icinga2-checks/check_matrix_synapse.py

151 lines
6.9 KiB
Python
Executable File

#!/usr/bin/env python3
import argparse
import sys
import time
import traceback
import numpy as np
import requests
from checker import nagios
from checker.synapse_grafana import get_avg_python_gc_time, get_event_send_time, get_outgoing_http_request_rate, get_waiting_for_db
parser = argparse.ArgumentParser(description='Process some integers.')
parser.add_argument('--grafana-server', required=True, help='Grafana server. Example: http://10.0.0.43:3000')
parser.add_argument('--synapse-server', required=True, help='Matrix Synapse server.')
parser.add_argument('--grafana-api-key', required=True)
parser.add_argument('--interval', default=15, type=int, help='Data interval in seconds.')
parser.add_argument('--range', default=2, type=int, help='Data range in minutes. Used for comparison and averaging.')
parser.add_argument('--type', required=True, choices=['gc-time', 'response-time', 'outgoing-http-rate', 'avg-send', 'db-lag'])
parser.add_argument('--warn', type=float, help='Manually set warn level.')
parser.add_argument('--crit', type=float, help='Manually set critical level.')
args = parser.parse_args()
# TODO: add warn suppoort
def main():
if args.type == 'gc-time':
# in seconds
python_gc_time_sum_MAX = 0.002 if not args.crit else args.crit
try:
raw_gc_time = get_avg_python_gc_time(args.grafana_api_key, args.interval, args.range, args.grafana_server)
if not len(raw_gc_time):
print(f"UNKNOWN: no data for average GC time")
sys.exit(nagios.UNKNOWN)
python_gc_time_sum = np.round(np.average(raw_gc_time), 5)
if python_gc_time_sum >= python_gc_time_sum_MAX:
print(f"CRITICAL: average GC time per collection is {python_gc_time_sum} sec. |'garbage-collection'={python_gc_time_sum}s;;;")
sys.exit(nagios.CRITICAL)
else:
print(f"OK: average GC time per collection is {python_gc_time_sum} sec. |'garbage-collection'={python_gc_time_sum}s;;;")
sys.exit(nagios.OK)
except Exception as e:
print(f'UNKNOWN: failed to check avg. GC time "{e}"')
print(traceback.format_exc())
sys.exit(nagios.UNKNOWN)
elif args.type == 'response-time':
response_time_MAX = 1 if not args.crit else args.crit
timeout = 10
try:
response_times = []
for i in range(10):
start = time.perf_counter()
try:
response = requests.post(args.synapse_server, timeout=timeout, verify=False)
except Exception as e:
print(f'UNKNOWN: failed to ping endpoint "{e}"')
print(traceback.format_exc())
sys.exit(nagios.UNKNOWN)
request_time = time.perf_counter() - start
response_times.append(np.round(request_time, 2))
time.sleep(1)
response_time = np.round(np.average(response_times), 2)
if response_time > response_time_MAX:
print(f"CRITICAL: response time is {response_time} sec. |'response-time'={response_time}s;;;")
sys.exit(nagios.CRITICAL)
else:
print(f"OK: response time is {response_time} sec. |'response-time'={response_time}s;;;")
sys.exit(nagios.OK)
except Exception as e:
print(f'UNKNOWN: failed to check response time "{e}"')
print(traceback.format_exc())
sys.exit(nagios.UNKNOWN)
elif args.type == 'outgoing-http-rate':
# outgoing req/sec
outgoing_http_request_rate_MAX = 10 if not args.crit else args.crit
try:
outgoing_http_request_rate = get_outgoing_http_request_rate(args.grafana_api_key, args.interval, args.range, args.grafana_server)
if not len(outgoing_http_request_rate.keys()):
print(f"UNKNOWN: no data for outgoing HTTP request rate")
sys.exit(nagios.UNKNOWN)
failed = {}
perf_data = '|'
for k, v in outgoing_http_request_rate.items():
perf_data = perf_data + f"'{k}'={v}s;;; "
if v > outgoing_http_request_rate_MAX:
failed[k] = v
if len(failed.keys()) > 0:
print(f'CRITICAL: outgoing HTTP request rate for {failed} req/sec.', perf_data)
sys.exit(nagios.CRITICAL)
print(f'OK: outgoing HTTP request rate is {outgoing_http_request_rate} req/sec.', perf_data)
sys.exit(nagios.OK)
except Exception as e:
print(f'UNKNOWN: failed to check outgoing HTTP request rate "{e}"')
print(traceback.format_exc())
sys.exit(nagios.UNKNOWN)
elif args.type == 'avg-send':
# Average send time in seconds
event_send_time_MAX = 1 if not args.crit else args.crit
try:
event_send_time = get_event_send_time(args.grafana_api_key, args.interval, args.range, args.grafana_server)
if not event_send_time:
print(f"UNKNOWN: no data for event send time")
sys.exit(nagios.UNKNOWN)
if event_send_time > event_send_time_MAX:
print(f"CRITICAL: average message send time is {event_send_time} sec. |'avg-send-time'={event_send_time}s;;;")
sys.exit(nagios.CRITICAL)
else:
print(f"OK: average message send time is {event_send_time} sec. |'avg-send-time'={event_send_time}s;;;")
sys.exit(nagios.OK)
except Exception as e:
print(f'UNKNOWN: failed to check average message send time "{e}"')
print(traceback.format_exc())
sys.exit(nagios.UNKNOWN)
elif args.type == 'db-lag':
db_lag_MAX = 0.01 if not args.crit else args.crit # in seconds
try:
db_lag, null_present, raw_data = get_waiting_for_db(args.grafana_api_key, args.interval, args.range, args.grafana_server)
if db_lag == None:
print(f"UNKNOWN: no data for DB lag")
sys.exit(nagios.UNKNOWN)
null_warn = f'Null data was present for this timeseries, value may be incorrect.\n{raw_data}' if null_present else ''
if db_lag > db_lag_MAX:
print(f"CRITICAL: DB lag is {db_lag} sec. {null_warn} |'db-lag'={db_lag}s;;;")
sys.exit(nagios.CRITICAL)
else:
print(f"OK: DB lag is {db_lag} sec. {null_warn} |'db-lag'={db_lag}s;;;")
sys.exit(nagios.OK)
except Exception as e:
print(f'UNKNOWN: failed to check DB lag "{e}"')
print(traceback.format_exc())
sys.exit(nagios.UNKNOWN)
else:
print('Wrong type')
sys.exit(nagios.UNKNOWN)
if __name__ == "__main__":
try:
main()
except Exception as e:
print(f'UNKNOWN: exception "{e}"')
print(traceback.format_exc())
sys.exit(nagios.UNKNOWN)