class MetricsService(Service):
"""
A service to provide a registry where metrics instruments can be registered and retrieved from.
It continuously reports metrics to the specified InfluxDB instance.
"""
def __init__(self,
influx_server: str,
influx_user: str,
influx_password: str,
influx_database: str,
host: str,
reporting_frequency: int = 10):
self._influx_server = influx_server
self._reporting_frequency = reporting_frequency
self._registry = HostMetricsRegistry(host)
self._reporter = InfluxReporter(registry=self._registry,
protocol='https',
port=443,
database=influx_database,
username=influx_user,
password=influx_password,
server=influx_server)
logger = get_extended_debug_logger(
'trinity.components.builtin.metrics.MetricsService')
@property
def registry(self) -> HostMetricsRegistry:
"""
Return the :class:`trinity.components.builtin.metrics.registry.HostMetricsRegistry` at which
metrics instruments can be registered and retrieved.
"""
return self._registry
async def run(self) -> None:
self.logger.info("Reporting metrics to %s", self._influx_server)
self.manager.run_daemon_task(self._continuously_report)
await self.manager.wait_finished()
async def _continuously_report(self) -> None:
async for _ in trio_utils.every(self._reporting_frequency):
self._reporter.report_now()
class BaseMetricsService(Service, MetricsServiceAPI):
"""
A service to provide a registry where metrics instruments can be registered and retrieved from.
It continuously reports metrics to the specified InfluxDB instance.
"""
MIN_SECONDS_BETWEEN_ERROR_LOGS = 60
def __init__(self, influx_server: str, influx_user: str,
influx_password: str, influx_database: str, host: str,
port: int, protocol: str, reporting_frequency: int):
self._unreported_error: Exception = None
self._last_time_reported: float = 0.0
self._influx_server = influx_server
self._reporting_frequency = reporting_frequency
self._registry = HostMetricsRegistry(host)
self._reporter = InfluxReporter(registry=self._registry,
database=influx_database,
username=influx_user,
password=influx_password,
protocol=protocol,
port=port,
server=influx_server)
logger = get_logger('trinity.components.builtin.metrics.MetricsService')
@property
def registry(self) -> HostMetricsRegistry:
"""
Return the :class:`trinity.components.builtin.metrics.registry.HostMetricsRegistry` at which
metrics instruments can be registered and retrieved.
"""
return self._registry
async def run(self) -> None:
self.logger.info("Reporting metrics to %s", self._influx_server)
self.manager.run_daemon_task(self.continuously_report)
await self.manager.wait_finished()
def report_now(self) -> None:
try:
self._reporter.report_now()
except (HTTPException, ConnectionError) as exc:
# This method is usually called every few seconds. If there's an issue with the
# connection we do not want to flood the log and tame down warnings.
# 1. We log the first instance of an exception immediately
# 2. We log follow up exceptions only after a minimum time has elapsed
# This means that we also might overwrite exceptions for different errors
if self._is_justified_to_log_error():
self._log_and_clear(exc)
else:
self._unreported_error = exc
else:
# If errors disappear, we want to make sure we eventually report the last instance
if self._unreported_error is not None and self._is_justified_to_log_error(
):
self._log_and_clear(self._unreported_error)
def _log_and_clear(self, error: Exception) -> None:
self.logger.warning("Unable to report metrics: %s", error)
self._unreported_error = None
self._last_time_reported = time.monotonic()
def _is_justified_to_log_error(self) -> bool:
return (self._last_time_reported == 0.0
or time.monotonic() - self._last_time_reported >
self.MIN_SECONDS_BETWEEN_ERROR_LOGS)
@abstractmethod
async def continuously_report(self) -> None:
...
