feat(opentelemetry): fastAPI tracing

The tests don't work yet since the webserver has to run in a thread and
the traces don't get exported from the thread with local json exporting.
However, with an export to an external server this should still work.
WIP

pyinfra/examples.py

@@ -2,17 +2,22 @@ from dynaconf import Dynaconf
 from fastapi import FastAPI
 from kn_utils.logging import logger
 
+from pyinfra.config.loader import validate_settings, get_all_validators
 from pyinfra.queue.callback import make_download_process_upload_callback, DataProcessor
 from pyinfra.queue.manager import QueueManager
-from pyinfra.webserver.prometheus import add_prometheus_endpoint, \
-    make_prometheus_processing_time_decorator_from_settings
+from pyinfra.utils.opentelemetry import setup_trace, instrument_pika, instrument_app
+from pyinfra.webserver.prometheus import (
+    add_prometheus_endpoint,
+    make_prometheus_processing_time_decorator_from_settings,
+)
 from pyinfra.webserver.utils import add_health_check_endpoint, create_webserver_thread_from_settings
 
 
 def start_queue_consumer_with_prometheus_and_health_endpoints(process_fn: DataProcessor, settings: Dynaconf):
     """Default serving logic for research services.
 
-    Supplies /health, /ready and /prometheus endpoints. The process_fn is monitored for processing time per call.
+    Supplies /health, /ready and /prometheus endpoints (if enabled). The process_fn is monitored for processing time per
+    call. Also traces the queue messages (if enabled).
     Workload is only received via queue messages. The message contains a file path to the data to be processed, which
     gets downloaded from the storage. The data and the message are then passed to the process_fn. The process_fn should
     return a json serializable object. This object is then uploaded to the storage. The response message is just the
@@ -20,15 +25,24 @@ def start_queue_consumer_with_prometheus_and_health_endpoints(process_fn: DataPr
 
     Adapt as needed.
     """
+    validate_settings(settings, get_all_validators())
+
     logger.info(f"Starting webserver and queue consumer...")
 
     app = FastAPI()
 
-    app = add_prometheus_endpoint(app)
-    process_fn = make_prometheus_processing_time_decorator_from_settings(settings)(process_fn)
-
     queue_manager = QueueManager(settings)
 
+    if settings.metrics.prometheus.enabled:
+        logger.info(f"Prometheus metrics enabled.")
+        app = add_prometheus_endpoint(app)
+        process_fn = make_prometheus_processing_time_decorator_from_settings(settings)(process_fn)
+
+    if settings.tracing.opentelemetry.enabled:
+        logger.info(f"OpenTelemetry tracing enabled.")
+        setup_trace(settings)
+        instrument_pika()
+
     app = add_health_check_endpoint(app, queue_manager.is_ready)
 
     webserver_thread = create_webserver_thread_from_settings(app, settings)

pyinfra/utils/opentelemetry.py

@@ -2,6 +2,7 @@ import json
 
 from dynaconf import Dynaconf
 from fastapi import FastAPI
+from kn_utils.logging import logger
 from opentelemetry import trace
 from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import OTLPSpanExporter
 from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
@@ -33,10 +34,9 @@ def setup_trace(settings: Dynaconf, service_name: str = None, exporter: SpanExpo
     exporter = exporter or get_exporter(settings)
 
     resource = Resource(attributes={"service.name": service_name})
-    provider = TracerProvider(resource=resource)
+    provider = TracerProvider(resource=resource, shutdown_on_exit=True)
 
     processor = BatchSpanProcessor(exporter)
-
     provider.add_span_processor(processor)
 
     # TODO: This produces a warning if trying to set the provider twice.
@@ -51,7 +51,7 @@ def get_exporter(settings: Dynaconf):
     if settings.tracing.opentelemetry.exporter == "json":
         return JsonSpanExporter()
     elif settings.tracing.opentelemetry.exporter == "otlp":
-        return OTLPSpanExporter(endpoint=settings.metrics.opentelemetry.endpoint)
+        return OTLPSpanExporter(endpoint=settings.tracing.opentelemetry.endpoint)
     elif settings.tracing.opentelemetry.exporter == "console":
         return ConsoleSpanExporter()
     else:

pyinfra/webserver/prometheus.py

@@ -36,16 +36,11 @@ def make_prometheus_processing_time_decorator_from_settings(
     postfix: str = "processing_time",
     registry: CollectorRegistry = REGISTRY,
 ) -> Decorator:
-    """Make a decorator for monitoring the processing time of a function. The decorator is only applied if the
-    prometheus metrics are enabled in the settings.
-    This, and other metrics should follow the convention
-        {product name}_{service name}_{processing step / parameter to monitor}.
+    """Make a decorator for monitoring the processing time of a function. This, and other metrics should follow the
+     convention {product name}_{service name}_{processing step / parameter to monitor}.
     """
     validate_settings(settings, validators=prometheus_validators)
 
