Handle issues introduced by OTLP gRPC protocol (#170)

In this commit, we are handling issues that arise from gRPC.
Essentially, if we build gRPC artifacts into our Docker image, it causes
the Docker image to only be compatible with applications built using the
same Python version. To solve this, we are doing two things: 1) we are
removing gRPC artifacts from the docker image and 2) we are changing the
default OTLP protocol to be HTTP. If customers attempt to set the
protocol as gRPC for ApplicationSignals, we will set the default
endpoint correctly.

Also we are changing Docker image to build with Python 3.11, which is
what we were originally doing when we encountered this issue (reference:
5b3ed74).
This is what the upstream does (see
[autoinstrumentation/python/Dockerfile](https://github.com/open-telemetry/opentelemetry-operator/blob/b5bb0ae34720d4be2d229dafecb87b61b37699b0/autoinstrumentation/python/Dockerfile)),
and having parity here is beneficial to us.

Testing:
* Create `app.py`:
```
from time import sleep
import boto3

try:
    boto3.client('s3').list_buckets()
except Exception:
    sleep(100)

```
* Run `./scripts/build_and_install_distro.sh`
* Run:
```
export OTEL_PYTHON_DISTRO="aws_distro"         
export OTEL_PYTHON_CONFIGURATOR="aws_configurator"
export OTEL_METRICS_EXPORTER="none"
unset OTEL_EXPORTER_OTLP_PROTOCOL
unset OTEL_AWS_APPLICATION_SIGNALS_ENABLED
```
* Run `opentelemetry-instrument python ./app.py`
```
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='localhost', port=4318): Max retries exceeded with url: /v1/traces (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7fc647cd5e50>: Failed to establish a new connection: [Errno 111] Connection refused'))
```
* Run `export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf;
opentelemetry-instrument python ./app.py`
```
requests.exceptions.ConnectionError: HTTPConnectionPool(host='localhost', port=4318): Max retries exceeded with url: /v1/traces (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f20ac96e490>: Failed to establish a new connection: [Errno 111] Connection refused'))
```
* Run `export OTEL_EXPORTER_OTLP_PROTOCOL=grpc; opentelemetry-instrument
python ./app.py`
```
Transient error StatusCode.UNAVAILABLE encountered while exporting traces to localhost:4317, retrying in 1s.
```
* Run
```
unset OTEL_EXPORTER_OTLP_PROTOCOL
export OTEL_METRIC_EXPORT_INTERVAL=1000
export OTEL_AWS_APPLICATION_SIGNALS_ENABLED=True
```
* Run `opentelemetry-instrument python ./app.py`
```
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='localhost', port=4316): Max retries exceeded with url: /v1/metrics (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f8f6c5aa5b0>: Failed to establish a new connection: [Errno 111] Connection refused'))
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='localhost', port=4318): Max retries exceeded with url: /v1/traces (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7f8f6dd9c9d0>: Failed to establish a new connection: [Errno 111] Connection refused'))
```
* Run `export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf;
opentelemetry-instrument python ./app.py`
```
urllib3.exceptions.MaxRetryError: HTTPConnectionPool(host='localhost', port=4318): Max retries exceeded with url: /v1/traces (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7ff5868efdc0>: Failed to establish a new connection: [Errno 111] Connection refused'))
requests.exceptions.ConnectionError: HTTPConnectionPool(host='localhost', port=4318): Max retries exceeded with url: /v1/traces (Caused by NewConnectionError('<urllib3.connection.HTTPConnection object at 0x7ff5868efdc0>: Failed to establish a new connection: [Errno 111] Connection refused'))
```
* Run `export OTEL_EXPORTER_OTLP_PROTOCOL=grpc; opentelemetry-instrument
python ./app.py`
```
Transient error StatusCode.UNAVAILABLE encountered while exporting metrics to localhost:4315, retrying in 1s.
Transient error StatusCode.UNAVAILABLE encountered while exporting traces to localhost:4317, retrying in 1s.
```

By submitting this pull request, I confirm that you can use, modify,
copy, and redistribute this contribution, under the terms of your
choice.

