def start_telemetry_callback(self) -> None:
"""
Run telemetry callback
:return:
"""
self.telemetry_callback = PeriodicCallback(self.send_telemetry, 250)
self.telemetry_callback.start()
async def create_session(self):
"""
Create consul session
:return:
"""
self.logger.info("Creating session")
checks = ["serfHealth"]
while True:
try:
self.session = await self.consul.session.create(
name=self.name,
checks=checks,
behavior="delete",
lock_delay=self.DEFAULT_CONSUL_LOCK_DELAY,
ttl=self.DEFAULT_CONSUL_SESSION_TTL,
)
break
except ConsulRepeatableErrors:
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
continue
self.logger.info("Session id: %s", self.session)
self.keep_alive_task = PeriodicCallback(
self.keep_alive, self.DEFAULT_CONSUL_SESSION_TTL * 1000 / 2
)
self.keep_alive_task.start()
async def start(self):
"""
Start all pending tasks
:return:
"""
self.thread_id = threading.get_ident()
self.resolver_expiration_task = PeriodicCallback(self.expire_resolvers, 10000)
self.resolver_expiration_task.start()
async def on_activate(self):
report_callback = PeriodicCallback(self.report, 10000)
report_callback.start()
check_callback = PeriodicCallback(self.check_channels, config.chwriter.batch_delay_ms)
check_callback.start()
self.logger.info("Sending records to %s", self.ch_address)
asyncio.create_task(self.subscribe_ch_streams())
asyncio.create_task(self.flush_data())
async def on_activate(self):
"""
Load rules from database after loading config
"""
self.logger.info("Using rule lookup solution: %s",
config.classifier.lookup_handler)
self.ruleset.load()
self.patternset.load()
self.load_triggers()
self.load_link_action()
self.load_handlers()
# Heat up MIB cache
MIBData.preload()
self.slot_number, self.total_slots = await self.acquire_slot()
await self.subscribe_stream("events.%s" % config.pool,
self.slot_number, self.on_event)
report_callback = PeriodicCallback(self.report, 1000)
report_callback.start()
async def on_activate(self):
# Listen sockets
server = SyslogServer(service=self)
for addr, port in server.iter_listen(config.syslogcollector.listen):
self.logger.info("Starting syslog server at %s:%s", addr, port)
try:
server.listen(port, addr)
except OSError as e:
metrics["error", ("type", "socket_listen_error")] += 1
self.logger.error("Failed to start syslog server at %s:%s: %s",
addr, port, e)
server.start()
# Report invalid sources every 60 seconds
self.logger.info("Stating invalid sources reporting task")
self.report_invalid_callback = PeriodicCallback(
self.report_invalid_sources, 60000)
self.report_invalid_callback.start()
# Start tracking changes
asyncio.get_running_loop().create_task(self.get_object_mappings())
class BaseService(object):
"""
Basic service implementation.
* on_change_<var> - subscribed to changes of config variable <var>
"""
# Service name
name = None
# Leader lock name
# Only one active instace per leader lock can be active
# at given moment
# Config variables can be expanded as %(varname)s
leader_lock_name = None
# Leader group name
# Only one service in leader group can be running at a time
# Config variables can be expanded as %(varname)s
# @todo: Deprecated, must be removed
leader_group_name = None
# Pooled service are used to distribute load between services.
# Pool name set in NOC_POOL parameter or --pool option.
# May be used in conjunction with leader_group_name
# to allow only one instance of services per node or datacenter
pooled = False
# Format string to set process name
# config variables can be expanded as %(name)s
process_name = "noc-%(name).10s"
# Connect to MongoDB on activate
use_mongo = False
# Initialize gettext and process *language* configuration
use_translation = False
# Initialize jinja2 templating engine
use_jinja = False
# Collect and send spans
use_telemetry = False
# Register traefik backend if not None
traefik_backend = None
# Traefik frontend rule
# i.e. PathPrefix:/api/<name>
traefik_frontend_rule = None
# Require DCS health status to be considered healthy
# Usually means resolution error to required services
# temporary leads service to unhealthy state
require_dcs_health = True
LOG_FORMAT = config.log_format
LOG_LEVELS = {
"critical": logging.CRITICAL,
"error": logging.ERROR,
"warning": logging.WARNING,
"info": logging.INFO,
"debug": logging.DEBUG,
}
DEFAULT_SHARDING_KEY = "managed_object"
SHARDING_KEYS = {"span": "ctx"}
# Timeout to wait NSQ writer is to close
NSQ_WRITER_CLOSE_TRY_TIMEOUT = 0.25
# Times to try to close NSQ writer
NSQ_WRITER_CLOSE_RETRIES = 5
class RegistrationError(Exception):
pass
def __init__(self):
set_service(self)
sys.excepthook = excepthook
self.loop: Optional[asyncio.BaseEventLoop] = None
self.logger = None
self.service_id = str(uuid.uuid4())
self.executors = {}
self.start_time = perf_counter()
self.pid = os.getpid()
self.nsq_readers = {} # handler -> Reader
self.nsq_writer = None
self.startup_ts = None
self.telemetry_callback = None
self.dcs = None
# Effective address and port
self.address = None
self.port = None
self.is_active = False
# Can be initialized in subclasses
self.scheduler = None
# NSQ Topics
# name -> TopicQueue()
self.topic_queues: Dict[str, TopicQueue] = {}
self.topic_queue_lock = threading.Lock()
# Liftbridge publisher
self.publish_queue: Optional[LiftBridgeQueue] = None
self.publisher_start_lock = threading.Lock()
#
self.active_subscribers = 0
self.subscriber_shutdown_waiter: Optional[asyncio.Event] = None
# Metrics partitions
self.n_metrics_partitions = len(config.clickhouse.cluster_topology.split(","))
#
self.metrics_key_lock = threading.Lock()
self.metrics_key_seq: int = 0
def create_parser(self) -> argparse.ArgumentParser:
"""
Return argument parser
"""
return argparse.ArgumentParser()
def add_arguments(self, parser: argparse.ArgumentParser) -> None:
"""
Apply additional parser arguments
"""
parser.add_argument(
"--node", action="store", dest="node", default=config.node, help="NOC node name"
)
parser.add_argument(
"--loglevel",
action="store",
choices=list(self.LOG_LEVELS),
dest="loglevel",
default=config.loglevel,
help="Logging level",
)
parser.add_argument(
"--instance",
action="store",
dest="instance",
type=int,
default=config.instance,
help="Instance number",
)
parser.add_argument(
"--debug",
action="store_true",
dest="debug",
default=False,
help="Dump additional debugging info",
)
parser.add_argument(
"--dcs",
action="store",
dest="dcs",
default=DEFAULT_DCS,
help="Distributed Coordinated Storage URL",
)
if self.pooled:
parser.add_argument(
"--pool", action="store", dest="pool", default=config.pool, help="NOC pool name"
)
@classmethod
def die(cls, msg: str = "") -> NoReturn:
"""
Dump message to stdout and terminate process with error code
"""
sys.stdout.write(str(msg) + "\n")
sys.stdout.flush()
os._exit(1)
def setup_logging(self, loglevel=None):
"""
Create new or setup existing logger
"""
# @todo: Duplicates config.setup_logging
if not loglevel:
loglevel = config.loglevel
logger = logging.getLogger()
