In the tags command, print percentages relative to profile total. (#929)

Before this change percentages were relative to the tag total and this sometimes causes confusion for users as it makes it look like the tag is set for all samples when it's not. Note that the percentage calculation as implemented does not attempt to be friendly to samples where a given tag has multiple values. Multiple values per tag per sample are discouraged and we should deprecate them. Maybe we should make pprof print a warning for such profiles.
google · Feb 8, 2025 · a522070 · a522070
1 parent fc31438
commit a522070
Show file tree

Hide file tree

Showing 8 changed files with 59 additions and 43 deletions.
diff --git a/internal/driver/driver_test.go b/internal/driver/driver_test.go
@@ -97,6 +97,7 @@ func TestParse(t *testing.T) {
 		{"weblist=line(1000|3000)$,addresses,flat", "cpu"},
 		{"tags,tagfocus=400kb:", "heap_request"},
 		{"tags,tagfocus=+400kb:", "heap_request"},
+		{"tags,relative_percentages,tagfocus=400kb:", "heap_request"},
 		{"dot", "long_name_funcs"},
 		{"text", "long_name_funcs"},
 	}

diff --git a/internal/driver/testdata/pprof.cpu.tags b/internal/driver/testdata/pprof.cpu.tags
@@ -1,13 +1,13 @@
- key1: Total 1.1s
-          1.0s (89.29%): tag1
-       100.0ms ( 8.93%): tag2
-        10.0ms ( 0.89%): tag3
-        10.0ms ( 0.89%): tag4
+ key1: Total 1.12s of 1.12s (  100%)
+          1s (89.29%): tag1
+       100ms ( 8.93%): tag2
+        10ms ( 0.89%): tag3
+        10ms ( 0.89%): tag4
 
- key2: Total 1.0s
-         1.0s (99.02%): tag1
-       10.0ms ( 0.98%): tag2
+ key2: Total 1.02s of 1.12s (91.07%)
+       1.01s (90.18%): tag1
+        10ms ( 0.89%): tag2
 
- key3: Total 100.0ms
-       100.0ms (  100%): tag2
+ key3: Total 100ms of 1.12s ( 8.93%)
+       100ms ( 8.93%): tag2
 
diff --git a/internal/driver/testdata/pprof.cpu.tags.focus.ignore b/internal/driver/testdata/pprof.cpu.tags.focus.ignore
@@ -1,6 +1,6 @@
- key1: Total 100.0ms
-       100.0ms (  100%): tag2
+ key1: Total 100ms of 1.12s ( 8.93%)
+       100ms ( 8.93%): tag2
 
- key3: Total 100.0ms
-       100.0ms (  100%): tag2
+ key3: Total 100ms of 1.12s ( 8.93%)
+       100ms ( 8.93%): tag2
 
diff --git a/internal/driver/testdata/pprof.heap.tags b/internal/driver/testdata/pprof.heap.tags
@@ -1,6 +1,6 @@
- bytes: Total 98.6MB
-          62.5MB (63.37%): 1.56MB
-          31.2MB (31.68%): 400kB
-           3.9MB ( 3.96%): 200kB
-        1000.0kB ( 0.99%): 100kB
+ bytes: Total 98.63MB of 98.63MB (  100%)
+        62.50MB (63.37%): 1.56MB
+        31.25MB (31.68%): 400kB
+         3.91MB ( 3.96%): 200kB
+         1000kB ( 0.99%): 100kB
 
diff --git a/internal/driver/testdata/pprof.heap.tags.unit b/internal/driver/testdata/pprof.heap.tags.unit
@@ -1,6 +1,6 @@
- bytes: Total 103424000.0B
-        65536000.0B (63.37%): 1638400B
-        32768000.0B (31.68%): 409600B
-         4096000.0B ( 3.96%): 204800B
-         1024000.0B ( 0.99%): 102400B
+ bytes: Total 103424000B of 103424000B (  100%)
+        65536000B (63.37%): 1638400B
+        32768000B (31.68%): 409600B
+         4096000B ( 3.96%): 204800B
+         1024000B ( 0.99%): 102400B
 
diff --git a/internal/driver/testdata/pprof.heap_request.relative_percentages.tags.focus b/internal/driver/testdata/pprof.heap_request.relative_percentages.tags.focus
@@ -0,0 +1,8 @@
+ bytes: Total 93.75MB of 93.75MB (  100%)
+        62.50MB (66.67%): 1.56MB
+        31.25MB (33.33%): 400kB
+
+ request: Total 93.75MB of 93.75MB (  100%)
+          62.50MB (66.67%): 1.56MB
+          31.25MB (33.33%): 400kB
+
diff --git a/internal/driver/testdata/pprof.heap_request.tags.focus b/internal/driver/testdata/pprof.heap_request.tags.focus
@@ -1,8 +1,8 @@
- bytes: Total 93.8MB
-        62.5MB (66.67%): 1.56MB
-        31.2MB (33.33%): 400kB
+ bytes: Total 93.75MB of 98.63MB (95.05%)
+        62.50MB (63.37%): 1.56MB
+        31.25MB (31.68%): 400kB
 
