// Copyright 2014 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package profile
import (
"errors"
"sort"
"strings"
)
func (p *Profile) decoder() []decoder {
return profileDecoder
}
// preEncode populates the unexported fields to be used by encode
// (with suffix X) from the corresponding exported fields. The
// exported fields are cleared up to facilitate testing.
func (p *Profile) preEncode() {
strings := make(map[string]int)
addString(strings, "")
for _, st := range p.SampleType {
st.typeX = addString(strings, st.Type)
st.unitX = addString(strings, st.Unit)
}
for _, s := range p.Sample {
s.labelX = nil
var keys []string
for k := range s.Label {
keys = append(keys, k)
}
sort.Strings(keys)
for _, k := range keys {
vs := s.Label[k]
for _, v := range vs {
s.labelX = append(s.labelX,
label{
keyX: addString(strings, k),
strX: addString(strings, v),
},
)
}
}
var numKeys []string
for k := range s.NumLabel {
numKeys = append(numKeys, k)
}
sort.Strings(numKeys)
for _, k := range numKeys {
keyX := addString(strings, k)
vs := s.NumLabel[k]
units := s.NumUnit[k]
for i, v := range vs {
var unitX int64
if len(units) != 0 {
unitX = addString(strings, units[i])
}
s.labelX = append(s.labelX,
label{
keyX: keyX,
numX: v,
unitX: unitX,
},
)
}
}
s.locationIDX = make([]uint64, len(s.Location))
for i, loc := range s.Location {
s.locationIDX[i] = loc.ID
}
}
for _, m := range p.Mapping {
m.fileX = addString(strings, m.File)
m.buildIDX = addString(strings, m.BuildID)
}
for _, l := range p.Location {
for i, ln := range l.Line {
if ln.Function != nil {
l.Line[i].functionIDX = ln.Function.ID
} else {
l.Line[i].functionIDX = 0
}
}
if l.Mapping != nil {
l.mappingIDX = l.Mapping.ID
} else {
l.mappingIDX = 0
}
}
for _, f := range p.Function {
f.nameX = addString(strings, f.Name)
f.systemNameX = addString(strings, f.SystemName)
f.filenameX = addString(strings, f.Filename)
}
p.dropFramesX = addString(strings, p.DropFrames)
p.keepFramesX = addString(strings, p.KeepFrames)
if pt := p.PeriodType; pt != nil {
pt.typeX = addString(strings, pt.Type)
pt.unitX = addString(strings, pt.Unit)
}
p.commentX = nil
for _, c := range p.Comments {
p.commentX = append(p.commentX, addString(strings, c))
}
p.defaultSampleTypeX = addString(strings, p.DefaultSampleType)
p.docURLX = addString(strings, p.DocURL)
p.stringTable = make([]string, len(strings))
for s, i := range strings {
p.stringTable[i] = s
}
}
func (p *Profile) encode(b *buffer) {
for _, x := range p.SampleType {
encodeMessage(b, 1, x)
}
for _, x := range p.Sample {
encodeMessage(b, 2, x)
}
for _, x := range p.Mapping {
encodeMessage(b, 3, x)
}
for _, x := range p.Location {
encodeMessage(b, 4, x)
}
for _, x := range p.Function {
encodeMessage(b, 5, x)
}
encodeStrings(b, 6, p.stringTable)
encodeInt64Opt(b, 7, p.dropFramesX)
encodeInt64Opt(b, 8, p.keepFramesX)
encodeInt64Opt(b, 9, p.TimeNanos)
encodeInt64Opt(b, 10, p.DurationNanos)
if pt := p.PeriodType; pt != nil && (pt.typeX != 0 || pt.unitX != 0) {
encodeMessage(b, 11, p.PeriodType)
}
encodeInt64Opt(b, 12, p.Period)
encodeInt64s(b, 13, p.commentX)
encodeInt64(b, 14, p.defaultSampleTypeX)
encodeInt64Opt(b, 15, p.docURLX)
}
var profileDecoder = []decoder{
nil, // 0
// repeated ValueType sample_type = 1
func(b *buffer, m message) error {
x := new(ValueType)
pp := m.(*Profile)
pp.SampleType = append(pp.SampleType, x)
return decodeMessage(b, x)
},
// repeated Sample sample = 2
func(b *buffer, m message) error {
x := new(Sample)
pp := m.(*Profile)
pp.Sample = append(pp.Sample, x)
return decodeMessage(b, x)
},
// repeated Mapping mapping = 3
func(b *buffer, m message) error {
x := new(Mapping)
pp := m.(*Profile)
pp.Mapping = append(pp.Mapping, x)
return decodeMessage(b, x)
},
// repeated Location location = 4
func(b *buffer, m message) error {
x := new(Location)
x.Line = b.tmpLines[:0] // Use shared space temporarily
pp := m.(*Profile)
pp.Location = append(pp.Location, x)
err := decodeMessage(b, x)
b.tmpLines = x.Line[:0]
// Copy to shrink size and detach from shared space.
x.Line = append([]Line(nil), x.Line...)
return err
},
// repeated Function function = 5
func(b *buffer, m message) error {
x := new(Function)
pp := m.(*Profile)
pp.Function = append(pp.Function, x)
return decodeMessage(b, x)
},
// repeated string string_table = 6
func(b *buffer, m message) error {
err := decodeStrings(b, &m.(*Profile).stringTable)
if err != nil {
return err
}
if m.(*Profile).stringTable[0] != "" {
return errors.New("string_table[0] must be ''")
}
return nil
},
// int64 drop_frames = 7
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).dropFramesX) },
// int64 keep_frames = 8
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).keepFramesX) },
// int64 time_nanos = 9
func(b *buffer, m message) error {
if m.(*Profile).TimeNanos != 0 {
return errConcatProfile
}
return decodeInt64(b, &m.(*Profile).TimeNanos)
},
// int64 duration_nanos = 10
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).DurationNanos) },
// ValueType period_type = 11
func(b *buffer, m message) error {
x := new(ValueType)
pp := m.(*Profile)
pp.PeriodType = x
return decodeMessage(b, x)
},
// 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) },
// string doc_link = 15;
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).docURLX) },
}
// postDecode takes the unexported fields populated by decode (with
// suffix X) and populates the corresponding exported fields.
// The unexported fields are cleared up to facilitate testing.
func (p *Profile) postDecode() error {
var err error
mappings := make(map[uint64]*Mapping, len(p.Mapping))
mappingIds := make([]*Mapping, len(p.Mapping)+1)
for _, m := range p.Mapping {
m.File, err = getString(p.stringTable, &m.fileX, err)
m.BuildID, err = getString(p.stringTable, &m.buildIDX, err)
if m.ID < uint64(len(mappingIds)) {
mappingIds[m.ID] = m
} else {
mappings[m.ID] = m
}
// If this a main linux kernel mapping with a relocation symbol suffix
// ("[kernel.kallsyms]_text"), extract said suffix.
// It is fairly hacky to handle at this level, but the alternatives appear even worse.
const prefix = "[kernel.kallsyms]"
if strings.HasPrefix(m.File, prefix) {
m.KernelRelocationSymbol = m.File[len(prefix):]
}
}
functions := make(map[uint64]*Function, len(p.Function))
functionIds := make([]*Function, len(p.Function)+1)
for _, f := range p.Function {
f.Name, err = getString(p.stringTable, &f.nameX, err)
f.SystemName, err = getString(p.stringTable, &f.systemNameX, err)
f.Filename, err = getString(p.stringTable, &f.filenameX, err)
if f.ID < uint64(len(functionIds)) {
functionIds[f.ID] = f
} else {
functions[f.ID] = f
}
}
locations := make(map[uint64]*Location, len(p.Location))
locationIds := make([]*Location, len(p.Location)+1)
for _, l := range p.Location {
if id := l.mappingIDX; id < uint64(len(mappingIds)) {
l.Mapping = mappingIds[id]
} else {
l.Mapping = mappings[id]
}
l.mappingIDX = 0
for i, ln := range l.Line {
if id := ln.functionIDX; id != 0 {
l.Line[i].functionIDX = 0
if id < uint64(len(functionIds)) {
l.Line[i].Function = functionIds[id]
} else {
l.Line[i].Function = functions[id]
}
}
}
if l.ID < uint64(len(locationIds)) {
locationIds[l.ID] = l
} else {
locations[l.ID] = l
}
}
for _, st := range p.SampleType {
st.Type, err = getString(p.stringTable, &st.typeX, err)
st.Unit, err = getString(p.stringTable, &st.unitX, err)
}
// Pre-allocate space for all locations.
numLocations := 0
for _, s := range p.Sample {
numLocations += len(s.locationIDX)
}
locBuffer := make([]*Location, numLocations)
for _, s := range p.Sample {
if len(s.labelX) > 0 {
labels := make(map[string][]string, len(s.labelX))
numLabels := make(map[string][]int64, len(s.labelX))
numUnits := make(map[string][]string, len(s.labelX))
for _, l := range s.labelX {
var key, value string
key, err = getString(p.stringTable, &l.keyX, err)
if l.strX != 0 {
value, err = getString(p.stringTable, &l.strX, err)
labels[key] = append(labels[key], value)
} else if l.numX != 0 || l.unitX != 0 {
numValues := numLabels[key]
units := numUnits[key]
if l.unitX != 0 {
var unit string
unit, err = getString(p.stringTable, &l.unitX, err)
units = padStringArray(units, len(numValues))
numUnits[key] = append(units, unit)
}
numLabels[key] = append(numLabels[key], l.numX)
}
}
if len(labels) > 0 {
s.Label = labels
}
if len(numLabels) > 0 {
s.NumLabel = numLabels
for key, units := range numUnits {
if len(units) > 0 {
numUnits[key] = padStringArray(units, len(numLabels[key]))
}
}
s.NumUnit = numUnits
}
}
s.Location = locBuffer[:len(s.locationIDX)]
locBuffer = locBuffer[len(s.locationIDX):]
for i, lid := range s.locationIDX {
if lid < uint64(len(locationIds)) {
s.Location[i] = locationIds[lid]
} else {
s.Location[i] = locations[lid]
}
}
s.locationIDX = nil
}
p.DropFrames, err = getString(p.stringTable, &p.dropFramesX, err)
p.KeepFrames, err = getString(p.stringTable, &p.keepFramesX, err)
if pt := p.PeriodType; pt == nil {
p.PeriodType = &ValueType{}
}
if pt := p.PeriodType; pt != nil {
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.DocURL, err = getString(p.stringTable, &p.docURLX, err)
p.stringTable = nil
return err
}
// padStringArray pads arr with enough empty strings to make arr
// length l when arr's length is less than l.
func padStringArray(arr []string, l int) []string {
if l <= len(arr) {
return arr
}
return append(arr, make([]string, l-len(arr))...)
}
func (p *ValueType) decoder() []decoder {
return valueTypeDecoder
}
func (p *ValueType) encode(b *buffer) {
encodeInt64Opt(b, 1, p.typeX)
encodeInt64Opt(b, 2, p.unitX)
}
var valueTypeDecoder = []decoder{
nil, // 0
// optional int64 type = 1
func(b *buffer, m message) error { return decodeInt64(b, &m.(*ValueType).typeX) },
// optional int64 unit = 2
func(b *buffer, m message) error { return decodeInt64(b, &m.(*ValueType).unitX) },
}
func (p *Sample) decoder() []decoder {
return sampleDecoder
}
func (p *Sample) encode(b *buffer) {
encodeUint64s(b, 1, p.locationIDX)
encodeInt64s(b, 2, p.Value)
for _, x := range p.labelX {
encodeMessage(b, 3, x)
}
}
var sampleDecoder = []decoder{
nil, // 0
// repeated uint64 location = 1
func(b *buffer, m message) error { return decodeUint64s(b, &m.(*Sample).locationIDX) },
// repeated int64 value = 2
func(b *buffer, m message) error { return decodeInt64s(b, &m.(*Sample).Value) },
// repeated Label label = 3
func(b *buffer, m message) error {
s := m.(*Sample)
n := len(s.labelX)
s.labelX = append(s.labelX, label{})
return decodeMessage(b, &s.labelX[n])
},
}
func (p label) decoder() []decoder {
return labelDecoder
}
func (p label) encode(b *buffer) {
encodeInt64Opt(b, 1, p.keyX)
encodeInt64Opt(b, 2, p.strX)
encodeInt64Opt(b, 3, p.numX)
encodeInt64Opt(b, 4, p.unitX)
}
var labelDecoder = []decoder{
nil, // 0
// optional int64 key = 1
func(b *buffer, m message) error { return decodeInt64(b, &m.(*label).keyX) },
// optional int64 str = 2
func(b *buffer, m message) error { return decodeInt64(b, &m.(*label).strX) },
// optional int64 num = 3
func(b *buffer, m message) error { return decodeInt64(b, &m.(*label).numX) },
// optional int64 num = 4
func(b *buffer, m message) error { return decodeInt64(b, &m.(*label).unitX) },
}
func (p *Mapping) decoder() []decoder {
return mappingDecoder
}
func (p *Mapping) encode(b *buffer) {
encodeUint64Opt(b, 1, p.ID)
encodeUint64Opt(b, 2, p.Start)
encodeUint64Opt(b, 3, p.Limit)
encodeUint64Opt(b, 4, p.Offset)
encodeInt64Opt(b, 5, p.fileX)
encodeInt64Opt(b, 6, p.buildIDX)
encodeBoolOpt(b, 7, p.HasFunctions)
encodeBoolOpt(b, 8, p.HasFilenames)
encodeBoolOpt(b, 9, p.HasLineNumbers)
encodeBoolOpt(b, 10, p.HasInlineFrames)
}
var mappingDecoder = []decoder{
nil, // 0
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Mapping).ID) }, // optional uint64 id = 1
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Mapping).Start) }, // optional uint64 memory_offset = 2
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Mapping).Limit) }, // optional uint64 memory_limit = 3
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Mapping).Offset) }, // optional uint64 file_offset = 4
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Mapping).fileX) }, // optional int64 filename = 5
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Mapping).buildIDX) }, // optional int64 build_id = 6
func(b *buffer, m message) error { return decodeBool(b, &m.(*Mapping).HasFunctions) }, // optional bool has_functions = 7
func(b *buffer, m message) error { return decodeBool(b, &m.(*Mapping).HasFilenames) }, // optional bool has_filenames = 8
func(b *buffer, m message) error { return decodeBool(b, &m.(*Mapping).HasLineNumbers) }, // optional bool has_line_numbers = 9
func(b *buffer, m message) error { return decodeBool(b, &m.(*Mapping).HasInlineFrames) }, // optional bool has_inline_frames = 10
}
func (p *Location) decoder() []decoder {
return locationDecoder
}
func (p *Location) encode(b *buffer) {
encodeUint64Opt(b, 1, p.ID)
encodeUint64Opt(b, 2, p.mappingIDX)
encodeUint64Opt(b, 3, p.Address)
for i := range p.Line {
encodeMessage(b, 4, &p.Line[i])
}
encodeBoolOpt(b, 5, p.IsFolded)
}
var locationDecoder = []decoder{
nil, // 0
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Location).ID) }, // optional uint64 id = 1;
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Location).mappingIDX) }, // optional uint64 mapping_id = 2;
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Location).Address) }, // optional uint64 address = 3;
func(b *buffer, m message) error { // repeated Line line = 4
pp := m.(*Location)
n := len(pp.Line)
pp.Line = append(pp.Line, Line{})
return decodeMessage(b, &pp.Line[n])
},
func(b *buffer, m message) error { return decodeBool(b, &m.(*Location).IsFolded) }, // optional bool is_folded = 5;
}
func (p *Line) decoder() []decoder {
return lineDecoder
}
func (p *Line) encode(b *buffer) {
encodeUint64Opt(b, 1, p.functionIDX)
encodeInt64Opt(b, 2, p.Line)
encodeInt64Opt(b, 3, p.Column)
}
var lineDecoder = []decoder{
nil, // 0
// optional uint64 function_id = 1
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Line).functionIDX) },
// optional int64 line = 2
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Line).Line) },
// optional int64 column = 3
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Line).Column) },
}
func (p *Function) decoder() []decoder {
return functionDecoder
}
func (p *Function) encode(b *buffer) {
encodeUint64Opt(b, 1, p.ID)
encodeInt64Opt(b, 2, p.nameX)
encodeInt64Opt(b, 3, p.systemNameX)
encodeInt64Opt(b, 4, p.filenameX)
encodeInt64Opt(b, 5, p.StartLine)
}
var functionDecoder = []decoder{
nil, // 0
// optional uint64 id = 1
func(b *buffer, m message) error { return decodeUint64(b, &m.(*Function).ID) },
// optional int64 function_name = 2
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Function).nameX) },
// optional int64 function_system_name = 3
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Function).systemNameX) },
// repeated int64 filename = 4
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Function).filenameX) },
// optional int64 start_line = 5
func(b *buffer, m message) error { return decodeInt64(b, &m.(*Function).StartLine) },
}
func addString(strings map[string]int, s string) int64 {
i, ok := strings[s]
if !ok {
i = len(strings)
strings[s] = i
}
return int64(i)
}
func getString(strings []string, strng *int64, err error) (string, error) {
if err != nil {
return "", err
}
s := int(*strng)
if s < 0 || s >= len(strings) {
return "", errMalformed
}
*strng = 0
return strings[s], nil
}
// Copyright 2014 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package profile
// Implements methods to filter samples from profiles.
import "regexp"
// FilterSamplesByName filters the samples in a profile and only keeps
// samples where at least one frame matches focus but none match ignore.
