chore(stat-detectors): Convert RCA to use p95 span time (#57409)

We want to use p95 instead of average because that's what the regression
issue is using. Also adds types for the data structures

src/sentry/api/endpoints/organization_events_root_cause_analysis.py

@@ -21,7 +21,7 @@ SNUBA_QUERY_LIMIT = 10000
 def query_spans(transaction, regression_breakpoint, params):
     selected_columns = [
         "count(span_id) as span_count",
-        "sumArray(spans_exclusive_time) as total_span_self_time",
+        "percentileArray(spans_exclusive_time, 0.95) as p95_self_time",
         "array_join(spans_op) as span_op",
         "array_join(spans_group) as span_group",
         # want a single event id to fetch from nodestore for the span description
@@ -34,7 +34,7 @@ def query_spans(transaction, regression_breakpoint, params):
         selected_columns=selected_columns,
         equations=[],
         query=f"transaction:{transaction}",
-        orderby=["span_op", "span_group", "total_span_self_time"],
+        orderby=["span_op", "span_group", "p95_self_time"],
         limit=SNUBA_QUERY_LIMIT,
         config=QueryBuilderConfig(
             auto_aggregations=True,

src/sentry/api/helpers/span_analysis.py

@@ -1,14 +1,23 @@
-# example data = list of dictionaries
+from typing import Any, List, TypedDict
 
-# data = [{'period': 1, 'k': 'browser', 'txn_count': 2978, 'sum_v': 2088325.678, 'count_v': 24069, 'avg_v': 86.76412306},
-# {'period': 1, 'k': 'http.client', 'txn_count': 2978, 'sum_v': 13444332.97, 'count_v': 78114, 'avg_v': 172.1116953},
-# {'period': 1, 'k': 'mark', 'txn_count': 2978, 'sum_v': 0, 'count_v': 5390, 'avg_v': 0},
-# {'period': 1, 'k': 'pageload', 'txn_count': 2978, 'sum_v': 1853836.58, 'count_v': 2978, 'avg_v': 622.5106045},
-# {'period': 1, 'k': 'paint', 'txn_count': 2978, 'sum_v': 0, 'count_v': 5151, 'avg_v': 0},
-# {'period': 1, 'k': 'resource.css', 'txn_count': 2978, 'sum_v': 449854.7821, 'count_v': 8926, 'avg_v': 50.39825029}]
 
+class Row(TypedDict):
+    span_op: str
+    span_group: str
+    transaction_count: int
+    p95_self_time: float
+    sample_event_id: str
+    span_count: int
+    period: str
 
-def span_analysis(data):
+
+class AugmentedData(Row):
+    span_key: str
+    relative_freq: float
+    score: float
+
+
+def span_analysis(data: List[Row]):
 
     # create a unique identifier for each span
     span_keys = [row["span_op"] + "," + row["span_group"] for row in data]
@@ -16,33 +25,32 @@ def span_analysis(data):
     # number of occurrences of a span/transaction
     count_col = [row["span_count"] for row in data]
     txn_count = [row["transaction_count"] for row in data]
+    p95_self_time = [row["p95_self_time"] for row in data]
 
-    # total self time of a span
-    sum_col = [row["total_span_self_time"] for row in data]
-
-    # add in three new fields
+    # add in two new fields
     # 1. relative freq - the avg number of times a span occurs per transaction
-    # 2. avg duration - average duration of a span (total self time / span count)
-    # 3. score - a nondescriptive metric to evaluate the span (relative freq * avg duration)
+    # 2. score - a nondescriptive metric to evaluate the span (relative freq * avg duration)
     relative_freq = [count_col[x] / txn_count[x] for x in range(len(count_col))]
-    avg_col = [sum_col[x] / count_col[x] for x in range(len(sum_col))]
-    score_col = [relative_freq[x] * avg_col[x] for x in range(len(relative_freq))]
-
-    for i in range(len(data)):
-        row = data[i]
-        row["relative_freq"] = relative_freq[i]
-        row["score"] = score_col[i]
-        row["avg_duration"] = avg_col[i]
-        row["span_key"] = span_keys[i]
+    score_col = [relative_freq[x] * p95_self_time[x] for x in range(len(relative_freq))]
+
+    data_frames: List[AugmentedData] = [
+        {
+            **data[i],  # type: ignore[misc]
+            "relative_freq": relative_freq[i],
+            "score": score_col[i],
+            "span_key": span_keys[i],
+        }
+        for i in range(len(data))
+    ]
 
     # create two dataframes for period 0 and 1 and keep only the same spans in both periods
-    span_data_p0 = {row["span_key"]: row for row in data if row["period"] == "before"}
-    span_data_p1 = {row["span_key"]: row for row in data if row["period"] == "after"}
+    span_data_p0 = {row["span_key"]: row for row in data_frames if row["period"] == "before"}
+    span_data_p1 = {row["span_key"]: row for row in data_frames if row["period"] == "after"}
 
     all_keys = set(span_data_p0.keys()).union(span_data_p1.keys())
 
     # merge the dataframes to do span analysis
-    problem_spans = []
+    problem_spans: List[Any] = []
 
     # Perform the join operation
     for key in all_keys:
@@ -53,27 +61,31 @@ def span_analysis(data):
         if row1 and row2:
             score_delta = row2["score"] - row1["score"]
             freq_delta = row2["relative_freq"] - row1["relative_freq"]
-            duration_delta = row2["avg_duration"] - row1["avg_duration"]
+            duration_delta = row2["p95_self_time"] - row1["p95_self_time"]
         elif row2:
             score_delta = row2["score"]
             freq_delta = row2["relative_freq"]
-            duration_delta = row2["avg_duration"]
+            duration_delta = row2["p95_self_time"]
             new_span = True
 
         # We're only interested in span changes if they positively impacted duration
         if score_delta > 0:
+            sample_event_id = row1 and row1["sample_event_id"] or row2 and row2["sample_event_id"]
+            if not sample_event_id:
+                continue
+
             problem_spans.append(
                 {
                     "span_op": key.split(",")[0],
                     "span_group": key.split(",")[1],
-                    "sample_event_id": row1["sample_event_id"] if row1 else row2["sample_event_id"],
+                    "sample_event_id": sample_event_id,
                     "score_delta": score_delta,
                     "freq_before": row1["relative_freq"] if row1 else 0,
-                    "freq_after": row2["relative_freq"],
+                    "freq_after": row2["relative_freq"] if row2 else 0,
                     "freq_delta": freq_delta,
                     "duration_delta": duration_delta,
-                    "duration_before": row1["avg_duration"] if row1 else 0,
-                    "duration_after": row2["avg_duration"],
+                    "duration_before": row1["p95_self_time"] if row1 else 0,
+                    "duration_after": row2["p95_self_time"] if row2 else 0,
                     "is_new_span": new_span,
                 }
             )

tests/sentry/api/endpoints/test_organization_root_cause_analysis.py

@@ -254,13 +254,13 @@ class OrganizationRootCauseAnalysisTest(MetricsAPIBaseTestCase):
                 "span_op": "django.middleware",
                 "span_group": "2b9cbb96dbf59baa",
                 "span_description": "middleware span",
-                "score_delta": 640.0,
+                "score_delta": 1578.0,
                 "freq_before": 1.0,
                 "freq_after": 3.0,
                 "freq_delta": 2.0,
-                "duration_delta": 173.33333333333334,
+                "duration_delta": 486.0,
                 "duration_before": 60.0,
-                "duration_after": 233.33333333333334,
+                "duration_after": 546.0,
                 "is_new_span": False,
             }
         ]