osbs opentelemetry

atomic_reactor/constants.py

@@ -76,6 +76,8 @@
 IMAGE_TYPE_OCI = 'oci'
 IMAGE_TYPE_OCI_TAR = 'oci-tar'
 
+OTEL_SERVICE_NAME = 'osbs'
+
 PLUGIN_KOJI_PROMOTE_PLUGIN_KEY = 'koji_promote'
 PLUGIN_KOJI_IMPORT_PLUGIN_KEY = 'koji_import'
 PLUGIN_KOJI_IMPORT_SOURCE_CONTAINER_PLUGIN_KEY = 'koji_import_source_container'

atomic_reactor/schemas/source_containers_user_params.json

@@ -10,6 +10,14 @@
     "kind": {"type": "string"},
     "koji_target": {"type": "string"},
     "koji_task_id": {"type": "integer"},
+    "opentelemetry_info": {
+      "type":  "object",
+      "properties": {
+        "traceparent": {"type":  ["string", "null"]},
+        "otel_url": {"type":  ["string", "null"]}
+      },
+      "additionalProperties": false
+    },
     "reactor_config_map": {"type": "string"},
     "scratch": {"type": "boolean"},
     "signing_intent": {"type": "string"},

atomic_reactor/schemas/user_params.json

@@ -28,6 +28,14 @@
     "koji_task_id": {"type": "integer"},
     "name": {"type": "string"},
     "operator_csv_modifications_url": {"type": "string"},
+    "opentelemetry_info": {
+      "type":  "object",
+      "properties": {
+        "traceparent": {"type":  ["string", "null"]},
+        "otel_url": {"type":  ["string", "null"]}
+      },
+      "additionalProperties": false
+    },
     "platform": {"type": "string"},
     "platforms": {
       "type": "array",

atomic_reactor/tasks/binary_container_build.py

@@ -17,11 +17,12 @@
 from json import JSONDecodeError
 
 from osbs.utils import ImageName
+from otel_extensions import instrumented, get_tracer
 
 from atomic_reactor import dirs
 from atomic_reactor import util
-from atomic_reactor.constants import REMOTE_HOST_MAX_RETRIES, REMOTE_HOST_RETRY_INTERVAL
-
+from atomic_reactor.constants import (REMOTE_HOST_MAX_RETRIES, REMOTE_HOST_RETRY_INTERVAL,
+                                      OTEL_SERVICE_NAME)
 from atomic_reactor.tasks.common import Task, TaskParams
 from atomic_reactor.utils import retries
 from atomic_reactor.utils import remote_host
@@ -103,23 +104,32 @@ def execute(self) -> Any:
             podman_remote = PodmanRemote.setup_for(
                 remote_resource, registries_authfile=get_authfile_path(config.registry)
             )
+            module_name = self.task_name + '_' + platform
+            tracer = get_tracer(module_name=module_name, service_name=OTEL_SERVICE_NAME)
+            with tracer.start_as_current_span("build_container") as span:
+                span.set_attribute('git_ref', self._params.user_params.get('git_ref'))
+                span.set_attribute('git_uri', self._params.user_params.get('git_uri'))
+                span.set_attribute('git_commit',
+                                   self._params.source.provider_params.get('git_commit', None))
+                koji_task_id = self._params.user_params.get('koji_task_id', None)
+                if koji_task_id:
+                    span.set_attribute('koji_task_id', koji_task_id)
+                # log the image+host for auditing purposes
+                logger.info("Building image=%s on host=%s", dest_tag, remote_resource.host.hostname)
+
+                output_lines = podman_remote.build_container(
+                    build_dir=build_dir,
+                    build_args=self.workflow_data.buildargs,
+                    dest_tag=dest_tag,
+                    flatpak=flatpak,
+                    memory_limit=config.remote_hosts.get("memory_limit"),
+                    podman_capabilities=config.remote_hosts.get("podman_capabilities")
+                )
+                for line in output_lines:
+                    logger.info(line.rstrip())
+                    build_log_file.write(line)
 