Dockerfile

@@ -3,20 +3,22 @@
 #   The packages are installed in the `/autoinstrumentation` directory. This is required as when instrumenting the pod by CWOperator,
 #   one init container will be created to copy all the content in `/autoinstrumentation` directory to app's container. Then
 #   update the `PYTHONPATH` environment variable accordingly. Then in the second stage, copy the directory to `/autoinstrumentation`.
-
-# Using Python 3.10 because we are utilizing the opentelemetry-exporter-otlp-proto-grpc exporter,
-# which relies on grpcio as a dependency. grpcio has strict dependencies on the OS and Python version.
-# Also mentioned in Docker build template in the upstream repository:
-# https://github.com/open-telemetry/opentelemetry-operator/blob/b5bb0ae34720d4be2d229dafecb87b61b37699b0/autoinstrumentation/python/requirements.txt#L2
-# For further details, please refer to: https://github.com/MicrosoftDocs/azure-docs/blob/main/articles/azure-functions/recover-python-functions.md#the-python-interpre[…]tions-python-worker
-FROM python:3.10 AS build
+FROM python:3.11 AS build
 
 WORKDIR /operator-build
 
 ADD aws-opentelemetry-distro/ ./aws-opentelemetry-distro/
 
 RUN mkdir workspace && pip install --target workspace ./aws-opentelemetry-distro
 
+# Remove opentelemetry-exporter-otlp-proto-grpc and grpcio, as grpcio has strict dependencies on the Python version and
+# will cause confusing failures if gRPC protocol is used. Now if gRPC protocol is requested by the user, instrumentation
+# will complain that grpc is not installed, which is more understandable. References:
+# * https://github.com/open-telemetry/opentelemetry-operator/blob/b5bb0ae34720d4be2d229dafecb87b61b37699b0/autoinstrumentation/python/requirements.txt#L2
+# * https://github.com/MicrosoftDocs/azure-docs/blob/main/articles/azure-functions/recover-python-functions.md#troubleshoot-cannot-import-cygrpc
+RUN pip uninstall opentelemetry-exporter-otlp-proto-grpc -y
+RUN pip uninstall grpcio -y
+
 FROM public.ecr.aws/amazonlinux/amazonlinux:minimal
 
 # Required to copy attribute files to distributed docker images

aws-opentelemetry-distro/src/amazon/opentelemetry/distro/aws_opentelemetry_configurator.py

@@ -17,7 +17,6 @@
 )
 from amazon.opentelemetry.distro.aws_span_metrics_processor_builder import AwsSpanMetricsProcessorBuilder
 from amazon.opentelemetry.distro.sampler.aws_xray_remote_sampler import AwsXRayRemoteSampler
-from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import OTLPMetricExporter as OTLPGrpcOTLPMetricExporter
 from opentelemetry.exporter.otlp.proto.http.metric_exporter import OTLPMetricExporter as OTLPHttpOTLPMetricExporter
 from opentelemetry.sdk._configuration import (
     _get_exporter_names,
@@ -274,17 +273,10 @@ def __new__(cls, *args, **kwargs):
     # pylint: disable=no-self-use
     def create_exporter(self):
         protocol = os.environ.get(
-            OTEL_EXPORTER_OTLP_METRICS_PROTOCOL, os.environ.get(OTEL_EXPORTER_OTLP_PROTOCOL, "grpc")
+            OTEL_EXPORTER_OTLP_METRICS_PROTOCOL, os.environ.get(OTEL_EXPORTER_OTLP_PROTOCOL, "http/protobuf")
         )
         _logger.debug("AWS Application Signals export protocol: %s", protocol)
 
-        application_signals_endpoint = os.environ.get(
-            APPLICATION_SIGNALS_EXPORTER_ENDPOINT_CONFIG,
-            os.environ.get(APP_SIGNALS_EXPORTER_ENDPOINT_CONFIG, "http://localhost:4315"),
-        )
-
-        _logger.debug("AWS Application Signals export endpoint: %s", application_signals_endpoint)
-
         temporality_dict: Dict[type, AggregationTemporality] = {}
         for typ in [
             Counter,
@@ -298,10 +290,27 @@ def create_exporter(self):
             temporality_dict[typ] = AggregationTemporality.DELTA
 
