007 add telemetry
This commit is contained in:
@@ -0,0 +1,129 @@
|
||||
"""OpenTelemetry metrics setup and recording helpers."""
|
||||
|
||||
from typing import Sequence
|
||||
|
||||
from opentelemetry import metrics
|
||||
from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import OTLPMetricExporter
|
||||
from opentelemetry.sdk.metrics import MeterProvider
|
||||
from opentelemetry.sdk.metrics.export import (
|
||||
MetricReader,
|
||||
PeriodicExportingMetricReader,
|
||||
)
|
||||
from opentelemetry.sdk.resources import Resource
|
||||
|
||||
_METER_PROVIDER: MeterProvider | None = None
|
||||
_METRICS: "ObservabilityMetrics" | None = None
|
||||
_METER_NAME = "g2s.observability"
|
||||
|
||||
|
||||
class ObservabilityMetrics:
|
||||
def __init__(
|
||||
self,
|
||||
meter_name: str,
|
||||
*,
|
||||
meter_provider: MeterProvider | None = None,
|
||||
) -> None:
|
||||
if meter_provider is None:
|
||||
meter = metrics.get_meter(meter_name)
|
||||
else:
|
||||
meter = meter_provider.get_meter(meter_name)
|
||||
self._request_count = meter.create_counter(
|
||||
name="http.server.request.count",
|
||||
unit="1",
|
||||
description="Total number of HTTP requests.",
|
||||
)
|
||||
self._error_count = meter.create_counter(
|
||||
name="http.server.error.count",
|
||||
unit="1",
|
||||
description="Total number of HTTP server errors.",
|
||||
)
|
||||
self._request_latency = meter.create_histogram(
|
||||
name="http.server.request.latency",
|
||||
unit="ms",
|
||||
description="HTTP request latency.",
|
||||
)
|
||||
self._provider_availability = meter.create_histogram(
|
||||
name="delivery.provider.availability",
|
||||
unit="1",
|
||||
description="Provider availability sample (1=available, 0=unavailable).",
|
||||
)
|
||||
|
||||
def record_http_request(
|
||||
self,
|
||||
*,
|
||||
method: str,
|
||||
route: str,
|
||||
status_code: int,
|
||||
duration_ms: float,
|
||||
is_error: bool,
|
||||
) -> None:
|
||||
attributes = {
|
||||
"http.method": method,
|
||||
"http.route": route,
|
||||
"http.status_code": status_code,
|
||||
}
|
||||
self._request_count.add(1, attributes)
|
||||
self._request_latency.record(duration_ms, attributes)
|
||||
if is_error:
|
||||
self._error_count.add(1, attributes)
|
||||
|
||||
def record_provider_availability(
|
||||
self,
|
||||
*,
|
||||
provider: str,
|
||||
is_available: bool,
|
||||
) -> None:
|
||||
self._provider_availability.record(
|
||||
1.0 if is_available else 0.0,
|
||||
{"provider": provider},
|
||||
)
|
||||
|
||||
|
||||
def setup_metrics(
|
||||
*,
|
||||
service_name: str,
|
||||
otlp_endpoint: str,
|
||||
meter_provider: MeterProvider | None = None,
|
||||
metric_readers: Sequence[MetricReader] | None = None,
|
||||
) -> MeterProvider:
|
||||
global _METER_PROVIDER, _METRICS
|
||||
|
||||
if meter_provider is None and _METER_PROVIDER is not None:
|
||||
return _METER_PROVIDER
|
||||
|
||||
if meter_provider is None:
|
||||
readers = list(metric_readers or ())
|
||||
if otlp_endpoint:
|
||||
metric_exporter = OTLPMetricExporter(
|
||||
endpoint=otlp_endpoint,
|
||||
insecure=otlp_endpoint.startswith("http://"),
|
||||
)
|
||||
readers.append(PeriodicExportingMetricReader(metric_exporter))
|
||||
meter_provider = MeterProvider(
|
||||
resource=Resource.create({"service.name": service_name}),
|
||||
metric_readers=readers,
|
||||
)
|
||||
|
||||
_METER_PROVIDER = meter_provider
|
||||
_METRICS = ObservabilityMetrics(
|
||||
_METER_NAME,
|
||||
meter_provider=meter_provider,
|
||||
)
|
||||
return meter_provider
|
||||
|
||||
|
||||
def get_metrics() -> ObservabilityMetrics:
|
||||
global _METRICS
|
||||
|
||||
if _METRICS is None:
|
||||
_METRICS = ObservabilityMetrics(_METER_NAME)
|
||||
return _METRICS
|
||||
|
||||
|
||||
def reset_metrics_for_tests() -> None:
|
||||
global _METER_PROVIDER, _METRICS
|
||||
|
||||
if _METER_PROVIDER is not None:
|
||||
_METER_PROVIDER.shutdown()
|
||||
_METER_PROVIDER = None
|
||||
_METRICS = None
|
||||
Reference in New Issue
Block a user