007 add telemetry
This commit is contained in:
@@ -0,0 +1,5 @@
|
||||
"""Observability setup and helpers."""
|
||||
|
||||
from app.observability.setup import setup_observability
|
||||
|
||||
__all__ = ["setup_observability"]
|
||||
@@ -0,0 +1,48 @@
|
||||
"""Structured logging setup with request and trace correlation."""
|
||||
|
||||
import logging
|
||||
from typing import Any
|
||||
|
||||
import structlog
|
||||
from opentelemetry import trace
|
||||
from opentelemetry.trace import SpanContext
|
||||
|
||||
|
||||
def configure_structured_logging(log_level: str) -> None:
|
||||
resolved_level = _resolve_log_level(log_level)
|
||||
logging.basicConfig(level=resolved_level, format="%(message)s")
|
||||
|
||||
structlog.configure(
|
||||
processors=[
|
||||
structlog.contextvars.merge_contextvars,
|
||||
structlog.stdlib.add_log_level,
|
||||
add_trace_context,
|
||||
structlog.processors.TimeStamper(fmt="iso", utc=True),
|
||||
structlog.processors.JSONRenderer(),
|
||||
],
|
||||
logger_factory=structlog.stdlib.LoggerFactory(),
|
||||
wrapper_class=structlog.make_filtering_bound_logger(resolved_level),
|
||||
cache_logger_on_first_use=True,
|
||||
)
|
||||
|
||||
|
||||
def add_trace_context(
|
||||
_: Any,
|
||||
__: str,
|
||||
event_dict: dict[str, Any],
|
||||
) -> dict[str, Any]:
|
||||
span = trace.get_current_span()
|
||||
span_context = span.get_span_context()
|
||||
if _is_valid_span_context(span_context):
|
||||
event_dict["trace_id"] = format(span_context.trace_id, "032x")
|
||||
event_dict["span_id"] = format(span_context.span_id, "016x")
|
||||
return event_dict
|
||||
|
||||
|
||||
def _resolve_log_level(log_level: str) -> int:
|
||||
candidate = log_level.strip().upper()
|
||||
return logging.getLevelNamesMapping().get(candidate, logging.INFO)
|
||||
|
||||
|
||||
def _is_valid_span_context(span_context: SpanContext) -> bool:
|
||||
return bool(span_context.is_valid and span_context.trace_id and span_context.span_id)
|
||||
@@ -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
|
||||
@@ -0,0 +1,26 @@
|
||||
"""High-level observability setup wiring."""
|
||||
|
||||
from fastapi import FastAPI
|
||||
|
||||
from app.config import ObservabilityConfig
|
||||
from app.observability.logging import configure_structured_logging
|
||||
from app.observability.metrics import setup_metrics
|
||||
from app.observability.tracing import (
|
||||
instrument_fastapi_app,
|
||||
instrument_httpx_client,
|
||||
setup_tracing,
|
||||
)
|
||||
|
||||
|
||||
def setup_observability(app: FastAPI, config: ObservabilityConfig) -> None:
|
||||
configure_structured_logging(config.log_level)
|
||||
setup_tracing(
|
||||
service_name=config.service_name,
|
||||
otlp_endpoint=config.otlp_endpoint,
|
||||
)
|
||||
setup_metrics(
|
||||
service_name=config.service_name,
|
||||
otlp_endpoint=config.otlp_endpoint,
|
||||
)
|
||||
instrument_fastapi_app(app)
|
||||
instrument_httpx_client()
|
||||
@@ -0,0 +1,80 @@
|
||||
"""OpenTelemetry tracing setup and instrumentation helpers."""
|
||||
|
||||
from typing import cast
|
||||
|
||||
from fastapi import FastAPI
|
||||
from opentelemetry import trace
|
||||
from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
|
||||
from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
|
||||
from opentelemetry.instrumentation.httpx import HTTPXClientInstrumentor
|
||||
from opentelemetry.sdk.resources import Resource
|
||||
from opentelemetry.sdk.trace import TracerProvider
|
||||
from opentelemetry.sdk.trace.export import BatchSpanProcessor
|
||||
from opentelemetry.trace import Tracer, TracerProvider as TraceAPIProvider
|
||||
|
||||
_TRACER_PROVIDER: TraceAPIProvider | None = None
|
||||
_HTTPX_INSTRUMENTED = False
|
||||
|
||||
|
||||
def setup_tracing(
|
||||
*,
|
||||
service_name: str,
|
||||
otlp_endpoint: str,
|
||||
tracer_provider: TraceAPIProvider | None = None,
|
||||
) -> TraceAPIProvider:
|
||||
global _TRACER_PROVIDER
|
||||
|
||||
if tracer_provider is None and _TRACER_PROVIDER is not None:
|
||||
return _TRACER_PROVIDER
|
||||
|
||||
if tracer_provider is None:
|
||||
tracer_provider = TracerProvider(
|
||||
resource=Resource.create({"service.name": service_name}),
|
||||
)
|
||||
if otlp_endpoint:
|
||||
span_exporter = OTLPSpanExporter(
|
||||
endpoint=otlp_endpoint,
|
||||
insecure=otlp_endpoint.startswith("http://"),
|
||||
)
|
||||
tracer_provider.add_span_processor(BatchSpanProcessor(span_exporter))
|
||||
|
||||
_TRACER_PROVIDER = tracer_provider
|
||||
return tracer_provider
|
||||
|
||||
|
||||
def instrument_fastapi_app(app: FastAPI) -> None:
|
||||
tracer_provider = _TRACER_PROVIDER
|
||||
if tracer_provider is None:
|
||||
return
|
||||
FastAPIInstrumentor.instrument_app(app, tracer_provider=tracer_provider)
|
||||
|
||||
|
||||
def instrument_httpx_client() -> None:
|
||||
global _HTTPX_INSTRUMENTED
|
||||
|
||||
tracer_provider = _TRACER_PROVIDER
|
||||
if tracer_provider is None:
|
||||
return
|
||||
|
||||
if _HTTPX_INSTRUMENTED:
|
||||
return
|
||||
HTTPXClientInstrumentor().instrument(tracer_provider=tracer_provider)
|
||||
_HTTPX_INSTRUMENTED = True
|
||||
|
||||
|
||||
def get_tracer(name: str) -> Tracer:
|
||||
tracer_provider = _TRACER_PROVIDER
|
||||
if tracer_provider is None:
|
||||
return trace.get_tracer(name)
|
||||
return cast(Tracer, tracer_provider.get_tracer(name))
|
||||
|
||||
|
||||
def reset_tracing_for_tests() -> None:
|
||||
global _TRACER_PROVIDER, _HTTPX_INSTRUMENTED
|
||||
|
||||
if _HTTPX_INSTRUMENTED:
|
||||
HTTPXClientInstrumentor().uninstrument()
|
||||
if isinstance(_TRACER_PROVIDER, TracerProvider):
|
||||
_TRACER_PROVIDER.shutdown()
|
||||
_TRACER_PROVIDER = None
|
||||
_HTTPX_INSTRUMENTED = False
|
||||
Reference in New Issue
Block a user