         if protocol == "http/protobuf":
+            application_signals_endpoint = os.environ.get(
+                APPLICATION_SIGNALS_EXPORTER_ENDPOINT_CONFIG,
+                os.environ.get(APP_SIGNALS_EXPORTER_ENDPOINT_CONFIG, "http://localhost:4316/v1/metrics"),
+            )
+            _logger.debug("AWS Application Signals export endpoint: %s", application_signals_endpoint)
             return OTLPHttpOTLPMetricExporter(
                 endpoint=application_signals_endpoint, preferred_temporality=temporality_dict
             )
         if protocol == "grpc":
+            # pylint: disable=import-outside-toplevel
+            # Delay import to only occur if gRPC specifically requested. Vended Docker image will not have gRPC bundled,
+            # so importing it at the class level can cause runtime failures.
+            from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import (
+                OTLPMetricExporter as OTLPGrpcOTLPMetricExporter,
+            )
+
+            application_signals_endpoint = os.environ.get(
+                APPLICATION_SIGNALS_EXPORTER_ENDPOINT_CONFIG,
+                os.environ.get(APP_SIGNALS_EXPORTER_ENDPOINT_CONFIG, "localhost:4315"),
+            )
+            _logger.debug("AWS Application Signals export endpoint: %s", application_signals_endpoint)
             return OTLPGrpcOTLPMetricExporter(
                 endpoint=application_signals_endpoint, preferred_temporality=temporality_dict
             )

aws-opentelemetry-distro/src/amazon/opentelemetry/distro/aws_opentelemetry_distro.py

@@ -5,12 +5,29 @@
 from amazon.opentelemetry.distro.patches._instrumentation_patch import apply_instrumentation_patches
 from opentelemetry.distro import OpenTelemetryDistro
 from opentelemetry.environment_variables import OTEL_PROPAGATORS, OTEL_PYTHON_ID_GENERATOR
-from opentelemetry.sdk.environment_variables import OTEL_EXPORTER_OTLP_METRICS_DEFAULT_HISTOGRAM_AGGREGATION
+from opentelemetry.sdk.environment_variables import (
+    OTEL_EXPORTER_OTLP_METRICS_DEFAULT_HISTOGRAM_AGGREGATION,
+    OTEL_EXPORTER_OTLP_PROTOCOL,
+)
 
 
 class AwsOpenTelemetryDistro(OpenTelemetryDistro):
     def _configure(self, **kwargs):
-        """
+        """Sets up default environment variables and apply patches
+
+        Set default OTEL_EXPORTER_OTLP_PROTOCOL to be HTTP. This must be run before super(), which attempts to set the
+        default to gRPC. If we run afterwards, we don't know if the default was set by base OpenTelemetryDistro or if it
+        was set by the user. We are setting to HTTP as gRPC does not work out of the box for the vended docker image,
+        due to gRPC having a strict dependency on the Python version the artifact was built for (OTEL observed this:
+        https://github.com/open-telemetry/opentelemetry-operator/blob/461ba68e80e8ac6bf2603eb353547cd026119ed2/autoinstrumentation/python/requirements.txt#L2-L3)
+
+        Also sets default OTEL_PROPAGATORS, OTEL_PYTHON_ID_GENERATOR, and
+        OTEL_EXPORTER_OTLP_METRICS_DEFAULT_HISTOGRAM_AGGREGATION to ensure good compatibility with X-Ray and Application
+        Signals.
+
+        Also applies patches to upstream instrumentation - usually these are stopgap measures until we can contribute
+        long-term changes to upstream.
+
         kwargs:
             apply_patches: bool - apply patches to upstream instrumentation. Default is True.
 
