Fixup accumulation for categorical histograms

Author: dpalmeiro

configs/adaptive-performance-power-saving-v1.yaml

@@ -16,7 +16,7 @@ histograms:
 
   # CPU time per process type
   - metrics.labeled_counter.power_cpu_time_per_process_type_ms:
-      aggregate: percentiles
+      aggregate: sum
 
   # Page load metrics
   - metrics.timing_distribution.performance_pageload_fcp

lib/analysis.py

@@ -288,7 +288,7 @@ def createNumericalTemplate():
 
 
 def createCategoricalTemplate():
-    template = {"desc": "", "labels": [], "counts": [], "sum": 0}
+    template = {"desc": "", "labels": [], "counts": [], "sample_counts": [], "sum": 0}
     return template
 
 
@@ -424,8 +424,10 @@ def processCategoricalMetrics(self, categorical_metrics):
             # Apply unified categorical analysis
             labels = data["bins"]
             counts = data["counts"]
+            sample_counts = data.get("sample_counts", [])
             result_location["labels"] = labels
             result_location["counts"] = counts
+            result_location["sample_counts"] = sample_counts
             total = sum(counts)
             result_location["sum"] = total
 

lib/parser.py

@@ -271,8 +271,13 @@ def annotateHistograms(config: Dict[str, Any], probeIndex: Dict[str, Any]) -> No
                     sys.exit(1)
 
                 if aggregate == "sum":
-                    # Sum mode: single scalar value
-                    config["histograms"][hist]["kind"] = "scalar"
+                    # Sum mode
+                    # For labeled_counter: keep per-label sums (use categorical kind for display)
+                    # For quantity: single scalar value
+                    if schema["type"] == "labeled_counter":
+                        config["histograms"][hist]["kind"] = "categorical"
+                    else:
+                        config["histograms"][hist]["kind"] = "scalar"
                 else:
                     # Percentiles mode: calculate median, p75, p95
                     # For labeled_counter, use labeled_percentiles (per-label tables)
@@ -386,8 +391,13 @@ def annotateHistograms(config: Dict[str, Any], probeIndex: Dict[str, Any]) -> No
                     sys.exit(1)
 
                 if aggregate == "sum":
-                    # Sum mode: single scalar value
-                    config["histograms"][hist]["kind"] = "scalar"
+                    # Sum mode
+                    # For labeled_counter: keep per-label sums (use categorical kind for display)
+                    # For quantity: single scalar value
+                    if schema["type"] == "labeled_counter":
+                        config["histograms"][hist]["kind"] = "categorical"
+                    else:
+                        config["histograms"][hist]["kind"] = "scalar"
                 else:
                     # Percentiles mode: calculate median, p75, p95
                     # For labeled_counter, use labeled_percentiles (per-label tables)

lib/report.py

@@ -204,6 +204,17 @@ def createSummarySection(self):
 
                     # Generate summary for categorical metrics
                     if data_type == "categorical":
+                        # Check if this is a labeled_counter with sum aggregate
+                        is_sum_aggregate = False
+                        for hist_key in self.data.get("histograms", {}):
+                            hist_name = hist_key.split(".")[-1]
+                            if hist_name == metric:
+                                is_sum_aggregate = (
+                                    self.data["histograms"][hist_key].get("aggregate")
+                                    == "sum"
+                                )
+                                break
+
                         branches = []
                         for branch in self.data["branches"]:
                             branch_name = (
@@ -223,23 +234,66 @@ def createSummarySection(self):
                                 ][metric]
                                 rows = []
                                 n_labels = len(metric_data["labels"])
-                                for i in range(n_labels):
+
+                                # For labeled_counter with sum aggregate and many labels,
+                                # only show top 2-3 by absolute uplift
+                                if is_sum_aggregate and n_labels > 10:
+                                    # Create list of (index, label, uplift) tuples
+                                    label_uplifts = [
+                                        (
+                                            i,
+                                            metric_data["labels"][i],
+                                            metric_data["uplift"][i],
+                                        )
+                                        for i in range(n_labels)
+                                    ]
+                                    # Sort by absolute uplift descending
+                                    label_uplifts.sort(
+                                        key=lambda x: abs(x[2]), reverse=True
+                                    )
+                                    # Take top 3
+                                    label_uplifts = label_uplifts[:3]
+                                    indices_to_show = [x[0] for x in label_uplifts]
+                                else:
+                                    indices_to_show = range(n_labels)
+
+                                for i in indices_to_show:
                                     label = metric_data["labels"][i]
                                     uplift = metric_data["uplift"][i]
 
