def add_aggregated_alert(self, match, rule):
""" Save a match as a pending aggregate alert to elasticsearch. """
if not rule['current_aggregate_id'] or rule['aggregate_alert_time'] < ts_to_dt(match[rule['timestamp_field']]):
# First match, set alert_time
match_time = ts_to_dt(match[rule['timestamp_field']])
alert_time = match_time + rule['aggregation']
rule['aggregate_alert_time'] = alert_time
agg_id = None
else:
# Already pending aggregation, use existing alert_time
alert_time = rule['aggregate_alert_time']
agg_id = rule['current_aggregate_id']
logging.info('Adding alert for %s to aggregation, next alert at %s' % (rule['name'], alert_time))
alert_body = self.get_alert_body(match, rule, False, alert_time)
if agg_id:
alert_body['aggregate_id'] = agg_id
res = self.writeback('elastalert', alert_body)
# If new aggregation, save _id
if res and not agg_id:
rule['current_aggregate_id'] = res['_id']
# Couldn't write the match to ES, save it in memory for now
if not res:
rule['agg_matches'].append(match)
return res
def is_silenced(self, rule_name):
""" Checks if rule_name is currently silenced. Returns false on exception. """
if rule_name in self.silence_cache:
if ts_now() < self.silence_cache[rule_name][0]:
return True
else:
return False
query = {'filter': {'term': {'rule_name': rule_name}},
'sort': {'until': {'order': 'desc'}}}
if self.writeback_es:
try:
res = self.writeback_es.search(index=self.writeback_index, doc_type='silence',
size=1, body=query, _source_include=['until', 'exponent'])
except ElasticsearchException as e:
self.handle_error("Error while querying for alert silence status: %s" % (e), {'rule': rule_name})
return False
if res['hits']['hits']:
until_ts = res['hits']['hits'][0]['_source']['until']
exponent = res['hits']['hits'][0]['_source'].get('exponent', 0)
self.silence_cache[rule_name] = (ts_to_dt(until_ts), exponent)
if ts_now() < ts_to_dt(until_ts):
return True
return False
def start(self):
""" Periodically go through each rule and run it """
starttime = self.args.start
if starttime:
try:
starttime = ts_to_dt(starttime)
except (TypeError, ValueError):
self.handle_error("%s is not a valid ISO 8601 timestamp (YYYY-MM-DDTHH:MM:SS+XX:00)" % (starttime))
exit(1)
while True:
# If writeback_es errored, it's disabled until the next query cycle
if not self.writeback_es:
self.writeback_es = self.new_elasticsearch(self.es_conn_config)
self.send_pending_alerts()
next_run = datetime.datetime.utcnow() + self.run_every
for rule in self.rules:
# Set endtime based on the rule's delay
delay = rule.get('query_delay')
if hasattr(self.args, 'end') and self.args.end:
endtime = ts_to_dt(self.args.end)
elif delay:
endtime = ts_now() - delay
else:
endtime = ts_now()
try:
num_matches = self.run_rule(rule, endtime, starttime)
except EAException as e:
self.handle_error("Error running rule %s: %s" % (rule['name'], e), {'rule': rule['name']})
else:
old_starttime = pretty_ts(rule.get('original_starttime'), rule.get('use_local_time'))
logging.info("Ran %s from %s to %s: %s query hits, %s matches,"
" %s alerts sent" % (rule['name'], old_starttime, pretty_ts(endtime, rule.get('use_local_time')),
self.num_hits, num_matches, self.alerts_sent))
self.alerts_sent = 0
self.remove_old_events(rule)
if next_run < datetime.datetime.utcnow():
