
file = alert.graph(type='html')
os.umask(oldmask)
text = DataOperations.getTitle(timelyMetric, analyticConfig)
# send email with graph attached
alert.email("", "", text, text, [file])
# log to syslog
alert.log(text)
def find_alerts(timelyMetric, analyticConfig, notebook=False):
if (analyticConfig.counter == True):
convertCounterToRate(timelyMetric, analyticConfig)
df = timelyMetric.getDataFrame()
graphDF = TimelyMetric.pivot(df, timelyMetric.metric, groupByColumn=analyticConfig.groupByColumn)
if analyticConfig.excludeColRegex is not None:
graphDF = graphDF.select(lambda x : not (re.search(analyticConfig.excludeColRegex, x)), axis=1)
if analyticConfig.includeColRegex is not None:
graphDF = graphDF.select(lambda x : re.search(analyticConfig.includeColRegex, x), axis=1)
if analyticConfig.sample is not None:
graphDF = TimelyMetric.resample(graphDF, analyticConfig.sample, how=analyticConfig.how)
graphDF_avg = pandas.DataFrame(graphDF, copy=True)
combined = pandas.DataFrame()
seriesConfig = {}
for i in graphDF_avg.columns:
col = str(i)
any_conditions_values = False
result_values = np.ones(graphDF[col].shape, bool)
if analyticConfig.orCondition:
result_values = np.zeros(graphDF[col].shape, bool)
any_conditions_average = False
result_average = np.ones(graphDF_avg[col].shape, bool)
if analyticConfig.orCondition:
result_average = np.zeros(graphDF_avg[col].shape, bool)
if analyticConfig.min_threshold is not None:
currCondition = graphDF[col].astype(float) < analyticConfig.min_threshold
result_values = addCondition(analyticConfig.orCondition, result_values, currCondition)
any_conditions_values = True
if analyticConfig.max_threshold is not None:
currCondition = graphDF[col].astype(float) > analyticConfig.max_threshold
result_values = addCondition(analyticConfig.orCondition, result_values, currCondition)
any_conditions_values = True
if analyticConfig.rolling_average_samples is not None:
graphDF_avg = TimelyMetric.rolling_average(graphDF_avg, col, rolling_average=analyticConfig.rolling_average_samples)
if analyticConfig.min_threshold_percentage is not None:
if analyticConfig.min_threshold_percentage >= 0:
multiple = 1.0 + (float(abs(analyticConfig.min_threshold_percentage)) / float(100))
else:
multiple = 1.0 - (float(abs(analyticConfig.min_threshold_percentage)) / float(100))
currCondition = graphDF[col].astype(float) < (graphDF_avg[col].astype(float) * multiple)
result_values = addCondition(analyticConfig.orCondition, result_values, currCondition)
any_conditions_values = True
if analyticConfig.max_threshold_percentage is not None:
if analyticConfig.max_threshold_percentage >= 0:
multiple = 1.0 + (float(abs(analyticConfig.max_threshold_percentage)) / float(100))
else:
multiple = 1.0 - (float(abs(analyticConfig.max_threshold_percentage)) / float(100))
currCondition = graphDF[col].astype(float) > (graphDF_avg[col].astype(float) * multiple)
result_values = addCondition(analyticConfig.orCondition, result_values, currCondition)
any_conditions_values = True
if analyticConfig.average_min_threshold is not None:
currCondition = graphDF_avg[col].astype(float) < analyticConfig.average_min_threshold
result_average = addCondition(analyticConfig.orCondition, result_average, currCondition)
any_conditions_average = True
if analyticConfig.average_max_threshold is not None:
currCondition = graphDF_avg[col].astype(float) > analyticConfig.average_max_threshold
result_average = addCondition(analyticConfig.orCondition, result_average, currCondition)
any_conditions_average = True
# if orCondition is AND and no exceptional conditions have been found, then result_values will be all True
if any_conditions_values == False:
result_values = np.zeros(graphDF[col].shape, bool)
exceptional_values = graphDF.loc[result_values, col]
# if orCondition is AND and no exceptional conditions have been found, then result_average will be all True
if any_conditions_average == False:
result_average = np.zeros(graphDF_avg[col].shape, bool)
exceptional_average = graphDF_avg.loc[result_average, col]
# only keep alerts that are in consecutive periods of length analyticConfig.min_alert_minutes
exceptional_values = keepConsecutiveAlerts(graphDF, exceptional_values, analyticConfig.min_alert_minutes)
exceptional_average = keepConsecutiveAlerts(graphDF_avg, exceptional_average, analyticConfig.min_alert_minutes)
# only evaluate the last analyticConfig.last_alert_minutes if set
if analyticConfig.last_alert_minutes is not None:
end = datetime.fromtimestamp(timelyMetric.timeDateRange.getEndMs() / 1000.00, DataOperations.utc)
recentEnoughBegin = end - timedelta(minutes=analyticConfig.last_alert_minutes)
