Add attrs to partial gran spans & tracking metrics

newrelic/core/application.py

@@ -1368,12 +1368,6 @@ def harvest(self, shutdown=False, flexible=False):
                                 spans_sampled = spans.num_samples
                                 internal_count_metric("Supportability/SpanEvent/TotalEventsSeen", spans_seen)
                                 internal_count_metric("Supportability/SpanEvent/TotalEventsSent", spans_sampled)
-                                if configuration.distributed_tracing.sampler.partial_granularity.enabled:
-                                    internal_count_metric(
-                                        f"Supportability/Python/PartialGranularity/{configuration.distributed_tracing.sampler.partial_granularity.type}",
-                                        1,
-                                    )
-
                                 stats.reset_span_events()
 
                     # Send error events

newrelic/core/attribute.py

@@ -87,6 +87,9 @@
     "message.routingKey",
     "messaging.destination.name",
     "messaging.system",
+    "nr.durations",
+    "nr.ids",
+    "nr.pg",
     "peer.address",
     "peer.hostname",
     "request.headers.accept",
@@ -125,6 +128,8 @@
     "span.kind",
 }
 
+SPAN_ERROR_ATTRIBUTES = {"error.class", "error.message", "error.expected"}
+
 
 MAX_NUM_USER_ATTRIBUTES = 128
 MAX_ATTRIBUTE_LENGTH = 255

newrelic/core/node_mixin.py

@@ -79,71 +79,91 @@ def span_event(
         if not partial_granularity_sampled:
             # intrinsics, user attrs, agent attrs
             return [i_attrs, u_attrs, a_attrs]
-        else:
-            if ct_exit_spans is None:
-                ct_exit_spans = {}
-
-            partial_granularity_type = settings.distributed_tracing.sampler.partial_granularity.type
-            exit_span_attrs_present = attribute.SPAN_ENTITY_RELATIONSHIP_ATTRIBUTES & set(a_attrs)
-            # If this is the entry node or an LLM span always return it.
-            if i_attrs.get("nr.entryPoint") or i_attrs["name"].startswith("Llm/"):
-                if partial_granularity_type == "reduced":
-                    return [i_attrs, u_attrs, a_attrs]
-                else:
-                    return [i_attrs, {}, {}]
-            # If the span is not an exit span, skip it by returning None.
-            if not exit_span_attrs_present:
-                return None
-            # If the span is an exit span and we are in reduced mode (meaning no attribute dropping),
-            # just return the exit span as is.
+
+        if ct_exit_spans is None:
+            ct_exit_spans = {"instrumented": 0, "kept": 0, "dropped_ids": 0}
+        ct_exit_spans["instrumented"] += 1
+
+        partial_granularity_type = settings.distributed_tracing.sampler.partial_granularity.type
+        a_attrs_set = set(a_attrs)
+        exit_span_attrs_present = attribute.SPAN_ENTITY_RELATIONSHIP_ATTRIBUTES & a_attrs_set
+        exit_span_error_attrs_present = attribute.SPAN_ERROR_ATTRIBUTES & a_attrs_set
+        # If this is an entry span, add `nr.pg` to indicate transaction is partial
+        # granularity sampled.
+        if i_attrs.get("nr.entryPoint"):
+            a_attrs["nr.pg"] = True
+        # If this is the entry node or an LLM span always return it.
+        if i_attrs.get("nr.entryPoint") or i_attrs["name"].startswith("Llm/"):
+            ct_exit_spans["kept"] += 1
             if partial_granularity_type == "reduced":
                 return [i_attrs, u_attrs, a_attrs]
             else:
-                a_minimized_attrs = attr_class({key: a_attrs[key] for key in exit_span_attrs_present})
-                # If we are in essential mode return the span with minimized attributes.
-                if partial_granularity_type == "essential":
-                    return [i_attrs, {}, a_minimized_attrs]
-                # If the span is an exit span but span compression (compact) is enabled,
-                # we need to check for uniqueness before returning it.
-                # Combine all the entity relationship attr values into a string to be
-                # used as the hash to check for uniqueness.
-                span_attrs = "".join([str(a_minimized_attrs[key]) for key in exit_span_attrs_present])
-                new_exit_span = span_attrs not in ct_exit_spans
-                # If this is a new exit span, add it to the known ct_exit_spans and
-                # return it.
-                if new_exit_span:
-                    # nr.ids is the list of span guids that share this unqiue exit span.
-                    a_minimized_attrs["nr.ids"] = []
-                    a_minimized_attrs["nr.durations"] = self.duration
-                    ct_exit_spans[span_attrs] = [i_attrs, a_minimized_attrs]
-                    return [i_attrs, {}, a_minimized_attrs]
-                # If this is an exit span we've already seen, add it's guid to the list
-                # of ids on the seen span, compute the new duration & start time, and
-                # return None.
+                return [i_attrs, {}, {key: a_attrs.get(key) for key in exit_span_error_attrs_present | {"nr.pg"}}]
+        # If the span is not an exit span, skip it by returning None.
+        if not exit_span_attrs_present:
+            return None
+        # If the span is an exit span and we are in reduced mode (meaning no attribute dropping),
+        # just return the exit span as is.
+        if partial_granularity_type == "reduced":
