def run(self, conf):
name = 'stalker-agent'
log_type = conf.get('log_type', 'syslog')
log_file = conf.get('log_path', '/var/log/stalker/stalker-runner.log')
if log_type == 'syslog':
logger = get_syslogger(conf, name)
else:
logger = get_logger(name, log_path=log_file)
def spawn_worker():
sr = StalkerRunner(conf)
while 1:
try:
sr.start()
except Exception as err:
logger.info(err)
worker_count = int(conf.get('workers', '1'))
def kill_children(*args):
"""Kills the entire process group."""
logger.error('SIGTERM received')
signal.signal(signal.SIGTERM, signal.SIG_IGN)
running[0] = False
os.killpg(0, signal.SIGTERM)
def hup(*args):
"""Shuts down the server, but allow running requests to complete"""
logger.error('SIGHUP received')
signal.signal(signal.SIGHUP, signal.SIG_IGN)
running[0] = False
running = [True]
signal.signal(signal.SIGTERM, kill_children)
signal.signal(signal.SIGHUP, hup)
children = []
while running[0]:
while len(children) < worker_count:
pid = os.fork()
if pid == 0:
signal.signal(signal.SIGHUP, signal.SIG_DFL)
signal.signal(signal.SIGTERM, signal.SIG_DFL)
spawn_worker()
logger.info('Child %d exiting normally' % os.getpid())
return
else:
logger.info('Started child %s' % pid)
children.append(pid)
try:
pid, status = os.wait()
if os.WIFEXITED(status) or os.WIFSIGNALED(status):
logger.error('Removing dead child %s' % pid)
if pid in children:
children.remove(pid)
except OSError, err:
if err.errno not in (errno.EINTR, errno.ECHILD):
raise
except KeyboardInterrupt:
break
def __init__(self, fullconf):
conf = fullconf['main']
self.fullconf = fullconf
log_file = conf.get('log_path', '/var/log/stalker/stalker-agent.log')
self.logger = get_logger('stalker_agent', log_path=log_file)
self.request_logger = FileLikeLogger(self.logger)
self.listen_addr = conf.get('listen_addr', '')
self.listen_port = int(conf.get('listen_port', '5050'))
self.ssl_crt_path = conf.get('ssl_cert', '/etc/stalker/ssl.crt')
self.ssl_key_path = conf.get('ssl_key', '/etc/stalker/ssl.key')
self.master_url = conf.get('master_url', 'http://localhost:5000')
if self.master_url.startswith('https://'):
self.master_scheme = 'https'
else:
self.master_scheme = 'http'
self.register_key = conf.get('register_key', 'itsamario')
self.check_key = conf.get('check_key', 'canhazstatus')
self.script_dir = conf.get('script_dir', '/etc/stalker/scripts')
self.default_interval = int(conf.get('default_interval', '300'))
self.default_priority = int(conf.get('default_priority', '1'))
self.scripts = {}
hostname_parts = int(conf.get('hostname_parts', '1'))
self.hostname = conf.get('hostname', '.'.join(getfqdn().split('.')[:hostname_parts]))
self.roles = [x.strip() for x in conf.get('roles',
'server').split(',')]
if not os.path.exists(self.script_dir):
raise Exception("No script dir: %s" % self.script_dir)
self._build_check_list()
self.logger.info('Found checks: %s' % self.scripts)
def __init__(self, conf):
self.conf = conf
self.name = 'stalker-runner-%d' % os.getpid()
log_type = conf.get('log_type', 'syslog')
log_file = conf.get('log_file', '/var/log/stalker/stalker-runner.log')
if log_type == 'syslog':
self.logger = get_syslogger(conf, self.name)
else:
self.logger = get_logger(self.name, log_path=log_file)
self.pool = eventlet.GreenPool()
self.check_key = conf.get('check_key', 'canhazstatus')
redis_host = conf.get('redis_host', '127.0.0.1')
