trace parts of osbs-client calls

expecially tasks with http requests

Signed-off-by: Harsh Modi <hmodi@redhat.com>

osbs/api.py

@@ -27,7 +27,7 @@
 from osbs.utils.labels import Labels
 # import utils in this way, so that we can mock standalone functions with flexmock
 from osbs import utils
-from osbs.utils.otel import get_current_traceparent
+from osbs.utils.otel import get_current_traceparent, init_otel
 
 
 def _load_pipeline_from_template(pipeline_run_path, substitutions):
@@ -255,6 +255,8 @@ def _get_binary_container_pipeline_data(self, *, buildtime_limit, user_params,
 
     @osbsapi
     def create_binary_container_build(self, **kwargs):
+        init_otel(otel_url=self.os_conf.get_otel_url(),
+                  traceparent=kwargs.get('traceparent', None))
         return self.create_binary_container_pipeline_run(**kwargs)
 
     @osbsapi
@@ -305,6 +307,7 @@ def create_binary_container_pipeline_run(self,
             # if the traceparent is not updated then child call will
             # be linked to the parent of current call
             traceparent = get_current_traceparent()
+            logger.info("updating traceparent in kwargs")
             kwargs.update({
                 'traceparent': traceparent
             })
@@ -361,6 +364,8 @@ def _get_source_container_pipeline_data(self, *, user_params, pipeline_run_name)
 
     @osbsapi
     def create_source_container_build(self, **kwargs):
+        init_otel(otel_url=self.os_conf.get_otel_url(),
+                  traceparent=kwargs.get('traceparent', None))
         return self.create_source_container_pipeline_run(**kwargs)
 
     @osbsapi
@@ -387,6 +392,7 @@ def create_source_container_pipeline_run(self,
             # if the traceparent is not updated then child call will
             # be linked to the parent of current call
             traceparent = get_current_traceparent()
+            logger.info("updating traceparent in kwargs")
             kwargs.update({
                 'traceparent': traceparent
             })

osbs/cli/main.py

@@ -14,6 +14,10 @@
 
 import sys
 import argparse
+
+from opentelemetry.sdk.environment_variables import OTEL_SERVICE_NAME
+from otel_extensions import instrumented, get_tracer
+
 from osbs import set_logging
 from osbs.api import OSBS
 from osbs.conf import Configuration
@@ -22,6 +26,7 @@
 from osbs.exceptions import (OsbsNetworkException, OsbsException, OsbsAuthException,
                              OsbsResponseException)
 from osbs.utils import UserWarningsStore
+from osbs.utils.otel import init_otel
 
 logger = logging.getLogger('osbs')
 
@@ -139,21 +144,26 @@ def cmd_build(args):
     if osbs.os_conf.get_flatpak():
         build_kwargs['flatpak'] = True
 
-    pipeline_run = osbs.create_binary_container_pipeline_run(**build_kwargs)
-
-    print_output(pipeline_run, export_metadata_file=args.export_metadata_file)
-
-    return_val = -1
-
-    if pipeline_run.has_succeeded():
-        return_val = 0
-    cleanup_used_resources = osbs.os_conf.get_cleanup_used_resources()
-    if cleanup_used_resources and pipeline_run.data is not None:
-        try:
-            logger.info("pipeline run removed: %s", pipeline_run.remove_pipeline_run())
-        except OsbsResponseException:
-            logger.error("failed to remove pipeline run %s", pipeline_run.pipeline_run_name)
-            raise
+    init_otel(otel_url=os_conf.get_otel_url(),
+              traceparent=args.traceparent)
+    span_name = 'binary_build'
+    tracer = get_tracer(module_name=span_name, service_name=OTEL_SERVICE_NAME)
+    with tracer.start_as_current_span(span_name):
+        pipeline_run = osbs.create_binary_container_pipeline_run(**build_kwargs)
+
+        print_output(pipeline_run, export_metadata_file=args.export_metadata_file)
+
+        return_val = -1
+
+        if pipeline_run.has_succeeded():
+            return_val = 0
+        cleanup_used_resources = osbs.os_conf.get_cleanup_used_resources()
+        if cleanup_used_resources and pipeline_run.data is not None:
+            try:
+                logger.info("pipeline run removed: %s", pipeline_run.remove_pipeline_run())
+            except OsbsResponseException:
+                logger.error("failed to remove pipeline run %s", pipeline_run.pipeline_run_name)
+                raise
     return return_val
 
 
@@ -181,21 +191,26 @@ def cmd_build_source_container(args):
     if args.traceparent:
         build_kwargs['traceparent'] = args.traceparent
 