if len(logger.handlers):
# Logger is already initialized
fmt = ErrorFormatter(self.LOG_FORMAT, None)
for h in logging.root.handlers:
if isinstance(h, logging.StreamHandler):
h.stream = sys.stdout
h.setFormatter(fmt)
logging.root.setLevel(loglevel)
else:
# Initialize logger
logging.basicConfig(stream=sys.stdout, format=self.LOG_FORMAT, level=loglevel)
self.logger = logging.getLogger(self.name)
logging.captureWarnings(True)
def setup_test_logging(self):
self.logger = logging.getLogger(self.name)
def setup_translation(self):
from noc.core.translation import set_translation, ugettext
set_translation(self.name, config.language)
if self.use_jinja:
from jinja2.defaults import DEFAULT_NAMESPACE
if "_" not in DEFAULT_NAMESPACE:
DEFAULT_NAMESPACE["_"] = ugettext
def on_change_loglevel(self, old_value, new_value):
if new_value not in self.LOG_LEVELS:
self.logger.error("Invalid loglevel '%s'. Ignoring", new_value)
return
self.logger.warning("Changing loglevel to %s", new_value)
logging.getLogger().setLevel(self.LOG_LEVELS[new_value])
def log_separator(self, symbol="*", length=72):
"""
Log a separator string to visually split log
"""
self.logger.warning(symbol * length)
if config.features.forensic:
self.logger.warning("[noc.core.forensic] [=Process restarted]")
def setup_signal_handlers(self):
"""
Set up signal handlers
"""
signal.signal(signal.SIGTERM, self.on_SIGTERM)
signal.signal(signal.SIGHUP, self.on_SIGHUP)
def set_proc_title(self):
"""
Set process title
"""
v = {"name": self.name, "instance": config.instance or "", "pool": config.pool or ""}
title = self.process_name % v
self.logger.debug("Setting process title to: %s", title)
setproctitle.setproctitle(title)
def start(self):
"""
Run main server loop
"""
self.startup_ts = perf_counter()
parser = self.create_parser()
self.add_arguments(parser)
options = parser.parse_args(sys.argv[1:])
cmd_options = vars(options)
cmd_options.pop("args", ())
# Bootstrap logging with --loglevel
self.setup_logging(cmd_options["loglevel"])
self.log_separator()
# Setup timezone
try:
self.logger.info("Setting timezone to %s", config.timezone)
setup_timezone()
except ValueError as e:
self.die(str(e))
# Setup title
self.set_proc_title()
# Setup signal handlers
self.setup_signal_handlers()
self.on_start()
# Starting IOLoop
self.is_active = True
if self.pooled:
self.logger.warning("Running service %s (pool: %s)", self.name, config.pool)
else:
self.logger.warning("Running service %s", self.name)
try:
setup_asyncio()
self.loop = asyncio.get_event_loop()
# Initialize DCS
self.dcs = get_dcs(cmd_options["dcs"])
# Activate service
self.loop.create_task(self.activate())
self.logger.warning("Starting IOLoop")
self.loop.run_forever()
except KeyboardInterrupt:
self.logger.warning("Interrupted by Ctrl+C")
except self.RegistrationError:
self.logger.info("Registration failed")
except Exception:
error_report()
finally:
if self.loop:
self.loop.create_task(self.deactivate())
self.logger.warning("Service %s has been terminated", self.name)
def get_event_loop(self) -> asyncio.BaseEventLoop:
return self.loop
def on_start(self):
"""
Reload config
"""
if self.use_translation:
self.setup_translation()
def stop(self):
self.logger.warning("Stopping")
self.loop.create_task(self.deactivate())
def on_SIGHUP(self, signo, frame):
# self.logger.warning("SIGHUP caught, rereading config")
pass
def on_SIGTERM(self, signo, frame):
self.logger.warning("SIGTERM caught, Stopping")
self.stop()
def get_service_address(self) -> Tuple[str, int]:
"""
Returns an (address, port) for HTTP service listener
"""
if self.address and self.port:
return self.address, self.port
if config.listen:
addr, port = config.listen.split(":")
port_tracker = config.instance
else:
addr, port = "auto", 0
port_tracker = 0
if addr == "auto":
addr = config.node
self.logger.info("Autodetecting address: auto -> %s", addr)
addr = config.node
port = int(port) + port_tracker
return addr, port
async def init_api(self):
"""
Initialize API routers and handlers
:return:
"""
raise NotImplementedError
async def shutdown_api(self):
"""
Stop API services
:return:
"""
raise NotImplementedError
async def activate(self):
"""
Initialize services before run
"""
self.logger.warning("Activating service")
if self.use_mongo:
from noc.core.mongo.connection import connect
connect()
await self.init_api()
#
if self.use_telemetry:
self.start_telemetry_callback()
self.loop.create_task(self.on_register())
async def deactivate(self):
if not self.is_active:
return
self.is_active = False
self.logger.info("Deactivating")
# Shutdown API
await self.shutdown_api()
# Release registration
if self.dcs:
self.logger.info("Deregistration")
await self.dcs.deregister()
# Shutdown schedulers
if self.scheduler:
try:
self.logger.info("Shutting down scheduler")
await self.scheduler.shutdown()
except asyncio.TimeoutError:
self.logger.info("Timed out when shutting down scheduler")
# Shutdown subscriptions
await self.shutdown_subscriptions()
# Shutdown executors
await self.shutdown_executors()
# Custom deactivation
await self.on_deactivate()
# Shutdown NSQ topics
await self.shutdown_topic_queues()
# Shutdown Liftbridge publisher
await self.shutdown_publisher()
# Continue deactivation
# Finally stop ioloop
self.dcs = None
self.logger.info("Stopping EventLoop")
self.loop.stop()
m = {}
apply_metrics(m)
apply_hists(m)
apply_quantiles(m)
self.logger.info("Post-mortem metrics: %s", m)
self.die("")
def get_register_tags(self):
tags = ["noc"]
if config.features.traefik:
if self.traefik_backend and self.traefik_frontend_rule:
tags += [
"traefik.tags=backend",
"traefik.backend=%s" % self.traefik_backend,
"traefik.frontend.rule=%s" % self.traefik_frontend_rule,
"traefik.backend.load-balancing=wrr",
]
weight = self.get_backend_weight()
if weight:
tags += ["traefik.backend.weight=%s" % weight]
limit = self.get_backend_limit()
if limit:
tags += ["traefik.backend.maxconn.amount=%s" % limit]
return tags
async def on_register(self):
addr, port = self.get_service_address()
r = await self.dcs.register(
self.name,
addr,
port,
pool=config.pool if self.pooled else None,
lock=self.get_leader_lock_name(),
tags=self.get_register_tags(),
)
if r:
# Finally call on_activate
await self.on_activate()
self.logger.info("Service is active (in %.2fms)", self.uptime() * 1000)
else:
raise self.RegistrationError()
async def on_activate(self):
"""
Called when service activated
"""
return
async def acquire_lock(self):
await self.dcs.acquire_lock("lock-%s" % self.name)
async def acquire_slot(self):
if self.pooled:
name = "%s-%s" % (self.name, config.pool)
else:
name = self.name
slot_number, total_slots = await self.dcs.acquire_slot(name, config.global_n_instances)
if total_slots <= 0:
self.die("Service misconfiguration detected: Invalid total_slots")
return slot_number, total_slots
async def on_deactivate(self):
return
def open_rpc(self, name, pool=None, sync=False, hints=None):
"""
Returns RPC proxy object.