// Returns true is the corresponding regexp matched at least one sample.
func (p *Profile) FilterSamplesByName(focus, ignore, hide, show *regexp.Regexp) (fm, im, hm, hnm bool) {
if focus == nil && ignore == nil && hide == nil && show == nil {
fm = true // Missing focus implies a match
return
}
focusOrIgnore := make(map[uint64]bool)
hidden := make(map[uint64]bool)
for _, l := range p.Location {
if ignore != nil && l.matchesName(ignore) {
im = true
focusOrIgnore[l.ID] = false
} else if focus == nil || l.matchesName(focus) {
fm = true
focusOrIgnore[l.ID] = true
}
if hide != nil && l.matchesName(hide) {
hm = true
l.Line = l.unmatchedLines(hide)
if len(l.Line) == 0 {
hidden[l.ID] = true
}
}
if show != nil {
l.Line = l.matchedLines(show)
if len(l.Line) == 0 {
hidden[l.ID] = true
} else {
hnm = true
}
}
}
s := make([]*Sample, 0, len(p.Sample))
for _, sample := range p.Sample {
if focusedAndNotIgnored(sample.Location, focusOrIgnore) {
if len(hidden) > 0 {
var locs []*Location
for _, loc := range sample.Location {
if !hidden[loc.ID] {
locs = append(locs, loc)
}
}
if len(locs) == 0 {
// Remove sample with no locations (by not adding it to s).
continue
}
sample.Location = locs
}
s = append(s, sample)
}
}
p.Sample = s
return
}
// ShowFrom drops all stack frames above the highest matching frame and returns
// whether a match was found. If showFrom is nil it returns false and does not
// modify the profile.
//
// Example: consider a sample with frames [A, B, C, B], where A is the root.
// ShowFrom(nil) returns false and has frames [A, B, C, B].
// ShowFrom(A) returns true and has frames [A, B, C, B].
// ShowFrom(B) returns true and has frames [B, C, B].
// ShowFrom(C) returns true and has frames [C, B].
// ShowFrom(D) returns false and drops the sample because no frames remain.
func (p *Profile) ShowFrom(showFrom *regexp.Regexp) (matched bool) {
if showFrom == nil {
return false
}
// showFromLocs stores location IDs that matched ShowFrom.
showFromLocs := make(map[uint64]bool)
// Apply to locations.
for _, loc := range p.Location {
if filterShowFromLocation(loc, showFrom) {
showFromLocs[loc.ID] = true
matched = true
}
}
// For all samples, strip locations after the highest matching one.
s := make([]*Sample, 0, len(p.Sample))
for _, sample := range p.Sample {
for i := len(sample.Location) - 1; i >= 0; i-- {
if showFromLocs[sample.Location[i].ID] {
sample.Location = sample.Location[:i+1]
s = append(s, sample)
break
}
}
}
p.Sample = s
return matched
}
// filterShowFromLocation tests a showFrom regex against a location, removes
// lines after the last match and returns whether a match was found. If the
// mapping is matched, then all lines are kept.
func filterShowFromLocation(loc *Location, showFrom *regexp.Regexp) bool {
if m := loc.Mapping; m != nil && showFrom.MatchString(m.File) {
return true
}
if i := loc.lastMatchedLineIndex(showFrom); i >= 0 {
loc.Line = loc.Line[:i+1]
return true
}
return false
}
// lastMatchedLineIndex returns the index of the last line that matches a regex,
// or -1 if no match is found.
func (loc *Location) lastMatchedLineIndex(re *regexp.Regexp) int {
for i := len(loc.Line) - 1; i >= 0; i-- {
if fn := loc.Line[i].Function; fn != nil {
if re.MatchString(fn.Name) || re.MatchString(fn.Filename) {
return i
}
}
}
return -1
}
// FilterTagsByName filters the tags in a profile and only keeps
// tags that match show and not hide.
func (p *Profile) FilterTagsByName(show, hide *regexp.Regexp) (sm, hm bool) {
matchRemove := func(name string) bool {
matchShow := show == nil || show.MatchString(name)
matchHide := hide != nil && hide.MatchString(name)
if matchShow {
sm = true
}
if matchHide {
hm = true
}
return !matchShow || matchHide
}
for _, s := range p.Sample {
for lab := range s.Label {
if matchRemove(lab) {
delete(s.Label, lab)
}
}
for lab := range s.NumLabel {
if matchRemove(lab) {
delete(s.NumLabel, lab)
}
}
}
return
}
// matchesName returns whether the location matches the regular
// expression. It checks any available function names, file names, and
// mapping object filename.
func (loc *Location) matchesName(re *regexp.Regexp) bool {
for _, ln := range loc.Line {
if fn := ln.Function; fn != nil {
if re.MatchString(fn.Name) || re.MatchString(fn.Filename) {
return true
}
}
}
if m := loc.Mapping; m != nil && re.MatchString(m.File) {
return true
}
return false
}
// unmatchedLines returns the lines in the location that do not match
// the regular expression.
func (loc *Location) unmatchedLines(re *regexp.Regexp) []Line {
if m := loc.Mapping; m != nil && re.MatchString(m.File) {
return nil
}
var lines []Line
for _, ln := range loc.Line {
if fn := ln.Function; fn != nil {
if re.MatchString(fn.Name) || re.MatchString(fn.Filename) {
continue
}
}
lines = append(lines, ln)
}
return lines
}
// matchedLines returns the lines in the location that match
// the regular expression.
func (loc *Location) matchedLines(re *regexp.Regexp) []Line {
if m := loc.Mapping; m != nil && re.MatchString(m.File) {
return loc.Line
}
var lines []Line
for _, ln := range loc.Line {
if fn := ln.Function; fn != nil {
if !re.MatchString(fn.Name) && !re.MatchString(fn.Filename) {
continue
}
}
lines = append(lines, ln)
}
return lines
}
// focusedAndNotIgnored looks up a slice of ids against a map of
// focused/ignored locations. The map only contains locations that are
// explicitly focused or ignored. Returns whether there is at least
// one focused location but no ignored locations.
func focusedAndNotIgnored(locs []*Location, m map[uint64]bool) bool {
var f bool
for _, loc := range locs {
if focus, focusOrIgnore := m[loc.ID]; focusOrIgnore {
if focus {
// Found focused location. Must keep searching in case there
// is an ignored one as well.
f = true
} else {
// Found ignored location. Can return false right away.
return false
}
}
}
return f
}
// TagMatch selects tags for filtering
type TagMatch func(s *Sample) bool
// FilterSamplesByTag removes all samples from the profile, except
// those that match focus and do not match the ignore regular
// expression.
func (p *Profile) FilterSamplesByTag(focus, ignore TagMatch) (fm, im bool) {
samples := make([]*Sample, 0, len(p.Sample))
for _, s := range p.Sample {
focused, ignored := true, false
if focus != nil {
focused = focus(s)
}
if ignore != nil {
ignored = ignore(s)
}
fm = fm || focused
im = im || ignored
if focused && !ignored {
samples = append(samples, s)
}
}
p.Sample = samples
return
}
// Copyright 2016 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package profile
import (
"fmt"
"strconv"
"strings"
)
// SampleIndexByName returns the appropriate index for a value of sample index.
// If numeric, it returns the number, otherwise it looks up the text in the
// profile sample types.
func (p *Profile) SampleIndexByName(sampleIndex string) (int, error) {
if sampleIndex == "" {
if dst := p.DefaultSampleType; dst != "" {
for i, t := range sampleTypes(p) {
if t == dst {
return i, nil
}
}
}
// By default select the last sample value
return len(p.SampleType) - 1, nil
}
if i, err := strconv.Atoi(sampleIndex); err == nil {
if i < 0 || i >= len(p.SampleType) {
return 0, fmt.Errorf("sample_index %s is outside the range [0..%d]", sampleIndex, len(p.SampleType)-1)
}
return i, nil
}
// Remove the inuse_ prefix to support legacy pprof options
// "inuse_space" and "inuse_objects" for profiles containing types
// "space" and "objects".
noInuse := strings.TrimPrefix(sampleIndex, "inuse_")
for i, t := range p.SampleType {
if t.Type == sampleIndex || t.Type == noInuse {
return i, nil
}
}
return 0, fmt.Errorf("sample_index %q must be one of: %v", sampleIndex, sampleTypes(p))
}
func sampleTypes(p *Profile) []string {
types := make([]string, len(p.SampleType))
for i, t := range p.SampleType {
types[i] = t.Type
}
return types
}
// Copyright 2014 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// This file implements parsers to convert java legacy profiles into
// the profile.proto format.
package profile
import (
"bytes"
"fmt"
"io"
"path/filepath"
"regexp"
"strconv"
"strings"
)
var (
attributeRx = regexp.MustCompile(`([\w ]+)=([\w ]+)`)
javaSampleRx = regexp.MustCompile(` *(\d+) +(\d+) +@ +([ x0-9a-f]*)`)
javaLocationRx = regexp.MustCompile(`^\s*0x([[:xdigit:]]+)\s+(.*)\s*$`)
javaLocationFileLineRx = regexp.MustCompile(`^(.*)\s+\((.+):(-?[[:digit:]]+)\)$`)
javaLocationPathRx = regexp.MustCompile(`^(.*)\s+\((.*)\)$`)
)
// javaCPUProfile returns a new Profile from profilez data.
// b is the profile bytes after the header, period is the profiling
// period, and parse is a function to parse 8-byte chunks from the
// profile in its native endianness.
func javaCPUProfile(b []byte, period int64, parse func(b []byte) (uint64, []byte)) (*Profile, error) {
p := &Profile{
Period: period * 1000,
PeriodType: &ValueType{Type: "cpu", Unit: "nanoseconds"},
SampleType: []*ValueType{{Type: "samples", Unit: "count"}, {Type: "cpu", Unit: "nanoseconds"}},
}
var err error
var locs map[uint64]*Location
if b, locs, err = parseCPUSamples(b, parse, false, p); err != nil {
return nil, err
}
if err = parseJavaLocations(b, locs, p); err != nil {
return nil, err
}
// Strip out addresses for better merge.
if err = p.Aggregate(true, true, true, true, false, false); err != nil {
return nil, err
}
return p, nil
}
// parseJavaProfile returns a new profile from heapz or contentionz
// data. b is the profile bytes after the header.
func parseJavaProfile(b []byte) (*Profile, error) {
h := bytes.SplitAfterN(b, []byte("\n"), 2)
if len(h) < 2 {
return nil, errUnrecognized
}
p := &Profile{
PeriodType: &ValueType{},
}
header := string(bytes.TrimSpace(h[0]))
var err error
var pType string
switch header {
case "--- heapz 1 ---":
pType = "heap"
case "--- contentionz 1 ---":
pType = "contention"
default:
return nil, errUnrecognized
}
if b, err = parseJavaHeader(pType, h[1], p); err != nil {
return nil, err
}
var locs map[uint64]*Location
if b, locs, err = parseJavaSamples(pType, b, p); err != nil {
return nil, err
}
if err = parseJavaLocations(b, locs, p); err != nil {
return nil, err
}
// Strip out addresses for better merge.
if err = p.Aggregate(true, true, true, true, false, false); err != nil {
return nil, err
}
return p, nil
}
// parseJavaHeader parses the attribute section on a java profile and
// populates a profile. Returns the remainder of the buffer after all
// attributes.
func parseJavaHeader(pType string, b []byte, p *Profile) ([]byte, error) {
nextNewLine := bytes.IndexByte(b, byte('\n'))
for nextNewLine != -1 {
line := string(bytes.TrimSpace(b[0:nextNewLine]))
if line != "" {
h := attributeRx.FindStringSubmatch(line)
if h == nil {
// Not a valid attribute, exit.
return b, nil
}
attribute, value := strings.TrimSpace(h[1]), strings.TrimSpace(h[2])
var err error
switch pType + "/" + attribute {
case "heap/format", "cpu/format", "contention/format":
if value != "java" {
return nil, errUnrecognized
}
case "heap/resolution":
p.SampleType = []*ValueType{
{Type: "inuse_objects", Unit: "count"},
{Type: "inuse_space", Unit: value},
}
case "contention/resolution":
p.SampleType = []*ValueType{
{Type: "contentions", Unit: "count"},
{Type: "delay", Unit: value},
}
case "contention/sampling period":
p.PeriodType = &ValueType{
Type: "contentions", Unit: "count",
}
if p.Period, err = strconv.ParseInt(value, 0, 64); err != nil {
return nil, fmt.Errorf("failed to parse attribute %s: %v", line, err)
}
case "contention/ms since reset":
millis, err := strconv.ParseInt(value, 0, 64)
if err != nil {
return nil, fmt.Errorf("failed to parse attribute %s: %v", line, err)
}
p.DurationNanos = millis * 1000 * 1000
default:
return nil, errUnrecognized
}
}
// Grab next line.
b = b[nextNewLine+1:]
nextNewLine = bytes.IndexByte(b, byte('\n'))
}
return b, nil
}
// parseJavaSamples parses the samples from a java profile and
// populates the Samples in a profile. Returns the remainder of the
// buffer after the samples.
func parseJavaSamples(pType string, b []byte, p *Profile) ([]byte, map[uint64]*Location, error) {
nextNewLine := bytes.IndexByte(b, byte('\n'))
locs := make(map[uint64]*Location)
for nextNewLine != -1 {
line := string(bytes.TrimSpace(b[0:nextNewLine]))
if line != "" {
sample := javaSampleRx.FindStringSubmatch(line)
if sample == nil {
// Not a valid sample, exit.
return b, locs, nil
}
// Java profiles have data/fields inverted compared to other
// profile types.
