feat(backend): add e2e useful tracing

Enabled tracing component inside git agent in order to have end to end
spans for a single request.
Added spans at most useful places such as message bus' execute_message
and database execute_query.

Signed-off-by: Fatih Acar <fatih@opsmill.com>

backend/infrahub/cli/git_agent.py

@@ -9,7 +9,7 @@
 from prometheus_client import start_http_server
 from rich.logging import RichHandler
 
-from infrahub import config
+from infrahub import __version__, config
 from infrahub.components import ComponentType
 from infrahub.core.initialization import initialization
 from infrahub.database import InfrahubDatabase, get_db
@@ -20,6 +20,7 @@
 from infrahub.services import InfrahubServices
 from infrahub.services.adapters.cache.redis import RedisCache
 from infrahub.services.adapters.message_bus.rabbitmq import RabbitMQMessageBus
+from infrahub.trace import configure_trace
 
 app = typer.Typer()
 
@@ -65,6 +66,15 @@ async def _start(debug: bool, port: int) -> None:
     client = await InfrahubClient.init(address=config.SETTINGS.main.internal_address, retry_on_failure=True, log=log)
     await client.branch.all()
 
+    # Initialize trace
+    if config.SETTINGS.trace.enable:
+        configure_trace(
+            service="infrahub-git-agent",
+            version=__version__,
+            exporter_endpoint=config.SETTINGS.trace.trace_endpoint,
+            exporter_protocol=config.SETTINGS.trace.exporter_protocol,
+        )
+
     # Initialize the lock
     initialize_lock()
 

backend/infrahub/database/__init__.py

@@ -15,6 +15,7 @@
 
 # from contextlib import asynccontextmanager
 from neo4j.exceptions import ClientError, ServiceUnavailable
+from otel_extensions import get_tracer
 
 from infrahub import config
 from infrahub.exceptions import DatabaseError
@@ -192,19 +193,22 @@ async def close(self):
     async def execute_query(
         self, query: str, params: Optional[Dict[str, Any]] = None, name: Optional[str] = "undefined"
     ) -> List[Record]:
-        with QUERY_EXECUTION_METRICS.labels(str(self._session_mode), name).time():
-            if self.is_transaction:
-                execution_method = await self.transaction()
-            else:
-                execution_method = await self.session()
-
-            try:
-                response = await execution_method.run(query=query, parameters=params)
-            except ServiceUnavailable as exc:
-                log.error("Database Service unavailable", error=str(exc))
-                raise DatabaseError(message="Unable to connect to the database") from exc
-
-            return [item async for item in response]
+        with get_tracer(__name__).start_as_current_span("execute_db_query") as span:
+            span.set_attribute("query", query)
+
+            with QUERY_EXECUTION_METRICS.labels(str(self._session_mode), name).time():
+                if self.is_transaction:
+                    execution_method = await self.transaction()
+                else:
+                    execution_method = await self.session()
+
+                try:
+                    response = await execution_method.run(query=query, parameters=params)
+                except ServiceUnavailable as exc:
+                    log.error("Database Service unavailable", error=str(exc))
+                    raise DatabaseError(message="Unable to connect to the database") from exc
+
+                return [item async for item in response]
 
     def render_list_comprehension(self, items: str, item_name: str) -> str:
         if self.db_type == DatabaseType.MEMGRAPH:

backend/infrahub/graphql/mutations/branch.py

@@ -4,6 +4,7 @@
 from graphene import Boolean, Field, InputObjectType, List, Mutation, String
 from graphql import GraphQLResolveInfo
 from infrahub_sdk.utils import extract_fields
+from otel_extensions import instrumented
 
 from infrahub import config, lock
 from infrahub.core import registry
@@ -44,6 +45,7 @@ class Arguments:
     object = Field(BranchType)
 