"""
if pool:
svc = "%s-%s" % (name, pool)
else:
svc = name
return RPCProxy(self, svc, sync=sync, hints=hints)
def get_mon_status(self):
return True
def get_mon_data(self):
"""
Returns monitoring data
"""
r = {
"status": self.get_mon_status(),
"service": self.name,
"instance": str(self.service_id),
"node": config.node,
"pid": self.pid,
# Current process uptime
"uptime": perf_counter() - self.start_time,
}
if self.pooled:
r["pool"] = config.pool
if self.executors:
for x in self.executors:
self.executors[x].apply_metrics(r)
apply_metrics(r)
for topic in self.topic_queues:
self.topic_queues[topic].apply_metrics(r)
if self.publish_queue:
self.publish_queue.apply_metrics(r)
apply_hists(r)
apply_quantiles(r)
return r
def iter_rpc_retry_timeout(self):
"""
Yield timeout to wait after unsuccessful RPC connection
"""
for t in config.rpc.retry_timeout.split(","):
yield float(t)
async def subscribe_stream(
self,
stream: str,
partition: int,
handler: Callable[
[Message],
Awaitable[None],
],
start_timestamp: Optional[float] = None,
start_position: StartPosition = StartPosition.RESUME,
cursor_id: Optional[str] = None,
auto_set_cursor: bool = True,
) -> None:
# @todo: Restart on failure
self.logger.info("Subscribing %s:%s", stream, partition)
cursor_id = cursor_id or self.name
try:
async with LiftBridgeClient() as client:
self.active_subscribers += 1
async for msg in client.subscribe(
stream=stream,
partition=partition,
start_position=start_position,
cursor_id=self.name,
start_timestamp=start_timestamp,
):
try:
await handler(msg)
except Exception as e:
self.logger.error("Failed to process message: %s", e)
if auto_set_cursor and cursor_id:
await client.set_cursor(
stream=stream,
partition=partition,
cursor_id=cursor_id,
offset=msg.offset,
)
if self.subscriber_shutdown_waiter:
break
finally:
self.active_subscribers -= 1
if self.subscriber_shutdown_waiter and not self.active_subscribers:
self.subscriber_shutdown_waiter.set()
async def subscribe(self, topic, channel, handler, raw=False, **kwargs):
"""
Subscribe message to channel
"""
def call_json_handler(message):
metrics[metric_in] += 1
try:
data = orjson.loads(message.body)
except ValueError as e:
metrics[metric_decode_fail] += 1
self.logger.debug("Cannot decode JSON message: %s", e)
return True # Broken message
if isinstance(data, dict):
with ErrorReport():
r = handler(message, **data)
else:
with ErrorReport():
r = handler(message, data)
if r:
metrics[metric_processed] += 1
elif message.is_async():
message.on("finish", on_finish)
else:
metrics[metric_deferred] += 1
return r
def call_raw_handler(message):
metrics[metric_in] += 1
with ErrorReport():
r = handler(message, message.body)
if r:
metrics[metric_processed] += 1
elif message.is_async():
message.on("finish", on_finish)
else:
metrics[metric_deferred] += 1
return r
def on_finish(*args, **kwargs):
metrics[metric_processed] += 1
t = topic.replace(".", "_")
metric_in = "nsq_msg_in_%s" % t
metric_decode_fail = "nsq_msg_decode_fail_%s" % t
metric_processed = "nsq_msg_processed_%s" % t
metric_deferred = "nsq_msg_deferred_%s" % t
lookupd = [str(a) for a in config.nsqlookupd.http_addresses]
self.logger.info("Subscribing to %s/%s (lookupd: %s)", topic, channel, ", ".join(lookupd))
self.nsq_readers[handler] = NSQReader(
message_handler=call_raw_handler if raw else call_json_handler,
topic=topic,
channel=channel,
lookupd_http_addresses=lookupd,
snappy=config.nsqd.compression == "snappy",
deflate=config.nsqd.compression == "deflate",
deflate_level=config.nsqd.compression_level
if config.nsqd.compression == "deflate"
else 6,
**kwargs,
)
def suspend_subscription(self, handler):
"""
Suspend subscription for handler
:param handler:
:return:
"""
self.logger.info("Suspending subscription for handler %s", handler)
self.nsq_readers[handler].set_max_in_flight(0)
def resume_subscription(self, handler):
"""
Resume subscription for handler
:param handler:
:return:
"""
self.logger.info("Resuming subscription for handler %s", handler)
self.nsq_readers[handler].set_max_in_flight(config.nsqd.max_in_flight)
def _init_publisher(self):
"""
Spin-up publisher and queue
:return:
"""
with self.publisher_start_lock:
if self.publish_queue:
return # Created in concurrent thread
self.publish_queue = LiftBridgeQueue(self.loop)
self.loop.create_task(self.publisher())
def publish(
self,
value: bytes,
stream: str,
partition: Optional[int] = None,
key: Optional[bytes] = None,
headers: Optional[Dict[str, bytes]] = None,
):
"""
Schedule publish request
:param value:
:param stream:
:param partition:
:param key:
:param headers:
:return:
"""
if not self.publish_queue:
self._init_publisher()
req = LiftBridgeClient.get_publish_request(
value=value,
stream=stream,
partition=partition,
key=key,
headers=headers,
auto_compress=bool(config.liftbridge.compression_method),
)
self.publish_queue.put(req)
async def publisher_guard(self):
while not self.publish_queue.to_shutdown:
try:
await self.publisher()
except Exception as e:
self.logger.error("Unhandled exception in liftbridge publisher: %s", e)
async def publisher(self):
async with LiftBridgeClient() as client:
while not self.publish_queue.to_shutdown:
req = await self.publish_queue.get(timeout=1)
if not req:
continue # Timeout or shutdown
try:
await client.publish_sync(req, wait_for_stream=True)
except LiftbridgeError as e:
self.logger.error("Failed to publish message: %s", e)
self.logger.error("Retry message")
await asyncio.sleep(1)
self.publish_queue.put(req, fifo=False)
def get_topic_queue(self, topic: str) -> TopicQueue:
q = self.topic_queues.get(topic)
if q:
return q
# Create when necessary
with self.topic_queue_lock:
q = self.topic_queues.get(topic)
if q:
return q # Created in concurrent task
q = TopicQueue(topic)
self.topic_queues[topic] = q