redis_port = int(conf.get('redis_port', '6379'))
redis_pass = conf.get('redis_password', '')
redis_usock = conf.get('redis_socket', None)
self.wq = conf.get('worker_id', 'worker1')
self.rc = redis.Redis(redis_host, redis_port, password=redis_pass,
unix_socket_path=redis_usock)
mongo_host = conf.get('mongo_host', '127.0.0.1')
mongo_port = int(conf.get('mongo_port', '27017'))
db_name = conf.get('db_name', 'stalkerweb')
self.c = MongoClient(host=mongo_host, port=mongo_port)
self.debug = False
self.db = self.c[db_name]
self.checks = self.db['checks']
self.state_log = self.db['state_log']
self.notifications = self.db['notifications']
self.host_window = int(conf.get('host_flood_window', '60'))
self.host_threshold = int(conf.get('host_flood_threshold', '5'))
self.flood_window = int(conf.get('dc_flood_window', '120'))
self.flood_threshold = int(conf.get('dc_flood_threshold', '100'))
self.flap_window = int(conf.get('flap_window', '1200'))
self.flap_threshold = int(conf.get('flap_threshold', '5'))
self.alert_threshold = int(conf.get('alert_threshold', '3'))
self.urlopen_timeout = int(conf.get('urlopen_timeout', '15'))
self.notify_plugins = {}
self._load_notification_plugins(conf)
self.statsd = StatsdEvent(conf, self.logger, 'stalker_runner.')
def __init__(self, conf):
self.conf = conf
log_file = conf.get('log_path', '/var/log/stalker/stalker-manager.log')
self.logger = get_logger('stalker_manager', log_path=log_file)
redis_host = conf.get('redis_host', '127.0.0.1')
redis_port = int(conf.get('redis_port', '6379'))
redis_pass = conf.get('redis_password', '')
redis_usock = conf.get('redis_socket', None)
self.wq = conf.get('qname', 'worker1')
self.rc = redis.Redis(redis_host, redis_port, password=redis_pass,
unix_socket_path=redis_usock)
mongo_host = conf.get('mongo_host', '127.0.0.1')
mongo_port = int(conf.get('mongo_port', '27017'))
db_name = conf.get('db_name', 'stalkerweb')
self.c = MongoClient(host=mongo_host, port=mongo_port)
self.db = self.c[db_name]
self.checks = self.db['checks']
self.notifications = self.db['notifications']
self.scan_interval = int(conf.get('scan_interval', '5'))
self.pause_file = conf.get('pause_file', '/tmp/.sm-pause')
self.shuffle_on_start = True
self.statsd = StatsdEvent(conf, self.logger, 'stalker_manager.')
self.metrics = {'checks': 0, 'pending': 0, 'suspended': 0,
'failing': 0, 'flapping': 0, 'qsize': 0}
from stalkerweb.auth import is_valid_login, login_required, remove_user
from stalkerweb.stutils import jsonify, genPrimaryKey64
from stalkerweb import app, rc, rdb
from stalker.stalker_utils import get_logger
from flask.ext.wtf import Form, Required, TextField, PasswordField, \
BooleanField
from werkzeug.contrib.cache import RedisCache
import rethinkdb as r
from rethinkdb.errors import RqlDriverError, RqlRuntimeError
VALID_STATES = ['alerting', 'pending', 'in_maintenance', 'suspended']
cache = RedisCache(host=app.config['REDIS_HOST'], port=app.config['REDIS_PORT'], default_timeout=app.config['CACHE_TTL'])
logger = get_logger(app.config['LOG_NAME'],
log_path=app.config['LOG_FILE'],
count=app.config['LOG_COUNT'])
class SignInForm(Form):
username = TextField(validators=[Required()])
password = PasswordField(validators=[Required()])
remember_me = BooleanField()
def _get_local_metrics():
metrics = {}
mkeys = ['checks', 'failing', 'flapping', 'pending', 'qsize',
'suspended']
try:
values = rc.mget(mkeys)