# We were processing for longer than our refresh interval
# This can happen if --start was specified with a large time period
# or if we are running too slow to process events in real time.
logging.warning("Querying from %s to %s took longer than %s!" % (old_starttime, endtime, self.run_every))
continue
# Only force starttime once
starttime = None
if not self.args.pin_rules:
self.load_rule_changes()
# Wait before querying again
sleep_for = (next_run - datetime.datetime.utcnow()).seconds
logging.info("Sleeping for %s seconds" % (sleep_for))
time.sleep(sleep_for)
def add_aggregated_alert(self, match, rule):
""" Save a match as a pending aggregate alert to elasticsearch. """
if (not rule['current_aggregate_id'] or
('aggregate_alert_time' in rule and rule['aggregate_alert_time'] < ts_to_dt(match[rule['timestamp_field']]))):
# Elastalert may have restarted while pending alerts exist
pending_alert = self.find_pending_aggregate_alert(rule)
if pending_alert:
alert_time = rule['aggregate_alert_time'] = ts_to_dt(pending_alert['_source']['alert_time'])
agg_id = rule['current_aggregate_id'] = pending_alert['_id']
elastalert_logger.info('Adding alert for %s to aggregation(id: %s), next alert at %s' % (rule['name'], agg_id, alert_time))
else:
# First match, set alert_time
match_time = ts_to_dt(match[rule['timestamp_field']])
alert_time = ''
if isinstance(rule['aggregation'], dict) and rule['aggregation'].get('schedule'):
croniter._datetime_to_timestamp = cronite_datetime_to_timestamp # For Python 2.6 compatibility
try:
iter = croniter(rule['aggregation']['schedule'], ts_now())
alert_time = unix_to_dt(iter.get_next())
except Exception as e:
self.handle_error("Error parsing aggregate send time Cron format %s" % (e), rule['aggregation']['schedule'])
else:
alert_time = match_time + rule['aggregation']
rule['aggregate_alert_time'] = alert_time
agg_id = None
elastalert_logger.info('New aggregation for %s. next alert at %s.' % (rule['name'], alert_time))
else:
# Already pending aggregation, use existing alert_time
alert_time = rule['aggregate_alert_time']
agg_id = rule['current_aggregate_id']
elastalert_logger.info('Adding alert for %s to aggregation(id: %s), next alert at %s' % (rule['name'], agg_id, alert_time))
alert_body = self.get_alert_body(match, rule, False, alert_time)
if agg_id:
alert_body['aggregate_id'] = agg_id
res = self.writeback('elastalert', alert_body)
# If new aggregation, save _id
if res and not agg_id:
rule['current_aggregate_id'] = res['_id']
# Couldn't write the match to ES, save it in memory for now
if not res:
rule['agg_matches'].append(match)
return res
def get_all_terms(self, args):
""" Performs a terms aggregation for each field to get every existing term. """
self.es = Elasticsearch(host=self.rules["es_host"], port=self.rules["es_port"])
window_size = datetime.timedelta(**self.rules.get("terms_window_size", {"days": 30}))
field_name = {"field": "", "size": 2147483647} # Integer.MAX_VALUE
query_template = {"aggs": {"values": {"terms": field_name}}}
if args and args.start:
end = ts_to_dt(args.start)
else:
end = ts_now()
start = end - window_size
if self.rules.get("use_strftime_index"):
index = format_index(self.rules["index"], start, end)
else:
index = self.rules["index"]
time_filter = {self.rules["timestamp_field"]: {"lte": dt_to_ts(end), "gte": dt_to_ts(start)}}
query_template["filter"] = {"bool": {"must": [{"range": time_filter}]}}
query = {"aggs": {"filtered": query_template}}
for field in self.fields:
field_name["field"] = field
res = self.es.search(body=query, index=index, ignore_unavailable=True, timeout=50)
if "aggregations" in res:
buckets = res["aggregations"]["filtered"]["values"]["buckets"]
keys = [bucket["key"] for bucket in buckets]
self.seen_values[field] = keys
elastalert_logger.info("Found %s unique values for %s" % (len(keys), field))
else:
self.seen_values[field] = []
elastalert_logger.info("Found no values for %s" % (field))
def get_all_terms(self, args):
""" Performs a terms aggregation for each field to get every existing term. """
self.es = Elasticsearch(host=self.rules['es_host'], port=self.rules['es_port'])
window_size = datetime.timedelta(**self.rules.get('terms_window_size', {'days': 30}))
field_name = {"field": "", "size": 2147483647} # Integer.MAX_VALUE
query_template = {"aggs": {"values": {"terms": field_name}}}
if args and args.start:
end = ts_to_dt(args.start)
else:
end = ts_now()
start = end - window_size
if self.rules.get('use_strftime_index'):
index = format_index(self.rules['index'], start, end)
else:
index = self.rules['index']
time_filter = {self.rules['timestamp_field']: {'lte': dt_to_ts(end), 'gte': dt_to_ts(start)}}
query_template['filter'] = {'bool': {'must': [{'range': time_filter}]}}
query = {'aggs': {'filtered': query_template}}
for field in self.fields:
field_name['field'] = field
res = self.es.search(body=query, index=index, ignore_unavailable=True, timeout=50)
if 'aggregations' in res:
buckets = res['aggregations']['filtered']['values']['buckets']
keys = [bucket['key'] for bucket in buckets]
self.seen_values[field] = keys
elastalert_logger.info('Found %s unique values for %s' % (len(keys), field))
else:
self.seen_values[field] = []
elastalert_logger.info('Found no values for %s' % (field))
def get_starttime(self, rule):
""" Query ES for the last time we ran this rule.
:param rule: The rule configuration.
:return: A timestamp or None.