self.loop.create_task(self.nsq_publisher_guard(q))
return q
async def nsq_publisher_guard(self, queue: TopicQueue):
while not queue.to_shutdown:
try:
await self.nsq_publisher(queue)
except Exception as e:
self.logger.error("Unhandled exception in NSQ publisher, restarting: %s", e)
async def nsq_publisher(self, queue: TopicQueue):
"""
Publisher for NSQ topic
:return:
"""
topic = queue.topic
self.logger.info("[nsq|%s] Starting NSQ publisher", topic)
while not queue.to_shutdown or not queue.is_empty():
# Message throttling. Wait and allow to collect more messages
await queue.wait(timeout=10, rate=config.nsqd.topic_mpub_rate)
# Get next batch up to `mpub_messages` messages or up to `mpub_size` size
messages = list(
queue.iter_get(
n=config.nsqd.mpub_messages,
size=config.nsqd.mpub_size,
total_overhead=4,
message_overhead=4,
)
)
if not messages:
continue
try:
self.logger.debug("[nsq|%s] Publishing %d messages", topic, len(messages))
await mpub(topic, messages, dcs=self.dcs)
except NSQPubError:
if queue.to_shutdown:
self.logger.debug(
"[nsq|%s] Failed to publish during shutdown. Dropping messages", topic
)
else:
# Return to queue
self.logger.info(
"[nsq|%s] Failed to publish. %d messages returned to queue",
topic,
len(messages),
)
queue.return_messages(messages)
del messages # Release memory
self.logger.info("[nsq|%s] Stopping NSQ publisher", topic)
# Queue is shut down and empty, notify
queue.notify_shutdown()
async def shutdown_executors(self):
if self.executors:
self.logger.info("Shutting down executors")
for x in self.executors:
try:
self.logger.info("Shutting down %s", x)
await self.executors[x].shutdown()
except asyncio.TimeoutError:
self.logger.info("Timed out when shutting down %s", x)
async def shutdown_subscriptions(self):
self.logger.info("Shutting down subscriptions")
self.subscriber_shutdown_waiter = asyncio.Event()
try:
await asyncio.wait_for(self.subscriber_shutdown_waiter.wait(), 10)
except asyncio.TimeoutError:
self.logger.info(
"Timed out when shutting down subscriptions. Some message may be still processing"
)
async def shutdown_publisher(self):
if self.publish_queue:
r = await self.publish_queue.drain(5.0)
if not r:
self.logger.info(
"Unclean shutdown of liftbridge queue. Up to %d messages may be lost",
self.publish_queue.qsize(),
)
self.publish_queue.shutdown()
async def shutdown_topic_queues(self):
# Issue shutdown
with self.topic_queue_lock:
has_topics = bool(self.topic_queues)
if has_topics:
self.logger.info("Shutting down topic queues")
for topic in self.topic_queues:
self.topic_queues[topic].shutdown()
# Wait for shutdown
while has_topics:
with self.topic_queue_lock:
topic = next(iter(self.topic_queues.keys()))
queue = self.topic_queues[topic]
del self.topic_queues[topic]
has_topics = bool(self.topic_queues)
try:
self.logger.info("Waiting shutdown of topic queue %s", topic)
await queue.wait_for_shutdown(5.0)
except asyncio.TimeoutError:
self.logger.info("Failed to shutdown topic queue %s: Timed out", topic)
def pub(self, topic, data, raw=False):
"""
Publish message to topic
:param topic: Topic name
:param data: Message to send. Message will be automatically
converted to JSON if *raw* is False, or passed as-is
otherwise
:param raw: True - pass message as-is, False - convert to JSON
"""
q = self.get_topic_queue(topic)
if raw:
q.put(data)
else:
for chunk in q.iter_encode_chunks(data):
q.put(chunk)
def mpub(self, topic, messages):
"""
Publish multiple messages to topic
"""
q = self.get_topic_queue(topic)
for m in messages:
for chunk in q.iter_encode_chunks(m):
q.put(chunk)
def get_executor(self, name: str) -> ThreadPoolExecutor:
"""
Return or start named executor
"""
executor = self.executors.get(name)
if not executor:
xt = "%s.%s_threads" % (self.name, name)
max_threads = config.get_parameter(xt)
self.logger.info(
"Starting threadpool executor %s (up to %d threads)", name, max_threads
)
executor = ThreadPoolExecutor(max_threads, name=name)
self.executors[name] = executor
return executor
def run_in_executor(
self, name: str, fn: Callable[[Any], T], *args: Any, **kwargs: Any
) -> asyncio.Future:
executor = self.get_executor(name)
return executor.submit(fn, *args, **kwargs)
@staticmethod
def _iter_metrics_raw_chunks(metrics: List[Dict[str, Any]]) -> Iterable[bytes]:
r: List[bytes] = []
size = 0
for mi in metrics:
jm = orjson.dumps(mi)
ljm = len(jm)
if size + ljm + 1 >= config.liftbridge.max_message_size:
yield b"\n".join(r)
r = []
size = 0
r.append(jm)
if size:
size += 1 + ljm
else:
size += ljm
if r:
yield b"\n".join(r)
def register_metrics(
self, table: str, metrics: List[Dict[str, Any]], key: Optional[int] = None
):
"""
Send collected metrics to `table`
Register metrics to send in non-clustered configuration.
Must be used via register_metrics only
:param fields: Table name
:param metrics: List of dicts containing metrics records
:param key: Sharding key, None for round-robin distribution
:return:
"""
if key is None:
with self.metrics_key_lock:
key = self.metrics_key_seq
self.metrics_key_seq += 1
for chunk in self._iter_metrics_raw_chunks(metrics):
self.publish(
chunk,
stream=f"ch.{table}",
partition=key % self.n_metrics_partitions,
)
def start_telemetry_callback(self) -> None:
"""
Run telemetry callback
:return:
"""
self.telemetry_callback = PeriodicCallback(self.send_telemetry, 250)
self.telemetry_callback.start()
async def send_telemetry(self):
"""
Publish telemetry data
:return:
"""
spans = get_spans()
if spans:
self.register_metrics("span", [span_to_dict(s) for s in spans])
def get_leader_lock_name(self):
if self.leader_lock_name:
return self.leader_lock_name % {"pool": config.pool}
return None
def get_backend_weight(self):
"""
Return backend weight for weighted load balancers
(i.e. traefik).
Return None for default weight
:return:
"""
return None
def get_backend_limit(self):
"""
Return backend connection limit for load balancers
(i.e. traefik)
Return None for no limits
:return:
"""
return None
def is_valid_health_check(self, service_id):
"""
Check received service id matches own service id
:param service_id:
:return:
"""
return not (
self.dcs
and self.dcs.health_check_service_id
and self.dcs.health_check_service_id != service_id
)
def get_health_status(self):
"""
Check service health to be reported to service registry
:return: (http code, message)
"""
if self.dcs and self.require_dcs_health:
# DCS is initialized
return self.dcs.get_status()
return 200, "OK"
def uptime(self):
if not self.startup_ts:
return 0
return perf_counter() - self.startup_ts
async def get_stream_partitions(self, stream: str) -> int:
"""
:param stream:
:return:
"""
async with LiftBridgeClient() as client:
while True:
meta = await client.fetch_metadata()
if meta.metadata:
for stream_meta in meta.metadata:
if stream_meta.name == stream:
if stream_meta.partitions:
return len(stream_meta.partitions)
break
# Cluster election in progress or cluster is misconfigured
self.logger.info("Stream '%s' has no active partitions. Waiting" % stream)
await asyncio.sleep(1)
class ConsulDCS(DCSBase):
"""
Consul-based DCS
URL format:
consul://<address>[:<port>]/<kv root>?token=<token>&check_interval=<...>&check_timeout=<...>&release_after=<...>
"""
DEFAULT_CONSUL_HOST = config.consul.host
DEFAULT_CONSUL_PORT = config.consul.port
DEFAULT_CONSUL_CHECK_INTERVAL = config.consul.check_interval
DEFAULT_CONSUL_CHECK_TIMEOUT = config.consul.connect_timeout
DEFAULT_CONSUL_RELEASE = "".join([str(config.consul.release), "s"])
DEFAULT_CONSUL_SESSION_TTL = config.consul.session_ttl
DEFAULT_CONSUL_LOCK_DELAY = config.consul.lock_delay
DEFAULT_CONSUL_RETRY_TIMEOUT = config.consul.retry_timeout
DEFAULT_CONSUL_KEEPALIVE_ATTEMPTS = config.consul.keepalive_attempts
EMPTY_HOLDER = ""
resolver_cls = ConsulResolver
def __init__(self, runner, url):
self.name = None
self.consul_host = self.DEFAULT_CONSUL_HOST
self.consul_port = self.DEFAULT_CONSUL_PORT
self.consul_prefix = "/"
self.consul_token = config.consul.token
self.check_interval = self.DEFAULT_CONSUL_CHECK_INTERVAL
self.check_timeout = self.DEFAULT_CONSUL_CHECK_TIMEOUT
self.release_after = self.DEFAULT_CONSUL_RELEASE
self.keepalive_attempts = self.DEFAULT_CONSUL_KEEPALIVE_ATTEMPTS
self.svc_name = None
self.svc_address = None
self.svc_port = None
self.svc_check_url = None
self.svc_id = None
self.session = None
self.slot_number = None
self.total_slots = None
super().__init__(runner, url)
self.consul = ConsulClient(host=self.consul_host,
port=self.consul_port,
token=self.consul_token)
self.session = None
self.keep_alive_task = None
self.service_watchers = {}
self.in_keep_alive = False
def parse_url(self, u):
if ":" in u.netloc:
self.consul_host, port = u.netloc.rsplit(":", 1)
self.consul_port = int(port)
else:
self.consul_host = u.netloc
self.consul_prefix = u.path[1:]
if self.consul_prefix.endswith("/"):
self.consul_prefix = self.consul_prefix[:-1]
for q in u.query.split("&"):
if not q or "=" not in q:
continue
k, v = q.split("=", 1)
v = unquote(v)
if k == "token":
self.consul_token = v
elif k == "check_interval":
self.check_interval = int(v)
elif k == "check_timeout":
self.check_timeout = int(v)
elif k == "release_after":
self.release_after = v
async def create_session(self):
"""
Create consul session
:return:
"""
self.logger.info("Creating session")
checks = ["serfHealth"]
while True:
try:
self.session = await self.consul.session.create(
name=self.name,
checks=checks,
behavior="delete",
lock_delay=self.DEFAULT_CONSUL_LOCK_DELAY,
ttl=self.DEFAULT_CONSUL_SESSION_TTL,
)
break
except ConsulRepeatableErrors:
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
continue
self.logger.info("Session id: %s", self.session)
self.keep_alive_task = PeriodicCallback(
self.keep_alive, self.DEFAULT_CONSUL_SESSION_TTL * 1000 / 2)
self.keep_alive_task.start()
async def destroy_session(self):
if self.session:
self.logger.info("Destroying session %s", self.session)
if self.keep_alive_task:
self.keep_alive_task.stop()
self.keep_alive_task = None
try:
await self.consul.session.destroy(self.session)
except ConsulRepeatableErrors:
pass # Ignore consul errors
self.session = None
async def register(self,
name,
address,
port,
pool=None,
lock=None,
tags=None):
if pool:
name = "%s-%s" % (name, pool)
self.name = name
if lock:
await self.acquire_lock(lock)
svc_id = self.session or str("svc-%s" % uuid.uuid4())
tags = tags[:] if tags else []
tags += [svc_id]
self.svc_check_url = "http://%s:%s/health/?service=%s" % (address,
port, svc_id)
self.health_check_service_id = svc_id
if config.features.consul_healthchecks:
checks = consul.Check.http(self.svc_check_url, self.check_interval,
"%ds" % self.check_timeout)
checks["DeregisterCriticalServiceAfter"] = self.release_after
else:
checks = []
if config.features.service_registration:
while True:
self.logger.info("Registering service %s: %s:%s (id=%s)", name,
address, port, svc_id)
try:
r = await self.consul.agent.service.register(
name=name,
service_id=svc_id,
address=address,
port=port,
tags=tags,
check=checks,
)
except ConsulRepeatableErrors as e:
metrics["error", ("type", "cant_register_consul")] += 1
self.logger.info("Cannot register service %s: %s", name, e)
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
continue
if r:
self.svc_id = svc_id
break
return r
else:
return True
async def deregister(self):
if self.session:
try:
await self.destroy_session()
except ConsulRepeatableErrors:
metrics["error",
("type", "cant_destroy_consul_session_soft")] += 1
except Exception as e:
metrics["error", ("type", "cant_destroy_consul_session")] += 1
self.logger.error("Cannot destroy session: %s", e)
if self.svc_id and config.features.service_registration:
try:
await self.consul.agent.service.deregister(self.svc_id)
except ConsulRepeatableErrors:
metrics["error", ("type", "cant_deregister_consul_soft")] += 1
except Exception as e:
metrics["error", ("type", "cant_deregister_consul")] += 1
self.logger.error("Cannot deregister service: %s", e)
self.svc_id = None
async def keep_alive(self):
metrics["dcs_consul_keepalives"] += 1
if self.in_keep_alive:
metrics["error", ("type", "dcs_consul_overlapped_keepalives")] += 1
return
self.in_keep_alive = True
try:
if self.session:
touched = False
for n in range(self.keepalive_attempts):
try:
await self.consul.session.renew(self.session)
self.logger.debug("Session renewed")
touched = True
break
except consul.base.NotFound as e:
self.logger.warning(
"Session lost by: '%s'. Forcing quit", e)
break
except ConsulRepeatableErrors as e:
self.logger.warning(
"Cannot refresh session due to ignorable error: %s",
e)
metrics["error",
("type", "dcs_consul_keepalive_retries")] += 1
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
if not touched:
self.logger.critical("Cannot refresh session, stopping")
if self.keep_alive_task:
self.keep_alive_task.stop()
self.keep_alive_task = None
self.kill()
elif self.keep_alive_task:
self.keep_alive_task.stop()
self.keep_alive_task = None
finally:
self.in_keep_alive = False
def get_lock_path(self, lock):
return self.consul_prefix + "/locks/" + lock
async def acquire_lock(self, name):
if not self.session:
await self.create_session()
key = self.get_lock_path(name)
index = None
while True:
self.logger.info("Acquiring lock: %s", key)
try:
status = await self.consul.kv.put(key=key,
value=self.session,
acquire=self.session,
token=self.consul_token)
if status:
break
else:
metrics["error",
("type", "dcs_consul_failed_get_lock")] += 1
self.logger.info("Failed to acquire lock")
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
except ConsulRepeatableErrors:
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
continue
# Waiting for lock release
while True:
try:
index, data = await self.consul.kv.get(
key=key, index=index, token=self.consul_token)
if not data:
index = None # Key has been deleted
await asyncio.sleep(self.DEFAULT_CONSUL_LOCK_DELAY *
(0.5 + random.random()))
break
except ConsulRepeatableErrors:
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
self.logger.info("Lock acquired")
async def get_slot_limit(self, name) -> Optional[int]:
"""
Return the current limit for given slot
:param name:
:return:
"""
manifest_path = "%s/slots/%s/manifest" % (self.consul_prefix, name)
while True:
self.logger.info("Attempting to get slot")
# Non-blocking for a first time
# Block until change every next try
try:
_, cv = await self.consul.kv.get(key=manifest_path, index=0)
if not cv:
return 0
return int(orjson.loads(cv["Value"]).get("Limit", 0))
except ConsulRepeatableErrors:
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
continue
async def acquire_slot(self, name, limit):
"""
Acquire shard slot
:param name: <service name>-<pool>
:param limit: Configured limit
:return: (slot number, number of instances)
"""
if not self.session:
await self.create_session()
if self.total_slots is not None:
return self.slot_number, self.total_slots
prefix = "%s/slots/%s" % (self.consul_prefix, name)
contender_path = "%s/%s" % (prefix, self.session)
contender_info = self.session
manifest_path = "%s/manifest" % prefix
self.logger.info("Writing contender slot info into %s", contender_path)
while True:
try:
status = await self.consul.kv.put(
key=contender_path,
value=contender_info,
acquire=self.session,
token=self.consul_token,
)
if status:
break
else:
metrics["error",
("type", "dcs_consul_failed_get_slot")] += 1
self.logger.info("Failed to write contender slot info")
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
except ConsulRepeatableErrors:
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
index = 0
cas = 0
while True:
self.logger.info("Attempting to get slot")
seen_sessions = set()
dead_contenders = set()
manifest = None
# Non-blocking for a first time
# Block until change every next try
try:
index, cv = await self.consul.kv.get(key=prefix,
index=index,
recurse=True)
except ConsulRepeatableErrors:
await asyncio.sleep(self.DEFAULT_CONSUL_RETRY_TIMEOUT)
continue
for e in cv:
if e["Key"] == manifest_path:
cas = e["ModifyIndex"]
# @todo: Handle errors
manifest = orjson.loads(e["Value"])
else:
if "Session" in e:
seen_sessions.add(e["Session"])
else:
dead_contenders.add(e["Key"])
if manifest:
total_slots = int(manifest.get("Limit", 0))
holders = [
h if h in seen_sessions else self.EMPTY_HOLDER
for h in manifest.get("Holders", [])
]
else:
self.logger.info("Initializing manifest")
total_slots = limit
holders = []
# Try to allocate slot
if len(holders) < total_slots:
# Available slots from the end
slot_number = len(holders)
holders += [self.session]
else:
# Try to reclaim slots in the middle
try:
slot_number = holders.index(self.EMPTY_HOLDER)
holders[slot_number] = self.session
except ValueError:
self.logger.info("All slots a busy, waiting")
continue
# Update manifest
self.logger.info("Attempting to acquire slot %s/%s", slot_number,
total_slots)
try:
r = await self.consul.kv.put(
key=manifest_path,
value=smart_text(
orjson.dumps({
"Limit": total_slots,
"Holders": holders
},
option=orjson.OPT_INDENT_2)),
cas=cas,
)
except ConsulRepeatableErrors as e:
self.logger.info("Cannot acquire slot: %s", e)
continue
if r:
self.logger.info("Acquired slot %s/%s", slot_number,
total_slots)
self.slot_number = slot_number
self.total_slots = total_slots
return slot_number, total_slots
self.logger.info("Cannot acquire slot: CAS changed, retry")
async def resolve_near(self,
name,
hint=None,
wait=True,
timeout=None,
full_result=False,
critical=False):
"""
Synchronous call to resolve nearby service
Commonly used for external services like databases
:param name: Service name
:param wait:
:param timeout:
:param full_result:
:param hint:
:param critical:
:return: address:port
"""
self.logger.info("Resolve near service %s", name)
index = 0
while True:
try:
index, services = await self.consul.health.service(
service=name,
index=index,
near="_agent",
token=self.consul_token,
passing=True)
except ConsulRepeatableErrors as e:
metrics["error",
("type", "dcs_consul_failed_resolve_near")] += 1
self.logger.info("Consul error: %s", e)
if critical:
metrics["error",
("type",
"dcs_consul_failed_resolve_critical_near")] += 1
self.set_faulty_status("Consul error: %s" % e)
time.sleep(CONSUL_NEAR_RETRY_TIMEOUT)
continue
if not services and wait:
metrics["error",
("type",
"dcs_consul_no_active_service %s" % name)] += 1
self.logger.info("No active service %s. Waiting", name)
if critical:
metrics["error",
("type",
"dcs_consul_no_active_critical_service %s" %
name)] += 1
self.set_faulty_status("No active service %s. Waiting" %
name)
time.sleep(CONSUL_NEAR_RETRY_TIMEOUT)
continue
r = []
for svc in services:
r += [
"%s:%s" % (
str(svc["Service"]["Address"]
or svc["Node"]["Address"]),
str(svc["Service"]["Port"]),
)
]
if not full_result:
break