-            # log the image+host for auditing purposes
-            logger.info("Building image=%s on host=%s", dest_tag, remote_resource.host.hostname)
-
-            output_lines = podman_remote.build_container(
-                build_dir=build_dir,
-                build_args=self.workflow_data.buildargs,
-                dest_tag=dest_tag,
-                flatpak=flatpak,
-                memory_limit=config.remote_hosts.get("memory_limit"),
-                podman_capabilities=config.remote_hosts.get("podman_capabilities")
-            )
-            for line in output_lines:
-                logger.info(line.rstrip())
-                build_log_file.write(line)
-
-            logger.info("Build finished successfully! Pushing image to %s", dest_tag)
+                logger.info("Build finished successfully! Pushing image to %s", dest_tag)
 
             image_size_limit = config.image_size_limit['binary_image']
             image_size = podman_remote.get_image_size(dest_tag)
@@ -135,6 +145,7 @@ def execute(self) -> Any:
 
             return remote_resource.host.hostname
 
+    @instrumented
     def acquire_remote_resource(self, remote_hosts_config: dict) -> remote_host.LockedResource:
         """Lock a build slot on a remote host."""
         logger.info("Acquiring a build slot on a remote host")
@@ -310,6 +321,7 @@ def build_container(
         else:
             yield last_line
 
+    @instrumented
     def get_image_size(self, dest_tag: ImageName) -> int:
         inspect_cmd = [*self._podman_remote_cmd, 'image', 'inspect', str(dest_tag)]
         try:
@@ -328,6 +340,7 @@ def get_image_size(self, dest_tag: ImageName) -> int:
                                f"{str(dest_tag)}") from e
         return image_size
 
+    @instrumented
     def push_container(self, dest_tag: ImageName, *, insecure: bool = False) -> None:
         """Push the built container (named dest_tag) to the registry (as dest_tag).
 

atomic_reactor/tasks/common.py

@@ -7,20 +7,28 @@
 """
 
 import abc
-import signal
 import json
+import logging
+import os
+import signal
 from dataclasses import dataclass
+from functools import cached_property
 from pathlib import Path
 from typing import Dict, Any, ClassVar, Generic, TypeVar, Optional
-from functools import cached_property
+
+from opentelemetry.instrumentation.requests import RequestsInstrumentor
+from otel_extensions import TelemetryOptions, init_telemetry_provider, get_tracer
 
 from atomic_reactor import config
 from atomic_reactor import dirs
 from atomic_reactor import inner
 from atomic_reactor import source
 from atomic_reactor import util
+from atomic_reactor.constants import OTEL_SERVICE_NAME
 from atomic_reactor.plugin import TaskCanceledException
 
+logger = logging.getLogger(__name__)
+
 
 def write_task_result(output_file, msg):
     with open(output_file, 'w') as f:
@@ -88,6 +96,7 @@ class Task(abc.ABC, Generic[ParamsT]):
     ignore_sigterm: ClassVar[bool] = False
     # Automatically save context data before exiting? (Note: do not use for parallel tasks)
     autosave_context_data: ClassVar[bool] = True
+    task_name = 'default'
 
     def __init__(self, params: ParamsT):
         """Initialize a Task."""
@@ -122,7 +131,35 @@ def run(self, *args, **kwargs):
             else:
                 signal.signal(signal.SIGTERM, self.throw_task_canceled_exception)
 
-            result = self.execute(*args, **kwargs)
+            opentelemetry_info = self._params.user_params.get('opentelemetry_info', {})
+            traceparent = opentelemetry_info.get('traceparent', None)
+            otel_url = opentelemetry_info.get('otel_url', None)
+
+            span_exporter = ''
+            otel_protocol = 'http/protobuf'
+            if not otel_url:
+                otel_protocol = 'custom'
+                span_exporter = '"opentelemetry.sdk.trace.export.ConsoleSpanExporter"'
+
+            if traceparent:
+                os.environ['TRACEPARENT'] = traceparent
+            logger.info('traceparent is set to %s', traceparent)
+            otel_options = TelemetryOptions(
+                OTEL_SERVICE_NAME=OTEL_SERVICE_NAME,
+                OTEL_EXPORTER_CUSTOM_SPAN_EXPORTER_TYPE=span_exporter,
+                OTEL_EXPORTER_OTLP_ENDPOINT=otel_url,
+                OTEL_EXPORTER_OTLP_PROTOCOL=otel_protocol,
+            )
+            init_telemetry_provider(otel_options)
+
+            RequestsInstrumentor().instrument()
+
+            span_name = self.task_name
+            if hasattr(self._params, 'platform'):
+                span_name += '_' + self._params.platform
+            tracer = get_tracer(module_name=span_name, service_name=OTEL_SERVICE_NAME)
+            with tracer.start_as_current_span(span_name):
+                result = self.execute(*args, **kwargs)
             if self._params.task_result:
                 write_task_result(self._params.task_result, json.dumps(result))
 

atomic_reactor/tasks/plugin_based.py

@@ -31,7 +31,6 @@ class PluginBasedTask(Task[ParamsT]):
     #   {"name": "add_filesystem", "args": {...}}
     # Refer to plugins.json schema for the details.
     plugins_conf: ClassVar[List[Dict[str, Any]]] = []
-    task_name = 'default'
 
     def prepare_workflow(self) -> inner.DockerBuildWorkflow:
         """Fully initialize the workflow instance to be used for running the list of plugins."""