"""
query = {'filter': {'term': {'rule_name': '%s' % (rule['name'])}},
'sort': {'@timestamp': {'order': 'desc'}}}
try:
if self.writeback_es:
res = self.writeback_es.search(index=self.writeback_index, doc_type='elastalert_status',
size=1, body=query, _source_include=['endtime', 'rule_name'])
if res['hits']['hits']:
endtime = ts_to_dt(res['hits']['hits'][0]['_source']['endtime'])
if ts_now() - endtime < self.old_query_limit:
return endtime
else:
logging.info("Found expired previous run for %s at %s" % (rule['name'], endtime))
return None
except (ElasticsearchException, KeyError) as e:
self.handle_error('Error querying for last run: %s' % (e), {'rule': rule['name']})
self.writeback_es = None
return None
def silence(self):
""" Silence an alert for a period of time. --silence and --rule must be passed as args. """
if self.debug:
logging.error('--silence not compatible with --debug')
exit(1)
if not self.args.rule:
logging.error('--silence must be used with --rule')
exit(1)
# With --rule, self.rules will only contain that specific rule
rule_name = self.rules[0]['name']
try:
unit, num = self.args.silence.split('=')
silence_time = datetime.timedelta(**{unit: int(num)})
# Double conversion to add tzinfo
silence_ts = ts_to_dt(dt_to_ts(silence_time + datetime.datetime.utcnow()))
except (ValueError, TypeError):
logging.error('%s is not a valid time period' % (self.args.silence))
exit(1)
if not self.set_realert(rule_name, silence_ts, 0):
logging.error('Failed to save silence command to elasticsearch')
exit(1)
logging.info('Success. %s will be silenced until %s' % (rule_name, silence_ts))
def get_match_str(self, match):
lt = self.rules.get('use_local_time')
starttime = pretty_ts(dt_to_ts(ts_to_dt(match[self.ts_field]) - self.rules['timeframe']), lt)
endtime = pretty_ts(match[self.ts_field], lt)
message = 'At least %d events occurred between %s and %s\n\n' % (self.rules['num_events'],
starttime,
endtime)
return message
def get_all_terms(self, args):
""" Performs a terms aggregation for each field to get every existing term. """
self.es = Elasticsearch(host=self.rules['es_host'], port=self.rules['es_port'], timeout=self.rules.get('es_conn_timeout', 50))
window_size = datetime.timedelta(**self.rules.get('terms_window_size', {'days': 30}))
field_name = {"field": "", "size": 2147483647} # Integer.MAX_VALUE
query_template = {"aggs": {"values": {"terms": field_name}}}
if args and args.start:
end = ts_to_dt(args.start)
else:
end = ts_now()
start = end - window_size
if self.rules.get('use_strftime_index'):
index = format_index(self.rules['index'], start, end)
else:
index = self.rules['index']
time_filter = {self.rules['timestamp_field']: {'lte': dt_to_ts(end), 'gte': dt_to_ts(start)}}
query_template['filter'] = {'bool': {'must': [{'range': time_filter}]}}
query = {'aggs': {'filtered': query_template}}
for field in self.fields:
# For composite keys, we will need to perform sub-aggregations
if type(field) == list:
level = query_template['aggs']
# Iterate on each part of the composite key and add a sub aggs clause to the elastic search query
for i, sub_field in enumerate(field):
level['values']['terms']['field'] = sub_field
if i < len(field) - 1:
# If we have more fields after the current one, then set up the next nested structure
level['values']['aggs'] = {'values': {'terms': copy.deepcopy(field_name)}}
level = level['values']['aggs']
else:
# For non-composite keys, only a single agg is needed
field_name['field'] = field
res = self.es.search(body=query, index=index, ignore_unavailable=True, timeout='50s')
if 'aggregations' in res:
buckets = res['aggregations']['filtered']['values']['buckets']
if type(field) == list:
# For composite keys, make the lookup based on all fields
# Make it a tuple since it can be hashed and used in dictionary lookups
self.seen_values[tuple(field)] = []
for bucket in buckets:
# We need to walk down the hierarchy and obtain the value at each level
self.seen_values[tuple(field)] += self.flatten_aggregation_hierarchy(bucket)
# If we don't have any results, it could either be because of the absence of any baseline data
# OR it may be because the composite key contained a non-primitive type. Either way, give the
# end-users a heads up to help them debug what might be going on.
if not self.seen_values[tuple(field)]:
elastalert_logger.warning((
'No results were found from all sub-aggregations. This can either indicate that there is '
'no baseline data OR that a non-primitive field was used in a composite key.'