-    if not settings.metrics.prometheus.enabled:
-        return identity
-
     processing_time_sum = Summary(
         f"{settings.metrics.prometheus.prefix}_{postfix}",
         "Summed up processing time per call.",

pyinfra/webserver/utils.py

@@ -8,11 +8,15 @@ from fastapi import FastAPI
 
 from pyinfra.config.validators import webserver_validators
 from pyinfra.config.loader import validate_settings
+from pyinfra.utils.opentelemetry import setup_trace, instrument_app
 
 
 def create_webserver_thread_from_settings(app: FastAPI, settings: Dynaconf) -> threading.Thread:
     validate_settings(settings, validators=webserver_validators)
 
+    if settings.tracing.opentelemetry.enabled:
+        return create_webserver_thread_with_tracing(app, settings)
+
     return create_webserver_thread(app=app, port=settings.webserver.port, host=settings.webserver.host)
 
 
@@ -25,6 +29,19 @@ def create_webserver_thread(app: FastAPI, port: int, host: str) -> threading.Thr
     return thread
 
 
+def create_webserver_thread_with_tracing(app: FastAPI, settings: Dynaconf) -> threading.Thread:
+
+    def inner():
+        setup_trace(settings)
+        instrument_app(app)
+        uvicorn.run(app, port=settings.webserver.port, host=settings.webserver.host, log_level=logging.WARNING)
+
+    thread = threading.Thread(target=inner)
+    thread.daemon = True
+
+    return thread
+
+
 HealthFunction = Callable[[], bool]
 
 

tests/unit_test/opentelemetry_test.py

@@ -1,28 +1,8 @@
 from time import sleep
 
 import pytest
-import requests
-from fastapi import FastAPI
 
-from pyinfra.utils.opentelemetry import get_exporter, setup_trace, instrument_pika, instrument_app
-from pyinfra.webserver.utils import create_webserver_thread_from_settings
-
-
-@pytest.fixture(scope="class")
-def app_with_tracing(settings):
-    app = FastAPI()
-
-    @app.get("/test")
-    def test():
-        return {"flat": "earth"}
-
-    instrument_app(app)
-
-    thread = create_webserver_thread_from_settings(app, settings)
-    thread.start()
-    sleep(1)
-    yield
-    thread.join(timeout=1)
+from pyinfra.utils.opentelemetry import get_exporter, setup_trace, instrument_pika
 
 
 @pytest.fixture(scope="session")
@@ -33,7 +13,6 @@ def exporter(settings):
 
 class TestOpenTelemetry:
     def test_queue_messages_are_traced(self, queue_manager, input_message, stop_message, settings, exporter):
-
         setup_trace(settings, exporter=exporter)
 
         instrument_pika()
@@ -53,11 +32,20 @@ class TestOpenTelemetry:
                 exported_trace["resource"]["attributes"]["service.name"] == settings.tracing.opentelemetry.service_name
             )
 
-    def test_webserver_requests_are_traced(self, settings, app_with_tracing, exporter):
-        settings.tracing.opentelemetry.exporter = "json"
-
-        setup_trace(settings, exporter=exporter)
-
-        requests.get(f"http://{settings.webserver.host}:{settings.webserver.port}/test")
-
-        print(exporter.traces)
+    # def test_webserver_requests_are_traced(self, settings):
+    #     settings.tracing.opentelemetry.exporter = "console"
+    #     settings.tracing.opentelemetry.enabled = True
+    #
+    #     app = FastAPI()
+    #
+    #     @app.get("/test")
+    #     def test():
+    #         return {"test": "test"}
+    #
+    #     thread = create_webserver_thread_from_settings(app, settings)
+    #     thread.start()
+    #     sleep(1)
+    #
+    #     requests.get(f"http://{settings.webserver.host}:{settings.webserver.port}/test")
+    #
+    #     thread.join(timeout=1)