@@ -19,13 +36,15 @@ def _configure(self, **kwargs):
             OTEL_EXPORTER_OTLP_METRICS_DEFAULT_HISTOGRAM_AGGREGATION environment variable. Need to work with upstream to
             make it to be configurable.
         """
+        os.environ.setdefault(OTEL_EXPORTER_OTLP_PROTOCOL, "http/protobuf")
+
         super(AwsOpenTelemetryDistro, self)._configure()
+
+        os.environ.setdefault(OTEL_PROPAGATORS, "xray,tracecontext,b3,b3multi")
+        os.environ.setdefault(OTEL_PYTHON_ID_GENERATOR, "xray")
         os.environ.setdefault(
             OTEL_EXPORTER_OTLP_METRICS_DEFAULT_HISTOGRAM_AGGREGATION, "base2_exponential_bucket_histogram"
         )
-        os.environ.setdefault(OTEL_PROPAGATORS, "xray,tracecontext,b3,b3multi")
-        os.environ.setdefault(OTEL_PYTHON_ID_GENERATOR, "xray")
 
-        # Apply patches to upstream instrumentation - usually stopgap measures until we can contribute long-term changes
         if kwargs.get("apply_patches", True):
             apply_instrumentation_patches()

aws-opentelemetry-distro/tests/amazon/opentelemetry/distro/test_aws_opentelementry_configurator.py

@@ -9,6 +9,7 @@
 from amazon.opentelemetry.distro.attribute_propagating_span_processor import AttributePropagatingSpanProcessor
 from amazon.opentelemetry.distro.aws_metric_attributes_span_exporter import AwsMetricAttributesSpanExporter
 from amazon.opentelemetry.distro.aws_opentelemetry_configurator import (
+    ApplicationSignalsExporterProvider,
     AwsOpenTelemetryConfigurator,
     _custom_import_sampler,
     _customize_exporter,
@@ -21,6 +22,9 @@
 from amazon.opentelemetry.distro.sampler._aws_xray_sampling_client import _AwsXRaySamplingClient
 from amazon.opentelemetry.distro.sampler.aws_xray_remote_sampler import AwsXRayRemoteSampler
 from opentelemetry.environment_variables import OTEL_LOGS_EXPORTER, OTEL_METRICS_EXPORTER, OTEL_TRACES_EXPORTER
+from opentelemetry.exporter.otlp.proto.common._internal.metrics_encoder import OTLPMetricExporterMixin
+from opentelemetry.exporter.otlp.proto.grpc.metric_exporter import OTLPMetricExporter as OTLPGrpcOTLPMetricExporter
+from opentelemetry.exporter.otlp.proto.http.metric_exporter import OTLPMetricExporter as OTLPHttpOTLPMetricExporter
 from opentelemetry.sdk.environment_variables import OTEL_TRACES_SAMPLER, OTEL_TRACES_SAMPLER_ARG
 from opentelemetry.sdk.resources import Resource
 from opentelemetry.sdk.trace import Span, SpanProcessor, Tracer, TracerProvider
@@ -29,18 +33,28 @@
 from opentelemetry.trace import get_tracer_provider
 