))
else:
keys = [bucket['key'] for bucket in buckets]
self.seen_values[field] = keys
elastalert_logger.info('Found %s unique values for %s' % (len(keys), field))
else:
self.seen_values[field] = []
elastalert_logger.info('Found no values for %s' % (field))
def get_match_str(self, match):
lt = self.rules.get('use_local_time')
starttime = pretty_ts(dt_to_ts(ts_to_dt(match[self.ts_field]) - self.rules['timeframe']), lt)
endtime = pretty_ts(match[self.ts_field], lt)
message = ('A maximum of %d unique %s(s) occurred since last alert or '
'between %s and %s\n\n' % (self.rules['max_cardinality'],
self.rules['cardinality_field'],
starttime, endtime))
return message
def get_match_str(self, match):
lt = self.rules.get('use_local_time')
match_ts = lookup_es_key(match, self.ts_field)
starttime = pretty_ts(dt_to_ts(ts_to_dt(match_ts) - self.rules['timeframe']), lt)
message = 'At least %d(%d) events occurred between %s and %s\n\n' % (self.rules['num_events'],
match['count'],
starttime,
endtime)
return message
def get_match_str(self, match):
ts = match[self.rules['timestamp_field']]
lt = self.rules.get('use_local_time')
message = 'An abnormally low number of events occurred around %s.\n' % (pretty_ts(ts, lt))
message += 'Between %s and %s, there were less than %s events.\n\n' % (
pretty_ts(dt_to_ts(ts_to_dt(ts) - self.rules['timeframe']), lt),
pretty_ts(ts, lt),
self.rules['threshold'])
return message
def get_match_str(self, match):
lt = self.rules.get("use_local_time")
starttime = pretty_ts(dt_to_ts(ts_to_dt(match[self.ts_field]) - self.rules["timeframe"]), lt)
endtime = pretty_ts(match[self.ts_field], lt)
message = "A maximum of %d unique %s(s) occurred since last alert or " "between %s and %s\n\n" % (
self.rules["max_cardinality"],
self.rules["cardinality_field"],
starttime,
endtime,
)
return message
def send_pending_alerts(self):
pending_alerts = self.find_recent_pending_alerts(self.alert_time_limit)
for alert in pending_alerts:
_id = alert['_id']
alert = alert['_source']
try:
rule_name = alert.pop('rule_name')
alert_time = alert.pop('alert_time')
match_body = alert.pop('match_body')
except KeyError:
# Malformed alert, drop it
continue
# Find original rule
for rule in self.rules:
if rule['name'] == rule_name:
break
else:
# Original rule is missing, keep alert for later if rule reappears
continue
# Set current_es for top_count_keys query
rule_es_conn_config = self.build_es_conn_config(rule)
self.current_es = self.new_elasticsearch(rule_es_conn_config)
self.current_es_addr = (rule['es_host'], rule['es_port'])
# Send the alert unless it's a future alert
if ts_now() > ts_to_dt(alert_time):
aggregated_matches = self.get_aggregated_matches(_id)
if aggregated_matches:
matches = [match_body] + [agg_match['match_body'] for agg_match in aggregated_matches]
self.alert(matches, rule, alert_time=alert_time)
if rule['current_aggregate_id'] == _id:
rule['current_aggregate_id'] = None
else:
self.alert([match_body], rule, alert_time=alert_time)
# Delete it from the index
try:
self.writeback_es.delete(index=self.writeback_index,
doc_type='elastalert',
id=_id)
except: # TODO: Give this a more relevant exception, try:except: is evil.
self.handle_error("Failed to delete alert %s at %s" % (_id, alert_time))
# Send in memory aggregated alerts
for rule in self.rules:
if rule['agg_matches']:
if ts_now() > rule['aggregate_alert_time']:
self.alert(rule['agg_matches'], rule)
rule['agg_matches'] = []
def send_pending_alerts(self):
pending_alerts = self.find_recent_pending_alerts(self.alert_time_limit)
for alert in pending_alerts:
_id = alert['_id']
alert = alert['_source']
try:
rule_name = alert.pop('rule_name')
alert_time = alert.pop('alert_time')
match_body = alert.pop('match_body')
except KeyError:
# Malformed alert, drop it
continue
agg_id = alert.get('aggregate_id', None)
if agg_id:
# Aggregated alerts will be taken care of by get_aggregated_matches
continue
# Find original rule
for rule in self.rules:
if rule['name'] == rule_name:
break
else:
# Original rule is missing, drop alert
continue
# Retry the alert unless it's a future alert
if ts_now() > ts_to_dt(alert_time):
aggregated_matches = self.get_aggregated_matches(_id)
if aggregated_matches:
matches = [match_body] + [agg_match['match_body'] for agg_match in aggregated_matches]
self.alert(matches, rule, alert_time=alert_time)
rule['current_aggregate_id'] = None
else:
self.alert([match_body], rule, alert_time=alert_time)
# Delete it from the index
try:
self.writeback_es.delete(index=self.writeback_index,
doc_type='elastalert',
id=_id)
except:
self.handle_error("Failed to delete alert %s at %s" % (_id, alert_time))
# Send in memory aggregated alerts
for rule in self.rules:
if rule['agg_matches']:
if ts_now() > rule['aggregate_alert_time']:
self.alert(rule['agg_matches'], rule)
rule['agg_matches'] = []
def get_match_str(self, match):
lt = self.rules.get('use_local_time')
starttime = (ts_to_dt(match[self.ts_field]) - self.rules['timeframe']).isoformat()
endtime = match[self.ts_field]
if 'max_cardinality' in self.rules:
message = ('A maximum of %d unique %s(s) occurred since last alert or between %s and %s\n\n' % (
self.rules['max_cardinality'],
self.rules['query_key'],
starttime, endtime))
else:
message = ('Less than %d unique %s(s) occurred since last alert or between %s and %s\n\n' % (
self.rules['min_cardinality'],
self.rules['query_key'],
starttime, endtime))
return message
def get_match_str(self, match):
ts = match[self.rules['timestamp_field']]
lt = self.rules.get('use_local_time')
try:
match_value = self.match_value[-1][:5]
except:
match_value = []
message = "Between %s and %s\n" % (pretty_ts(dt_to_ts(ts_to_dt(ts) - self.rules['timeframe']), lt), pretty_ts(ts, lt))
message += "%s(%s) %s %s\nmatch value:\n\t%s...\n\n" % (
self.rules['stat'],
self.rules['stat_field'],
self.rules['stat_type'],
self.rules['threshold'],
'\n\t'.join(match_value)
)
return message
def run_all_rules(self):
""" Run each rule one time """
# If writeback_es errored, it's disabled until the next query cycle
if not self.writeback_es:
self.writeback_es = self.new_elasticsearch(self.es_conn_config)
self.send_pending_alerts()
next_run = datetime.datetime.utcnow() + self.run_every
for rule in self.rules:
# Set endtime based on the rule's delay
delay = rule.get('query_delay')
if hasattr(self.args, 'end') and self.args.end:
endtime = ts_to_dt(self.args.end)
elif delay:
endtime = ts_now() - delay
else:
endtime = ts_now()
try:
num_matches = self.run_rule(rule, endtime, self.starttime)
except EAException as e:
self.handle_error("Error running rule %s: %s" % (rule['name'], e), {'rule': rule['name']})
else:
old_starttime = pretty_ts(rule.get('original_starttime'), rule.get('use_local_time'))
logging.info("Ran %s from %s to %s: %s query hits, %s matches,"
" %s alerts sent" % (rule['name'], old_starttime, pretty_ts(endtime, rule.get('use_local_time')),
self.num_hits, num_matches, self.alerts_sent))
self.alerts_sent = 0
self.remove_old_events(rule)
if next_run < datetime.datetime.utcnow():