                                     # Skip small uplifts for enumerated histograms with many labels
-                                    if n_labels > 5 and abs(uplift) < 0.05:
+                                    if (
+                                        not is_sum_aggregate
+                                        and n_labels > 5
+                                        and abs(uplift) < 0.05
+                                    ):
                                         continue
 
                                     weight = "font-weight:normal;"
+                                    color = ""
                                     if abs(uplift) >= 10:
                                         effect = "Large"
                                         weight = "font-weight:bold;"
+                                        color = (
+                                            "color:red;"
+                                            if uplift > 0
+                                            else "color:green;"
+                                        )
                                     elif abs(uplift) >= 5:
                                         effect = "Medium"
                                         weight = "font-weight:bold;"
+                                        color = (
+                                            "color:red;"
+                                            if uplift > 0
+                                            else "color:green;"
+                                        )
                                     elif abs(uplift) >= 2:
                                         effect = "Small"
+                                        color = (
+                                            "color:red;"
+                                            if uplift > 0
+                                            else "color:green;"
+                                        )
                                     else:
                                         effect = "None"
 
@@ -255,6 +309,7 @@ def createSummarySection(self):
                                             "uplift": uplift_desc,
                                             "effect": effect,
                                             "weight": weight,
+                                            "color": color,
                                             "style": f"background:{row_background};",
                                         }
                                     )
@@ -1154,36 +1209,51 @@ def createCategoricalComparison(self, segment, metric, metric_type):
 
         n_elem = len(self.data[control][segment][metric_type][metric]["counts"])
 
-        # Get unit from metric config if available
+        # Get unit and aggregate type from metric config if available
         unit = ""
+        aggregate = None
         for hist_key in self.data.get("histograms", {}):
             hist_name = hist_key.split(".")[-1]
             if hist_name == metric:
                 unit = self.data["histograms"][hist_key].get("unit", "")
+                aggregate = self.data["histograms"][hist_key].get("aggregate")
                 break
-        if n_elem <= 10:
+        if n_elem <= 20:
+            # For metrics with <= 20 categories (like labeled counters), show all
             indices = set(range(0, n_elem))
+        else:
+            # For metrics with many categories, only show significant changes
+            for branch in self.data["branches"]:
+                if branch == control:
+                    continue
+                uplift = self.data[branch][segment][metric_type][metric]["uplift"]
 
-        for branch in self.data["branches"]:
-            if branch == control:
-                continue
-            uplift = self.data[branch][segment][metric_type][metric]["uplift"]
-
-            for i in range(len(uplift)):
-                # Show categories with > 1% change
-                if abs(uplift[i]) > 1:
-                    indices.add(i)
+                for i in range(len(uplift)):
+                    # Show categories with > 1% change
+                    if abs(uplift[i]) > 1:
+                        indices.add(i)
 
         datasets = []
         for branch in self.data["branches"]:
             counts_branch = [
                 float(self.data[branch][segment][metric_type][metric]["counts"][i])
                 for i in indices
             ]
+            sample_counts_branch = []
+            if "sample_counts" in self.data[branch][segment][metric_type][metric]:
+                sample_counts_list = self.data[branch][segment][metric_type][metric][
+                    "sample_counts"
+                ]
+                if sample_counts_list:
+                    sample_counts_branch = [
+                        float(sample_counts_list[i]) for i in indices
+                    ]
+
             datasets.append(
                 {
                     "branch": branch,
                     "counts": counts_branch,
+                    "sample_counts": sample_counts_branch,
                 }
             )
 