 
-# This class setup Tracer Provider Globally, which can only set once
-# if there is another setup for tracer provider, may cause issue
 class TestAwsOpenTelemetryConfigurator(TestCase):
+    """Tests AwsOpenTelemetryConfigurator and AwsOpenTelemetryDistro
+
+    NOTE: This class setup Tracer Provider Globally, which can only be set once. If there is another setup for tracer
+    provider, it may cause issues for those tests.
+    """
+
     @classmethod
     def setUpClass(cls):
-        os.environ.setdefault(OTEL_TRACES_EXPORTER, "none")
-        os.environ.setdefault(OTEL_METRICS_EXPORTER, "none")
-        os.environ.setdefault(OTEL_LOGS_EXPORTER, "none")
-        os.environ.setdefault(OTEL_TRACES_SAMPLER, "traceidratio")
-        os.environ.setdefault(OTEL_TRACES_SAMPLER_ARG, "0.01")
+        # Run AwsOpenTelemetryDistro to set up environment, then validate expected env values.
         aws_open_telemetry_distro: AwsOpenTelemetryDistro = AwsOpenTelemetryDistro()
         aws_open_telemetry_distro.configure(apply_patches=False)
+        validate_distro_environ()
+
+        # Overwrite exporter configs to keep tests clean, set sampler configs for tests
+        os.environ[OTEL_TRACES_EXPORTER] = "none"
+        os.environ[OTEL_METRICS_EXPORTER] = "none"
+        os.environ[OTEL_LOGS_EXPORTER] = "none"
+        os.environ[OTEL_TRACES_SAMPLER] = "traceidratio"
+        os.environ[OTEL_TRACES_SAMPLER_ARG] = "0.01"
+
+        # Run configurator and get trace provider
         aws_otel_configurator: AwsOpenTelemetryConfigurator = AwsOpenTelemetryConfigurator()
         aws_otel_configurator.configure()
         cls.tracer_provider: TracerProvider = get_tracer_provider()
@@ -249,3 +263,40 @@ def test_customize_span_processors(self):
         second_processor: SpanProcessor = mock_tracer_provider.add_span_processor.call_args_list[1].args[0]
         self.assertIsInstance(second_processor, AwsSpanMetricsProcessor)
         os.environ.pop("OTEL_AWS_APPLICATION_SIGNALS_ENABLED", None)
+
+    def test_application_signals_exporter_provider(self):
+        # Check default protocol - HTTP, as specified by AwsOpenTelemetryDistro.
+        exporter: OTLPMetricExporterMixin = ApplicationSignalsExporterProvider().create_exporter()
+        self.assertIsInstance(exporter, OTLPHttpOTLPMetricExporter)
+        self.assertEqual("http://localhost:4316/v1/metrics", exporter._endpoint)
+
+        # Overwrite protocol to gRPC.
+        os.environ["OTEL_EXPORTER_OTLP_PROTOCOL"] = "grpc"
+        exporter: SpanExporter = ApplicationSignalsExporterProvider().create_exporter()
+        self.assertIsInstance(exporter, OTLPGrpcOTLPMetricExporter)
+        self.assertEqual("localhost:4315", exporter._endpoint)
+
+        # Overwrite protocol back to HTTP.
+        os.environ["OTEL_EXPORTER_OTLP_PROTOCOL"] = "http/protobuf"
+        exporter: SpanExporter = ApplicationSignalsExporterProvider().create_exporter()
+        self.assertIsInstance(exporter, OTLPHttpOTLPMetricExporter)
+        self.assertEqual("http://localhost:4316/v1/metrics", exporter._endpoint)
+
+
+def validate_distro_environ():
+    tc: TestCase = TestCase()
+    # Set by OpenTelemetryDistro
+    tc.assertEqual("otlp", os.environ.get("OTEL_TRACES_EXPORTER"))
+    tc.assertEqual("otlp", os.environ.get("OTEL_METRICS_EXPORTER"))
+
+    # Set by AwsOpenTelemetryDistro
+    tc.assertEqual("http/protobuf", os.environ.get("OTEL_EXPORTER_OTLP_PROTOCOL"))
+    tc.assertEqual(
+        "base2_exponential_bucket_histogram", os.environ.get("OTEL_EXPORTER_OTLP_METRICS_DEFAULT_HISTOGRAM_AGGREGATION")
+    )
+    tc.assertEqual("xray,tracecontext,b3,b3multi", os.environ.get("OTEL_PROPAGATORS"))
+    tc.assertEqual("xray", os.environ.get("OTEL_PYTHON_ID_GENERATOR"))
+
+    # Not set
+    tc.assertEqual(None, os.environ.get("OTEL_TRACES_SAMPLER"))
+    tc.assertEqual(None, os.environ.get("OTEL_TRACES_SAMPLER_ARG"))

contract-tests/tests/test/amazon/base/contract_test_base.py

@@ -90,6 +90,7 @@ def setUp(self) -> None:
             .with_env("OTEL_METRIC_EXPORT_INTERVAL", "50")
             .with_env("OTEL_AWS_APPLICATION_SIGNALS_ENABLED", "true")
             .with_env("OTEL_METRICS_EXPORTER", "none")
+            .with_env("OTEL_EXPORTER_OTLP_PROTOCOL", "grpc")
             .with_env("OTEL_BSP_SCHEDULE_DELAY", "1")
             .with_env("OTEL_AWS_APPLICATION_SIGNALS_EXPORTER_ENDPOINT", f"http://collector:{_MOCK_COLLECTOR_PORT}")
             .with_env("OTEL_EXPORTER_OTLP_TRACES_ENDPOINT", f"http://collector:{_MOCK_COLLECTOR_PORT}")