# We were processing for longer than our refresh interval
# This can happen if --start was specified with a large time period
# or if we are running too slow to process events in real time.
logging.warning("Querying from %s to %s took longer than %s!" % (old_starttime, endtime, self.run_every))
# Only force starttime once
self.starttime = None
if not self.args.pin_rules:
self.load_rule_changes()
def start(self):
""" Periodically go through each rule and run it """
if self.starttime:
try:
self.starttime = ts_to_dt(self.starttime)
except (TypeError, ValueError):
self.handle_error("%s is not a valid ISO 8601 timestamp (YYYY-MM-DDTHH:MM:SS+XX:00)" % (self.starttime))
exit(1)
self.running = True
while self.running:
next_run = datetime.datetime.utcnow() + self.run_every
self.run_all_rules()
if next_run < datetime.datetime.utcnow():
continue
# Wait before querying again
sleep_duration = (next_run - datetime.datetime.utcnow()).seconds
self.sleep_for(sleep_duration)
def get_all_terms(self, args):
""" Performs a terms aggregation for each field to get every existing term. """
self.es = Elasticsearch(host=self.rules['es_host'],
port=self.rules['es_port'],
timeout=self.rules.get('es_conn_timeout', 50))
window_size = datetime.timedelta(
**self.rules.get('terms_window_size', {'days': 30}))
field_name = {"field": "", "size": 2147483647} # Integer.MAX_VALUE
query_template = {"aggs": {"values": {"terms": field_name}}}
if args and args.start:
end = ts_to_dt(args.start)
else:
end = ts_now()
start = end - window_size
if self.rules.get('use_strftime_index'):
index = format_index(self.rules['index'], start, end)
else:
index = self.rules['index']
time_filter = {
self.rules['timestamp_field']: {
'lte': dt_to_ts(end),
'gte': dt_to_ts(start)
}
}
query_template['filter'] = {'bool': {'must': [{'range': time_filter}]}}
query = {'aggs': {'filtered': query_template}}
for field in self.fields:
field_name['field'] = field
res = self.es.search(body=query,
index=index,
ignore_unavailable=True,
timeout='50s')
if 'aggregations' in res:
buckets = res['aggregations']['filtered']['values']['buckets']
keys = [bucket['key'] for bucket in buckets]
self.seen_values[field] = keys
elastalert_logger.info('Found %s unique values for %s' %
(len(keys), field))
else:
self.seen_values[field] = []
elastalert_logger.info('Found no values for %s' % (field))
def create_default_title(self, matches):
subject = '%s: %d matches found - %s' % \
(self.rule['name'], len(matches),
pretty_ts(ts_to_dt(self.pipeline['alert_time'])))
return subject
def alert(self, matches, rule, alert_time=None):
""" Send out an alert.
:param matches: A list of matches.
:param rule: A rule configuration.