class Metrics(Service):
logger = logging.getLogger('alexandria.metrics.Metrics')
def __init__(self,
host: str,
client: ClientAPI,
kademlia: KademliaAPI,
influx_server: str,
influx_user: str,
influx_password: str,
influx_database: str,
influx_port: int = 443,
influx_protocol: str = 'https',
reporting_frequency: int = 10,
process_collection_frequency: int = 3):
self._influx_server = influx_server
self._reporting_frequency = reporting_frequency
self._process_collection_frequency = process_collection_frequency
self._registry = HostMetricsRegistry(host)
self._reporter = InfluxReporter(
registry=self._registry,
protocol=influx_protocol,
port=influx_port,
database=influx_database,
username=influx_user,
password=influx_password,
server=influx_server
)
self.client = client
self.kademlia = kademlia
@classmethod
def from_cli_args(cls,
args: Namespace,
client: ClientAPI,
kademlia: KademliaAPI,
) -> 'Metrics':
return cls(
host=args.metrics_host,
client=client,
kademlia=kademlia,
influx_server=args.metrics_influx_server,
influx_user=args.metrics_influx_user,
influx_password=args.metrics_influx_password,
influx_database=args.metrics_influx_database,
influx_port=args.metrics_influx_port,
influx_protocol=args.metrics_influx_protocol,
reporting_frequency=args.metrics_reporting_frequency,
)
async def run(self) -> None:
self.manager.run_daemon_task(
self._continuously_report,
self._reporting_frequency,
)
self.manager.run_daemon_task(
self._collect_system_metrics,
self._process_collection_frequency,
)
self.manager.run_daemon_task(
self._report_routing_table_stats,
10,
)
self.manager.run_daemon_task(
self._report_content_manager_stats,
10,
)
self.logger.info('Metrics started')
for payload_type in PAYLOAD_TYPES:
self.manager.run_daemon_task(self._report_inbound_message_stats, payload_type)
self.manager.run_daemon_task(self._report_event, self.client.events.session_created, 'events/session-created') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.session_idle, 'events/session-idle') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.handshake_complete, 'events/handshake-complete') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.handshake_timeout, 'events/handshake-timeout') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.datagram_received, 'datagram/inbound') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.datagram_sent, 'datagram/outbound') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_ping, 'messages/outbound/Ping') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_pong, 'messages/outbound/Pong') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_find_nodes, 'messages/outbound/FindNodes') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_found_nodes, 'messages/outbound/FoundNodes') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_advertise, 'messages/outbound/Advertise') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_ack, 'messages/outbound/Ack') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_locate, 'messages/outbound/Locate') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_locations, 'messages/outbound/Locations') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_retrieve, 'messages/outbound/Retrieve') # noqa: E501
self.manager.run_daemon_task(self._report_event, self.client.events.sent_chunk, 'messages/outbound/Chunk') # noqa: E501
await self.manager.wait_finished()
async def _continuously_report(self, frequency: int) -> None:
async for _ in every(frequency):
self._reporter.report_now()
async def _report_event(self, event: EventAPI[Any], suffix: str) -> None:
counter = self._registry.counter(f'alexandria.{suffix}.counter')
meter = self._registry.meter(f'alexandria.{suffix}.meter')
async with event.subscribe() as subscription:
async for _ in subscription:
counter.inc()
meter.mark()
async def _report_routing_table_stats(self, frequency: int) -> None:
size_gauge = self._registry.gauge('alexandria.dht/routing-table/total-nodes.gauge')
async for _ in every(frequency):
stats = self.kademlia.routing_table.get_stats()
size_gauge.set_value(stats.total_nodes)