- request: Total 93.8MB
-          62.5MB (66.67%): 1.56MB
-          31.2MB (33.33%): 400kB
+ request: Total 93.75MB of 98.63MB (95.05%)
+          62.50MB (63.37%): 1.56MB
+          31.25MB (31.68%): 400kB
 
diff --git a/internal/report/report.go b/internal/report/report.go
@@ -699,21 +699,26 @@ func printTags(w io.Writer, rpt *Report) error {
 	p := rpt.prof
 
 	o := rpt.options
-	formatTag := func(v int64, key string) string {
-		return measurement.ScaledLabel(v, key, o.OutputUnit)
+	formatTag := func(v int64, unit string) string {
+		return measurement.ScaledLabel(v, unit, o.OutputUnit)
 	}
 
-	// Hashtable to keep accumulate tags as key,value,count.
+	// Accumulate tags as key,value,count.
 	tagMap := make(map[string]map[string]int64)
+	// Note that we assume single value per tag per sample. Multiple values are
+	// encodable in the format but are discouraged.
+	tagTotalMap := make(map[string]int64)
 	for _, s := range p.Sample {
+		sampleValue := o.SampleValue(s.Value)
 		for key, vals := range s.Label {
 			for _, val := range vals {
 				valueMap, ok := tagMap[key]
 				if !ok {
 					valueMap = make(map[string]int64)
 					tagMap[key] = valueMap
 				}
-				valueMap[val] += o.SampleValue(s.Value)
+				valueMap[val] += sampleValue
+				tagTotalMap[key] += sampleValue
 			}
 		}
 		for key, vals := range s.NumLabel {
@@ -725,7 +730,8 @@ func printTags(w io.Writer, rpt *Report) error {
 					valueMap = make(map[string]int64)
 					tagMap[key] = valueMap
 				}
-				valueMap[val] += o.SampleValue(s.Value)
+				valueMap[val] += sampleValue
+				tagTotalMap[key] += sampleValue
 			}
 		}
 	}
@@ -736,22 +742,23 @@ func printTags(w io.Writer, rpt *Report) error {
 	}
 	tabw := tabwriter.NewWriter(w, 0, 0, 1, ' ', tabwriter.AlignRight)
 	for _, tagKey := range graph.SortTags(tagKeys, true) {
-		var total int64
 		key := tagKey.Name
 		tags := make([]*graph.Tag, 0, len(tagMap[key]))
 		for t, c := range tagMap[key] {
-			total += c
 			tags = append(tags, &graph.Tag{Name: t, Flat: c})
 		}
 
-		f, u := measurement.Scale(total, o.SampleUnit, o.OutputUnit)
-		fmt.Fprintf(tabw, "%s:\t Total %.1f%s\n", key, f, u)
+		tagTotal, profileTotal := tagTotalMap[key], rpt.Total()
+		if profileTotal > 0 {
+			fmt.Fprintf(tabw, "%s:\t Total %s of %s (%s)\n", key, rpt.formatValue(tagTotal), rpt.formatValue(profileTotal), measurement.Percentage(tagTotal, profileTotal))
+		} else {
+			fmt.Fprintf(tabw, "%s:\t Total %s of %s\n", key, rpt.formatValue(tagTotal), rpt.formatValue(profileTotal))
+		}
 		for _, t := range graph.SortTags(tags, true) {
-			f, u := measurement.Scale(t.FlatValue(), o.SampleUnit, o.OutputUnit)
-			if total > 0 {
-				fmt.Fprintf(tabw, " \t%.1f%s (%s):\t %s\n", f, u, measurement.Percentage(t.FlatValue(), total), t.Name)
+			if profileTotal > 0 {
+				fmt.Fprintf(tabw, " \t%s (%s):\t %s\n", rpt.formatValue(t.FlatValue()), measurement.Percentage(t.FlatValue(), profileTotal), t.Name)
 			} else {
-				fmt.Fprintf(tabw, " \t%.1f%s:\t %s\n", f, u, t.Name)
+				fmt.Fprintf(tabw, " \t%s:\t %s\n", rpt.formatValue(t.FlatValue()), t.Name)
 			}
 		}
 		fmt.Fprintln(tabw)