"""
if alert_time is None:
alert_time = ts_now()
# Compute top count keys
if rule.get('top_count_keys'):
for match in matches:
if 'query_key' in rule and rule['query_key'] in match:
qk = match[rule['query_key']]
else:
qk = None
start = ts_to_dt(match[rule['timestamp_field']]) - rule.get('timeframe', datetime.timedelta(minutes=10))
end = ts_to_dt(match[rule['timestamp_field']]) + datetime.timedelta(minutes=10)
keys = rule.get('top_count_keys')
counts = self.get_top_counts(rule, start, end, keys, rule.get('top_count_number'), qk)
match.update(counts)
# Generate a kibana dashboard for the first match
if rule.get('generate_kibana_link') or rule.get('use_kibana_dashboard'):
try:
if rule.get('generate_kibana_link'):
kb_link = self.generate_kibana_db(rule, matches[0])
else:
kb_link = self.use_kibana_link(rule, matches[0])
except EAException as e:
self.handle_error("Could not generate kibana dash for %s match: %s" % (rule['name'], e))
else:
if kb_link:
matches[0]['kibana_link'] = kb_link
for enhancement in rule['match_enhancements']:
for match in matches:
try:
enhancement.process(match)
except EAException as e:
self.handle_error("Error running match enhancement: %s" % (e), {'rule': rule['name']})
# Don't send real alerts in debug mode
if self.debug:
alerter = DebugAlerter(rule)
alerter.alert(matches)
return
# Run the alerts
alert_sent = False
alert_exception = None
for alert in rule['alert']:
try:
alert.alert(matches)
except EAException as e:
self.handle_error('Error while running alert %s: %s' % (alert.get_info()['type'], e), {'rule': rule['name']})
alert_exception = str(e)
else:
self.alerts_sent += 1
alert_sent = True
# Write the alert(s) to ES
agg_id = None
for match in matches:
alert_body = self.get_alert_body(match, rule, alert_sent, alert_time, alert_exception)
# Set all matches to aggregate together
if agg_id:
alert_body['aggregate_id'] = agg_id
res = self.writeback('elastalert', alert_body)
if res and not agg_id:
agg_id = res['_id']
def send_alert(self, matches, rule, alert_time=None):
""" Send out an alert.
:param matches: A list of matches.
:param rule: A rule configuration.
"""
if alert_time is None:
alert_time = ts_now()
# Compute top count keys
if rule.get('top_count_keys'):
for match in matches:
if 'query_key' in rule and rule['query_key'] in match:
qk = match[rule['query_key']]
else:
qk = None
start = ts_to_dt(match[rule['timestamp_field']]) - rule.get('timeframe', datetime.timedelta(minutes=10))
end = ts_to_dt(match[rule['timestamp_field']]) + datetime.timedelta(minutes=10)
keys = rule.get('top_count_keys')
counts = self.get_top_counts(rule, start, end, keys, qk=qk)
match.update(counts)
# Generate a kibana3 dashboard for the first match
if rule.get('generate_kibana_link') or rule.get('use_kibana_dashboard'):
try:
if rule.get('generate_kibana_link'):
kb_link = self.generate_kibana_db(rule, matches[0])
else:
kb_link = self.use_kibana_link(rule, matches[0])
except EAException as e:
self.handle_error("Could not generate kibana dash for %s match: %s" % (rule['name'], e))
else:
if kb_link:
matches[0]['kibana_link'] = kb_link
if rule.get('use_kibana4_dashboard'):
kb_link = self.generate_kibana4_db(rule, matches[0])
if kb_link:
matches[0]['kibana_link'] = kb_link
#if not rule.get('run_enhancements_first'):
# for enhancement in rule.get('match_enhancements'):
# valid_matches = []
# for match in matches:
# try:
# enhancement.process(match)
# valid_matches.append(match)
# except DropMatchException as e:
# pass
# except EAException as e:
# self.handle_error("Error running match enhancement: %s" % (e), {'rule': rule['name']})
# matches = valid_matches
# if not matches:
# return None
# Don't send real alerts in debug mode
if self.debug:
alerter = DebugAlerter(rule)
alerter.alert(matches)
return None
# Run the alerts
alert_sent = False
alert_exception = None
# Alert.pipeline is a single object shared between every alerter
# This allows alerters to pass objects and data between themselves
alert_pipeline = {"alert_time": alert_time}
for alert in rule.get("actions"):
alert.pipeline = alert_pipeline
try:
alert.alert(matches)
except EAException as e:
self.handle_error('Error while running alert %s: %s' % (alert.get_info()['type'], e), {'rule': rule['name']})
alert_exception = str(e)
else:
self.alerts_sent += 1
alert_sent = True
def replace_ts(self, hits, rule):
for hit in hits:
hit['_source'][rule['timestamp_field']] = ts_to_dt(hit['_source'][rule['timestamp_field']])
def unwrap_interval_buckets(self, timestamp, query_key, interval_buckets):
for interval_data in interval_buckets:
# Use bucket key here instead of start_time for more accurate match timestamp
self.check_matches(ts_to_dt(interval_data['key_as_string']),
query_key, interval_data)
def get_all_terms(self, args):
""" Performs a terms aggregation for each field to get every existing term. """
self.es = elasticsearch_client(self.rules)
window_size = datetime.timedelta(**self.rules.get('terms_window_size', {'days': 30}))
field_name = {"field": "", "size": 2147483647} # Integer.MAX_VALUE
query_template = {"aggs": {"values": {"terms": field_name}}}