     @classmethod
+    @instrumented
     async def mutate(cls, root: dict, info: GraphQLResolveInfo, data: BranchCreateInput, background_execution=False):
         db: InfrahubDatabase = info.context.get("infrahub_database")
 

backend/infrahub/message_bus/operations/__init__.py

@@ -1,5 +1,7 @@
 import json
 
+from otel_extensions import get_tracer
+
 from infrahub.log import get_logger
 from infrahub.message_bus import InfrahubResponse, messages
 from infrahub.message_bus.operations import check, event, finalize, git, refresh, requests, send, transform, trigger
@@ -53,19 +55,22 @@
 
 
 async def execute_message(routing_key: str, message_body: bytes, service: InfrahubServices):
-    message_data = json.loads(message_body)
-    message = messages.MESSAGE_MAP[routing_key](**message_data)
-    message.set_log_data(routing_key=routing_key)
-    try:
-        await COMMAND_MAP[routing_key](message=message, service=service)
-    except Exception as exc:  # pylint: disable=broad-except
-        if message.reply_requested:
-            response = InfrahubResponse(passed=False, response_class="rpc_error", response_data={"error": str(exc)})
-            await service.reply(message=response, initiator=message)
-            return
-        if message.reached_max_retries:
-            log.error("Message failed after maximum number of retries", error=str(exc))
-            await set_check_status(message, conclusion="failure", service=service)
-            return
-        message.increase_retry_count()
-        await service.send(message, delay=MessageTTL.FIVE)
+    with get_tracer(__name__).start_as_current_span("execute_message") as span:
+        span.set_attribute("routing_key", routing_key)
+
+        message_data = json.loads(message_body)
+        message = messages.MESSAGE_MAP[routing_key](**message_data)
+        message.set_log_data(routing_key=routing_key)
+        try:
+            await COMMAND_MAP[routing_key](message=message, service=service)
+        except Exception as exc:  # pylint: disable=broad-except
+            if message.reply_requested:
+                response = InfrahubResponse(passed=False, response_class="rpc_error", response_data={"error": str(exc)})
+                await service.reply(message=response, initiator=message)
+                return
+            if message.reached_max_retries:
+                log.error("Message failed after maximum number of retries", error=str(exc))
+                await set_check_status(message, conclusion="failure", service=service)
+                return
+            message.increase_retry_count()
+            await service.send(message, delay=MessageTTL.FIVE)

backend/infrahub/server.py

@@ -12,7 +12,7 @@
 from fastapi.staticfiles import StaticFiles
 from fastapi.templating import Jinja2Templates
 from infrahub_sdk.timestamp import TimestampFormatError
-from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor
+from opentelemetry.instrumentation.fastapi import FastAPIInstrumentor, Span
 from pydantic import ValidationError
 from starlette_exporter import PrometheusMiddleware, handle_metrics
 
@@ -32,7 +32,7 @@
 from infrahub.services import InfrahubServices, services
 from infrahub.services.adapters.cache.redis import RedisCache
 from infrahub.services.adapters.message_bus.rabbitmq import RabbitMQMessageBus
-from infrahub.trace import add_span_exception, configure_trace, get_traceid, get_tracer
+from infrahub.trace import add_span_exception, configure_trace, get_traceid
 from infrahub.worker import WORKER_IDENTITY
 
 
@@ -46,8 +46,8 @@ async def app_initialization(application: FastAPI) -> None:
     # Initialize trace
     if config.SETTINGS.trace.enable:
         configure_trace(
+            service="infrahub-server",
             version=__version__,
-            exporter_type=config.SETTINGS.trace.exporter_type,
             exporter_endpoint=config.SETTINGS.trace.trace_endpoint,
             exporter_protocol=config.SETTINGS.trace.exporter_protocol,
         )
@@ -101,8 +101,13 @@ async def lifespan(application: FastAPI):
     redoc_url="/api/redoc",
 )
 
-FastAPIInstrumentor().instrument_app(app, excluded_urls=".*/metrics")
-tracer = get_tracer()
+
+def server_request_hook(span: Span, scope: dict):  # pylint: disable=unused-argument
+    if span and span.is_recording():
+        span.set_attribute("worker", WORKER_IDENTITY)
+
+
+FastAPIInstrumentor().instrument_app(app, excluded_urls=".*/metrics", server_request_hook=server_request_hook)
 