var err error
value1, value2, value3 := sample[2], sample[1], sample[3]
addrs, err := parseHexAddresses(value3)
if err != nil {
return nil, nil, fmt.Errorf("malformed sample: %s: %v", line, err)
}
var sloc []*Location
for _, addr := range addrs {
loc := locs[addr]
if locs[addr] == nil {
loc = &Location{
Address: addr,
}
p.Location = append(p.Location, loc)
locs[addr] = loc
}
sloc = append(sloc, loc)
}
s := &Sample{
Value: make([]int64, 2),
Location: sloc,
}
if s.Value[0], err = strconv.ParseInt(value1, 0, 64); err != nil {
return nil, nil, fmt.Errorf("parsing sample %s: %v", line, err)
}
if s.Value[1], err = strconv.ParseInt(value2, 0, 64); err != nil {
return nil, nil, fmt.Errorf("parsing sample %s: %v", line, err)
}
switch pType {
case "heap":
const javaHeapzSamplingRate = 524288 // 512K
if s.Value[0] == 0 {
return nil, nil, fmt.Errorf("parsing sample %s: second value must be non-zero", line)
}
s.NumLabel = map[string][]int64{"bytes": {s.Value[1] / s.Value[0]}}
s.Value[0], s.Value[1] = scaleHeapSample(s.Value[0], s.Value[1], javaHeapzSamplingRate)
case "contention":
if period := p.Period; period != 0 {
s.Value[0] = s.Value[0] * p.Period
s.Value[1] = s.Value[1] * p.Period
}
}
p.Sample = append(p.Sample, s)
}
// Grab next line.
b = b[nextNewLine+1:]
nextNewLine = bytes.IndexByte(b, byte('\n'))
}
return b, locs, nil
}
// parseJavaLocations parses the location information in a java
// profile and populates the Locations in a profile. It uses the
// location addresses from the profile as both the ID of each
// location.
func parseJavaLocations(b []byte, locs map[uint64]*Location, p *Profile) error {
r := bytes.NewBuffer(b)
fns := make(map[string]*Function)
for {
line, err := r.ReadString('\n')
if err != nil {
if err != io.EOF {
return err
}
if line == "" {
break
}
}
if line = strings.TrimSpace(line); line == "" {
continue
}
jloc := javaLocationRx.FindStringSubmatch(line)
if len(jloc) != 3 {
continue
}
addr, err := strconv.ParseUint(jloc[1], 16, 64)
if err != nil {
return fmt.Errorf("parsing sample %s: %v", line, err)
}
loc := locs[addr]
if loc == nil {
// Unused/unseen
continue
}
var lineFunc, lineFile string
var lineNo int64
if fileLine := javaLocationFileLineRx.FindStringSubmatch(jloc[2]); len(fileLine) == 4 {
// Found a line of the form: "function (file:line)"
lineFunc, lineFile = fileLine[1], fileLine[2]
if n, err := strconv.ParseInt(fileLine[3], 10, 64); err == nil && n > 0 {
lineNo = n
}
} else if filePath := javaLocationPathRx.FindStringSubmatch(jloc[2]); len(filePath) == 3 {
// If there's not a file:line, it's a shared library path.
// The path isn't interesting, so just give the .so.
lineFunc, lineFile = filePath[1], filepath.Base(filePath[2])
} else if strings.Contains(jloc[2], "generated stub/JIT") {
lineFunc = "STUB"
} else {
// Treat whole line as the function name. This is used by the
// java agent for internal states such as "GC" or "VM".
lineFunc = jloc[2]
}
fn := fns[lineFunc]
if fn == nil {
fn = &Function{
Name: lineFunc,
SystemName: lineFunc,
Filename: lineFile,
}
fns[lineFunc] = fn
p.Function = append(p.Function, fn)
}
loc.Line = []Line{
{
Function: fn,
Line: lineNo,
},
}
loc.Address = 0
}
p.remapLocationIDs()
p.remapFunctionIDs()
p.remapMappingIDs()
return nil
}
// Copyright 2014 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// This file implements parsers to convert legacy profiles into the
// profile.proto format.
package profile
import (
"bufio"
"bytes"
"fmt"
"io"
"math"
"regexp"
"strconv"
"strings"
)
var (
countStartRE = regexp.MustCompile(`\A(\S+) profile: total \d+\z`)
countRE = regexp.MustCompile(`\A(\d+) @(( 0x[0-9a-f]+)+)\z`)
heapHeaderRE = regexp.MustCompile(`heap profile: *(\d+): *(\d+) *\[ *(\d+): *(\d+) *\] *@ *(heap[_a-z0-9]*)/?(\d*)`)
heapSampleRE = regexp.MustCompile(`(-?\d+): *(-?\d+) *\[ *(\d+): *(\d+) *] @([ x0-9a-f]*)`)
contentionSampleRE = regexp.MustCompile(`(\d+) *(\d+) @([ x0-9a-f]*)`)
hexNumberRE = regexp.MustCompile(`0x[0-9a-f]+`)
growthHeaderRE = regexp.MustCompile(`heap profile: *(\d+): *(\d+) *\[ *(\d+): *(\d+) *\] @ growthz?`)
fragmentationHeaderRE = regexp.MustCompile(`heap profile: *(\d+): *(\d+) *\[ *(\d+): *(\d+) *\] @ fragmentationz?`)
threadzStartRE = regexp.MustCompile(`--- threadz \d+ ---`)
threadStartRE = regexp.MustCompile(`--- Thread ([[:xdigit:]]+) \(name: (.*)/(\d+)\) stack: ---`)
// Regular expressions to parse process mappings. Support the format used by Linux /proc/.../maps and other tools.
// Recommended format:
// Start End object file name offset(optional) linker build id
// 0x40000-0x80000 /path/to/binary (@FF00) abc123456
spaceDigits = `\s+[[:digit:]]+`
hexPair = `\s+[[:xdigit:]]+:[[:xdigit:]]+`
oSpace = `\s*`
// Capturing expressions.
cHex = `(?:0x)?([[:xdigit:]]+)`
cHexRange = `\s*` + cHex + `[\s-]?` + oSpace + cHex + `:?`
cSpaceString = `(?:\s+(\S+))?`
cSpaceHex = `(?:\s+([[:xdigit:]]+))?`
cSpaceAtOffset = `(?:\s+\(@([[:xdigit:]]+)\))?`
cPerm = `(?:\s+([-rwxp]+))?`
procMapsRE = regexp.MustCompile(`^` + cHexRange + cPerm + cSpaceHex + hexPair + spaceDigits + cSpaceString)
briefMapsRE = regexp.MustCompile(`^` + cHexRange + cPerm + cSpaceString + cSpaceAtOffset + cSpaceHex)
// Regular expression to parse log data, of the form:
// ... file:line] msg...
logInfoRE = regexp.MustCompile(`^[^\[\]]+:[0-9]+]\s`)
)
func isSpaceOrComment(line string) bool {
trimmed := strings.TrimSpace(line)
return len(trimmed) == 0 || trimmed[0] == '#'
}
// parseGoCount parses a Go count profile (e.g., threadcreate or
// goroutine) and returns a new Profile.
func parseGoCount(b []byte) (*Profile, error) {
s := bufio.NewScanner(bytes.NewBuffer(b))
// Skip comments at the beginning of the file.
for s.Scan() && isSpaceOrComment(s.Text()) {
}
if err := s.Err(); err != nil {
return nil, err
}
m := countStartRE.FindStringSubmatch(s.Text())
if m == nil {
return nil, errUnrecognized
}
profileType := m[1]
p := &Profile{
PeriodType: &ValueType{Type: profileType, Unit: "count"},
Period: 1,
SampleType: []*ValueType{{Type: profileType, Unit: "count"}},
}
locations := make(map[uint64]*Location)
for s.Scan() {
line := s.Text()
if isSpaceOrComment(line) {
continue
}
if strings.HasPrefix(line, "---") {
break
}
m := countRE.FindStringSubmatch(line)
if m == nil {
return nil, errMalformed
}
n, err := strconv.ParseInt(m[1], 0, 64)
if err != nil {
return nil, errMalformed
}
fields := strings.Fields(m[2])
locs := make([]*Location, 0, len(fields))
for _, stk := range fields {
addr, err := strconv.ParseUint(stk, 0, 64)
if err != nil {
return nil, errMalformed
}
// Adjust all frames by -1 to land on top of the call instruction.
addr--
loc := locations[addr]
if loc == nil {
loc = &Location{
Address: addr,
}
locations[addr] = loc
p.Location = append(p.Location, loc)
}
locs = append(locs, loc)
}
p.Sample = append(p.Sample, &Sample{
Location: locs,
Value: []int64{n},
})
}
if err := s.Err(); err != nil {
return nil, err
}
if err := parseAdditionalSections(s, p); err != nil {
return nil, err
}
return p, nil
}
// remapLocationIDs ensures there is a location for each address
// referenced by a sample, and remaps the samples to point to the new
// location ids.
func (p *Profile) remapLocationIDs() {
seen := make(map[*Location]bool, len(p.Location))
var locs []*Location
for _, s := range p.Sample {
for _, l := range s.Location {
if seen[l] {
continue
}
l.ID = uint64(len(locs) + 1)
locs = append(locs, l)
seen[l] = true
}
}
p.Location = locs
}
func (p *Profile) remapFunctionIDs() {
seen := make(map[*Function]bool, len(p.Function))
var fns []*Function
for _, l := range p.Location {
for _, ln := range l.Line {
fn := ln.Function
if fn == nil || seen[fn] {
continue
}
fn.ID = uint64(len(fns) + 1)
fns = append(fns, fn)
seen[fn] = true
}
}
p.Function = fns
}
// remapMappingIDs matches location addresses with existing mappings
// and updates them appropriately. This is O(N*M), if this ever shows
// up as a bottleneck, evaluate sorting the mappings and doing a
// binary search, which would make it O(N*log(M)).
func (p *Profile) remapMappingIDs() {
// Some profile handlers will incorrectly set regions for the main
// executable if its section is remapped. Fix them through heuristics.
if len(p.Mapping) > 0 {
// Remove the initial mapping if named '/anon_hugepage' and has a
// consecutive adjacent mapping.
if m := p.Mapping[0]; strings.HasPrefix(m.File, "/anon_hugepage") {
if len(p.Mapping) > 1 && m.Limit == p.Mapping[1].Start {
p.Mapping = p.Mapping[1:]
}
}
}
// Subtract the offset from the start of the main mapping if it
// ends up at a recognizable start address.
if len(p.Mapping) > 0 {
const expectedStart = 0x400000
if m := p.Mapping[0]; m.Start-m.Offset == expectedStart {
m.Start = expectedStart
m.Offset = 0
}
}
// Associate each location with an address to the corresponding
// mapping. Create fake mapping if a suitable one isn't found.
var fake *Mapping
nextLocation:
for _, l := range p.Location {
a := l.Address
if l.Mapping != nil || a == 0 {
continue
}
for _, m := range p.Mapping {
if m.Start <= a && a < m.Limit {
l.Mapping = m
continue nextLocation
}
}
// Work around legacy handlers failing to encode the first
// part of mappings split into adjacent ranges.
for _, m := range p.Mapping {
if m.Offset != 0 && m.Start-m.Offset <= a && a < m.Start {
m.Start -= m.Offset
m.Offset = 0
l.Mapping = m
continue nextLocation
}
}
// If there is still no mapping, create a fake one.
// This is important for the Go legacy handler, which produced
// no mappings.
if fake == nil {
fake = &Mapping{
ID: 1,
Limit: ^uint64(0),
}
p.Mapping = append(p.Mapping, fake)
}
l.Mapping = fake
}
// Reset all mapping IDs.
for i, m := range p.Mapping {
m.ID = uint64(i + 1)
}
}
var cpuInts = []func([]byte) (uint64, []byte){
get32l,
get32b,
get64l,
get64b,
}
func get32l(b []byte) (uint64, []byte) {
if len(b) < 4 {
return 0, nil
}
return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24, b[4:]
}
func get32b(b []byte) (uint64, []byte) {
if len(b) < 4 {
return 0, nil
}
return uint64(b[3]) | uint64(b[2])<<8 | uint64(b[1])<<16 | uint64(b[0])<<24, b[4:]
}
func get64l(b []byte) (uint64, []byte) {
if len(b) < 8 {
return 0, nil
}
return uint64(b[0]) | uint64(b[1])<<8 | uint64(b[2])<<16 | uint64(b[3])<<24 | uint64(b[4])<<32 | uint64(b[5])<<40 | uint64(b[6])<<48 | uint64(b[7])<<56, b[8:]
}
func get64b(b []byte) (uint64, []byte) {
if len(b) < 8 {
return 0, nil
}
return uint64(b[7]) | uint64(b[6])<<8 | uint64(b[5])<<16 | uint64(b[4])<<24 | uint64(b[3])<<32 | uint64(b[2])<<40 | uint64(b[1])<<48 | uint64(b[0])<<56, b[8:]
}
// parseCPU parses a profilez legacy profile and returns a newly
// populated Profile.
//
// The general format for profilez samples is a sequence of words in
// binary format. The first words are a header with the following data:
//
// 1st word -- 0
// 2nd word -- 3
// 3rd word -- 0 if a c++ application, 1 if a java application.
// 4th word -- Sampling period (in microseconds).
// 5th word -- Padding.
func parseCPU(b []byte) (*Profile, error) {
var parse func([]byte) (uint64, []byte)
var n1, n2, n3, n4, n5 uint64
for _, parse = range cpuInts {
var tmp []byte
n1, tmp = parse(b)
n2, tmp = parse(tmp)
n3, tmp = parse(tmp)
n4, tmp = parse(tmp)
n5, tmp = parse(tmp)
if tmp != nil && n1 == 0 && n2 == 3 && n3 == 0 && n4 > 0 && n5 == 0 {
b = tmp
return cpuProfile(b, int64(n4), parse)
}
if tmp != nil && n1 == 0 && n2 == 3 && n3 == 1 && n4 > 0 && n5 == 0 {
b = tmp
return javaCPUProfile(b, int64(n4), parse)
}
}
return nil, errUnrecognized
}
// cpuProfile returns a new Profile from C++ profilez data.
// b is the profile bytes after the header, period is the profiling
// period, and parse is a function to parse 8-byte chunks from the
// profile in its native endianness.
func cpuProfile(b []byte, period int64, parse func(b []byte) (uint64, []byte)) (*Profile, error) {
p := &Profile{
Period: period * 1000,
PeriodType: &ValueType{Type: "cpu", Unit: "nanoseconds"},
SampleType: []*ValueType{
{Type: "samples", Unit: "count"},
{Type: "cpu", Unit: "nanoseconds"},
},
}
var err error
if b, _, err = parseCPUSamples(b, parse, true, p); err != nil {
return nil, err
}
// If *most* samples have the same second-to-the-bottom frame, it
// strongly suggests that it is an uninteresting artifact of
// measurement -- a stack frame pushed by the signal handler. The
// bottom frame is always correct as it is picked up from the signal
// structure, not the stack. Check if this is the case and if so,
// remove.
// Remove up to two frames.
maxiter := 2
// Allow one different sample for this many samples with the same
// second-to-last frame.
similarSamples := 32
margin := len(p.Sample) / similarSamples
for iter := 0; iter < maxiter; iter++ {
addr1 := make(map[uint64]int)
for _, s := range p.Sample {
if len(s.Location) > 1 {
a := s.Location[1].Address
addr1[a] = addr1[a] + 1
}
}
for id1, count := range addr1 {
if count >= len(p.Sample)-margin {
// Found uninteresting frame, strip it out from all samples
for _, s := range p.Sample {
if len(s.Location) > 1 && s.Location[1].Address == id1 {
s.Location = append(s.Location[:1], s.Location[2:]...)