if args and args.start:
end = ts_to_dt(args.start)
else:
end = ts_now()
start = end - window_size
step = datetime.timedelta(**self.rules.get('window_step_size', {'days': 1}))
for field in self.fields:
tmp_start = start
tmp_end = min(start + step, end)
time_filter = {self.rules['timestamp_field']: {'lt': dt_to_ts(tmp_end), 'gte': dt_to_ts(tmp_start)}}
query_template['filter'] = {'bool': {'must': [{'range': time_filter}]}}
query = {'aggs': {'filtered': query_template}}
# For composite keys, we will need to perform sub-aggregations
if type(field) == list:
self.seen_values.setdefault(tuple(field), [])
level = query_template['aggs']
# Iterate on each part of the composite key and add a sub aggs clause to the elastic search query
for i, sub_field in enumerate(field):
level['values']['terms']['field'] = add_raw_postfix(sub_field)
if i < len(field) - 1:
# If we have more fields after the current one, then set up the next nested structure
level['values']['aggs'] = {'values': {'terms': copy.deepcopy(field_name)}}
level = level['values']['aggs']
else:
self.seen_values.setdefault(field, [])
# For non-composite keys, only a single agg is needed
field_name['field'] = add_raw_postfix(field)
# Query the entire time range in small chunks
while tmp_start < end:
if self.rules.get('use_strftime_index'):
index = format_index(self.rules['index'], tmp_start, tmp_end)
else:
index = self.rules['index']
res = self.es.search(body=query, index=index, ignore_unavailable=True, timeout='50s')
if 'aggregations' in res:
buckets = res['aggregations']['filtered']['values']['buckets']
if type(field) == list:
# For composite keys, make the lookup based on all fields
# Make it a tuple since it can be hashed and used in dictionary lookups
for bucket in buckets:
# We need to walk down the hierarchy and obtain the value at each level
self.seen_values[tuple(field)] += self.flatten_aggregation_hierarchy(bucket)
else:
keys = [bucket['key'] for bucket in buckets]
self.seen_values[field] += keys
else:
self.seen_values.setdefault(field, [])
if tmp_start == tmp_end:
break
tmp_start = tmp_end
tmp_end = min(tmp_start + step, end)
time_filter[self.rules['timestamp_field']] = {'lt': dt_to_ts(tmp_end), 'gte': dt_to_ts(tmp_start)}
for key, values in self.seen_values.iteritems():
if not values:
if type(key) == tuple:
# If we don't have any results, it could either be because of the absence of any baseline data
# OR it may be because the composite key contained a non-primitive type. Either way, give the
# end-users a heads up to help them debug what might be going on.
elastalert_logger.warning((
'No results were found from all sub-aggregations. This can either indicate that there is '
'no baseline data OR that a non-primitive field was used in a composite key.'
))
else:
elastalert_logger.info('Found no values for %s' % (field))
continue
self.seen_values[key] = list(set(values))
elastalert_logger.info('Found %s unique values for %s' % (len(values), key))
def get_match_str(self, match):
ts = match[self.rules['timestamp_field']]
lt = self.rules.get('use_local_time')
message = 'An abnormally low number of events occurred around %s.\n' % (pretty_ts(ts, lt))
message += 'Between %s and %s, there were less than %s events.\n\n' % (pretty_ts(dt_to_ts(ts_to_dt(ts) - self.rules['timeframe']), lt),
pretty_ts(ts, lt),
self.rules['threshold'])
return message
def alert(self, matches, rule, alert_time=None):
""" Send out an alert.
:param matches: A list of matches.
:param rule: A rule configuration.
"""
if alert_time is None:
alert_time = ts_now()
# Compute top count keys
if rule.get('top_count_keys'):
for match in matches:
if 'query_key' in rule and rule['query_key'] in match:
qk = match[rule['query_key']]
else:
qk = None
start = ts_to_dt(match[rule['timestamp_field']]) - rule.get(
'timeframe', datetime.timedelta(minutes=10))
end = ts_to_dt(
match[rule['timestamp_field']]) + datetime.timedelta(
minutes=10)
keys = rule.get('top_count_keys')
counts = self.get_top_counts(rule, start, end, keys,
rule.get('top_count_number'), qk)
match.update(counts)
# Generate a kibana dashboard for the first match
if rule.get('generate_kibana_link') or rule.get(
'use_kibana_dashboard'):
try:
if rule.get('generate_kibana_link'):
kb_link = self.generate_kibana_db(rule, matches[0])
else:
kb_link = self.use_kibana_link(rule, matches[0])
except EAException as e:
self.handle_error(
"Could not generate kibana dash for %s match: %s" %
(rule['name'], e))
else:
if kb_link:
matches[0]['kibana_link'] = kb_link
for enhancement in rule['match_enhancements']:
for match in matches:
try:
enhancement.process(match)
except EAException as e:
self.handle_error(
"Error running match enhancement: %s" % (e),
{'rule': rule['name']})
# Don't send real alerts in debug mode
if self.debug:
alerter = DebugAlerter(rule)
alerter.alert(matches)
return
# Run the alerts
alert_sent = False
alert_exception = None
for alert in rule['alert']:
try:
alert.alert(matches)
except EAException as e:
self.handle_error(
'Error while running alert %s: %s' %
(alert.get_info()['type'], e), {'rule': rule['name']})