 FRONTEND_DIRECTORY = os.environ.get("INFRAHUB_FRONTEND_DIRECTORY", os.path.abspath("frontend"))
 FRONTEND_ASSET_DIRECTORY = f"{FRONTEND_DIRECTORY}/dist/assets"
@@ -121,15 +126,17 @@ async def lifespan(application: FastAPI):
 async def logging_middleware(request: Request, call_next: Callable[[Request], Awaitable[Response]]) -> Response:
     clear_log_context()
     request_id = correlation_id.get()
-    with tracer.start_as_current_span("processing request " + request_id):
-        trace_id = get_traceid()
-        set_log_data(key="request_id", value=request_id)
-        set_log_data(key="app", value="infrahub.api")
-        set_log_data(key="worker", value=WORKER_IDENTITY)
-        if trace_id:
-            set_log_data(key="trace_id", value=trace_id)
-        response = await call_next(request)
-        return response
+
+    set_log_data(key="request_id", value=request_id)
+    set_log_data(key="app", value="infrahub.api")
+    set_log_data(key="worker", value=WORKER_IDENTITY)
+
+    trace_id = get_traceid()
+    if trace_id:
+        set_log_data(key="trace_id", value=trace_id)
+
+    response = await call_next(request)
+    return response
 
 
 @app.middleware("http")

backend/infrahub/services/adapters/message_bus/__init__.py

@@ -2,14 +2,45 @@
 
 from typing import TYPE_CHECKING, Optional
 
+import opentelemetry.instrumentation.aio_pika.span_builder
+from opentelemetry.instrumentation.aio_pika import AioPikaInstrumentor
+from opentelemetry.semconv.trace import SpanAttributes
+
 if TYPE_CHECKING:
-    from aio_pika.abc import AbstractExchange
+    from aio_pika.abc import AbstractChannel, AbstractExchange
+    from opentelemetry.instrumentation.aio_pika.span_builder import SpanBuilder
 
     from infrahub.message_bus import InfrahubMessage, InfrahubResponse
     from infrahub.message_bus.types import MessageTTL
     from infrahub.services import InfrahubServices
 
 
+AioPikaInstrumentor().instrument()
+
+
+# TODO: remove this once https://github.com/open-telemetry/opentelemetry-python-contrib/issues/1835 is resolved
+def patch_spanbuilder_set_channel() -> None:
+    """
+    The default SpanBuilder.set_channel does not work with aio_pika 9.1 and the refactored connection
+    attribute
+    """
+
+    def set_channel(self: SpanBuilder, channel: AbstractChannel) -> None:
+        if hasattr(channel, "_connection"):
+            url = channel._connection.url
+            self._attributes.update(
+                {
+                    SpanAttributes.NET_PEER_NAME: url.host,
+                    SpanAttributes.NET_PEER_PORT: url.port,
+                }
+            )
+
+    opentelemetry.instrumentation.aio_pika.span_builder.SpanBuilder.set_channel = set_channel  # type: ignore[misc]
+
+
+patch_spanbuilder_set_channel()
+
+
 class InfrahubMessageBus:
     # This exchange attribute should be removed when the InfrahubRpcClient
     # class has been removed

backend/infrahub/services/adapters/message_bus/rabbitmq.py

@@ -6,6 +6,7 @@
 
 import aio_pika
 from infrahub_sdk import UUIDT
+from opentelemetry import context, propagate
 
 from infrahub import config
 from infrahub.components import ComponentType
@@ -183,17 +184,28 @@ async def subscribe(self) -> None:
             async for message in qiterator:
                 try:
                     async with message.process(requeue=False):
+                        # auto instrumentation not supported yet for RPCs, do it ourselves...
+                        token = None
+                        headers = message.headers or {}
+                        ctx = propagate.extract(headers)
+                        if ctx is not None:
+                            token = context.attach(ctx)
+
                         clear_log_context()
-                        if message.routing_key in messages.MESSAGE_MAP:
-                            await execute_message(
-                                routing_key=message.routing_key, message_body=message.body, service=self.service
-                            )
-                        else:
-                            self.service.log.error(
-                                "Unhandled routing key for message",
-                                routing_key=message.routing_key,
-                                message=message.body,
-                            )
+                        try:
+                            if message.routing_key in messages.MESSAGE_MAP:
+                                await execute_message(
+                                    routing_key=message.routing_key, message_body=message.body, service=self.service
+                                )
+                            else:
+                                self.service.log.error(
+                                    "Unhandled routing key for message",
+                                    routing_key=message.routing_key,
+                                    message=message.body,
+                                )
+                        finally:
+                            if token is not None:
+                                context.detach(token)
 