}
}
break
}
}
}
if err := p.ParseMemoryMap(bytes.NewBuffer(b)); err != nil {
return nil, err
}
cleanupDuplicateLocations(p)
return p, nil
}
func cleanupDuplicateLocations(p *Profile) {
// The profile handler may duplicate the leaf frame, because it gets
// its address both from stack unwinding and from the signal
// context. Detect this and delete the duplicate, which has been
// adjusted by -1. The leaf address should not be adjusted as it is
// not a call.
for _, s := range p.Sample {
if len(s.Location) > 1 && s.Location[0].Address == s.Location[1].Address+1 {
s.Location = append(s.Location[:1], s.Location[2:]...)
}
}
}
// parseCPUSamples parses a collection of profilez samples from a
// profile.
//
// profilez samples are a repeated sequence of stack frames of the
// form:
//
// 1st word -- The number of times this stack was encountered.
// 2nd word -- The size of the stack (StackSize).
// 3rd word -- The first address on the stack.
// ...
// StackSize + 2 -- The last address on the stack
//
// The last stack trace is of the form:
//
// 1st word -- 0
// 2nd word -- 1
// 3rd word -- 0
//
// Addresses from stack traces may point to the next instruction after
// each call. Optionally adjust by -1 to land somewhere on the actual
// call (except for the leaf, which is not a call).
func parseCPUSamples(b []byte, parse func(b []byte) (uint64, []byte), adjust bool, p *Profile) ([]byte, map[uint64]*Location, error) {
locs := make(map[uint64]*Location)
for len(b) > 0 {
var count, nstk uint64
count, b = parse(b)
nstk, b = parse(b)
if b == nil || nstk > uint64(len(b)/4) {
return nil, nil, errUnrecognized
}
var sloc []*Location
addrs := make([]uint64, nstk)
for i := 0; i < int(nstk); i++ {
addrs[i], b = parse(b)
}
if count == 0 && nstk == 1 && addrs[0] == 0 {
// End of data marker
break
}
for i, addr := range addrs {
if adjust && i > 0 {
addr--
}
loc := locs[addr]
if loc == nil {
loc = &Location{
Address: addr,
}
locs[addr] = loc
p.Location = append(p.Location, loc)
}
sloc = append(sloc, loc)
}
p.Sample = append(p.Sample,
&Sample{
Value: []int64{int64(count), int64(count) * p.Period},
Location: sloc,
})
}
// Reached the end without finding the EOD marker.
return b, locs, nil
}
// parseHeap parses a heapz legacy or a growthz profile and
// returns a newly populated Profile.
func parseHeap(b []byte) (p *Profile, err error) {
s := bufio.NewScanner(bytes.NewBuffer(b))
if !s.Scan() {
if err := s.Err(); err != nil {
return nil, err
}
return nil, errUnrecognized
}
p = &Profile{}
sampling := ""
hasAlloc := false
line := s.Text()
p.PeriodType = &ValueType{Type: "space", Unit: "bytes"}
if header := heapHeaderRE.FindStringSubmatch(line); header != nil {
sampling, p.Period, hasAlloc, err = parseHeapHeader(line)
if err != nil {
return nil, err
}
} else if header = growthHeaderRE.FindStringSubmatch(line); header != nil {
p.Period = 1
} else if header = fragmentationHeaderRE.FindStringSubmatch(line); header != nil {
p.Period = 1
} else {
return nil, errUnrecognized
}
if hasAlloc {
// Put alloc before inuse so that default pprof selection
// will prefer inuse_space.
p.SampleType = []*ValueType{
{Type: "alloc_objects", Unit: "count"},
{Type: "alloc_space", Unit: "bytes"},
{Type: "inuse_objects", Unit: "count"},
{Type: "inuse_space", Unit: "bytes"},
}
} else {
p.SampleType = []*ValueType{
{Type: "objects", Unit: "count"},
{Type: "space", Unit: "bytes"},
}
}
locs := make(map[uint64]*Location)
for s.Scan() {
line := strings.TrimSpace(s.Text())
if isSpaceOrComment(line) {
continue
}
if isMemoryMapSentinel(line) {
break
}
value, blocksize, addrs, err := parseHeapSample(line, p.Period, sampling, hasAlloc)
if err != nil {
return nil, err
}
var sloc []*Location
for _, addr := range addrs {
// Addresses from stack traces point to the next instruction after
// each call. Adjust by -1 to land somewhere on the actual call.
addr--
loc := locs[addr]
if locs[addr] == nil {
loc = &Location{
Address: addr,
}
p.Location = append(p.Location, loc)
locs[addr] = loc
}
sloc = append(sloc, loc)
}
p.Sample = append(p.Sample, &Sample{
Value: value,
Location: sloc,
NumLabel: map[string][]int64{"bytes": {blocksize}},
})
}
if err := s.Err(); err != nil {
return nil, err
}
if err := parseAdditionalSections(s, p); err != nil {
return nil, err
}
return p, nil
}
func parseHeapHeader(line string) (sampling string, period int64, hasAlloc bool, err error) {
header := heapHeaderRE.FindStringSubmatch(line)
if header == nil {
return "", 0, false, errUnrecognized
}
if len(header[6]) > 0 {
if period, err = strconv.ParseInt(header[6], 10, 64); err != nil {
return "", 0, false, errUnrecognized
}
}
if (header[3] != header[1] && header[3] != "0") || (header[4] != header[2] && header[4] != "0") {
hasAlloc = true
}
switch header[5] {
case "heapz_v2", "heap_v2":
return "v2", period, hasAlloc, nil
case "heapprofile":
return "", 1, hasAlloc, nil
case "heap":
return "v2", period / 2, hasAlloc, nil
default:
return "", 0, false, errUnrecognized
}
}
// parseHeapSample parses a single row from a heap profile into a new Sample.
func parseHeapSample(line string, rate int64, sampling string, includeAlloc bool) (value []int64, blocksize int64, addrs []uint64, err error) {
sampleData := heapSampleRE.FindStringSubmatch(line)
if len(sampleData) != 6 {
return nil, 0, nil, fmt.Errorf("unexpected number of sample values: got %d, want 6", len(sampleData))
}
// This is a local-scoped helper function to avoid needing to pass
// around rate, sampling and many return parameters.
addValues := func(countString, sizeString string, label string) error {
count, err := strconv.ParseInt(countString, 10, 64)
if err != nil {
return fmt.Errorf("malformed sample: %s: %v", line, err)
}
size, err := strconv.ParseInt(sizeString, 10, 64)
if err != nil {
return fmt.Errorf("malformed sample: %s: %v", line, err)
}
if count == 0 && size != 0 {
return fmt.Errorf("%s count was 0 but %s bytes was %d", label, label, size)
}
if count != 0 {
blocksize = size / count
if sampling == "v2" {
count, size = scaleHeapSample(count, size, rate)
}
}
value = append(value, count, size)
return nil
}
if includeAlloc {
if err := addValues(sampleData[3], sampleData[4], "allocation"); err != nil {
return nil, 0, nil, err
}
}
if err := addValues(sampleData[1], sampleData[2], "inuse"); err != nil {
return nil, 0, nil, err
}
addrs, err = parseHexAddresses(sampleData[5])
if err != nil {
return nil, 0, nil, fmt.Errorf("malformed sample: %s: %v", line, err)
}
return value, blocksize, addrs, nil
}
// parseHexAddresses extracts hex numbers from a string, attempts to convert
// each to an unsigned 64-bit number and returns the resulting numbers as a
// slice, or an error if the string contains hex numbers which are too large to
// handle (which means a malformed profile).
func parseHexAddresses(s string) ([]uint64, error) {
hexStrings := hexNumberRE.FindAllString(s, -1)
var addrs []uint64
for _, s := range hexStrings {
if addr, err := strconv.ParseUint(s, 0, 64); err == nil {
addrs = append(addrs, addr)
} else {
return nil, fmt.Errorf("failed to parse as hex 64-bit number: %s", s)
}
}
return addrs, nil
}
// scaleHeapSample adjusts the data from a heapz Sample to
// account for its probability of appearing in the collected
// data. heapz profiles are a sampling of the memory allocations
// requests in a program. We estimate the unsampled value by dividing
// each collected sample by its probability of appearing in the
// profile. heapz v2 profiles rely on a poisson process to determine
// which samples to collect, based on the desired average collection
// rate R. The probability of a sample of size S to appear in that
// profile is 1-exp(-S/R).
func scaleHeapSample(count, size, rate int64) (int64, int64) {
if count == 0 || size == 0 {
return 0, 0
}
if rate <= 1 {
// if rate==1 all samples were collected so no adjustment is needed.
// if rate<1 treat as unknown and skip scaling.
return count, size
}
avgSize := float64(size) / float64(count)
scale := 1 / (1 - math.Exp(-avgSize/float64(rate)))
return int64(float64(count) * scale), int64(float64(size) * scale)
}
// parseContention parses a mutex or contention profile. There are 2 cases:
// "--- contentionz " for legacy C++ profiles (and backwards compatibility)
// "--- mutex:" or "--- contention:" for profiles generated by the Go runtime.
func parseContention(b []byte) (*Profile, error) {
s := bufio.NewScanner(bytes.NewBuffer(b))
if !s.Scan() {
if err := s.Err(); err != nil {
return nil, err
}
return nil, errUnrecognized
}
switch l := s.Text(); {
case strings.HasPrefix(l, "--- contentionz "):
case strings.HasPrefix(l, "--- mutex:"):
case strings.HasPrefix(l, "--- contention:"):
default:
return nil, errUnrecognized
}
p := &Profile{
PeriodType: &ValueType{Type: "contentions", Unit: "count"},
Period: 1,
SampleType: []*ValueType{
{Type: "contentions", Unit: "count"},
{Type: "delay", Unit: "nanoseconds"},
},
}
var cpuHz int64
// Parse text of the form "attribute = value" before the samples.
const delimiter = "="
for s.Scan() {
line := s.Text()
if line = strings.TrimSpace(line); isSpaceOrComment(line) {
continue
}
if strings.HasPrefix(line, "---") {
break
}
attr := strings.SplitN(line, delimiter, 2)
if len(attr) != 2 {
break
}
key, val := strings.TrimSpace(attr[0]), strings.TrimSpace(attr[1])
var err error
switch key {
case "cycles/second":
if cpuHz, err = strconv.ParseInt(val, 0, 64); err != nil {
return nil, errUnrecognized
}
case "sampling period":
if p.Period, err = strconv.ParseInt(val, 0, 64); err != nil {
return nil, errUnrecognized
}
case "ms since reset":
ms, err := strconv.ParseInt(val, 0, 64)
if err != nil {
return nil, errUnrecognized
}
p.DurationNanos = ms * 1000 * 1000
case "format":
// CPP contentionz profiles don't have format.
return nil, errUnrecognized
case "resolution":
// CPP contentionz profiles don't have resolution.
return nil, errUnrecognized
case "discarded samples":
default:
return nil, errUnrecognized
}
}
if err := s.Err(); err != nil {
return nil, err
}
locs := make(map[uint64]*Location)
for {
line := strings.TrimSpace(s.Text())
if strings.HasPrefix(line, "---") {
break
}
if !isSpaceOrComment(line) {
value, addrs, err := parseContentionSample(line, p.Period, cpuHz)
if err != nil {
return nil, err
}
var sloc []*Location
for _, addr := range addrs {
// Addresses from stack traces point to the next instruction after
// each call. Adjust by -1 to land somewhere on the actual call.
addr--
loc := locs[addr]
if locs[addr] == nil {
loc = &Location{
Address: addr,
}
p.Location = append(p.Location, loc)
locs[addr] = loc
}
sloc = append(sloc, loc)
}
p.Sample = append(p.Sample, &Sample{
Value: value,
Location: sloc,
})
}
if !s.Scan() {
break
}
}
if err := s.Err(); err != nil {
return nil, err
}
if err := parseAdditionalSections(s, p); err != nil {
return nil, err
}
return p, nil
}
// parseContentionSample parses a single row from a contention profile
// into a new Sample.
func parseContentionSample(line string, period, cpuHz int64) (value []int64, addrs []uint64, err error) {
sampleData := contentionSampleRE.FindStringSubmatch(line)
if sampleData == nil {
return nil, nil, errUnrecognized
}
v1, err := strconv.ParseInt(sampleData[1], 10, 64)
if err != nil {
return nil, nil, fmt.Errorf("malformed sample: %s: %v", line, err)
}
v2, err := strconv.ParseInt(sampleData[2], 10, 64)
if err != nil {
return nil, nil, fmt.Errorf("malformed sample: %s: %v", line, err)
}
// Unsample values if period and cpuHz are available.
// - Delays are scaled to cycles and then to nanoseconds.
// - Contentions are scaled to cycles.
if period > 0 {
if cpuHz > 0 {
cpuGHz := float64(cpuHz) / 1e9
v1 = int64(float64(v1) * float64(period) / cpuGHz)
}
v2 = v2 * period
}
value = []int64{v2, v1}
addrs, err = parseHexAddresses(sampleData[3])
if err != nil {
return nil, nil, fmt.Errorf("malformed sample: %s: %v", line, err)
}
return value, addrs, nil
}
// parseThread parses a Threadz profile and returns a new Profile.
func parseThread(b []byte) (*Profile, error) {
s := bufio.NewScanner(bytes.NewBuffer(b))
// Skip past comments and empty lines seeking a real header.
for s.Scan() && isSpaceOrComment(s.Text()) {
}
line := s.Text()
if m := threadzStartRE.FindStringSubmatch(line); m != nil {
// Advance over initial comments until first stack trace.
for s.Scan() {
if line = s.Text(); isMemoryMapSentinel(line) || strings.HasPrefix(line, "-") {
break
}
}
} else if t := threadStartRE.FindStringSubmatch(line); len(t) != 4 {
return nil, errUnrecognized
}
p := &Profile{
SampleType: []*ValueType{{Type: "thread", Unit: "count"}},
PeriodType: &ValueType{Type: "thread", Unit: "count"},
Period: 1,
}
locs := make(map[uint64]*Location)
// Recognize each thread and populate profile samples.
for !isMemoryMapSentinel(line) {
if strings.HasPrefix(line, "---- no stack trace for") {
break
}
if t := threadStartRE.FindStringSubmatch(line); len(t) != 4 {
return nil, errUnrecognized
}
var addrs []uint64
var err error
line, addrs, err = parseThreadSample(s)
if err != nil {
return nil, err
}
if len(addrs) == 0 {
// We got a --same as previous threads--. Bump counters.
if len(p.Sample) > 0 {
s := p.Sample[len(p.Sample)-1]
s.Value[0]++
}
continue
}
var sloc []*Location
for i, addr := range addrs {
// Addresses from stack traces point to the next instruction after
// each call. Adjust by -1 to land somewhere on the actual call
// (except for the leaf, which is not a call).
if i > 0 {
addr--
}
loc := locs[addr]
if locs[addr] == nil {
loc = &Location{
Address: addr,
}
p.Location = append(p.Location, loc)
locs[addr] = loc
}
sloc = append(sloc, loc)
}
p.Sample = append(p.Sample, &Sample{
Value: []int64{1},
Location: sloc,
})
}
if err := parseAdditionalSections(s, p); err != nil {
return nil, err
}
cleanupDuplicateLocations(p)
return p, nil
}
// parseThreadSample parses a symbolized or unsymbolized stack trace.
// Returns the first line after the traceback, the sample (or nil if
// it hits a 'same-as-previous' marker) and an error.
func parseThreadSample(s *bufio.Scanner) (nextl string, addrs []uint64, err error) {
var line string
sameAsPrevious := false
for s.Scan() {
line = strings.TrimSpace(s.Text())
if line == "" {
continue
}
if strings.HasPrefix(line, "---") {
break
}
if strings.Contains(line, "same as previous thread") {
sameAsPrevious = true
continue
}
curAddrs, err := parseHexAddresses(line)
if err != nil {
return "", nil, fmt.Errorf("malformed sample: %s: %v", line, err)
}
addrs = append(addrs, curAddrs...)
}
if err := s.Err(); err != nil {
return "", nil, err
}
if sameAsPrevious {
return line, nil, nil
}
return line, addrs, nil
}
// parseAdditionalSections parses any additional sections in the
// profile, ignoring any unrecognized sections.
func parseAdditionalSections(s *bufio.Scanner, p *Profile) error {
for !isMemoryMapSentinel(s.Text()) && s.Scan() {
}
if err := s.Err(); err != nil {
return err
}
return p.ParseMemoryMapFromScanner(s)
}
// ParseProcMaps parses a memory map in the format of /proc/self/maps.
// ParseMemoryMap should be called after setting on a profile to
// associate locations to the corresponding mapping based on their
// address.
func ParseProcMaps(rd io.Reader) ([]*Mapping, error) {
s := bufio.NewScanner(rd)
return parseProcMapsFromScanner(s)
}
func parseProcMapsFromScanner(s *bufio.Scanner) ([]*Mapping, error) {
var mapping []*Mapping
var attrs []string
const delimiter = "="
r := strings.NewReplacer()
for s.Scan() {
line := r.Replace(removeLoggingInfo(s.Text()))
m, err := parseMappingEntry(line)
if err != nil {
if err == errUnrecognized {
// Recognize assignments of the form: attr=value, and replace
// $attr with value on subsequent mappings.
if attr := strings.SplitN(line, delimiter, 2); len(attr) == 2 {
attrs = append(attrs, "$"+strings.TrimSpace(attr[0]), strings.TrimSpace(attr[1]))
r = strings.NewReplacer(attrs...)
}
// Ignore any unrecognized entries
continue
}
return nil, err
}
if m == nil {
continue
}
mapping = append(mapping, m)
}
if err := s.Err(); err != nil {
return nil, err
}
return mapping, nil
}
// removeLoggingInfo detects and removes log prefix entries generated
// by the glog package. If no logging prefix is detected, the string
// is returned unmodified.
func removeLoggingInfo(line string) string {
if match := logInfoRE.FindStringIndex(line); match != nil {
return line[match[1]:]
}
return line
}
// ParseMemoryMap parses a memory map in the format of
// /proc/self/maps, and overrides the mappings in the current profile.
// It renumbers the samples and locations in the profile correspondingly.
func (p *Profile) ParseMemoryMap(rd io.Reader) error {
return p.ParseMemoryMapFromScanner(bufio.NewScanner(rd))
}
// ParseMemoryMapFromScanner parses a memory map in the format of
// /proc/self/maps or a variety of legacy format, and overrides the
// mappings in the current profile. It renumbers the samples and
// locations in the profile correspondingly.
func (p *Profile) ParseMemoryMapFromScanner(s *bufio.Scanner) error {
mapping, err := parseProcMapsFromScanner(s)
if err != nil {
return err
}
p.Mapping = append(p.Mapping, mapping...)
p.massageMappings()
p.remapLocationIDs()
p.remapFunctionIDs()
p.remapMappingIDs()
return nil
}
func parseMappingEntry(l string) (*Mapping, error) {
var start, end, perm, file, offset, buildID string
if me := procMapsRE.FindStringSubmatch(l); len(me) == 6 {
start, end, perm, offset, file = me[1], me[2], me[3], me[4], me[5]
} else if me := briefMapsRE.FindStringSubmatch(l); len(me) == 7 {
start, end, perm, file, offset, buildID = me[1], me[2], me[3], me[4], me[5], me[6]
} else {
return nil, errUnrecognized
}
var err error
mapping := &Mapping{
File: file,
BuildID: buildID,
}
if perm != "" && !strings.Contains(perm, "x") {
// Skip non-executable entries.
return nil, nil
}
if mapping.Start, err = strconv.ParseUint(start, 16, 64); err != nil {
return nil, errUnrecognized
}
if mapping.Limit, err = strconv.ParseUint(end, 16, 64); err != nil {
return nil, errUnrecognized
}
if offset != "" {
if mapping.Offset, err = strconv.ParseUint(offset, 16, 64); err != nil {
return nil, errUnrecognized
}
}
return mapping, nil
}
var memoryMapSentinels = []string{
"--- Memory map: ---",
"MAPPED_LIBRARIES:",
}
// isMemoryMapSentinel returns true if the string contains one of the
// known sentinels for memory map information.
func isMemoryMapSentinel(line string) bool {
for _, s := range memoryMapSentinels {
if strings.Contains(line, s) {
return true
}
}
return false
}
func (p *Profile) addLegacyFrameInfo() {
switch {
case isProfileType(p, heapzSampleTypes):
p.DropFrames, p.KeepFrames = allocRxStr, allocSkipRxStr
case isProfileType(p, contentionzSampleTypes):
p.DropFrames, p.KeepFrames = lockRxStr, ""
default:
p.DropFrames, p.KeepFrames = cpuProfilerRxStr, ""
}
}
var heapzSampleTypes = [][]string{
{"allocations", "size"}, // early Go pprof profiles
{"objects", "space"},
{"inuse_objects", "inuse_space"},
{"alloc_objects", "alloc_space"},
{"alloc_objects", "alloc_space", "inuse_objects", "inuse_space"}, // Go pprof legacy profiles
}
var contentionzSampleTypes = [][]string{
{"contentions", "delay"},
}
func isProfileType(p *Profile, types [][]string) bool {
st := p.SampleType
nextType:
for _, t := range types {
if len(st) != len(t) {
continue
}
for i := range st {
if st[i].Type != t[i] {
continue nextType
}
}
return true
}
return false
}
var allocRxStr = strings.Join([]string{
// POSIX entry points.
`calloc`,
`cfree`,
`malloc`,
`free`,
`memalign`,
`do_memalign`,
`(__)?posix_memalign`,
`pvalloc`,
`valloc`,
`realloc`,
// TC malloc.
`tcmalloc::.*`,
`tc_calloc`,
`tc_cfree`,
`tc_malloc`,
`tc_free`,
`tc_memalign`,
`tc_posix_memalign`,
`tc_pvalloc`,
`tc_valloc`,
`tc_realloc`,
`tc_new`,
`tc_delete`,
`tc_newarray`,
`tc_deletearray`,
`tc_new_nothrow`,
`tc_newarray_nothrow`,
// Memory-allocation routines on OS X.
`malloc_zone_malloc`,
`malloc_zone_calloc`,
`malloc_zone_valloc`,
`malloc_zone_realloc`,
`malloc_zone_memalign`,
`malloc_zone_free`,
// Go runtime
`runtime\..*`,
// Other misc. memory allocation routines
`BaseArena::.*`,
`(::)?do_malloc_no_errno`,
`(::)?do_malloc_pages`,
`(::)?do_malloc`,
`DoSampledAllocation`,
`MallocedMemBlock::MallocedMemBlock`,
`_M_allocate`,
`__builtin_(vec_)?delete`,
`__builtin_(vec_)?new`,
`__gnu_cxx::new_allocator::allocate`,
`__libc_malloc`,
`__malloc_alloc_template::allocate`,
`allocate`,
`cpp_alloc`,
`operator new(\[\])?`,
`simple_alloc::allocate`,
}, `|`)
var allocSkipRxStr = strings.Join([]string{
// Preserve Go runtime frames that appear in the middle/bottom of
// the stack.
`runtime\.panic`,
`runtime\.reflectcall`,
`runtime\.call[0-9]*`,
}, `|`)
var cpuProfilerRxStr = strings.Join([]string{
`ProfileData::Add`,
`ProfileData::prof_handler`,
`CpuProfiler::prof_handler`,
`__pthread_sighandler`,
`__restore`,
}, `|`)
var lockRxStr = strings.Join([]string{
`RecordLockProfileData`,
`(base::)?RecordLockProfileData.*`,
`(base::)?SubmitMutexProfileData.*`,
`(base::)?SubmitSpinLockProfileData.*`,
`(base::Mutex::)?AwaitCommon.*`,
`(base::Mutex::)?Unlock.*`,
`(base::Mutex::)?UnlockSlow.*`,
`(base::Mutex::)?ReaderUnlock.*`,
`(base::MutexLock::)?~MutexLock.*`,
`(Mutex::)?AwaitCommon.*`,
`(Mutex::)?Unlock.*`,
`(Mutex::)?UnlockSlow.*`,
`(Mutex::)?ReaderUnlock.*`,
`(MutexLock::)?~MutexLock.*`,
`(SpinLock::)?Unlock.*`,
`(SpinLock::)?SlowUnlock.*`,
`(SpinLockHolder::)?~SpinLockHolder.*`,
}, `|`)
// Copyright 2014 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package profile
import (
"encoding/binary"
"fmt"
"slices"
"sort"
"strconv"
"strings"
)
// Compact performs garbage collection on a profile to remove any
// unreferenced fields. This is useful to reduce the size of a profile
// after samples or locations have been removed.
func (p *Profile) Compact() *Profile {
p, _ = Merge([]*Profile{p})
return p
}
// Merge merges all the profiles in profs into a single Profile.
// Returns a new profile independent of the input profiles. The merged
// profile is compacted to eliminate unused samples, locations,
// functions and mappings. Profiles must have identical profile sample
// and period types or the merge will fail. profile.Period of the
// resulting profile will be the maximum of all profiles, and
// profile.TimeNanos will be the earliest nonzero one. Merges are
// associative with the caveat of the first profile having some
// specialization in how headers are combined. There may be other
// subtleties now or in the future regarding associativity.
func Merge(srcs []*Profile) (*Profile, error) {
if len(srcs) == 0 {
return nil, fmt.Errorf("no profiles to merge")
}
p, err := combineHeaders(srcs)
if err != nil {
return nil, err
}
pm := &profileMerger{
p: p,
samples: make(map[sampleKey]*Sample, len(srcs[0].Sample)),
locations: make(map[locationKey]*Location, len(srcs[0].Location)),
functions: make(map[functionKey]*Function, len(srcs[0].Function)),
mappings: make(map[mappingKey]*Mapping, len(srcs[0].Mapping)),
}
for _, src := range srcs {
// Clear the profile-specific hash tables
pm.locationsByID = makeLocationIDMap(len(src.Location))
pm.functionsByID = make(map[uint64]*Function, len(src.Function))
pm.mappingsByID = make(map[uint64]mapInfo, len(src.Mapping))
if len(pm.mappings) == 0 && len(src.Mapping) > 0 {
// The Mapping list has the property that the first mapping
// represents the main binary. Take the first Mapping we see,
// otherwise the operations below will add mappings in an
// arbitrary order.
pm.mapMapping(src.Mapping[0])
}
for _, s := range src.Sample {
if !isZeroSample(s) {
pm.mapSample(s)
}
}
}
if slices.ContainsFunc(p.Sample, isZeroSample) {
// If there are any zero samples, re-merge the profile to GC
// them.
return Merge([]*Profile{p})
}
return p, nil
}
// Normalize normalizes the source profile by multiplying each value in profile by the
// ratio of the sum of the base profile's values of that sample type to the sum of the
// source profile's value of that sample type.
func (p *Profile) Normalize(pb *Profile) error {
if err := p.compatible(pb); err != nil {
return err
}
baseVals := make([]int64, len(p.SampleType))
for _, s := range pb.Sample {
for i, v := range s.Value {
baseVals[i] += v
}
}
srcVals := make([]int64, len(p.SampleType))
for _, s := range p.Sample {
for i, v := range s.Value {
srcVals[i] += v
}
}
normScale := make([]float64, len(baseVals))
for i := range baseVals {
if srcVals[i] == 0 {
normScale[i] = 0.0
} else {
normScale[i] = float64(baseVals[i]) / float64(srcVals[i])
}
}
p.ScaleN(normScale)
return nil
}
func isZeroSample(s *Sample) bool {
for _, v := range s.Value {
if v != 0 {
return false
}
}
return true
}
type profileMerger struct {
p *Profile
// Memoization tables within a profile.
locationsByID locationIDMap
functionsByID map[uint64]*Function
mappingsByID map[uint64]mapInfo
// Memoization tables for profile entities.
samples map[sampleKey]*Sample
locations map[locationKey]*Location
functions map[functionKey]*Function
mappings map[mappingKey]*Mapping
}
type mapInfo struct {
m *Mapping
offset int64
}
func (pm *profileMerger) mapSample(src *Sample) *Sample {
// Check memoization table
k := pm.sampleKey(src)
if ss, ok := pm.samples[k]; ok {
for i, v := range src.Value {
ss.Value[i] += v
}
return ss
}
// Make new sample.
s := &Sample{
Location: make([]*Location, len(src.Location)),
Value: make([]int64, len(src.Value)),
Label: make(map[string][]string, len(src.Label)),
NumLabel: make(map[string][]int64, len(src.NumLabel)),
NumUnit: make(map[string][]string, len(src.NumLabel)),
}
for i, l := range src.Location {
s.Location[i] = pm.mapLocation(l)
}
for k, v := range src.Label {
vv := make([]string, len(v))
copy(vv, v)
s.Label[k] = vv
}
for k, v := range src.NumLabel {
u := src.NumUnit[k]
vv := make([]int64, len(v))
uu := make([]string, len(u))
copy(vv, v)
copy(uu, u)
s.NumLabel[k] = vv
s.NumUnit[k] = uu
}
copy(s.Value, src.Value)
pm.samples[k] = s
pm.p.Sample = append(pm.p.Sample, s)
return s
}
func (pm *profileMerger) sampleKey(sample *Sample) sampleKey {
// Accumulate contents into a string.
var buf strings.Builder
buf.Grow(64) // Heuristic to avoid extra allocs
// encode a number
putNumber := func(v uint64) {
var num [binary.MaxVarintLen64]byte
n := binary.PutUvarint(num[:], v)
buf.Write(num[:n])
}
// encode a string prefixed with its length.
putDelimitedString := func(s string) {
putNumber(uint64(len(s)))
buf.WriteString(s)
}
for _, l := range sample.Location {
// Get the location in the merged profile, which may have a different ID.
if loc := pm.mapLocation(l); loc != nil {
putNumber(loc.ID)
}
}
putNumber(0) // Delimiter
for _, l := range sortedKeys1(sample.Label) {
putDelimitedString(l)
values := sample.Label[l]
putNumber(uint64(len(values)))
for _, v := range values {
putDelimitedString(v)
}
}
for _, l := range sortedKeys2(sample.NumLabel) {
putDelimitedString(l)
values := sample.NumLabel[l]
putNumber(uint64(len(values)))
for _, v := range values {
putNumber(uint64(v))
}
units := sample.NumUnit[l]
putNumber(uint64(len(units)))
for _, v := range units {
putDelimitedString(v)
}
}
return sampleKey(buf.String())
}
type sampleKey string
// sortedKeys1 returns the sorted keys found in a string->[]string map.
//
// Note: this is currently non-generic since github pprof runs golint,
// which does not support generics. When that issue is fixed, it can
// be merged with sortedKeys2 and made into a generic function.
func sortedKeys1(m map[string][]string) []string {
if len(m) == 0 {
return nil
}
keys := make([]string, 0, len(m))
for k := range m {
keys = append(keys, k)
}
sort.Strings(keys)
return keys
}
// sortedKeys2 returns the sorted keys found in a string->[]int64 map.
//
// Note: this is currently non-generic since github pprof runs golint,
// which does not support generics. When that issue is fixed, it can
// be merged with sortedKeys1 and made into a generic function.
func sortedKeys2(m map[string][]int64) []string {
if len(m) == 0 {
return nil
}
keys := make([]string, 0, len(m))
for k := range m {
keys = append(keys, k)
}
sort.Strings(keys)
return keys
}
func (pm *profileMerger) mapLocation(src *Location) *Location {
if src == nil {
return nil
}
if l := pm.locationsByID.get(src.ID); l != nil {
return l
}
mi := pm.mapMapping(src.Mapping)
l := &Location{
ID: uint64(len(pm.p.Location) + 1),
Mapping: mi.m,
Address: uint64(int64(src.Address) + mi.offset),
Line: make([]Line, len(src.Line)),
IsFolded: src.IsFolded,
}
for i, ln := range src.Line {
l.Line[i] = pm.mapLine(ln)
}
// Check memoization table. Must be done on the remapped location to
// account for the remapped mapping ID.
k := l.key()
if ll, ok := pm.locations[k]; ok {
pm.locationsByID.set(src.ID, ll)
return ll
}
pm.locationsByID.set(src.ID, l)
pm.locations[k] = l
pm.p.Location = append(pm.p.Location, l)
return l
}
// key generates locationKey to be used as a key for maps.
func (l *Location) key() locationKey {
key := locationKey{
addr: l.Address,
isFolded: l.IsFolded,
}
if l.Mapping != nil {
// Normalizes address to handle address space randomization.
key.addr -= l.Mapping.Start
key.mappingID = l.Mapping.ID
}
lines := make([]string, len(l.Line)*3)
for i, line := range l.Line {
if line.Function != nil {
lines[i*2] = strconv.FormatUint(line.Function.ID, 16)
}
lines[i*2+1] = strconv.FormatInt(line.Line, 16)
lines[i*2+2] = strconv.FormatInt(line.Column, 16)
}
key.lines = strings.Join(lines, "|")
return key
}
type locationKey struct {
addr, mappingID uint64
lines string
isFolded bool
}
func (pm *profileMerger) mapMapping(src *Mapping) mapInfo {
if src == nil {
return mapInfo{}
}
if mi, ok := pm.mappingsByID[src.ID]; ok {
return mi
}
// Check memoization tables.
mk := src.key()
if m, ok := pm.mappings[mk]; ok {
mi := mapInfo{m, int64(m.Start) - int64(src.Start)}
pm.mappingsByID[src.ID] = mi
return mi
}
m := &Mapping{
ID: uint64(len(pm.p.Mapping) + 1),
Start: src.Start,
Limit: src.Limit,
Offset: src.Offset,
File: src.File,
KernelRelocationSymbol: src.KernelRelocationSymbol,
BuildID: src.BuildID,
HasFunctions: src.HasFunctions,
HasFilenames: src.HasFilenames,
HasLineNumbers: src.HasLineNumbers,
HasInlineFrames: src.HasInlineFrames,
}
pm.p.Mapping = append(pm.p.Mapping, m)
// Update memoization tables.
pm.mappings[mk] = m
mi := mapInfo{m, 0}
pm.mappingsByID[src.ID] = mi
return mi
}
// key generates encoded strings of Mapping to be used as a key for
// maps.
func (m *Mapping) key() mappingKey {
// Normalize addresses to handle address space randomization.
// Round up to next 4K boundary to avoid minor discrepancies.
const mapsizeRounding = 0x1000
size := m.Limit - m.Start
size = size + mapsizeRounding - 1
size = size - (size % mapsizeRounding)
key := mappingKey{
size: size,
offset: m.Offset,
}
switch {
case m.BuildID != "":
key.buildIDOrFile = m.BuildID
case m.File != "":
key.buildIDOrFile = m.File
default:
// A mapping containing neither build ID nor file name is a fake mapping. A
// key with empty buildIDOrFile is used for fake mappings so that they are
// treated as the same mapping during merging.
}
return key
}
type mappingKey struct {
size, offset uint64
buildIDOrFile string
}
func (pm *profileMerger) mapLine(src Line) Line {
ln := Line{
Function: pm.mapFunction(src.Function),
Line: src.Line,
Column: src.Column,
}
return ln
}
func (pm *profileMerger) mapFunction(src *Function) *Function {
if src == nil {
return nil
}
if f, ok := pm.functionsByID[src.ID]; ok {
return f
}
k := src.key()
if f, ok := pm.functions[k]; ok {
pm.functionsByID[src.ID] = f
return f
}
f := &Function{
ID: uint64(len(pm.p.Function) + 1),
Name: src.Name,
SystemName: src.SystemName,
Filename: src.Filename,
StartLine: src.StartLine,
}
pm.functions[k] = f
pm.functionsByID[src.ID] = f
pm.p.Function = append(pm.p.Function, f)
return f
}
// key generates a struct to be used as a key for maps.
func (f *Function) key() functionKey {
return functionKey{
f.StartLine,
f.Name,
f.SystemName,
f.Filename,
}
}
type functionKey struct {
startLine int64
name, systemName, fileName string
}
// combineHeaders checks that all profiles can be merged and returns
// their combined profile.
func combineHeaders(srcs []*Profile) (*Profile, error) {
for _, s := range srcs[1:] {
if err := srcs[0].compatible(s); err != nil {
return nil, err
}
}
var timeNanos, durationNanos, period int64
var comments []string
seenComments := map[string]bool{}
var docURL string
var defaultSampleType string
for _, s := range srcs {
if timeNanos == 0 || s.TimeNanos < timeNanos {
timeNanos = s.TimeNanos
}
durationNanos += s.DurationNanos
if period == 0 || period < s.Period {
period = s.Period
}
for _, c := range s.Comments {
if seen := seenComments[c]; !seen {
comments = append(comments, c)
seenComments[c] = true
}
}
if defaultSampleType == "" {
defaultSampleType = s.DefaultSampleType
}
if docURL == "" {
docURL = s.DocURL
}
}
p := &Profile{
SampleType: make([]*ValueType, len(srcs[0].SampleType)),
DropFrames: srcs[0].DropFrames,
KeepFrames: srcs[0].KeepFrames,
TimeNanos: timeNanos,
DurationNanos: durationNanos,
PeriodType: srcs[0].PeriodType,
Period: period,
Comments: comments,
DefaultSampleType: defaultSampleType,
DocURL: docURL,
}
copy(p.SampleType, srcs[0].SampleType)
return p, nil
}
// compatible determines if two profiles can be compared/merged.
// returns nil if the profiles are compatible; otherwise an error with
// details on the incompatibility.
func (p *Profile) compatible(pb *Profile) error {
if !equalValueType(p.PeriodType, pb.PeriodType) {
return fmt.Errorf("incompatible period types %v and %v", p.PeriodType, pb.PeriodType)
}
if len(p.SampleType) != len(pb.SampleType) {
return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType)
}
for i := range p.SampleType {
if !equalValueType(p.SampleType[i], pb.SampleType[i]) {
return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType)
}
}
return nil
}
// equalValueType returns true if the two value types are semantically
// equal. It ignores the internal fields used during encode/decode.
func equalValueType(st1, st2 *ValueType) bool {
return st1.Type == st2.Type && st1.Unit == st2.Unit
}
// locationIDMap is like a map[uint64]*Location, but provides efficiency for
// ids that are densely numbered, which is often the case.
type locationIDMap struct {
dense []*Location // indexed by id for id < len(dense)
sparse map[uint64]*Location // indexed by id for id >= len(dense)
}
func makeLocationIDMap(n int) locationIDMap {
return locationIDMap{
dense: make([]*Location, n),
sparse: map[uint64]*Location{},
}
}
func (lm locationIDMap) get(id uint64) *Location {
if id < uint64(len(lm.dense)) {
return lm.dense[int(id)]
}
return lm.sparse[id]
}
func (lm locationIDMap) set(id uint64, loc *Location) {
if id < uint64(len(lm.dense)) {
lm.dense[id] = loc
return
}
lm.sparse[id] = loc
}
// CompatibilizeSampleTypes makes profiles compatible to be compared/merged. It
// keeps sample types that appear in all profiles only and drops/reorders the
// sample types as necessary.
//
// In the case of sample types order is not the same for given profiles the
// order is derived from the first profile.
//
// Profiles are modified in-place.
//
// It returns an error if the sample type's intersection is empty.
func CompatibilizeSampleTypes(ps []*Profile) error {
sTypes := commonSampleTypes(ps)
if len(sTypes) == 0 {
return fmt.Errorf("profiles have empty common sample type list")
}
for _, p := range ps {
if err := compatibilizeSampleTypes(p, sTypes); err != nil {
return err
}
}
return nil
}
// commonSampleTypes returns sample types that appear in all profiles in the
// order how they ordered in the first profile.
func commonSampleTypes(ps []*Profile) []string {
if len(ps) == 0 {
return nil
}
sTypes := map[string]int{}
for _, p := range ps {
for _, st := range p.SampleType {
sTypes[st.Type]++
}
}
var res []string
for _, st := range ps[0].SampleType {
if sTypes[st.Type] == len(ps) {
res = append(res, st.Type)
}
}
return res
}
// compatibilizeSampleTypes drops sample types that are not present in sTypes
// list and reorder them if needed.
//
// It sets DefaultSampleType to sType[0] if it is not in sType list.
//
// It assumes that all sample types from the sTypes list are present in the
// given profile otherwise it returns an error.
func compatibilizeSampleTypes(p *Profile, sTypes []string) error {
if len(sTypes) == 0 {
return fmt.Errorf("sample type list is empty")
}
defaultSampleType := sTypes[0]
reMap, needToModify := make([]int, len(sTypes)), false
for i, st := range sTypes {
if st == p.DefaultSampleType {
defaultSampleType = p.DefaultSampleType
}
idx := searchValueType(p.SampleType, st)
if idx < 0 {
return fmt.Errorf("%q sample type is not found in profile", st)
}
reMap[i] = idx
if idx != i {
needToModify = true
}
}
if !needToModify && len(sTypes) == len(p.SampleType) {
return nil
}
p.DefaultSampleType = defaultSampleType
oldSampleTypes := p.SampleType
p.SampleType = make([]*ValueType, len(sTypes))
for i, idx := range reMap {
p.SampleType[i] = oldSampleTypes[idx]
}
values := make([]int64, len(sTypes))
for _, s := range p.Sample {
for i, idx := range reMap {
values[i] = s.Value[idx]
}
s.Value = s.Value[:len(values)]
copy(s.Value, values)
}
return nil
}
func searchValueType(vts []*ValueType, s string) int {
for i, vt := range vts {
if vt.Type == s {
return i
}
}
return -1
}
// Copyright 2014 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// Package profile provides a representation of profile.proto and
// methods to encode/decode profiles in this format.
package profile
import (
"bytes"
"compress/gzip"
"fmt"
"io"
"math"
"path/filepath"
"regexp"
"slices"
"sort"
"strings"
"sync"
"time"
)
// Profile is an in-memory representation of profile.proto.
type Profile struct {
SampleType []*ValueType
DefaultSampleType string
Sample []*Sample
Mapping []*Mapping
Location []*Location
Function []*Function
Comments []string
DocURL string
DropFrames string
KeepFrames string
TimeNanos int64
DurationNanos int64
PeriodType *ValueType
Period int64
// The following fields are modified during encoding and copying,
// so are protected by a Mutex.
encodeMu sync.Mutex
commentX []int64
docURLX int64
dropFramesX int64
keepFramesX int64
stringTable []string
defaultSampleTypeX int64
}
// ValueType corresponds to Profile.ValueType
type ValueType struct {
Type string // cpu, wall, inuse_space, etc
Unit string // seconds, nanoseconds, bytes, etc
typeX int64
unitX int64
}
// Sample corresponds to Profile.Sample
type Sample struct {
Location []*Location
Value []int64
// Label is a per-label-key map to values for string labels.
//
// In general, having multiple values for the given label key is strongly
// discouraged - see docs for the sample label field in profile.proto. The
// main reason this unlikely state is tracked here is to make the
// decoding->encoding roundtrip not lossy. But we expect that the value
// slices present in this map are always of length 1.
Label map[string][]string
// NumLabel is a per-label-key map to values for numeric labels. See a note
// above on handling multiple values for a label.
NumLabel map[string][]int64
// NumUnit is a per-label-key map to the unit names of corresponding numeric
// label values. The unit info may be missing even if the label is in
// NumLabel, see the docs in profile.proto for details. When the value is
// slice is present and not nil, its length must be equal to the length of
// the corresponding value slice in NumLabel.
NumUnit map[string][]string
locationIDX []uint64
labelX []label
}
// label corresponds to Profile.Label
type label struct {
keyX int64
// Exactly one of the two following values must be set
strX int64
numX int64 // Integer value for this label
// can be set if numX has value
unitX int64
}
// Mapping corresponds to Profile.Mapping
type Mapping struct {
ID uint64
Start uint64
Limit uint64
Offset uint64
File string
BuildID string
HasFunctions bool
HasFilenames bool
HasLineNumbers bool
HasInlineFrames bool
fileX int64
buildIDX int64
// Name of the kernel relocation symbol ("_text" or "_stext"), extracted from File.
// For linux kernel mappings generated by some tools, correct symbolization depends
// on knowing which of the two possible relocation symbols was used for `Start`.
// This is given to us as a suffix in `File` (e.g. "[kernel.kallsyms]_stext").
//
// Note, this public field is not persisted in the proto. For the purposes of
// copying / merging / hashing profiles, it is considered subsumed by `File`.
KernelRelocationSymbol string
}
// Location corresponds to Profile.Location
type Location struct {
ID uint64
Mapping *Mapping
Address uint64
Line []Line
IsFolded bool
mappingIDX uint64
}
// Line corresponds to Profile.Line
type Line struct {
Function *Function
Line int64
Column int64
functionIDX uint64
}
// Function corresponds to Profile.Function
type Function struct {
ID uint64
Name string
SystemName string
Filename string
StartLine int64
nameX int64
systemNameX int64
filenameX int64
}
// Parse parses a profile and checks for its validity. The input
// may be a gzip-compressed encoded protobuf or one of many legacy
// profile formats which may be unsupported in the future.
func Parse(r io.Reader) (*Profile, error) {
data, err := io.ReadAll(r)
if err != nil {
return nil, err
}
return ParseData(data)
}
// ParseData parses a profile from a buffer and checks for its
// validity.
func ParseData(data []byte) (*Profile, error) {
var p *Profile
var err error
if len(data) >= 2 && data[0] == 0x1f && data[1] == 0x8b {
gz, err := gzip.NewReader(bytes.NewBuffer(data))
if err == nil {
data, err = io.ReadAll(gz)
}
if err != nil {
return nil, fmt.Errorf("decompressing profile: %v", err)
}
}
if p, err = ParseUncompressed(data); err != nil && err != errNoData && err != errConcatProfile {
p, err = parseLegacy(data)
}
if err != nil {
return nil, fmt.Errorf("parsing profile: %v", err)
}
if err := p.CheckValid(); err != nil {
return nil, fmt.Errorf("malformed profile: %v", err)
}
return p, nil
}
var errUnrecognized = fmt.Errorf("unrecognized profile format")
var errMalformed = fmt.Errorf("malformed profile format")
var errNoData = fmt.Errorf("empty input file")
var errConcatProfile = fmt.Errorf("concatenated profiles detected")
func parseLegacy(data []byte) (*Profile, error) {
parsers := []func([]byte) (*Profile, error){
parseCPU,
parseHeap,
parseGoCount, // goroutine, threadcreate
parseThread,
parseContention,
parseJavaProfile,
}
for _, parser := range parsers {
p, err := parser(data)
if err == nil {
p.addLegacyFrameInfo()
return p, nil
}
if err != errUnrecognized {
return nil, err
}
}
return nil, errUnrecognized
}
// ParseUncompressed parses an uncompressed protobuf into a profile.
func ParseUncompressed(data []byte) (*Profile, error) {
if len(data) == 0 {
return nil, errNoData
}
p := &Profile{}
if err := unmarshal(data, p); err != nil {
return nil, err
}
if err := p.postDecode(); err != nil {
return nil, err
}
return p, nil
}
var libRx = regexp.MustCompile(`([.]so$|[.]so[._][0-9]+)`)
// massageMappings applies heuristic-based changes to the profile
// mappings to account for quirks of some environments.
func (p *Profile) massageMappings() {
// Merge adjacent regions with matching names, checking that the offsets match
if len(p.Mapping) > 1 {
mappings := []*Mapping{p.Mapping[0]}
for _, m := range p.Mapping[1:] {
lm := mappings[len(mappings)-1]
if adjacent(lm, m) {
lm.Limit = m.Limit
if m.File != "" {
lm.File = m.File
}
if m.BuildID != "" {
lm.BuildID = m.BuildID
}
p.updateLocationMapping(m, lm)
continue
}
mappings = append(mappings, m)
}
p.Mapping = mappings
}
// Use heuristics to identify main binary and move it to the top of the list of mappings
for i, m := range p.Mapping {
file := strings.TrimSpace(strings.Replace(m.File, "(deleted)", "", -1))
if len(file) == 0 {
continue
}
if len(libRx.FindStringSubmatch(file)) > 0 {
continue
}
if file[0] == '[' {
continue
}
// Swap what we guess is main to position 0.
p.Mapping[0], p.Mapping[i] = p.Mapping[i], p.Mapping[0]
break
}
// Keep the mapping IDs neatly sorted
for i, m := range p.Mapping {
m.ID = uint64(i + 1)
}
}
// adjacent returns whether two mapping entries represent the same
// mapping that has been split into two. Check that their addresses are adjacent,
// and if the offsets match, if they are available.
func adjacent(m1, m2 *Mapping) bool {
if m1.File != "" && m2.File != "" {
if m1.File != m2.File {
return false
}
}
if m1.BuildID != "" && m2.BuildID != "" {
if m1.BuildID != m2.BuildID {
return false
}
}
if m1.Limit != m2.Start {
return false
}
if m1.Offset != 0 && m2.Offset != 0 {
offset := m1.Offset + (m1.Limit - m1.Start)
if offset != m2.Offset {
return false
}
}
return true
}
func (p *Profile) updateLocationMapping(from, to *Mapping) {
for _, l := range p.Location {
if l.Mapping == from {
l.Mapping = to
}
}
}
func serialize(p *Profile) []byte {
p.encodeMu.Lock()
p.preEncode()
b := marshal(p)
p.encodeMu.Unlock()
return b
}
// Write writes the profile as a gzip-compressed marshaled protobuf.
func (p *Profile) Write(w io.Writer) error {
zw := gzip.NewWriter(w)
defer zw.Close()
_, err := zw.Write(serialize(p))
return err
}
// WriteUncompressed writes the profile as a marshaled protobuf.
func (p *Profile) WriteUncompressed(w io.Writer) error {
_, err := w.Write(serialize(p))
return err
}
// CheckValid tests whether the profile is valid. Checks include, but are
// not limited to:
// - len(Profile.Sample[n].value) == len(Profile.value_unit)
// - Sample.id has a corresponding Profile.Location
func (p *Profile) CheckValid() error {
// Check that sample values are consistent
sampleLen := len(p.SampleType)
if sampleLen == 0 && len(p.Sample) != 0 {
return fmt.Errorf("missing sample type information")
}
for _, s := range p.Sample {
if s == nil {
return fmt.Errorf("profile has nil sample")
}
if len(s.Value) != sampleLen {
return fmt.Errorf("mismatch: sample has %d values vs. %d types", len(s.Value), len(p.SampleType))
}
for _, l := range s.Location {
if l == nil {
return fmt.Errorf("sample has nil location")
}
}
}
// Check that all mappings/locations/functions are in the tables
// Check that there are no duplicate ids
mappings := make(map[uint64]*Mapping, len(p.Mapping))
for _, m := range p.Mapping {
if m == nil {
return fmt.Errorf("profile has nil mapping")
}
if m.ID == 0 {
return fmt.Errorf("found mapping with reserved ID=0")
}
if mappings[m.ID] != nil {
return fmt.Errorf("multiple mappings with same id: %d", m.ID)
}
mappings[m.ID] = m
}
functions := make(map[uint64]*Function, len(p.Function))
for _, f := range p.Function {
if f == nil {
return fmt.Errorf("profile has nil function")
}
if f.ID == 0 {
return fmt.Errorf("found function with reserved ID=0")
}
if functions[f.ID] != nil {
return fmt.Errorf("multiple functions with same id: %d", f.ID)
}
functions[f.ID] = f
}
locations := make(map[uint64]*Location, len(p.Location))
for _, l := range p.Location {
if l == nil {
return fmt.Errorf("profile has nil location")
}
if l.ID == 0 {
return fmt.Errorf("found location with reserved id=0")
}
if locations[l.ID] != nil {
return fmt.Errorf("multiple locations with same id: %d", l.ID)
}
locations[l.ID] = l
if m := l.Mapping; m != nil {
if m.ID == 0 || mappings[m.ID] != m {
return fmt.Errorf("inconsistent mapping %p: %d", m, m.ID)
}
}
for _, ln := range l.Line {
f := ln.Function
if f == nil {
return fmt.Errorf("location id: %d has a line with nil function", l.ID)
}
if f.ID == 0 || functions[f.ID] != f {
return fmt.Errorf("inconsistent function %p: %d", f, f.ID)
}
}
}
return nil
}
// Aggregate merges the locations in the profile into equivalence
// classes preserving the request attributes. It also updates the
// samples to point to the merged locations.
func (p *Profile) Aggregate(inlineFrame, function, filename, linenumber, columnnumber, address bool) error {
for _, m := range p.Mapping {
m.HasInlineFrames = m.HasInlineFrames && inlineFrame
m.HasFunctions = m.HasFunctions && function
m.HasFilenames = m.HasFilenames && filename
m.HasLineNumbers = m.HasLineNumbers && linenumber
}
// Aggregate functions
if !function || !filename {
for _, f := range p.Function {
if !function {
f.Name = ""
f.SystemName = ""
}
if !filename {
f.Filename = ""
}
}
}
// Aggregate locations
if !inlineFrame || !address || !linenumber || !columnnumber {
for _, l := range p.Location {
if !inlineFrame && len(l.Line) > 1 {
l.Line = l.Line[len(l.Line)-1:]
}
if !linenumber {
for i := range l.Line {
l.Line[i].Line = 0
l.Line[i].Column = 0
}
}
if !columnnumber {
for i := range l.Line {
l.Line[i].Column = 0
}
}
if !address {
l.Address = 0
}
}
}
return p.CheckValid()
}
// NumLabelUnits returns a map of numeric label keys to the units
// associated with those keys and a map of those keys to any units
// that were encountered but not used.
// Unit for a given key is the first encountered unit for that key. If multiple
// units are encountered for values paired with a particular key, then the first
// unit encountered is used and all other units are returned in sorted order
// in map of ignored units.
// If no units are encountered for a particular key, the unit is then inferred
// based on the key.
func (p *Profile) NumLabelUnits() (map[string]string, map[string][]string) {
numLabelUnits := map[string]string{}
ignoredUnits := map[string]map[string]bool{}
encounteredKeys := map[string]bool{}
// Determine units based on numeric tags for each sample.
for _, s := range p.Sample {
for k := range s.NumLabel {
encounteredKeys[k] = true
for _, unit := range s.NumUnit[k] {
if unit == "" {
continue
}
if wantUnit, ok := numLabelUnits[k]; !ok {
numLabelUnits[k] = unit
} else if wantUnit != unit {
if v, ok := ignoredUnits[k]; ok {
v[unit] = true
} else {
ignoredUnits[k] = map[string]bool{unit: true}
}
}
}
}
}
// Infer units for keys without any units associated with
// numeric tag values.
for key := range encounteredKeys {
unit := numLabelUnits[key]
if unit == "" {
switch key {
case "alignment", "request":
numLabelUnits[key] = "bytes"
default:
numLabelUnits[key] = key
}
}
}
// Copy ignored units into more readable format
unitsIgnored := make(map[string][]string, len(ignoredUnits))
for key, values := range ignoredUnits {
units := make([]string, len(values))
i := 0
for unit := range values {
units[i] = unit
i++
}
sort.Strings(units)
unitsIgnored[key] = units
}
return numLabelUnits, unitsIgnored
}
// String dumps a text representation of a profile. Intended mainly
// for debugging purposes.
func (p *Profile) String() string {
ss := make([]string, 0, len(p.Comments)+len(p.Sample)+len(p.Mapping)+len(p.Location))
for _, c := range p.Comments {
ss = append(ss, "Comment: "+c)
}
if url := p.DocURL; url != "" {
ss = append(ss, fmt.Sprintf("Doc: %s", url))
}
if pt := p.PeriodType; pt != nil {
ss = append(ss, fmt.Sprintf("PeriodType: %s %s", pt.Type, pt.Unit))
}
ss = append(ss, fmt.Sprintf("Period: %d", p.Period))
if p.TimeNanos != 0 {
ss = append(ss, fmt.Sprintf("Time: %v", time.Unix(0, p.TimeNanos)))
}
if p.DurationNanos != 0 {
ss = append(ss, fmt.Sprintf("Duration: %.4v", time.Duration(p.DurationNanos)))
}
ss = append(ss, "Samples:")
var sh1 string
for _, s := range p.SampleType {
dflt := ""
if s.Type == p.DefaultSampleType {
dflt = "[dflt]"
}
sh1 = sh1 + fmt.Sprintf("%s/%s%s ", s.Type, s.Unit, dflt)
}
ss = append(ss, strings.TrimSpace(sh1))
for _, s := range p.Sample {
ss = append(ss, s.string())
}
ss = append(ss, "Locations")
for _, l := range p.Location {
ss = append(ss, l.string())
}
ss = append(ss, "Mappings")
for _, m := range p.Mapping {
ss = append(ss, m.string())
}
return strings.Join(ss, "\n") + "\n"
}
// string dumps a text representation of a mapping. Intended mainly
// for debugging purposes.
func (m *Mapping) string() string {
bits := ""
if m.HasFunctions {
bits = bits + "[FN]"
}
if m.HasFilenames {
bits = bits + "[FL]"
}
if m.HasLineNumbers {
bits = bits + "[LN]"
}
if m.HasInlineFrames {
bits = bits + "[IN]"
}
return fmt.Sprintf("%d: %#x/%#x/%#x %s %s %s",
m.ID,
m.Start, m.Limit, m.Offset,
m.File,
m.BuildID,
bits)
}
// string dumps a text representation of a location. Intended mainly
// for debugging purposes.
func (l *Location) string() string {
ss := []string{}
locStr := fmt.Sprintf("%6d: %#x ", l.ID, l.Address)
if m := l.Mapping; m != nil {
locStr = locStr + fmt.Sprintf("M=%d ", m.ID)
}
if l.IsFolded {
locStr = locStr + "[F] "
}
if len(l.Line) == 0 {
ss = append(ss, locStr)
}
for li := range l.Line {
lnStr := "??"
if fn := l.Line[li].Function; fn != nil {
lnStr = fmt.Sprintf("%s %s:%d:%d s=%d",
fn.Name,
fn.Filename,
l.Line[li].Line,
l.Line[li].Column,
fn.StartLine)
if fn.Name != fn.SystemName {
lnStr = lnStr + "(" + fn.SystemName + ")"
}
}
ss = append(ss, locStr+lnStr)
// Do not print location details past the first line
locStr = " "
}
return strings.Join(ss, "\n")
}
// string dumps a text representation of a sample. Intended mainly
// for debugging purposes.
func (s *Sample) string() string {
ss := []string{}
var sv string
for _, v := range s.Value {
sv = fmt.Sprintf("%s %10d", sv, v)
}
sv = sv + ": "
for _, l := range s.Location {
sv = sv + fmt.Sprintf("%d ", l.ID)
}
ss = append(ss, sv)
const labelHeader = " "
if len(s.Label) > 0 {
ss = append(ss, labelHeader+labelsToString(s.Label))
}
if len(s.NumLabel) > 0 {
ss = append(ss, labelHeader+numLabelsToString(s.NumLabel, s.NumUnit))
}
return strings.Join(ss, "\n")
}
// labelsToString returns a string representation of a
// map representing labels.
func labelsToString(labels map[string][]string) string {
ls := []string{}
for k, v := range labels {
ls = append(ls, fmt.Sprintf("%s:%v", k, v))
}
sort.Strings(ls)
return strings.Join(ls, " ")
}
// numLabelsToString returns a string representation of a map
// representing numeric labels.
func numLabelsToString(numLabels map[string][]int64, numUnits map[string][]string) string {
ls := []string{}
for k, v := range numLabels {
units := numUnits[k]
var labelString string
if len(units) == len(v) {
values := make([]string, len(v))
for i, vv := range v {
values[i] = fmt.Sprintf("%d %s", vv, units[i])
}
labelString = fmt.Sprintf("%s:%v", k, values)
} else {
labelString = fmt.Sprintf("%s:%v", k, v)
}
ls = append(ls, labelString)
}
sort.Strings(ls)
return strings.Join(ls, " ")
}
// SetLabel sets the specified key to the specified value for all samples in the
// profile.
func (p *Profile) SetLabel(key string, value []string) {
for _, sample := range p.Sample {
if sample.Label == nil {
sample.Label = map[string][]string{key: value}
} else {
sample.Label[key] = value
}
}
}
// RemoveLabel removes all labels associated with the specified key for all
// samples in the profile.
func (p *Profile) RemoveLabel(key string) {
for _, sample := range p.Sample {
delete(sample.Label, key)
}
}
// HasLabel returns true if a sample has a label with indicated key and value.
func (s *Sample) HasLabel(key, value string) bool {
return slices.Contains(s.Label[key], value)
}
// SetNumLabel sets the specified key to the specified value for all samples in the
// profile. "unit" is a slice that describes the units that each corresponding member
// of "values" is measured in (e.g. bytes or seconds). If there is no relevant
// unit for a given value, that member of "unit" should be the empty string.
// "unit" must either have the same length as "value", or be nil.
func (p *Profile) SetNumLabel(key string, value []int64, unit []string) {
for _, sample := range p.Sample {
if sample.NumLabel == nil {
sample.NumLabel = map[string][]int64{key: value}
} else {
sample.NumLabel[key] = value
}
if sample.NumUnit == nil {
sample.NumUnit = map[string][]string{key: unit}
} else {
sample.NumUnit[key] = unit
}
}
}
// RemoveNumLabel removes all numerical labels associated with the specified key for all
// samples in the profile.
func (p *Profile) RemoveNumLabel(key string) {
for _, sample := range p.Sample {
delete(sample.NumLabel, key)
delete(sample.NumUnit, key)
}
}
// DiffBaseSample returns true if a sample belongs to the diff base and false
// otherwise.
func (s *Sample) DiffBaseSample() bool {
return s.HasLabel("pprof::base", "true")
}
// Scale multiplies all sample values in a profile by a constant and keeps
// only samples that have at least one non-zero value.
func (p *Profile) Scale(ratio float64) {
if ratio == 1 {
return
}
ratios := make([]float64, len(p.SampleType))
for i := range p.SampleType {
ratios[i] = ratio
}
p.ScaleN(ratios)
}
// ScaleN multiplies each sample values in a sample by a different amount
// and keeps only samples that have at least one non-zero value.
func (p *Profile) ScaleN(ratios []float64) error {
if len(p.SampleType) != len(ratios) {
return fmt.Errorf("mismatched scale ratios, got %d, want %d", len(ratios), len(p.SampleType))
}
allOnes := true
for _, r := range ratios {
if r != 1 {
allOnes = false
break
}
}
if allOnes {
return nil
}
fillIdx := 0
for _, s := range p.Sample {
keepSample := false
for i, v := range s.Value {
if ratios[i] != 1 {
val := int64(math.Round(float64(v) * ratios[i]))
s.Value[i] = val
keepSample = keepSample || val != 0
}
}
if keepSample {
p.Sample[fillIdx] = s
fillIdx++
}
}
p.Sample = p.Sample[:fillIdx]
return nil
}
// HasFunctions determines if all locations in this profile have
// symbolized function information.
func (p *Profile) HasFunctions() bool {
for _, l := range p.Location {
if l.Mapping != nil && !l.Mapping.HasFunctions {
return false
}
}
return true
}
// HasFileLines determines if all locations in this profile have
// symbolized file and line number information.
func (p *Profile) HasFileLines() bool {
for _, l := range p.Location {
if l.Mapping != nil && (!l.Mapping.HasFilenames || !l.Mapping.HasLineNumbers) {
return false
}
}
return true
}
// Unsymbolizable returns true if a mapping points to a binary for which
// locations can't be symbolized in principle, at least now. Examples are
// "[vdso]", "[vsyscall]" and some others, see the code.
func (m *Mapping) Unsymbolizable() bool {
name := filepath.Base(m.File)
switch {
case strings.HasPrefix(name, "["):
case strings.HasPrefix(name, "linux-vdso"):
case strings.HasPrefix(m.File, "/dev/dri/"):
case m.File == "//anon":
case m.File == "":
case strings.HasPrefix(m.File, "/memfd:"):
default:
return false
}
return true
}
// Copy makes a fully independent copy of a profile.
func (p *Profile) Copy() *Profile {
pp := &Profile{}
if err := unmarshal(serialize(p), pp); err != nil {
panic(err)
}
if err := pp.postDecode(); err != nil {
panic(err)
}
return pp
}
// Copyright 2014 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// This file is a simple protocol buffer encoder and decoder.
// The format is described at
// https://developers.google.com/protocol-buffers/docs/encoding
//
// A protocol message must implement the message interface:
// decoder() []decoder
// encode(*buffer)
//
// The decode method returns a slice indexed by field number that gives the
// function to decode that field.
// The encode method encodes its receiver into the given buffer.
//
// The two methods are simple enough to be implemented by hand rather than
// by using a protocol compiler.
//
// See profile.go for examples of messages implementing this interface.
//
// There is no support for groups, message sets, or "has" bits.
package profile
import (
"errors"
"fmt"
)
type buffer struct {
field int // field tag
typ int // proto wire type code for field
u64 uint64
data []byte
tmp [16]byte
tmpLines []Line // temporary storage used while decoding "repeated Line".
}
type decoder func(*buffer, message) error
type message interface {
decoder() []decoder
encode(*buffer)
}
func marshal(m message) []byte {
var b buffer
m.encode(&b)
return b.data
}
func encodeVarint(b *buffer, x uint64) {
for x >= 128 {
b.data = append(b.data, byte(x)|0x80)
x >>= 7
}
b.data = append(b.data, byte(x))
}
func encodeLength(b *buffer, tag int, len int) {
encodeVarint(b, uint64(tag)<<3|2)
encodeVarint(b, uint64(len))
}
func encodeUint64(b *buffer, tag int, x uint64) {
// append varint to b.data
encodeVarint(b, uint64(tag)<<3)
encodeVarint(b, x)
}
func encodeUint64s(b *buffer, tag int, x []uint64) {
if len(x) > 2 {
// Use packed encoding
n1 := len(b.data)
for _, u := range x {
encodeVarint(b, u)
}
n2 := len(b.data)
encodeLength(b, tag, n2-n1)
n3 := len(b.data)
copy(b.tmp[:], b.data[n2:n3])
copy(b.data[n1+(n3-n2):], b.data[n1:n2])
copy(b.data[n1:], b.tmp[:n3-n2])
return
}
for _, u := range x {
encodeUint64(b, tag, u)
}
}
func encodeUint64Opt(b *buffer, tag int, x uint64) {
if x == 0 {
return
}
encodeUint64(b, tag, x)
}
func encodeInt64(b *buffer, tag int, x int64) {
u := uint64(x)
encodeUint64(b, tag, u)
}
func encodeInt64s(b *buffer, tag int, x []int64) {
if len(x) > 2 {
// Use packed encoding
n1 := len(b.data)
for _, u := range x {
encodeVarint(b, uint64(u))
}
n2 := len(b.data)
encodeLength(b, tag, n2-n1)
n3 := len(b.data)
copy(b.tmp[:], b.data[n2:n3])
copy(b.data[n1+(n3-n2):], b.data[n1:n2])
copy(b.data[n1:], b.tmp[:n3-n2])
return
}
for _, u := range x {
encodeInt64(b, tag, u)
}
}
func encodeInt64Opt(b *buffer, tag int, x int64) {
if x == 0 {
return
}
encodeInt64(b, tag, x)
}
func encodeString(b *buffer, tag int, x string) {
encodeLength(b, tag, len(x))
b.data = append(b.data, x...)
}
func encodeStrings(b *buffer, tag int, x []string) {
for _, s := range x {
encodeString(b, tag, s)
}
}
func encodeBool(b *buffer, tag int, x bool) {
if x {
encodeUint64(b, tag, 1)
} else {
encodeUint64(b, tag, 0)
}
}
func encodeBoolOpt(b *buffer, tag int, x bool) {
if x {
encodeBool(b, tag, x)
}
}
func encodeMessage(b *buffer, tag int, m message) {
n1 := len(b.data)
m.encode(b)
n2 := len(b.data)
encodeLength(b, tag, n2-n1)
n3 := len(b.data)
copy(b.tmp[:], b.data[n2:n3])
copy(b.data[n1+(n3-n2):], b.data[n1:n2])
copy(b.data[n1:], b.tmp[:n3-n2])
}
func unmarshal(data []byte, m message) (err error) {
b := buffer{data: data, typ: 2}
return decodeMessage(&b, m)
}
func le64(p []byte) uint64 {
return uint64(p[0]) | uint64(p[1])<<8 | uint64(p[2])<<16 | uint64(p[3])<<24 | uint64(p[4])<<32 | uint64(p[5])<<40 | uint64(p[6])<<48 | uint64(p[7])<<56
}
func le32(p []byte) uint32 {
return uint32(p[0]) | uint32(p[1])<<8 | uint32(p[2])<<16 | uint32(p[3])<<24
}
func decodeVarint(data []byte) (uint64, []byte, error) {
var u uint64
for i := 0; ; i++ {
if i >= 10 || i >= len(data) {
return 0, nil, errors.New("bad varint")
}
u |= uint64(data[i]&0x7F) << uint(7*i)
if data[i]&0x80 == 0 {
return u, data[i+1:], nil
}
}
}
func decodeField(b *buffer, data []byte) ([]byte, error) {
x, data, err := decodeVarint(data)
if err != nil {
return nil, err
}
b.field = int(x >> 3)
b.typ = int(x & 7)
b.data = nil
b.u64 = 0
switch b.typ {
case 0:
b.u64, data, err = decodeVarint(data)
if err != nil {
return nil, err
}
case 1:
if len(data) < 8 {
return nil, errors.New("not enough data")
}
b.u64 = le64(data[:8])
data = data[8:]
case 2:
var n uint64
n, data, err = decodeVarint(data)
if err != nil {
return nil, err
}
if n > uint64(len(data)) {
return nil, errors.New("too much data")
}
b.data = data[:n]
data = data[n:]
case 5:
if len(data) < 4 {
return nil, errors.New("not enough data")
}
b.u64 = uint64(le32(data[:4]))
data = data[4:]
default:
return nil, fmt.Errorf("unknown wire type: %d", b.typ)
}
return data, nil
}
func checkType(b *buffer, typ int) error {
if b.typ != typ {
return errors.New("type mismatch")
}
return nil
}
func decodeMessage(b *buffer, m message) error {
if err := checkType(b, 2); err != nil {
return err
}
dec := m.decoder()
data := b.data
for len(data) > 0 {
// pull varint field# + type
var err error
data, err = decodeField(b, data)
if err != nil {
return err
}
if b.field >= len(dec) || dec[b.field] == nil {
continue
}
if err := dec[b.field](b, m); err != nil {
return err
}
}
return nil
}
func decodeInt64(b *buffer, x *int64) error {
if err := checkType(b, 0); err != nil {
return err
}
*x = int64(b.u64)
return nil
}
func decodeInt64s(b *buffer, x *[]int64) error {
if b.typ == 2 {
// Packed encoding
data := b.data
for len(data) > 0 {
var u uint64
var err error
if u, data, err = decodeVarint(data); err != nil {
return err
}
*x = append(*x, int64(u))
}
return nil
}
var i int64
if err := decodeInt64(b, &i); err != nil {
return err
}
*x = append(*x, i)
return nil
}
func decodeUint64(b *buffer, x *uint64) error {
if err := checkType(b, 0); err != nil {
return err
}
*x = b.u64
return nil
}
func decodeUint64s(b *buffer, x *[]uint64) error {
if b.typ == 2 {
data := b.data
// Packed encoding
for len(data) > 0 {
var u uint64
var err error
if u, data, err = decodeVarint(data); err != nil {
return err
}
*x = append(*x, u)
}
return nil
}
var u uint64
if err := decodeUint64(b, &u); err != nil {
return err
}
*x = append(*x, u)
return nil
}
func decodeString(b *buffer, x *string) error {
if err := checkType(b, 2); err != nil {
return err
}
*x = string(b.data)
return nil
}
func decodeStrings(b *buffer, x *[]string) error {
var s string
if err := decodeString(b, &s); err != nil {
return err
}
*x = append(*x, s)
return nil
}
func decodeBool(b *buffer, x *bool) error {
if err := checkType(b, 0); err != nil {
return err
}
if int64(b.u64) == 0 {
*x = false
} else {
*x = true
}
return nil
}
// Copyright 2014 Google Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
// Implements methods to remove frames from profiles.
package profile
import (
"fmt"
"regexp"
"slices"
"strings"
)
var (
reservedNames = []string{"(anonymous namespace)", "operator()"}
bracketRx = func() *regexp.Regexp {
var quotedNames []string
for _, name := range append(reservedNames, "(") {
quotedNames = append(quotedNames, regexp.QuoteMeta(name))
}
return regexp.MustCompile(strings.Join(quotedNames, "|"))
}()
)
// simplifyFunc does some primitive simplification of function names.
func simplifyFunc(f string) string {
// Account for leading '.' on the PPC ELF v1 ABI.
funcName := strings.TrimPrefix(f, ".")
// Account for unsimplified names -- try to remove the argument list by trimming
// starting from the first '(', but skipping reserved names that have '('.
for _, ind := range bracketRx.FindAllStringSubmatchIndex(funcName, -1) {
foundReserved := slices.Contains(reservedNames, funcName[ind[0]:ind[1]])
if !foundReserved {
funcName = funcName[:ind[0]]
break
}
}
return funcName
}
// Prune removes all nodes beneath a node matching dropRx, and not
// matching keepRx. If the root node of a Sample matches, the sample
// will have an empty stack.
func (p *Profile) Prune(dropRx, keepRx *regexp.Regexp) {
prune := make(map[uint64]bool)
pruneBeneath := make(map[uint64]bool)
// simplifyFunc can be expensive, so cache results.
// Note that the same function name can be encountered many times due
// different lines and addresses in the same function.
pruneCache := map[string]bool{} // Map from function to whether or not to prune
pruneFromHere := func(s string) bool {
if r, ok := pruneCache[s]; ok {
return r
}
funcName := simplifyFunc(s)
if dropRx.MatchString(funcName) {
if keepRx == nil || !keepRx.MatchString(funcName) {
pruneCache[s] = true
return true
}
}
pruneCache[s] = false
return false
}
for _, loc := range p.Location {
var i int
for i = len(loc.Line) - 1; i >= 0; i-- {
if fn := loc.Line[i].Function; fn != nil && fn.Name != "" {
if pruneFromHere(fn.Name) {
break
}
}
}
if i >= 0 {
// Found matching entry to prune.
pruneBeneath[loc.ID] = true
// Remove the matching location.
if i == len(loc.Line)-1 {
// Matched the top entry: prune the whole location.
prune[loc.ID] = true
} else {
loc.Line = loc.Line[i+1:]
}
}
}
// Prune locs from each Sample
for _, sample := range p.Sample {
// Scan from the root to the leaves to find the prune location.
// Do not prune frames before the first user frame, to avoid
// pruning everything.
foundUser := false
for i := len(sample.Location) - 1; i >= 0; i-- {
id := sample.Location[i].ID
if !prune[id] && !pruneBeneath[id] {
foundUser = true
continue
}
if !foundUser {
continue
}
if prune[id] {
sample.Location = sample.Location[i+1:]
break
}
if pruneBeneath[id] {
sample.Location = sample.Location[i:]
break
}
}
}
}
// RemoveUninteresting prunes and elides profiles using built-in
// tables of uninteresting function names.
func (p *Profile) RemoveUninteresting() error {
var keep, drop *regexp.Regexp
var err error
if p.DropFrames != "" {
if drop, err = regexp.Compile("^(" + p.DropFrames + ")$"); err != nil {
return fmt.Errorf("failed to compile regexp %s: %v", p.DropFrames, err)
}
if p.KeepFrames != "" {
if keep, err = regexp.Compile("^(" + p.KeepFrames + ")$"); err != nil {
return fmt.Errorf("failed to compile regexp %s: %v", p.KeepFrames, err)
}
}
p.Prune(drop, keep)
}
return nil
}
// PruneFrom removes all nodes beneath the lowest node matching dropRx, not including itself.
//
// Please see the example below to understand this method as well as
// the difference from Prune method.
//
// A sample contains Location of [A,B,C,B,D] where D is the top frame and there's no inline.
//
// PruneFrom(A) returns [A,B,C,B,D] because there's no node beneath A.
// Prune(A, nil) returns [B,C,B,D] by removing A itself.
//
// PruneFrom(B) returns [B,C,B,D] by removing all nodes beneath the first B when scanning from the bottom.
// Prune(B, nil) returns [D] because a matching node is found by scanning from the root.
func (p *Profile) PruneFrom(dropRx *regexp.Regexp) {
pruneBeneath := make(map[uint64]bool)
for _, loc := range p.Location {
for i := 0; i < len(loc.Line); i++ {
if fn := loc.Line[i].Function; fn != nil && fn.Name != "" {
funcName := simplifyFunc(fn.Name)
if dropRx.MatchString(funcName) {
// Found matching entry to prune.
pruneBeneath[loc.ID] = true
loc.Line = loc.Line[i:]
break
}
}
}
}
// Prune locs from each Sample
for _, sample := range p.Sample {
// Scan from the bottom leaf to the root to find the prune location.
for i, loc := range sample.Location {
if pruneBeneath[loc.ID] {
sample.Location = sample.Location[i:]
break
}
}
}
}