def run(cls) -> Dict:
timeout = conf.get(HEALTH_CHECK_WORKER_TIMEOUT)
arg = random.randint(1, 10)
result = None
try:
task_result = celery_app.send_task(
cls.WORKER_HEALTH_TASK,
args=[arg, arg],
expires=timeout,
)
task_result.get(timeout=timeout)
if task_result.result != health_task(arg, arg):
result = Result(
message='Service returned wrong health result.',
severity=Result.WARNING
)
except IOError:
result = Result(
message='Service has an "IOError".',
severity=Result.ERROR
)
except Exception as e:
result = Result(
message='Service has an "{}" error.'.format(e),
severity=Result.ERROR
)
if not result:
result = Result()
return {cls.WORKER_NAME: result}
def run(cls):
try:
disk = psutil.disk_usage('/')
if disk.percent >= cls.ERROR_THRESHOLD:
result = Result(
message=
'{host} {percent}% disk usage exceeds error {disk_usage}%'.
format(host=cls.HOST,
percent=disk.percent,
disk_usage=cls.ERROR_THRESHOLD),
severity=Result.ERROR)
elif disk.percent >= cls.WARNING_THRESHOLD:
result = Result(
message=
'{host} {percent}% disk usage exceeds warning {disk_usage}%'
.format(host=cls.HOST,
percent=disk.percent,
disk_usage=cls.ERROR_THRESHOLD),
severity=Result.ERROR)
else:
result = Result()
except ValueError:
result = Result(message='An error was raised, try again later.',
severity=Result.ERROR)
return {'DISK': result}
def check():
"""Open and close the broker channel."""
try:
# Context to release connection
with Connection(conf.get('CELERY_BROKER_URL')) as conn:
conn.connect()
except ConnectionRefusedError:
return Result(
message='Service unable to connect, "Connection was refused".',
severity=Result.ERROR)
except AccessRefused:
return Result(
message='Service unable to connect, "Authentication error".',
severity=Result.ERROR)
except IOError:
return Result(message='Service has an "IOError".',
severity=Result.ERROR)
except Exception as e:
return Result(message='Service has an "{}" error.'.format(e),
severity=Result.ERROR)
else:
return Result()
def redis_health(connection):
try:
info = connection.info()
return Result(message='Service is healthy, db size {}'.format(info['used_memory']))
except redis.exceptions.ConnectionError:
return Result(message='Service unable to connect, "Connection error".',
severity=Result.ERROR)
except Exception as e:
return Result(message='Service unable to connect, encountered error "{}".'.format(e),
severity=Result.ERROR)
def run(cls) -> Dict:
response = safe_request('{}/healthz'.format(get_local_stream()), 'GET')
status_code = response.status_code
if status_code == 200:
result = Result()
else:
result = Result(
message='Service is not healthy, response {}'.format(
status_code),
severity=Result.ERROR)
return {'STREAMS': result}
def run(cls):
response = safe_request('', 'GET')
status_code = response.status_code
if status_code == 200:
result = Result()
else:
result = Result(
message='Service is not healthy, response {}'.format(
status_code),
severity=Result.ERROR)
return {'STREAMS': result}
def run(cls) -> Dict:
response = safe_request('{}/_health'.format(get_settings_ws_api_url()),
'GET')
status_code = response.status_code
if status_code == 200:
result = Result()
else:
result = Result(
message='Service is not healthy, response {}'.format(
status_code),
severity=Result.ERROR)
return {'STREAMS': result}
def run(cls):
results = {}
result = cls.redis_health(RedisEphemeralTokens.connection())
if not result.is_healthy:
results['REDIS_EPH_TOKENS'] = result
result = cls.redis_health(RedisSessions.connection())
if not result.is_healthy:
results['REDIS_SESSIONS'] = result
result = cls.redis_health(RedisTTL.connection())
if not result.is_healthy:
results['REDIS_TTL'] = result
result = cls.redis_health(RedisToStream.connection())
if not result.is_healthy:
results['REDIS_TO_STREAM'] = result
result = cls.redis_health(RedisJobContainers.connection())
if not result.is_healthy:
results['REDIS_CONTAINERS'] = result
if not results:
results = {'REDIS': Result()}
return results
def pg_health():
try:
with connection.cursor() as cursor:
cursor.execute('SELECT 1; -- Healthcheck')
health = cursor.fetchone()[0] == 1
if health:
cursor.execute(
"select pg_database_size('postgres') as size")
size = cursor.fetchone()
return Result(
message='Service is healthy, db size {}'.format(size))
return Result(message='Service is not working.',
severity=Result.WARNING)
except Exception as e:
return Result(
message='Service unable to connect, encountered "{}" error.'.
format(e),
severity=Result.ERROR)
def run(cls) -> Dict:
try:
cache.set('health_check', 'test', 1)
if cache.get('health_check') != 'test':
result = Result(message='Cache key does not match.',
severity=Result.ERROR)
else:
result = Result()
except CacheKeyWarning:
result = Result(message='Cache key warning.',
severity=Result.ERROR)
except ValueError:
result = Result(message='Cache raised a ValueError.',
severity=Result.ERROR)
except ConnectionError:
result = Result(message='Cache raised a ConnectionError.',
severity=Result.ERROR)
return {'CACHE': result}
def run(cls):
try:
memory = psutil.virtual_memory()
available = '{:n}'.format(int(memory.available / 1024 / 1024))
if memory.available < (cls.ERROR_THRESHOLD * 1024 * 1024):
result = Result(
message='Memory available ({}) is below error threshold ({})'
.format(available, cls.ERROR_THRESHOLD),
severity=Result.ERROR)
elif memory.available < (cls.ERROR_THRESHOLD * 1024 * 1024):
result = Result(
message=
'Memory available ({}) is below warning threshold ({})'.
format(available, cls.WARNING_THRESHOLD),
severity=Result.WARNING)
else:
result = Result()
except ValueError:
result = Result(message='An error was raised, try again later.',
severity=Result.ERROR)
return {'MEMORY': result}