self.logger.info("Resolved near service %s to %s", name, r)
if critical:
self.clear_faulty_status()
return r
class DCSBase(object):
# Resolver class
resolver_cls = None
# HTTP code to be returned by /health endpoint when service is healthy
HEALTH_OK_HTTP_CODE = 200
# HTTP code to be returned by /health endpoint when service is unhealthy
# and must be temporary removed from resolver
HEALTH_FAILED_HTTP_CODE = 429
def __init__(self, runner, url):
self.runner = runner
self.logger = logging.getLogger(__name__)
self.url = url
self.parse_url(urlparse(url))
# service -> resolver instances
self.resolvers = {}
self.resolvers_lock = threading.Lock()
self.resolver_expiration_task = None
self.health_check_service_id = None
self.status = True
self.status_message = ""
self.thread_id = None
def parse_url(self, u):
pass
async def start(self):
"""
Start all pending tasks
:return:
"""
self.thread_id = threading.get_ident()
self.resolver_expiration_task = PeriodicCallback(self.expire_resolvers, 10000)
self.resolver_expiration_task.start()
def stop(self):
"""
Stop all pending tasks
:return:
"""
if self.resolver_expiration_task:
self.resolver_expiration_task.stop()
self.resolver_expiration_task = None
# Stop all resolvers
with self.resolvers_lock:
for svc in self.resolvers:
r = self.resolvers[svc]
self.logger.info("Stopping resolver for service %s", svc)
r.stop()
self.resolvers = {}
async def register(self, name, address, port, pool=None, lock=None, tags=None):
"""
Register service
:param name:
:param address:
:param port:
:param pool:
:param lock:
:param tags: List of extra tags
:return:
"""
raise NotImplementedError()
def kill(self):
self.logger.info("Shooting self with SIGTERM")
os.kill(os.getpid(), signal.SIGTERM)
async def get_slot_limit(self, name) -> Optional[int]:
"""
Return the current limit for given slot
:param name:
:return:
"""
raise NotImplementedError()
async def acquire_slot(self, name, limit):
"""
Acquire shard slot
:param name: <service name>-<pool>
:param limit: Configured limit
:return: (slot number, number of instances)
"""
raise NotImplementedError()
async def get_resolver(self, name, critical=False, near=False, track=True):
def run_resolver(res):
loop = asyncio.get_running_loop()
loop.call_soon(loop.create_task, res.start())
if track:
with self.resolvers_lock:
resolver = self.resolvers.get((name, critical, near))
if not resolver:
self.logger.info("Running resolver for service %s", name)
resolver = self.resolver_cls(self, name, critical=critical, near=near)
self.resolvers[name, critical, near] = resolver
run_resolver(resolver)
else:
# One-time resolver
resolver = self.resolver_cls(self, name, critical=critical, near=near, track=False)
run_resolver(resolver)
return resolver
async def resolve(
self,
name,
hint=None,
wait=True,
timeout=None,
full_result=False,
critical=False,
near=False,
track=True,
):
async def wrap():
resolver = await self.get_resolver(name, critical=critical, near=near, track=track)
r = await resolver.resolve(
hint=hint, wait=wait, timeout=timeout, full_result=full_result
)
return r
return await self.runner.trampoline(wrap())
async def expire_resolvers(self):
with self.resolvers_lock:
for svc in list(self.resolvers):
r = self.resolvers[svc]
if r.is_expired():
self.logger.info("Stopping expired resolver for service %s", svc)
r.stop()
del self.resolvers[svc]
def resolve_sync(self, name, hint=None, wait=True, timeout=None, full_result=False):
"""
Returns *hint* when service is active or new service
instance,
:param name:
:param hint:
:param full_result:
:return:
"""
async def _resolve():
r = await self.resolve(
name, hint=hint, wait=wait, timeout=timeout, full_result=full_result
)
return r
return run_sync(_resolve)
async def resolve_near(
self, name, hint=None, wait=True, timeout=None, full_result=False, critical=False
):
"""
Synchronous call to resolve nearby service
Commonly used for external services like databases
:param name: Service name
:return: address:port
"""
raise NotImplementedError()
def get_status(self):
if self.status:
return self.HEALTH_OK_HTTP_CODE, "OK"
else:
return self.HEALTH_FAILED_HTTP_CODE, self.status_message
def set_faulty_status(self, message):
if self.status or self.status_message != message:
self.logger.info("Set faulty status to: %s", message)
self.status = False
self.status_message = message
def clear_faulty_status(self):
if not self.status:
self.logger.info("Clearing faulty status")
self.status = True
self.status_message = ""
class SyslogCollectorService(TornadoService):
name = "syslogcollector"
pooled = True
process_name = "noc-%(name).10s-%(pool).5s"
def __init__(self):
super().__init__()
self.mappings_callback = None
self.report_invalid_callback = None
self.source_configs = {} # id -> SourceConfig
self.address_configs = {} # address -> SourceConfig
self.invalid_sources = defaultdict(int) # ip -> count
self.pool_partitions: Dict[str, int] = {}
async def on_activate(self):
# Listen sockets
server = SyslogServer(service=self)
for addr, port in server.iter_listen(config.syslogcollector.listen):
self.logger.info("Starting syslog server at %s:%s", addr, port)
try:
server.listen(port, addr)
except OSError as e:
metrics["error", ("type", "socket_listen_error")] += 1
self.logger.error("Failed to start syslog server at %s:%s: %s",
addr, port, e)
server.start()
# Report invalid sources every 60 seconds
self.logger.info("Stating invalid sources reporting task")
self.report_invalid_callback = PeriodicCallback(
self.report_invalid_sources, 60000)
self.report_invalid_callback.start()
# Start tracking changes
asyncio.get_running_loop().create_task(self.get_object_mappings())
async def get_pool_partitions(self, pool: str) -> int:
parts = self.pool_partitions.get(pool)
if not parts:
parts = await self.get_stream_partitions("events.%s" % pool)
self.pool_partitions[pool] = parts
return parts
def lookup_config(self, address: str) -> Optional[SourceConfig]:
"""
Returns object id for given address or None when
unknown source
"""
cfg = self.address_configs.get(address)
if cfg:
return cfg
# Register invalid event source
if self.address_configs:
self.invalid_sources[address] += 1
metrics["error", ("type", "object_not_found")] += 1
return None
def register_message(self, cfg: SourceConfig, timestamp: int, message: str,
facility: int, severity: int) -> None:
"""
Spool message to be sent
"""
if cfg.process_events:
# Send to classifier
metrics["events_out"] += 1
self.publish(
orjson.dumps({
"ts": timestamp,
"object": cfg.id,