async def _report_inbound_message_stats(self, payload_type: Type[ssz.Serializable]) -> None:
name = payload_type.__name__
counter = self._registry.counter(f'alexandria.messages/inbound/{name}.counter')
meter = self._registry.meter(f'alexandria.messages/inbound/{name}.meter')
async with self.client.message_dispatcher.subscribe(payload_type) as subscription:
async for payload in subscription:
counter.inc()
meter.mark()
async def _report_content_manager_stats(self, frequency: int) -> None:
gauge = self._registry.gauge
durable_db_item_count_gauge = gauge('alexandria.content/durable-db/item-count.gauge')
ephemeral_db_item_count_gauge = gauge('alexandria.content/ephemeral-db/item-count.gauge')
ephemeral_db_capacity_gauge = gauge('alexandria.content/ephemeral-db/capacity.gauge')
ephemeral_db_size_gauge = gauge('alexandria.content/ephemeral-db/size.gauge')
ephemeral_index_capacity_gauge = gauge('alexandria.content/ephemeral-index/capacity.gauge')
ephemeral_index_size_gauge = gauge('alexandria.content/ephemeral-index/size.gauge')
cache_db_item_count_gauge = gauge('alexandria.content/cache-db/item-count.gauge')
cache_db_capacity_gauge = gauge('alexandria.content/cache-db/capacity.gauge')
cache_db_size_gauge = gauge('alexandria.content/cache-db/size.gauge')
cache_index_capacity_gauge = gauge('alexandria.content/cache-index/capacity.gauge')
cache_index_size_gauge = gauge('alexandria.content/cache-index/size.gauge')
async for _ in every(frequency):
stats = self.kademlia.content_manager.get_stats()
durable_db_item_count_gauge.set_value(stats.durable_item_count)
ephemeral_db_item_count_gauge.set_value(stats.ephemeral_db_count)
ephemeral_db_capacity_gauge.set_value(stats.ephemeral_db_capacity)
ephemeral_db_size_gauge.set_value(
stats.ephemeral_db_total_capacity - stats.ephemeral_db_capacity
)
ephemeral_index_capacity_gauge.set_value(stats.ephemeral_index_capacity)
ephemeral_index_size_gauge.set_value(
stats.ephemeral_index_total_capacity - stats.ephemeral_index_capacity
)
cache_db_item_count_gauge.set_value(stats.cache_db_count)
cache_db_capacity_gauge.set_value(stats.cache_db_capacity)
cache_db_size_gauge.set_value(stats.cache_db_total_capacity - stats.cache_db_capacity)
cache_index_capacity_gauge.set_value(stats.cache_index_capacity)
cache_index_size_gauge.set_value(
stats.cache_index_total_capacity - stats.cache_index_capacity
)
async def _collect_system_metrics(self, frequency: int) -> None:
cpu_sysload_gauge = self._registry.gauge('alexandria.system/cpu/sysload.gauge')
cpu_syswait_gauge = self._registry.gauge('alexandria.system/cpu/syswait.gauge')
memory_used_gauge = self._registry.gauge('alexandria.system/memory/used.gauge')
memory_free_gauge = self._registry.gauge('alexandria.system/memory/free.gauge')
disk_readdata_meter = self._registry.meter('alexandria.system/disk/readdata.meter')
disk_writedata_meter = self._registry.meter('alexandria.system/disk/writedata.meter')
network_in_packets_meter = self._registry.meter('alexandria.network/in/packets/total.meter')
network_out_packets_meter = self._registry.meter('alexandria.network/out/packets/total.meter') # noqa: E501
previous = read_system_stats()
async for _ in every(frequency, initial_delay=frequency):
current = read_system_stats()
global_time = current.cpu_stats.global_time - previous.cpu_stats.global_time
cpu_sysload_gauge.set_value(global_time / frequency)
global_wait = current.cpu_stats.global_wait_io - previous.cpu_stats.global_wait_io
cpu_syswait_gauge.set_value(global_wait / frequency)
memory_used_gauge.set_value(current.memory_stats.used)
memory_free_gauge.set_value(current.memory_stats.free)
read_bytes = current.disk_stats.read_bytes - previous.disk_stats.read_bytes
disk_readdata_meter.mark(read_bytes)
write_bytes = current.disk_stats.write_bytes - previous.disk_stats.write_bytes
disk_writedata_meter.mark(write_bytes)
in_packets = current.network_stats.in_packets - previous.network_stats.in_packets
network_in_packets_meter.mark(in_packets)
out_packets = current.network_stats.out_packets - previous.network_stats.out_packets
network_out_packets_meter.mark(out_packets)
previous = current