                 except Exception:  # pylint: disable=broad-except
                     self.service.log.exception("Processing error for message %r" % message)

backend/infrahub/trace.py

@@ -1,18 +1,6 @@
 from opentelemetry import trace
-from opentelemetry.exporter.otlp.proto.grpc.trace_exporter import (
-    OTLPSpanExporter as GRPCSpanExporter,
-)
-from opentelemetry.exporter.otlp.proto.http.trace_exporter import (
-    OTLPSpanExporter as HTTPSpanExporter,
-)
-from opentelemetry.sdk.resources import Resource
-from opentelemetry.sdk.trace import TracerProvider
-from opentelemetry.sdk.trace.export import BatchSpanProcessor, ConsoleSpanExporter
 from opentelemetry.trace import StatusCode
-
-
-def get_tracer(name: str = "infrahub") -> trace.Tracer:
-    return trace.get_tracer(name)
+from otel_extensions import TelemetryOptions, init_telemetry_provider
 
 
 def get_current_span_with_context() -> trace.Span:
@@ -54,42 +42,10 @@ def add_span_exception(exception: Exception) -> None:
         current_span.record_exception(exception)
 
 
-def create_tracer_provider(
-    version: str, exporter_type: str, exporter_endpoint: str = None, exporter_protocol: str = None
-) -> TracerProvider:
-    # Create a BatchSpanProcessor exporter based on the type
-    if exporter_type == "console":
-        exporter = ConsoleSpanExporter()
-    elif exporter_type == "otlp":
-        if not exporter_endpoint:
-            raise ValueError("Exporter type is set to otlp but endpoint is not set")
-        if exporter_protocol == "http/protobuf":
-            exporter = HTTPSpanExporter(endpoint=exporter_endpoint)
-        elif exporter_protocol == "grpc":
-            exporter = GRPCSpanExporter(endpoint=exporter_endpoint)
-    else:
-        raise ValueError("Exporter type unsupported by Infrahub")
-
-    # Resource can be required for some backends, e.g. Jaeger
-    resource = Resource(attributes={"service.name": "infrahub", "service.version": version})
-    span_processor = BatchSpanProcessor(exporter)
-    tracer_provider = TracerProvider(resource=resource)
-    tracer_provider.add_span_processor(span_processor)
-
-    return tracer_provider
-
-
-def configure_trace(
-    version: str, exporter_type: str, exporter_endpoint: str = None, exporter_protocol: str = None
-) -> None:
-    # Create a trace provider with the exporter
-    tracer_provider = create_tracer_provider(
-        version=version,
-        exporter_type=exporter_type,
-        exporter_endpoint=exporter_endpoint,
-        exporter_protocol=exporter_protocol,
+def configure_trace(service: str, version: str, exporter_endpoint: str = None, exporter_protocol: str = None) -> None:
+    options = TelemetryOptions(
+        OTEL_SERVICE_NAME=service,
+        OTEL_EXPORTER_OTLP_ENDPOINT=exporter_endpoint,
+        OTEL_EXPORTER_OTLP_PROTOCOL=exporter_protocol,
     )
-    tracer_provider.get_tracer(__name__)
-
-    # Register the trace provider
-    trace.set_tracer_provider(tracer_provider)
+    init_telemetry_provider(options, **{"service.version": version})

development/docker-compose.override.yml.tmp

@@ -3,7 +3,7 @@ version: "3.4"
 services:
   # --------------------------------------------------------------------------------
   #  - Prometheus to collect all metrics endpoints
-  #  - Tempo to receive traces
+  #  - Tempo or Jaeger to receive traces
   #  - Grafana to visualize these metrics
   #  - Loki to receive logs from promtail
   #  - Promtail to parse logs from different source
@@ -43,6 +43,13 @@ services:
     ports:
       - "3200:3200"
 
+  # jaeger:
+  #   image: jaegertracing/all-in-one:1.53
+  #   environment:
+  #     COLLECTOR_ZIPKIN_HOST_PORT: ":9411"
+  #   ports:
+  #     - "16686:16686"
+
   prometheus:
     image: prom/prometheus:latest
     volumes: