runtime/pprof: introduce "allocs" profile

The Go's heap profile contains four kinds of samples
(inuse_space, inuse_objects, alloc_space, and alloc_objects).
The pprof tool by default chooses the inuse_space (the bytes
of live, in-use objects). When analyzing the current memory
usage the choice of inuse_space as the default may be useful,
but in some cases, users are more interested in analyzing the
total allocation statistics throughout the program execution.
For example, when we analyze the memory profile from benchmark
or program test run, we are more likely interested in the whole
allocation history than the live heap snapshot at the end of
the test or benchmark.

The pprof tool provides flags to control which sample type
to be used for analysis. However, it is one of the less-known
features of pprof and we believe it's better to choose the
right type of samples as the default when producing the profile.

This CL introduces a new type of profile, "allocs", which is
the same as the "heap" profile but marks the alloc_space
as the default type unlike heap profiles that use inuse_space
as the default type.

'go test -memprofile=...' command is changed to use the new
"allocs" profile type instead of the traditional "heap" profile.

Fixes golang#24443

Change-Id: I012dd4b6dcacd45644d7345509936b8380b6fbd9
Reviewed-on: https://go-review.googlesource.com/102696
Run-TryBot: Hyang-Ah Hana Kim <[email protected]>
Reviewed-by: Russ Cox <[email protected]>

src/cmd/go/internal/test/test.go

@@ -331,14 +331,13 @@ profile the tests during execution:
 	    Writes test binary as -c would.
 
 	-memprofile mem.out
-	    Write a memory profile to the file after all tests have passed.
+	    Write an allocation profile to the file after all tests have passed.
 	    Writes test binary as -c would.
 
 	-memprofilerate n
-	    Enable more precise (and expensive) memory profiles by setting
-	    runtime.MemProfileRate. See 'go doc runtime.MemProfileRate'.
-	    To profile all memory allocations, use -test.memprofilerate=1
-	    and pass --alloc_space flag to the pprof tool.
+	    Enable more precise (and expensive) memory allocation profiles by
+	    setting runtime.MemProfileRate. See 'go doc runtime.MemProfileRate'.
+	    To profile all memory allocations, use -test.memprofilerate=1.
 
 	-mutexprofile mutex.out
 	    Write a mutex contention profile to the specified file

src/runtime/pprof/internal/profile/encode.go

@@ -197,6 +197,10 @@ var profileDecoder = []decoder{
 	},
 	// repeated int64 period = 12
 	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).Period) },
+	// repeated int64 comment = 13
+	func(b *buffer, m message) error { return decodeInt64s(b, &m.(*Profile).commentX) },
+	// int64 defaultSampleType = 14
+	func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).defaultSampleTypeX) },
 }
 
 // postDecode takes the unexported fields populated by decode (with
@@ -278,6 +282,14 @@ func (p *Profile) postDecode() error {
 		pt.Type, err = getString(p.stringTable, &pt.typeX, err)
 		pt.Unit, err = getString(p.stringTable, &pt.unitX, err)
 	}
+	for _, i := range p.commentX {
+		var c string
+		c, err = getString(p.stringTable, &i, err)
+		p.Comments = append(p.Comments, c)
+	}
+
+	p.commentX = nil
+	p.DefaultSampleType, err = getString(p.stringTable, &p.defaultSampleTypeX, err)
 	p.stringTable = nil
 	return nil
 }

src/runtime/pprof/internal/profile/profile.go

@@ -22,11 +22,13 @@ import (
 
 // Profile is an in-memory representation of profile.proto.
 type Profile struct {
-	SampleType []*ValueType
-	Sample     []*Sample
-	Mapping    []*Mapping
-	Location   []*Location
-	Function   []*Function
+	SampleType        []*ValueType
+	DefaultSampleType string
+	Sample            []*Sample
+	Mapping           []*Mapping
+	Location          []*Location
+	Function          []*Function
+	Comments          []string
 
 	DropFrames string
 	KeepFrames string
@@ -36,9 +38,11 @@ type Profile struct {
 	PeriodType    *ValueType
 	Period        int64
 
-	dropFramesX int64
-	keepFramesX int64
-	stringTable []string
+	commentX           []int64
+	dropFramesX        int64
+	keepFramesX        int64
+	stringTable        []string
+	defaultSampleTypeX int64
 }
 
 // ValueType corresponds to Profile.ValueType

src/runtime/pprof/pprof.go

@@ -99,7 +99,8 @@ import (
 // Each Profile has a unique name. A few profiles are predefined:
 //
 //	goroutine    - stack traces of all current goroutines
-//	heap         - a sampling of all heap allocations
+//	heap         - a sampling of memory allocations of live objects
+//	allocs       - a sampling of all past memory allocations
 //	threadcreate - stack traces that led to the creation of new OS threads
 //	block        - stack traces that led to blocking on synchronization primitives
 //	mutex        - stack traces of holders of contended mutexes
@@ -114,6 +115,16 @@ import (
 // all known allocations. This exception helps mainly in programs running
 // without garbage collection enabled, usually for debugging purposes.
 //
+// The heap profile tracks both the allocation sites for all live objects in
+// the application memory and for all objects allocated since the program start.
+// Pprof's -inuse_space, -inuse_objects, -alloc_space, and -alloc_objects
+// flags select which to display, defaulting to -inuse_space (live objects,
+// scaled by size).
+//
+// The allocs profile is the same as the heap profile but changes the default
+// pprof display to -alloc_space, the total number of bytes allocated since
+// the program began (including garbage-collected bytes).
+//
 // The CPU profile is not available as a Profile. It has a special API,
 // the StartCPUProfile and StopCPUProfile functions, because it streams
 // output to a writer during profiling.
@@ -150,6 +161,12 @@ var heapProfile = &Profile{
 	write: writeHeap,
 }
 
+var allocsProfile = &Profile{
+	name:  "allocs",
+	count: countHeap, // identical to heap profile
+	write: writeAlloc,
+}
+
 var blockProfile = &Profile{
 	name:  "block",
 	count: countBlock,
@@ -170,6 +187,7 @@ func lockProfiles() {
 			"goroutine":    goroutineProfile,
 			"threadcreate": threadcreateProfile,
 			"heap":         heapProfile,
+			"allocs":       allocsProfile,
 			"block":        blockProfile,
 			"mutex":        mutexProfile,
 		}
@@ -511,6 +529,16 @@ func countHeap() int {
 
 // writeHeap writes the current runtime heap profile to w.
 func writeHeap(w io.Writer, debug int) error {
+	return writeHeapInternal(w, debug, "")
+}
+
+// writeAlloc writes the current runtime heap profile to w
+// with the total allocation space as the default sample type.
+func writeAlloc(w io.Writer, debug int) error {
+	return writeHeapInternal(w, debug, "alloc_space")
+}
+
+func writeHeapInternal(w io.Writer, debug int, defaultSampleType string) error {
 	var memStats *runtime.MemStats
 	if debug != 0 {
 		// Read mem stats first, so that our other allocations
@@ -541,7 +569,7 @@ func writeHeap(w io.Writer, debug int) error {
 	}
 
 	if debug == 0 {
-		return writeHeapProto(w, p, int64(runtime.MemProfileRate))
+		return writeHeapProto(w, p, int64(runtime.MemProfileRate), defaultSampleType)
 	}
 
 	sort.Slice(p, func(i, j int) bool { return p[i].InUseBytes() > p[j].InUseBytes() })

src/runtime/pprof/proto.go

@@ -54,18 +54,20 @@ type memMap struct {
 
 const (
 	// message Profile
-	tagProfile_SampleType    = 1  // repeated ValueType
-	tagProfile_Sample        = 2  // repeated Sample
-	tagProfile_Mapping       = 3  // repeated Mapping
-	tagProfile_Location      = 4  // repeated Location
-	tagProfile_Function      = 5  // repeated Function
-	tagProfile_StringTable   = 6  // repeated string
-	tagProfile_DropFrames    = 7  // int64 (string table index)
-	tagProfile_KeepFrames    = 8  // int64 (string table index)
-	tagProfile_TimeNanos     = 9  // int64
-	tagProfile_DurationNanos = 10 // int64
-	tagProfile_PeriodType    = 11 // ValueType (really optional string???)
-	tagProfile_Period        = 12 // int64
+	tagProfile_SampleType        = 1  // repeated ValueType
+	tagProfile_Sample            = 2  // repeated Sample
+	tagProfile_Mapping           = 3  // repeated Mapping
+	tagProfile_Location          = 4  // repeated Location
+	tagProfile_Function          = 5  // repeated Function
+	tagProfile_StringTable       = 6  // repeated string
+	tagProfile_DropFrames        = 7  // int64 (string table index)
+	tagProfile_KeepFrames        = 8  // int64 (string table index)
+	tagProfile_TimeNanos         = 9  // int64
+	tagProfile_DurationNanos     = 10 // int64
+	tagProfile_PeriodType        = 11 // ValueType (really optional string???)
+	tagProfile_Period            = 12 // int64
+	tagProfile_Comment           = 13 // repeated int64
+	tagProfile_DefaultSampleType = 14 // int64
 
 	// message ValueType
 	tagValueType_Type = 1 // int64 (string table index)

src/runtime/pprof/proto_test.go

@@ -63,7 +63,7 @@ func TestConvertCPUProfileEmpty(t *testing.T) {
 		{Type: "cpu", Unit: "nanoseconds"},
 	}
 
-	checkProfile(t, p, 2000*1000, periodType, sampleType, nil)
+	checkProfile(t, p, 2000*1000, periodType, sampleType, nil, "")
 }
 
 func f1() { f1() }
@@ -130,18 +130,23 @@ func TestConvertCPUProfile(t *testing.T) {
 			{ID: 4, Mapping: map2, Address: addr2 + 1},
 		}},
 	}
-	checkProfile(t, p, period, periodType, sampleType, samples)
+	checkProfile(t, p, period, periodType, sampleType, samples, "")
 }
 
-func checkProfile(t *testing.T, p *profile.Profile, period int64, periodType *profile.ValueType, sampleType []*profile.ValueType, samples []*profile.Sample) {
+func checkProfile(t *testing.T, p *profile.Profile, period int64, periodType *profile.ValueType, sampleType []*profile.ValueType, samples []*profile.Sample, defaultSampleType string) {
+	t.Helper()
+
 	if p.Period != period {
-		t.Fatalf("p.Period = %d, want %d", p.Period, period)
+		t.Errorf("p.Period = %d, want %d", p.Period, period)
 	}
 	if !reflect.DeepEqual(p.PeriodType, periodType) {
-		t.Fatalf("p.PeriodType = %v\nwant = %v", fmtJSON(p.PeriodType), fmtJSON(periodType))
+		t.Errorf("p.PeriodType = %v\nwant = %v", fmtJSON(p.PeriodType), fmtJSON(periodType))
 	}
 	if !reflect.DeepEqual(p.SampleType, sampleType) {
-		t.Fatalf("p.SampleType = %v\nwant = %v", fmtJSON(p.SampleType), fmtJSON(sampleType))
+		t.Errorf("p.SampleType = %v\nwant = %v", fmtJSON(p.SampleType), fmtJSON(sampleType))
+	}
+	if defaultSampleType != p.DefaultSampleType {
+		t.Errorf("p.DefaultSampleType = %v\nwant = %v", p.DefaultSampleType, defaultSampleType)
 	}
 	// Clear line info since it is not in the expected samples.
 	// If we used f1 and f2 above, then the samples will have line info.

src/runtime/pprof/protomem.go

@@ -12,14 +12,17 @@ import (
 )
 
 // writeHeapProto writes the current heap profile in protobuf format to w.
-func writeHeapProto(w io.Writer, p []runtime.MemProfileRecord, rate int64) error {
+func writeHeapProto(w io.Writer, p []runtime.MemProfileRecord, rate int64, defaultSampleType string) error {
 	b := newProfileBuilder(w)
 	b.pbValueType(tagProfile_PeriodType, "space", "bytes")
 	b.pb.int64Opt(tagProfile_Period, rate)
 	b.pbValueType(tagProfile_SampleType, "alloc_objects", "count")
 	b.pbValueType(tagProfile_SampleType, "alloc_space", "bytes")
 	b.pbValueType(tagProfile_SampleType, "inuse_objects", "count")
 	b.pbValueType(tagProfile_SampleType, "inuse_space", "bytes")
+	if defaultSampleType != "" {
+		b.pb.int64Opt(tagProfile_DefaultSampleType, b.stringIndex(defaultSampleType))
+	}
 
 	values := []int64{0, 0, 0, 0}
 	var locs []uint64

src/runtime/pprof/protomem_test.go

@@ -14,7 +14,6 @@ import (
 func TestConvertMemProfile(t *testing.T) {
 	addr1, addr2, map1, map2 := testPCs(t)
 
-	var buf bytes.Buffer
 	// MemProfileRecord stacks are return PCs, so add one to the
 	// addresses recorded in the "profile". The proto profile
 	// locations are call PCs, so conversion will subtract one
@@ -27,15 +26,6 @@ func TestConvertMemProfile(t *testing.T) {
 		{AllocBytes: 512 * 1024, FreeBytes: 512 * 1024, AllocObjects: 1, FreeObjects: 1, Stack0: [32]uintptr{a1 + 1, a1 + 2, a2 + 3}},
 	}
 
-	if err := writeHeapProto(&buf, rec, rate); err != nil {
-		t.Fatalf("writing profile: %v", err)
-	}
-
-	p, err := profile.Parse(&buf)
-	if err != nil {
-		t.Fatalf("profile.Parse: %v", err)
-	}
-
 	periodType := &profile.ValueType{Type: "space", Unit: "bytes"}
 	sampleType := []*profile.ValueType{
 		{Type: "alloc_objects", Unit: "count"},
@@ -70,5 +60,25 @@ func TestConvertMemProfile(t *testing.T) {
 			NumLabel: map[string][]int64{"bytes": {829411}},
 		},
 	}
-	checkProfile(t, p, rate, periodType, sampleType, samples)
+	for _, tc := range []struct {
+		name              string
+		defaultSampleType string
+	}{
+		{"heap", ""},
+		{"allocs", "alloc_space"},
+	} {
+		t.Run(tc.name, func(t *testing.T) {
+			var buf bytes.Buffer
+			if err := writeHeapProto(&buf, rec, rate, tc.defaultSampleType); err != nil {
+				t.Fatalf("writing profile: %v", err)
+			}
+
+			p, err := profile.Parse(&buf)
+			if err != nil {
+				t.Fatalf("profile.Parse: %v", err)
+			}
+
+			checkProfile(t, p, rate, periodType, sampleType, samples, tc.defaultSampleType)
+		})
+	}
 }

src/testing/internal/testdeps/deps.go

@@ -46,10 +46,6 @@ func (TestDeps) StopCPUProfile() {
 	pprof.StopCPUProfile()
 }
 
-func (TestDeps) WriteHeapProfile(w io.Writer) error {
-	return pprof.WriteHeapProfile(w)
-}
-
 func (TestDeps) WriteProfileTo(name string, w io.Writer, debug int) error {
 	return pprof.Lookup(name).WriteTo(w, debug)
 }

src/testing/testing.go

@@ -260,8 +260,8 @@ var (
 	coverProfile         = flag.String("test.coverprofile", "", "write a coverage profile to `file`")
 	matchList            = flag.String("test.list", "", "list tests, examples, and benchmarks matching `regexp` then exit")
 	match                = flag.String("test.run", "", "run only tests and examples matching `regexp`")
-	memProfile           = flag.String("test.memprofile", "", "write a memory profile to `file`")
-	memProfileRate       = flag.Int("test.memprofilerate", 0, "set memory profiling `rate` (see runtime.MemProfileRate)")
+	memProfile           = flag.String("test.memprofile", "", "write an allocation profile to `file`")
+	memProfileRate       = flag.Int("test.memprofilerate", 0, "set memory allocation profiling `rate` (see runtime.MemProfileRate)")
 	cpuProfile           = flag.String("test.cpuprofile", "", "write a cpu profile to `file`")
 	blockProfile         = flag.String("test.blockprofile", "", "write a goroutine blocking profile to `file`")
 	blockProfileRate     = flag.Int("test.blockprofilerate", 1, "set blocking profile `rate` (see runtime.SetBlockProfileRate)")
@@ -909,7 +909,6 @@ type matchStringOnly func(pat, str string) (bool, error)
 func (f matchStringOnly) MatchString(pat, str string) (bool, error)   { return f(pat, str) }
 func (f matchStringOnly) StartCPUProfile(w io.Writer) error           { return errMain }
 func (f matchStringOnly) StopCPUProfile()                             {}
-func (f matchStringOnly) WriteHeapProfile(w io.Writer) error          { return errMain }
 func (f matchStringOnly) WriteProfileTo(string, io.Writer, int) error { return errMain }
 func (f matchStringOnly) ImportPath() string                          { return "" }
 func (f matchStringOnly) StartTestLog(io.Writer)                      {}
@@ -949,7 +948,6 @@ type testDeps interface {
 	StopCPUProfile()
 	StartTestLog(io.Writer)
 	StopTestLog() error
-	WriteHeapProfile(io.Writer) error
 	WriteProfileTo(string, io.Writer, int) error
 }
 
@@ -1188,7 +1186,7 @@ func (m *M) writeProfiles() {
 			os.Exit(2)
 		}
 		runtime.GC() // materialize all statistics
-		if err = m.deps.WriteHeapProfile(f); err != nil {
+		if err = m.deps.WriteProfileTo("allocs", f, 0); err != nil {
 			fmt.Fprintf(os.Stderr, "testing: can't write %s: %s\n", *memProfile, err)
 			os.Exit(2)
 		}