"data": {
"source": "syslog",
"collector": config.pool,
"message": message
},
}),
stream=cfg.stream,
partition=cfg.partition,
)
if cfg.archive_events and cfg.bi_id:
# Archive message
metrics["events_archived"] += 1
now = datetime.datetime.now()
ts = now.strftime("%Y-%m-%d %H:%M:%S")
date = ts.split(" ")[0]
self.register_metrics(
"syslog",
[{
"date": date,
"ts": ts,
"managed_object": cfg.bi_id,
"facility": facility,
"severity": severity,
"message": message,
}],
)
async def get_object_mappings(self):
"""
Subscribe and track datastream changes
"""
# Register RPC aliases
client = SysologDataStreamClient("cfgsyslog", service=self)
# Track stream changes
while True:
self.logger.info("Starting to track object mappings")
try:
await client.query(
limit=config.syslogcollector.ds_limit,
filters=["pool(%s)" % config.pool],
block=1,
)
except NOCError as e:
self.logger.info("Failed to get object mappings: %s", e)
await asyncio.sleep(1)
async def report_invalid_sources(self):
"""
Report invalid event sources
"""
if not self.invalid_sources:
return
total = sum(self.invalid_sources[s] for s in self.invalid_sources)
self.logger.info(
"Dropping %d messages with invalid sources: %s",
total,
", ".join("%s: %s" % (s, self.invalid_sources[s])
for s in self.invalid_sources),
)
self.invalid_sources = defaultdict(int)
async def update_source(self, data):
# Get old config
old_cfg = self.source_configs.get(data["id"])
if old_cfg:
old_addresses = set(old_cfg.addresses)
else:
old_addresses = set()
# Get pool and sharding information
fm_pool = data.get("fm_pool", None) or config.pool
num_partitions = await self.get_pool_partitions(fm_pool)
# Build new config
cfg = SourceConfig(
id=data["id"],
addresses=tuple(data["addresses"]),
bi_id=data.get("bi_id"), # For backward compatibility
process_events=data.get("process_events",
True), # For backward compatibility
archive_events=data.get("archive_events", False),
stream="events.%s" % fm_pool,
partition=int(data["id"]) % num_partitions,
)
new_addresses = set(cfg.addresses)
# Add new addresses, update remaining
for addr in new_addresses:
self.address_configs[addr] = cfg
# Revoke stale addresses
for addr in old_addresses - new_addresses:
del self.address_configs[addr]
# Update configs
self.source_configs[data["id"]] = cfg
# Update metrics
metrics["sources_changed"] += 1
async def delete_source(self, id):
cfg = self.source_configs.get(id)
if not cfg:
return
for addr in cfg.addresses:
del self.address_configs[addr]
del self.source_configs[id]
metrics["sources_deleted"] += 1
class TrapCollectorService(TornadoService):
name = "trapcollector"
leader_group_name = "trapcollector-%(dc)s-%(node)s"
pooled = True
process_name = "noc-%(name).10s-%(pool).5s"
def __init__(self):
super().__init__()
self.mappings_callback = None
self.report_invalid_callback = None
self.source_configs: Dict[str, SourceConfig] = {} # id -> SourceConfig
self.address_configs = {} # address -> SourceConfig
self.invalid_sources = defaultdict(int) # ip -> count
self.pool_partitions: Dict[str, int] = {}
async def on_activate(self):
# Listen sockets
server = TrapServer(service=self)
for addr, port in server.iter_listen(config.trapcollector.listen):
self.logger.info("Starting SNMP Trap server at %s:%s", addr, port)
try:
server.listen(port, addr)
except OSError as e:
metrics["error", ("type", "socket_listen_error")] += 1
self.logger.error(
"Failed to start SNMP Trap server at %s:%s: %s", addr,
port, e)
server.start()
# Report invalid sources every 60 seconds
self.logger.info("Stating invalid sources reporting task")
self.report_invalid_callback = PeriodicCallback(
self.report_invalid_sources, 60000)
self.report_invalid_callback.start()
# Start tracking changes
asyncio.get_running_loop().create_task(self.get_object_mappings())
async def get_pool_partitions(self, pool: str) -> int:
parts = self.pool_partitions.get(pool)
if not parts:
parts = await self.get_stream_partitions("events.%s" % pool)
self.pool_partitions[pool] = parts
return parts
def lookup_config(self, address: str) -> Optional[SourceConfig]:
"""
Returns object config for given address or None when
unknown source
"""
cfg = self.address_configs.get(address)
if cfg:
return cfg
# Register invalid event source
if self.address_configs:
self.invalid_sources[address] += 1
metrics["error", ("type", "object_not_found")] += 1
return None
def register_message(self, cfg: SourceConfig, timestamp: int,
data: Dict[str, Any]):
"""
Spool message to be sent
"""
metrics["events_out"] += 1
self.publish(
orjson.dumps({
"ts": timestamp,
"object": cfg.id,
"data": data,
}),
stream=cfg.stream,
partition=cfg.partition,
)
async def get_object_mappings(self):
"""
Coroutine to request object mappings
"""
self.logger.info("Starting to track object mappings")
client = TrapDataStreamClient("cfgtrap", service=self)
# Track stream changes
while True:
try:
await client.query(limit=config.trapcollector.ds_limit,
filters=["pool(%s)" % config.pool],
block=1)
except NOCError as e:
self.logger.info("Failed to get object mappings: %s", e)
await asyncio.sleep(1)
async def report_invalid_sources(self):
"""
Report invalid event sources
"""
if not self.invalid_sources:
return
total = sum(self.invalid_sources[s] for s in self.invalid_sources)
self.logger.info(
"Dropping %d messages with invalid sources: %s",
total,
", ".join("%s: %s" % (s, self.invalid_sources[s])
for s in self.invalid_sources),
)
self.invalid_sources = defaultdict(int)
async def update_source(self, data):
# Get old config
old_cfg = self.source_configs.get(data["id"])
if old_cfg:
old_addresses = set(old_cfg.addresses)
else:
old_addresses = set()
# Get pool and sharding information
fm_pool = data.get("fm_pool", None) or config.pool
num_partitions = await self.get_pool_partitions(fm_pool)
# Build new config
cfg = SourceConfig(
id=data["id"],
addresses=tuple(data["addresses"]),
stream="events.%s" % fm_pool,
partition=int(data["id"]) % num_partitions,
)
new_addresses = set(cfg.addresses)
# Add new addresses, update remaining
for addr in new_addresses:
self.address_configs[addr] = cfg
# Revoke stale addresses
for addr in old_addresses - new_addresses:
del self.address_configs[addr]
# Update configs
self.source_configs[data["id"]] = cfg
# Update metrics
metrics["sources_changed"] += 1
async def delete_source(self, id):
cfg = self.source_configs.get(id)
if not cfg:
return
for addr in cfg.addresses:
del self.address_configs[addr]
del self.source_configs[id]
metrics["sources_deleted"] += 1