otel-requirements.in

@@ -0,0 +1,8 @@
+calver
+hatchling
+opentelemetry-api
+opentelemetry-exporter-otlp
+opentelemetry-instrumentation-requests
+opentelemetry-sdk
+otel-extensions
+urllib3 < 2.0

otel-requirements.txt

@@ -0,0 +1,112 @@
+#
+# This file is autogenerated by pip-compile with Python 3.8
+# by the following command:
+#
+#    pip-compile --output-file=otel-requirements.txt otel-requirements.in
+#
+backoff==2.2.1
+    # via
+    #   opentelemetry-exporter-otlp-proto-grpc
+    #   opentelemetry-exporter-otlp-proto-http
+calver==2022.6.26
+    # via -r otel-requirements.in
+certifi==2023.7.22
+    # via requests
+charset-normalizer==3.2.0
+    # via requests
+deprecated==1.2.14
+    # via
+    #   opentelemetry-api
+    #   opentelemetry-exporter-otlp-proto-grpc
+    #   opentelemetry-exporter-otlp-proto-http
+editables==0.5
+    # via hatchling
+googleapis-common-protos==1.60.0
+    # via
+    #   opentelemetry-exporter-otlp-proto-grpc
+    #   opentelemetry-exporter-otlp-proto-http
+grpcio==1.57.0
+    # via opentelemetry-exporter-otlp-proto-grpc
+hatchling==1.18.0
+    # via -r otel-requirements.in
+idna==3.4
+    # via requests
+importlib-metadata==6.8.0
+    # via opentelemetry-api
+opentelemetry-api==1.19.0
+    # via
+    #   -r otel-requirements.in
+    #   opentelemetry-exporter-otlp-proto-grpc
+    #   opentelemetry-exporter-otlp-proto-http
+    #   opentelemetry-instrumentation
+    #   opentelemetry-instrumentation-requests
+    #   opentelemetry-sdk
+    #   otel-extensions
+opentelemetry-exporter-otlp==1.19.0
+    # via -r otel-requirements.in
+opentelemetry-exporter-otlp-proto-common==1.19.0
+    # via
+    #   opentelemetry-exporter-otlp-proto-grpc
+    #   opentelemetry-exporter-otlp-proto-http
+opentelemetry-exporter-otlp-proto-grpc==1.19.0
+    # via opentelemetry-exporter-otlp
+opentelemetry-exporter-otlp-proto-http==1.19.0
+    # via opentelemetry-exporter-otlp
+opentelemetry-instrumentation==0.40b0
+    # via opentelemetry-instrumentation-requests
+opentelemetry-instrumentation-requests==0.40b0
+    # via -r otel-requirements.in
+opentelemetry-proto==1.19.0
+    # via
+    #   opentelemetry-exporter-otlp-proto-common
+    #   opentelemetry-exporter-otlp-proto-grpc
+    #   opentelemetry-exporter-otlp-proto-http
+opentelemetry-sdk==1.19.0
+    # via
+    #   -r otel-requirements.in
+    #   opentelemetry-exporter-otlp-proto-grpc
+    #   opentelemetry-exporter-otlp-proto-http
+    #   otel-extensions
+opentelemetry-semantic-conventions==0.40b0
+    # via
+    #   opentelemetry-instrumentation-requests
+    #   opentelemetry-sdk
+opentelemetry-util-http==0.40b0
+    # via opentelemetry-instrumentation-requests
+otel-extensions==0.2.5
+    # via -r otel-requirements.in
+packaging==23.1
+    # via hatchling
+pathspec==0.11.2
+    # via hatchling
+pluggy==1.3.0
+    # via hatchling
+protobuf==4.24.2
+    # via
+    #   googleapis-common-protos
+    #   opentelemetry-proto
+pydantic==1.10.12
+    # via otel-extensions
+requests==2.31.0
+    # via opentelemetry-exporter-otlp-proto-http
+tomli==2.0.1
+    # via hatchling
+trove-classifiers==2023.8.7
+    # via hatchling
+typing-extensions==4.7.1
+    # via
+    #   opentelemetry-sdk
+    #   pydantic
+urllib3==1.26.16
+    # via
+    #   -r otel-requirements.in
+    #   requests
+wrapt==1.15.0
+    # via
+    #   deprecated
+    #   opentelemetry-instrumentation
+zipp==3.16.2
+    # via importlib-metadata
+
+# The following packages are considered to be unsafe in a requirements file:
+# setuptools