exceptional_values = exceptional_values.ix[recentEnoughBegin:end]
exceptional_average = exceptional_average.ix[recentEnoughBegin:end]
anyValueExceptions = exceptional_values.size > 0
anyAverageExceptions = exceptional_average.size > 0
if (analyticConfig.display.lower() == "all") or (analyticConfig.display.lower() == "alerts" and anyValueExceptions):
combined[col] = graphDF[col]
if analyticConfig.rolling_average_samples is not None:
if (analyticConfig.display.lower() == "all"):
combined[col + '_avg'] = graphDF_avg[col]
else:
if (anyAverageExceptions):
combined[col + '_avg'] = graphDF_avg[col]
if (anyValueExceptions and (analyticConfig.min_threshold_percentage is not None or analyticConfig.max_threshold_percentage is not None)):
combined[col + '_avg'] = graphDF_avg[col]
if ((analyticConfig.display.lower() == "all") or (analyticConfig.display.lower() == "alerts" and anyValueExceptions)):
combined[col + '_warn'] = exceptional_values.dropna()
seriesConfig[col + '_warn'] = {
"mode" : "markers",
"marker" : {
"symbol" : "hash-open",
"color" : "red"
}
}
if ((analyticConfig.display.lower() == "all") or (analyticConfig.display.lower() == "alerts" and anyAverageExceptions)):
combined[col + '_avg_warn'] = exceptional_average.dropna()
seriesConfig[col + '_avg_warn'] = {
"mode" : "markers",
"marker" : {
"symbol" : "hash-open",
"color" : "red"
}
}
timelyAlert = None
if not combined.empty:
alertAnalyticConfig = TimelyAnalyticConfiguration(analyticConfig)
if alertAnalyticConfig.groupByColumn is None:
alertAnalyticConfig.groupByColumn = timelyMetric.metric + "_obs"
combined = TimelyMetric.unpivot(combined, timelyMetric.metric, groupByColumn=alertAnalyticConfig.groupByColumn)
combined = combined.sort_index()
combined['date'] = combined.index.values
combined = combined.sort_values(['date', alertAnalyticConfig.groupByColumn])
combined = combined.drop(['date'], 1)
combined = combined.dropna()
combined = DataOperations.ensureMinSeriesLength(combined, alertAnalyticConfig.groupByColumn)
message = DataOperations.getTitle(timelyMetric, analyticConfig, separator=', ')
timelyAlert = TimelyAlert(timelyMetric, combined, message, seriesConfig, alertAnalyticConfig, notebook)
return timelyAlert
def find_alerts(timelyMetric, analyticConfig, notebook=False):
df = timelyMetric.getDataFrame()
graphDF = TimelyMetric.pivot(df, timelyMetric.metric, groupByColumn=analyticConfig.groupByColumn)
if analyticConfig.excludeColRegex is not None:
graphDF = graphDF.select(lambda x : not (re.search(analyticConfig.excludeColRegex, x)), axis=1)
if analyticConfig.includeColRegex is not None:
graphDF = graphDF.select(lambda x : re.search(analyticConfig.includeColRegex, x), axis=1)
if analyticConfig.sample is not None:
graphDF = TimelyMetric.resample(graphDF, analyticConfig.sample, how=analyticConfig.how)
graphDF_avg = pandas.DataFrame(graphDF, copy=True)
combined = pandas.DataFrame()
now = Timestamp.now()
seriesConfig = {}
for i in graphDF_avg.columns:
col = str(i)
any_conditions_values = False
result_values = np.ones(graphDF[col].shape, bool)
if analyticConfig.orCondition:
result_values = np.zeros(graphDF[col].shape, bool)
any_conditions_average = False
result_average = np.ones(graphDF_avg[col].shape, bool)
if analyticConfig.orCondition:
result_average = np.zeros(graphDF_avg[col].shape, bool)
if analyticConfig.min_threshold is not None:
currCondition = graphDF[col].astype(float) < analyticConfig.min_threshold
result_values = addCondition(analyticConfig.orCondition, result_values, currCondition)
any_conditions_values = True
if analyticConfig.max_threshold is not None:
currCondition = graphDF[col].astype(float) > analyticConfig.max_threshold
result_values = addCondition(analyticConfig.orCondition, result_values, currCondition)
any_conditions_values = True
if analyticConfig.rolling_average_samples is not None:
graphDF_avg = TimelyMetric.rolling_average(graphDF_avg, col, rolling_average=analyticConfig.rolling_average_samples)
if analyticConfig.alert_percentage is not None:
if analyticConfig.alert_percentage > 0:
multiple = 1.0 + (float(abs(analyticConfig.alert_percentage)) / float(100))
currCondition = graphDF[col].astype(float) > (graphDF_avg[col].astype(float) * multiple)
result_values = addCondition(analyticConfig.orCondition, result_values, currCondition)
any_conditions_values = True
if analyticConfig.alert_percentage < 0:
multiple = 1.0 - (float(abs(analyticConfig.alert_percentage)) / float(100))
if multiple > 0:
currCondition = graphDF[col].astype(float) < (graphDF_avg[col].astype(float) * multiple)
result_values = addCondition(analyticConfig.orCondition, result_values, currCondition)
any_conditions_values = True
if analyticConfig.average_min_threshold is not None:
currCondition = graphDF_avg[col].astype(float) < analyticConfig.average_min_threshold
result_average = addCondition(analyticConfig.orCondition, result_average, currCondition)
any_conditions_average = True
if analyticConfig.average_max_threshold is not None:
currCondition = graphDF_avg[col].astype(float) > analyticConfig.average_max_threshold
result_average = addCondition(analyticConfig.orCondition, result_average, currCondition)
any_conditions_average = True
# if orCondition is AND and no exceptional conditions have been found, then result_values will be all True
if any_conditions_values == False:
result_values = np.zeros(graphDF[col].shape, bool)
exceptional_values = graphDF.loc[result_values, col]
# if orCondition is AND and no exceptional conditions have been found, then result_average will be all True
if any_conditions_average == False:
result_average = np.zeros(graphDF_avg[col].shape, bool)
exceptional_average = graphDF_avg.loc[result_average, col]
# only evaluate the last analyticConfig.last_alert_minutes if set
if analyticConfig.last_alert_minutes is not None:
recentEnoughBegin = now - timedelta(minutes=analyticConfig.last_alert_minutes)
exceptional_values = exceptional_values.ix[recentEnoughBegin:now]
exceptional_average = exceptional_average.ix[recentEnoughBegin:now]
# only keep alerts that are in consecutive periods of length analyticConfig.min_alert_minutes
exceptional_values = keepConsecutiveAlerts(graphDF, exceptional_values, analyticConfig.min_alert_minutes)
exceptional_average = keepConsecutiveAlerts(graphDF_avg, exceptional_average, analyticConfig.min_alert_minutes)
anyValueExceptions = exceptional_values.size > 0
anyAverageExceptions = exceptional_average.size > 0
if (analyticConfig.display.lower() == "all") or (analyticConfig.display.lower() == "alerts" and anyValueExceptions):
combined[col] = graphDF[col]
if analyticConfig.rolling_average_samples is not None:
if (analyticConfig.display.lower() == "all") or (analyticConfig.display.lower() == "alerts" and (anyAverageExceptions or analyticConfig.alert_percentage is not None)):
combined[col+'_avg'] = graphDF_avg[col]
if ((analyticConfig.display.lower() == "all") or (analyticConfig.display.lower() == "alerts" and anyValueExceptions)):
combined[col + '_warn'] = exceptional_values.dropna()
seriesConfig[col + '_warn'] = {
"mode" : "markers",
"marker" : {
"symbol" : "hash-open",
"color" : "red"
}
}
if ((analyticConfig.display.lower() == "all") or (analyticConfig.display.lower() == "alerts" and anyAverageExceptions)):
combined[col + '_avg_warn'] = exceptional_average.dropna()
seriesConfig[col + '_avg_warn'] = {
"mode" : "markers",
"marker" : {
"symbol" : "hash-open",
"color" : "red"
}
}
timelyAlert = None
if not combined.empty:
alertAnalyticConfig = TimelyAnalyticConfiguration(analyticConfig)
if alertAnalyticConfig.groupByColumn is None:
alertAnalyticConfig.groupByColumn = timelyMetric.metric + "_obs"
combined = TimelyMetric.unpivot(combined, timelyMetric.metric, groupByColumn=alertAnalyticConfig.groupByColumn)
combined = combined.sort_index()
combined['date'] = combined.index.values
combined = combined.sort_values(['date', analyticConfig.groupByColumn])
combined = combined.drop(['date'], 1)
combined = combined.dropna()
combined = DataOperations.ensureMinSeriesLength(combined, alertAnalyticConfig.groupByColumn)
message = DataOperations.getTitle(timelyMetric, analyticConfig, separator=', ')
timelyAlert = TimelyAlert(timelyMetric, combined, message, seriesConfig, alertAnalyticConfig, notebook)
return timelyAlert
def graph(self, type="png"):
graphConfig = {}
graphConfig["title"] = DataOperations.getTitle(self.timelyMetric, self.analyticConfig)
return TimelyMetric.graph(self.analyticConfig, self.dataFrame, self.timelyMetric, seriesConfig=self.seriesConfig, graphConfig=graphConfig, notebook=self.notebook, type=type)
'how' : 'mean',
'rolling_average_period' : '12 hours',
'min_threshold' : None,
'average_min_threshold' : None,
'max_threshold' : None,
'average_max_threshold' : None,
'alert_percentage' : 25,
'boolean' : 'and',
'min_alert_period' : '5 minutes',
'last_alert' : '1 hour',
'display' : 'all',
'output_dir' : '/path/to/output'
})
alert = TimelyAnalytic.find_alerts(timelyMetric, analyticConfig)
if alert is not None:
# write graph to file
oldmask = os.umask(022)
file = alert.graph(type='html')
os.umask(oldmask)
text = DataOperations.getTitle(timelyMetric, analyticConfig)
# send email with graph attached
alert.email("", "", text, text, [file])
# log to syslog
alert.log(text)