+            ct_exit_spans["kept"] += 1
+            return [i_attrs, u_attrs, a_attrs]
+        else:
+            a_minimized_attrs = attr_class(
+                {key: a_attrs[key] for key in (exit_span_attrs_present | exit_span_error_attrs_present)}
+            )
+            # If we are in essential mode return the span with minimized attributes.
+            if partial_granularity_type == "essential":
+                ct_exit_spans["kept"] += 1
+                return [i_attrs, {}, a_minimized_attrs]
+            # If the span is an exit span but span compression (compact) is enabled,
+            # we need to check for uniqueness before returning it.
+            # Combine all the entity relationship attr values into a string to be
+            # used as the hash to check for uniqueness.
+            span_attrs = "".join([str(a_minimized_attrs[key]) for key in exit_span_attrs_present])
+            new_exit_span = span_attrs not in ct_exit_spans
+            # If this is a new exit span, add it to the known ct_exit_spans and
+            # return it.
+            if new_exit_span:
+                # nr.ids is the list of span guids that share this unqiue exit span.
+                a_minimized_attrs["nr.ids"] = []
+                a_minimized_attrs["nr.durations"] = self.duration
+                ct_exit_spans[span_attrs] = [i_attrs, a_minimized_attrs]
+                ct_exit_spans["kept"] += 1
+                return [i_attrs, {}, a_minimized_attrs]
+            # If this is an exit span we've already seen, add the error attributes
+            # (last occurring error takes precedence), add it's guid to the list
+            # of ids on the seen span, compute the new duration & start time, and
+            # return None.
+            ct_exit_spans[span_attrs][1].update(
+                attr_class({key: a_minimized_attrs[key] for key in exit_span_error_attrs_present})
+            )
+            # Max size for `nr.ids` = 1024. Max length = 63 (each span id is 16 bytes + 8 bytes for list type).
+            if len(ct_exit_spans[span_attrs][1]["nr.ids"]) < 63:
                 ct_exit_spans[span_attrs][1]["nr.ids"].append(self.guid)
-                # Max size for `nr.ids` = 1024. Max length = 63 (each span id is 16 bytes + 8 bytes for list type).
-                ct_exit_spans[span_attrs][1]["nr.ids"] = ct_exit_spans[span_attrs][1]["nr.ids"][:63]
-                # Compute the new start and end time for all compressed spans and use
-                # that to set the duration for all compressed spans.
-                current_start_time = ct_exit_spans[span_attrs][0]["timestamp"]
-                current_end_time = (
-                    ct_exit_spans[span_attrs][0]["timestamp"] / 1000 + ct_exit_spans[span_attrs][1]["nr.durations"]
-                )
-                new_start_time = i_attrs["timestamp"]
-                new_end_time = i_attrs["timestamp"] / 1000 + i_attrs["duration"]
-                set_start_time = min(new_start_time, current_start_time)
-                # If the new span starts after the old span's end time or the new span
-                # ends before the current span starts; add the durations.
-                if current_end_time < new_start_time / 1000 or new_end_time < current_start_time / 1000:
-                    set_duration = ct_exit_spans[span_attrs][1]["nr.durations"] + i_attrs["duration"]
-                # Otherwise, if the new and old span's overlap in time, use the newest
-                # end time and subtract the start time from it to calculate the new
-                # duration.
-                else:
-                    set_duration = max(current_end_time, new_end_time) - set_start_time / 1000
-                ct_exit_spans[span_attrs][0]["timestamp"] = set_start_time
-                ct_exit_spans[span_attrs][1]["nr.durations"] = set_duration
-                return None
+            else:
+                ct_exit_spans["dropped_ids"] += 1
+
+            ct_exit_spans[span_attrs][1]["nr.ids"] = ct_exit_spans[span_attrs][1]["nr.ids"][:63]
+            # Compute the new start and end time for all compressed spans and use
+            # that to set the duration for all compressed spans.
+            current_start_time = ct_exit_spans[span_attrs][0]["timestamp"]
+            current_end_time = (
+                ct_exit_spans[span_attrs][0]["timestamp"] / 1000 + ct_exit_spans[span_attrs][1]["nr.durations"]
+            )
+            new_start_time = i_attrs["timestamp"]
+            new_end_time = i_attrs["timestamp"] / 1000 + i_attrs["duration"]
+            set_start_time = min(new_start_time, current_start_time)
+            # If the new span starts after the old span's end time or the new span
+            # ends before the current span starts; add the durations.
+            if current_end_time < new_start_time / 1000 or new_end_time < current_start_time / 1000:
+                set_duration = ct_exit_spans[span_attrs][1]["nr.durations"] + i_attrs["duration"]
+            # Otherwise, if the new and old span's overlap in time, use the newest
+            # end time and subtract the start time from it to calculate the new
+            # duration.
+            else:
+                set_duration = max(current_end_time, new_end_time) - set_start_time / 1000
+            ct_exit_spans[span_attrs][0]["timestamp"] = set_start_time
+            ct_exit_spans[span_attrs][1]["nr.durations"] = set_duration
 
     def span_events(
         self,
@@ -162,13 +182,11 @@ def span_events(
             partial_granularity_sampled=partial_granularity_sampled,
             ct_exit_spans=ct_exit_spans,
         )
-        ct_exit_spans["instrumented"] += 1
         parent_id = parent_guid
         if span:  # span will be None if the span is an inprocess span or repeated exit span.
-            ct_exit_spans["kept"] += 1
             yield span
             # Compressed spans are always reparented onto the entry span.
-            if not settings.distributed_tracing.sampler.partial_granularity.type == "compact" or span[0].get(
+            if settings.distributed_tracing.sampler.partial_granularity.type != "compact" or span[0].get(
                 "nr.entryPoint"
             ):
                 parent_id = self.guid
@@ -181,9 +199,7 @@ def span_events(
                 partial_granularity_sampled=partial_granularity_sampled,
                 ct_exit_spans=ct_exit_spans,
             ):
-                ct_exit_spans["instrumented"] += 1
                 if event:  # event will be None if the span is an inprocess span or repeated exit span.
-                    ct_exit_spans["kept"] += 1
                     yield event
 
 

newrelic/core/stats_engine.py

@@ -1192,18 +1192,26 @@ def record_transaction(self, transaction):
                     self._span_events.add(event, priority=transaction.priority)
                 if transaction.partial_granularity_sampled:
                     partial_gran_type = settings.distributed_tracing.sampler.partial_granularity.type
-                    self.record_custom_metrics(
-                        [
-                            (
-                                f"Supportability/DistributedTrace/PartialGranularity/{partial_gran_type}/Span/Instrumented",
-                                {"count": transaction.instrumented},
-                            ),
-                            (
-                                f"Supportability/DistributedTrace/PartialGranularity/{partial_gran_type}/Span/Kept",
-                                {"count": transaction.kept},
-                            ),
-                        ]
+                    self.record_custom_metric(
+                        f"Supportability/Python/PartialGranularity/{partial_gran_type}", {"count": 1}
                     )
+                    instrumented = getattr(transaction, "instrumented", 0)
+                    if instrumented:
+                        self.record_custom_metric(
+                            f"Supportability/DistributedTrace/PartialGranularity/{partial_gran_type}/Span/Instrumented",
+                            {"count": instrumented},
+                        )
+                    kept = getattr(transaction, "kept", 0)
+                    if instrumented:
+                        self.record_custom_metric(
+                            f"Supportability/DistributedTrace/PartialGranularity/{partial_gran_type}/Span/Kept",
+                            {"count": kept},
+                        )
+                    dropped_ids = getattr(transaction, "dropped_ids", 0)
+                    if dropped_ids:
+                        self.record_custom_metric(
+                            "Supportability/Python/PartialGranularity/NrIds/Dropped", {"count": dropped_ids}
+                        )
 
         # Merge in log events
 

newrelic/core/transaction_node.py

@@ -634,7 +634,7 @@ def span_events(self, settings, attr_class=dict):
                 ("priority", self.priority),
             )
         )
-        ct_exit_spans = {"instrumented": 0, "kept": 0}
+        ct_exit_spans = {"instrumented": 0, "kept": 0, "dropped_ids": 0}
         yield from self.root.span_events(
             settings,
             base_attrs,
@@ -646,6 +646,8 @@ def span_events(self, settings, attr_class=dict):
         # If this transaction is partial granularity sampled, record the number of spans
         # instrumented and the number of spans kept to monitor cost savings of partial
         # granularity tracing.
+        # Also record the number of span ids dropped (fragmentation) in compact mode.
         if self.partial_granularity_sampled:
             self.instrumented = ct_exit_spans["instrumented"]
             self.kept = ct_exit_spans["kept"]
+            self.dropped_ids = ct_exit_spans["dropped_ids"]