alert_exception = str(e)
else:
self.alerts_sent += 1
alert_sent = True
# Write the alert(s) to ES
agg_id = None
for match in matches:
alert_body = self.get_alert_body(match, rule, alert_sent,
alert_time, alert_exception)
# Set all matches to aggregate together
if agg_id:
alert_body['aggregate_id'] = agg_id
res = self.writeback('elastalert', alert_body)
if res and not agg_id:
agg_id = res['_id']
def start(self):
""" Periodically go through each rule and run it """
starttime = self.args.start
if starttime:
try:
starttime = ts_to_dt(starttime)
except (TypeError, ValueError):
self.handle_error(
"%s is not a valid ISO 8601 timestamp (YYYY-MM-DDTHH:MM:SS+XX:00)"
% (starttime))
exit(1)
while True:
# If writeback_es errored, it's disabled until the next query cycle
if not self.writeback_es:
self.writeback_es = Elasticsearch(host=self.es_host,
port=self.es_port)
self.send_pending_alerts()
next_run = datetime.datetime.utcnow() + self.run_every
for rule in self.rules:
# Set endtime based on the rule's delay
delay = rule.get('query_delay')
if hasattr(self.args, 'end') and self.args.end:
endtime = ts_to_dt(self.args.end)
elif delay:
endtime = ts_now() - delay
else:
endtime = ts_now()
try:
num_matches = self.run_rule(rule, endtime, starttime)
except EAException as e:
self.handle_error(
"Error running rule %s: %s" % (rule['name'], e),
{'rule': rule['name']})
else:
old_starttime = pretty_ts(rule.get('original_starttime'),
rule.get('use_local_time'))
logging.info(
"Ran %s from %s to %s: %s query hits, %s matches,"
" %s alerts sent" %
(rule['name'], old_starttime,
pretty_ts(endtime, rule.get('use_local_time')),
self.num_hits, num_matches, self.alerts_sent))
self.alerts_sent = 0
self.remove_old_events(rule)
if next_run < datetime.datetime.utcnow():
# We were processing for longer than our refresh interval
# This can happen if --start was specified with a large time period
# or if we are running too slow to process events in real time.
logging.warning("Querying from %s to %s took longer than %s!" %
(old_starttime, endtime, self.run_every))
continue
# Only force starttime once
starttime = None
if not self.args.pin_rules:
self.load_rule_changes()
# Wait before querying again
sleep_for = (next_run - datetime.datetime.utcnow()).seconds
logging.info("Sleeping for %s seconds" % (sleep_for))
time.sleep(sleep_for)
def send_alert(self, matches, rule, alert_time=None):
""" Send out an alert.
:param matches: A list of matches.
:param rule: A rule configuration.
"""
if alert_time is None:
alert_time = ts_now()
# Compute top count keys
if rule.get('top_count_keys'):
for match in matches:
if 'query_key' in rule and rule['query_key'] in match:
qk = match[rule['query_key']]
else:
qk = None
start = ts_to_dt(match[rule['timestamp_field']]) - rule.get(
'timeframe', datetime.timedelta(minutes=10))
end = ts_to_dt(
match[rule['timestamp_field']]) + datetime.timedelta(
minutes=10)
keys = rule.get('top_count_keys')
counts = self.get_top_counts(rule, start, end, keys, qk=qk)
match.update(counts)
# Generate a kibana3 dashboard for the first match
if rule.get('generate_kibana_link') or rule.get(
'use_kibana_dashboard'):
try:
if rule.get('generate_kibana_link'):
kb_link = self.generate_kibana_db(rule, matches[0])
else:
kb_link = self.use_kibana_link(rule, matches[0])
except EAException as e:
self.handle_error(
"Could not generate kibana dash for %s match: %s" %
(rule['name'], e))
else:
if kb_link:
matches[0]['kibana_link'] = kb_link
if rule.get('use_kibana4_dashboard'):
kb_link = self.generate_kibana4_db(rule, matches[0])
if kb_link:
matches[0]['kibana_link'] = kb_link
#if not rule.get('run_enhancements_first'):
# for enhancement in rule.get('match_enhancements'):
# valid_matches = []
# for match in matches:
# try:
# enhancement.process(match)
# valid_matches.append(match)
# except DropMatchException as e:
# pass
# except EAException as e:
# self.handle_error("Error running match enhancement: %s" % (e), {'rule': rule['name']})
# matches = valid_matches
# if not matches:
# return None
# Don't send real alerts in debug mode
if self.debug:
alerter = DebugAlerter(rule)
alerter.alert(matches)
return None
# Run the alerts
alert_sent = False
alert_exception = None
# Alert.pipeline is a single object shared between every alerter
# This allows alerters to pass objects and data between themselves
alert_pipeline = {"alert_time": alert_time}
for alert in rule.get("actions"):
alert.pipeline = alert_pipeline
try:
alert.alert(matches)
except EAException as e:
self.handle_error(
'Error while running alert %s: %s' %
(alert.get_info()['type'], e), {'rule': rule['name']})
alert_exception = str(e)
else:
self.alerts_sent += 1
alert_sent = True
def replace_ts(self, hits, rule):
for hit in hits:
hit['_source'][rule['timestamp_field']] = ts_to_dt(
hit['_source'][rule['timestamp_field']])
def unwrap_interval_buckets(self, timestamp, query_key, interval_buckets):
for interval_data in interval_buckets:
# Use bucket key here instead of start_time for more accurate match timestamp
self.check_matches(ts_to_dt(interval_data['key_as_string']), query_key, interval_data)