@@ -1217,6 +1287,19 @@ def createCategoricalComparison(self, segment, metric, metric_type):
                 else:
                     count_display = f"{count_num:,}"
 
+                # Get sample count if available
+                sample_count = None
+                if "sample_counts" in dataset and dataset["sample_counts"]:
+                    sample_count_num = int(dataset["sample_counts"][idx])
+                    if sample_count_num >= 1000000000:
+                        sample_count = f"{sample_count_num / 1000000000:.1f}B"
+                    elif sample_count_num >= 1000000:
+                        sample_count = f"{sample_count_num / 1000000:.1f}M"
+                    elif sample_count_num >= 1000:
+                        sample_count = f"{sample_count_num / 1000:.1f}K"
+                    else:
+                        sample_count = f"{sample_count_num:,}"
+
                 # Uplift only for non-control branches
                 uplift = None
                 if j > 0 and len(datasets) > 1 and "uplift" in datasets[1]:
@@ -1226,6 +1309,7 @@ def createCategoricalComparison(self, segment, metric, metric_type):
                     {
                         "branch_name": dataset["branch"],
                         "count": count_display,
+                        "sample_count": sample_count,
                         "uplift": uplift,
                     }
                 )
@@ -1246,6 +1330,7 @@ def createCategoricalComparison(self, segment, metric, metric_type):
             "metric": metric,
             "segment": segment,
             "unit": unit,
+            "aggregate": aggregate,
         }
         self.doc(t.render(context))
 

lib/telemetry.py

@@ -468,16 +468,18 @@ def collectResultsFromQuery_OS_segments(
             df = histograms[histogram]
             if segment == "All":
                 subset = (
-                    df[df["branch"] == branch][["bucket", "counts"]]
+                    df[df["branch"] == branch][["bucket", "counts", "sample_count"]]
                     .groupby(["bucket"])
                     .sum()
                 )
                 buckets = list(subset.index)
                 counts = list(subset["counts"])
+                sample_counts = list(subset.get("sample_count", [0] * len(buckets)))
             else:
                 subset = df[(df["segment"] == segment) & (df["branch"] == branch)]
                 buckets = list(subset["bucket"])
                 counts = list(subset["counts"])
+                sample_counts = list(subset.get("sample_count", [0] * len(buckets)))
 
             # Some clients report bucket sizes that are not real, and these buckets
             # end up having 1-5 samples in them.  Filter these out entirely.
@@ -491,6 +493,8 @@ def collectResultsFromQuery_OS_segments(
                 for i in sorted(remove, reverse=True):
                     del buckets[i]
                     del counts[i]
+                    if sample_counts:
+                        del sample_counts[i]
 
             # Add labels to the buckets for categorical histograms.
             if self.config["histograms"][histogram]["kind"] == "categorical":
@@ -501,15 +505,23 @@ def collectResultsFromQuery_OS_segments(
                     if len(labels) == (len(buckets) - 1) and counts[-1] == 0:
                         del buckets[-1]
                         del counts[-1]
+                        if sample_counts:
+                            del sample_counts[-1]
 
                     # Create a bucket->count mapping from query results
                     bucket_to_count = dict(zip(buckets, counts))
+                    bucket_to_sample_count = (
+                        dict(zip(buckets, sample_counts)) if sample_counts else {}
+                    )
 
                     # Remap counts to match label order, filling missing labels with 0
                     new_counts = []
+                    new_sample_counts = []
                     for label in labels:
                         new_counts.append(bucket_to_count.get(label, 0))
+                        new_sample_counts.append(bucket_to_sample_count.get(label, 0))
                     counts = new_counts
+                    sample_counts = new_sample_counts
 