-    pipeline_run = osbs.create_source_container_pipeline_run(**build_kwargs)
-
-    print_output(pipeline_run, export_metadata_file=args.export_metadata_file)
-
-    return_val = -1
-
-    if pipeline_run.has_succeeded():
-        return_val = 0
-    cleanup_used_resources = osbs.os_conf.get_cleanup_used_resources()
-    if cleanup_used_resources and pipeline_run.data is not None:
-        try:
-            logger.info("pipeline run removed: %s", pipeline_run.remove_pipeline_run())
-        except OsbsResponseException:
-            logger.error("failed to remove pipeline run %s", pipeline_run.pipeline_run_name)
-            raise
+    init_otel(otel_url=os_conf.get_otel_url(),
+              traceparent=args.traceparent)
+    span_name = 'source_build'
+    tracer = get_tracer(module_name=span_name, service_name=OTEL_SERVICE_NAME)
+    with tracer.start_as_current_span(span_name):
+        pipeline_run = osbs.create_source_container_pipeline_run(**build_kwargs)
+
+        print_output(pipeline_run, export_metadata_file=args.export_metadata_file)
+
+        return_val = -1
+
+        if pipeline_run.has_succeeded():
+            return_val = 0
+        cleanup_used_resources = osbs.os_conf.get_cleanup_used_resources()
+        if cleanup_used_resources and pipeline_run.data is not None:
+            try:
+                logger.info("pipeline run removed: %s", pipeline_run.remove_pipeline_run())
+            except OsbsResponseException:
+                logger.error("failed to remove pipeline run %s", pipeline_run.pipeline_run_name)
+                raise
     return return_val
 
 
@@ -426,6 +441,7 @@ def main():
         set_logging(level=logging.INFO)
 
     return_value = -1
+
     try:
         return_value = args.func(args)
     except AttributeError:

osbs/constants.py

@@ -74,6 +74,8 @@
 # number of seconds to wait, before retrying on openshift not found
 OS_NOT_FOUND_MAX_WAIT = 1
 
+OTEL_SERVICE_NAME = "osbs"
+
 ISOLATED_RELEASE_FORMAT = re.compile(r'^\d+\.\d+(\..+)?$')
 RELEASE_LABEL_FORMAT = re.compile(r"""^\d+             # First character must be a digit
                                       ([._]?           # allow separators between groups

osbs/tekton.py

@@ -14,10 +14,11 @@
 from typing import Dict, List, Tuple, Callable, Any
 from datetime import datetime
 
+from otel_extensions import instrumented
 
 from osbs.exceptions import OsbsResponseException, OsbsAuthException, OsbsException
 from osbs.constants import (DEFAULT_NAMESPACE, SERVICEACCOUNT_SECRET, SERVICEACCOUNT_TOKEN,
-                            SERVICEACCOUNT_CACRT)
+                            SERVICEACCOUNT_CACRT, OTEL_SERVICE_NAME)
 from osbs.osbs_http import HttpSession
 from osbs.kerberos_ccache import kerberos_ccache_init
 from osbs.utils import retry_on_conflict
@@ -422,6 +423,7 @@ def pipeline_run_url(self):
             )
         return self._pipeline_run_url
 
+    @instrumented
     def start_pipeline_run(self):
         if not self.input_data:
             raise OsbsException("No input data provided for pipeline run to start")
@@ -445,6 +447,7 @@ def start_pipeline_run(self):
         )
         return response.json()
 
+    @instrumented
     def remove_pipeline_run(self):
         url = self.os.build_url(
             self.api_path,
@@ -458,6 +461,7 @@ def remove_pipeline_run(self):
         return response.json()
 
     @retry_on_conflict
+    @instrumented
     def cancel_pipeline_run(self):
         data = copy.deepcopy(self.minimal_data)
         data['spec']['status'] = 'CancelledRunFinally'
@@ -479,6 +483,7 @@ def cancel_pipeline_run(self):
             raise OsbsException(exc_msg)
         return response_json
 
+    @instrumented
     def get_info(self, wait=False):
         if wait:
             self.wait_for_start()
@@ -731,6 +736,7 @@ def load_result(result: Dict[str, str]) -> Tuple[str, Any]:
             name: value for name, value in map(load_result, pipeline_results) if value is not None
         }
 
+    @instrumented
     def wait_for_start(self):
         """
         https://tekton.dev/docs/pipelines/pipelineruns/#monitoring-execution-status
@@ -843,6 +849,7 @@ def _get_logs_stream(self):
                     except StopIteration:
                         del streaming_task_runs[pipeline_task_name]
 
+    @instrumented
     def get_logs(self, follow=False, wait=False):
         if wait or follow:
             return self._get_logs_stream()

osbs/utils/otel.py

@@ -20,6 +20,7 @@
 
 
 def init_otel(otel_url: Optional[str], traceparent: Optional[str]):
+    logger.info("Initializing otel with traceparent %s", traceparent)
     span_exporter = ''
     otel_protocol = 'http/protobuf'
     if not otel_url:
@@ -35,11 +36,15 @@ def init_otel(otel_url: Optional[str], traceparent: Optional[str]):
         OTEL_EXPORTER_OTLP_PROTOCOL=otel_protocol,
     )
     init_telemetry_provider(otel_options)
+    if 'TRACEPARENT' in os.environ:
+        del os.environ['TRACEPARENT']
     RequestsInstrumentor().instrument()
+    logger.info("Initialization complete")
 
 
-def get_current_traceparent(**kwargs):
+def get_current_traceparent():
     tracecontext = trace.get_current_span().get_span_context()
     traceparent = (f'00-{format_trace_id(tracecontext.trace_id)}-'
                    f'{format_span_id(tracecontext.span_id)}-01')
+    logger.info("current traceparent is %s", traceparent)
     return traceparent