icinga2-checks/check_matrix_msg.py

import argparse
import asyncio
import sys
import time
import traceback

from nio import JoinResponse, RoomSendResponse

from checker import nagios, print_icinga2_check_status
from checker.synapse_client import login_and_cache


async def main(args):
    start_time = time.time()
    client = await login_and_cache(args.username, args.password, args.homeserver, auth_file=args.auth_file)

    join_response = await client.join(args.room)
    if not isinstance(join_response, JoinResponse):
        print_icinga2_check_status(f'failed to join room!\n{join_response}', nagios.STATE_CRIT)
        sys.exit(nagios.STATE_CRIT)

    time.sleep(1)

    test_msg = f'TEST MESSAGE AT {time.time()}'

    send_response = await client.room_send(
        room_id=args.room,
        message_type="m.room.message",
        content={"msgtype": "m.text", "body": test_msg},
    )

    if not isinstance(send_response, RoomSendResponse):
        print_icinga2_check_status(f'failed to send message!\n{send_response}', nagios.STATE_CRIT)
        sys.exit(nagios.STATE_CRIT)

    time.sleep(1)

    event_id = send_response.event_id

    found_msg = False
    for i in range(args.tries):
        sync_response = await client.sync(30000)
        room_timeline = sync_response.rooms.join[args.room].timeline
        if any(event.event_id == event_id for event in room_timeline.events):
            found_msg = True
            break

    await client.close()

    if not found_msg:
        print_icinga2_check_status(f'failed to find our message in the room timeline', nagios.STATE_CRIT)
        sys.exit(nagios.STATE_CRIT)

    elapsed_time = round((time.time() - start_time) - 2, 2)
    perfdata = {
        'elapsed': {
            'value': elapsed_time,
            'warn': args.warn,
            'crit': args.critical,
            'min': 0,
            'unit': 's'
        }
    }

    if elapsed_time >= args.critical:
        return_code = nagios.STATE_CRIT
        return_msg = f'message send time was {elapsed_time} seconds'
    elif elapsed_time >= args.warn:
        return_code = nagios.STATE_WARN
        return_msg = f'message send time was {elapsed_time} seconds'
    else:
        return_code = nagios.STATE_OK
        return_msg = f'message send successful, took {elapsed_time} seconds'

    print_icinga2_check_status(return_msg, return_code, perfdata=perfdata)
    sys.exit(return_code)


if __name__ == "__main__":
    parser = argparse.ArgumentParser(description='Test sending message to a Matrix room.')
    parser.add_argument('-s', '--homeserver', required=True, help='Your Matrix server. Example: https://your-homeserver.org')
    parser.add_argument('-u', '--username', required=True, help='Matrix username.')
    parser.add_argument('-p', '--password', required=True, help='Matrix password.')
    parser.add_argument('-r', '--room', required=True, help='Room to send a message to.')
    parser.add_argument('-a', '--auth-file', help="File to cache the bot's login details to.")
    parser.add_argument('-t', '--tries', type=int, default=10, help="How many times to check for the test message.")
    parser.add_argument('--warn', type=int, default=20, help='Warn level for message send time.')
    parser.add_argument('--critical', type=int, default=30, help='Critical level for message send time.')
    args = parser.parse_args()

    try:
        asyncio.run(main(args))
    except Exception as e:
        print(f"UNKNOWN: exception\n{e}")
        print(traceback.format_exc())
        sys.exit(nagios.UNKNOWN)