                     # Use labels as bucket names
                     buckets = labels
@@ -520,20 +532,33 @@ def collectResultsFromQuery_OS_segments(
                 maxBucket = self.config["histograms"][histogram]["max"]
                 remove = []
                 maxBucketCount = 0
+                maxBucketSampleCount = 0
                 for i, x in enumerate(buckets):
                     if x >= maxBucket:
                         remove.append(i)
                         maxBucketCount = maxBucketCount + counts[i]
+                        if sample_counts:
+                            maxBucketSampleCount = (
+                                maxBucketSampleCount + sample_counts[i]
+                            )
                 for i in sorted(remove, reverse=True):
                     del buckets[i]
                     del counts[i]
+                    if sample_counts:
+                        del sample_counts[i]
                 buckets.append(maxBucket)
                 counts.append(maxBucketCount)
+                if sample_counts:
+                    sample_counts.append(maxBucketSampleCount)
 
             assert len(buckets) == len(counts)
             results[branch][segment]["histograms"][histogram] = {}
             results[branch][segment]["histograms"][histogram]["bins"] = buckets
             results[branch][segment]["histograms"][histogram]["counts"] = counts
+            if sample_counts:
+                results[branch][segment]["histograms"][histogram][
+                    "sample_counts"
+                ] = sample_counts
             print(f"    segment={segment} len(histogram: {histogram}) = ", len(buckets))
 
         for metric in self.config["pageload_event_metrics"]:

lib/templates/html/categorical.html

@@ -40,7 +40,7 @@
           grace: '50%',
           beginAtZero: true,
           title: {
-            text: "Count{% if unit %} ({{unit}}){% endif %}",
+            text: "{% if aggregate == 'sum' %}Sum{% else %}Count{% endif %}{% if unit %} ({{unit}}){% endif %}",
             display: true
           }
         },
@@ -57,6 +57,22 @@
   <div style="margin-bottom: 20px;">
     <h5 style="margin: 5px 0; color: #555; font-weight: bold;">{{row.label}}</h5>
     <table border="1" cellpadding="0" cellspacing="0" class="stat-table" style="table-layout: fixed; width: 100%;">
+{% if aggregate == 'sum' %}
+      <colgroup>
+        <col style="width: 30%;">
+        <col style="width: 20%;">
+        <col style="width: 30%;">
+        <col style="width: 20%;">
+      </colgroup>
+      <thead>
+        <tr>
+          <th style="font-size: 11px; padding: 6px;">Branch</th>
+          <th style="font-size: 11px; padding: 6px;">N</th>
+          <th style="font-size: 11px; padding: 6px;">Sum{% if unit %} ({{unit}}){% endif %}</th>
+          <th style="font-size: 11px; padding: 6px;">Uplift (%)</th>
+        </tr>
+      </thead>
+{% else %}
       <colgroup>
         <col style="width: 35%;">
         <col style="width: 30%;">
@@ -69,10 +85,14 @@ <h5 style="margin: 5px 0; color: #555; font-weight: bold;">{{row.label}}</h5>
           <th style="font-size: 11px; padding: 6px;">Uplift (%)</th>
         </tr>
       </thead>
+{% endif %}
       <tbody>
 {% for branch_row in row.branch_rows %}
         <tr style="background-color: {% if forloop.counter0|divisibleby:2 %}#f9f9f9{% else %}#ffffff{% endif %};">
           <td style="text-align: left; padding: 6px; font-size: 13px;">{{branch_row.branch_name}}</td>
+{% if aggregate == 'sum' %}
+          <td style="padding: 6px; font-size: 13px;">{% if branch_row.sample_count %}{{branch_row.sample_count}}{% else %}-{% endif %}</td>
+{% endif %}
           <td style="padding: 6px; font-size: 13px;">{{branch_row.count}}</td>
           <td style="padding: 6px; font-size: 13px;{% if branch_row.uplift is not none and branch_row.uplift >= 10 %} background-color: #ffcdd2; color: #c62828;{% elif branch_row.uplift is not none and branch_row.uplift <= -10 %} background-color: #c8e6c9; color: #2e7d32;{% endif %}">
 {% if branch_row.uplift is not none %}