diff options
Diffstat (limited to '')
| -rw-r--r-- | src/internal/profile/encode.go | 482 | ||||
| -rw-r--r-- | src/internal/profile/filter.go | 158 | ||||
| -rw-r--r-- | src/internal/profile/legacy_profile.go | 1266 | ||||
| -rw-r--r-- | src/internal/profile/merge.go | 461 | ||||
| -rw-r--r-- | src/internal/profile/profile.go | 613 | ||||
| -rw-r--r-- | src/internal/profile/profile_test.go | 79 | ||||
| -rw-r--r-- | src/internal/profile/proto.go | 363 | ||||
| -rw-r--r-- | src/internal/profile/proto_test.go | 67 | ||||
| -rw-r--r-- | src/internal/profile/prune.go | 97 |
9 files changed, 3586 insertions, 0 deletions
diff --git a/src/internal/profile/encode.go b/src/internal/profile/encode.go new file mode 100644 index 0000000..af31933 --- /dev/null +++ b/src/internal/profile/encode.go @@ -0,0 +1,482 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package profile + +import ( + "errors" + "fmt" + "sort" +) + +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 { + vs := s.NumLabel[k] + for _, v := range vs { + s.labelX = append(s.labelX, + Label{ + keyX: addString(strings, k), + numX: v, + }, + ) + } + } + s.locationIDX = nil + for _, l := range s.Location { + s.locationIDX = append(s.locationIDX, l.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.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) +} + +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) + pp := m.(*Profile) + pp.Location = append(pp.Location, x) + return decodeMessage(b, x) + }, + // 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 + }, + // repeated int64 drop_frames = 7 + func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).dropFramesX) }, + // repeated int64 keep_frames = 8 + func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).keepFramesX) }, + // repeated int64 time_nanos = 9 + func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).TimeNanos) }, + // repeated int64 duration_nanos = 10 + func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).DurationNanos) }, + // optional string period_type = 11 + func(b *buffer, m message) error { + x := new(ValueType) + pp := m.(*Profile) + pp.PeriodType = x + return decodeMessage(b, x) + }, + // repeated int64 period = 12 + func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).Period) }, + // repeated int64 comment = 13 + func(b *buffer, m message) error { return decodeInt64s(b, &m.(*Profile).commentX) }, + // int64 defaultSampleType = 14 + func(b *buffer, m message) error { return decodeInt64(b, &m.(*Profile).defaultSampleTypeX) }, +} + +// postDecode takes the unexported fields populated by decode (with +// 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) + for _, m := range p.Mapping { + m.File, err = getString(p.stringTable, &m.fileX, err) + m.BuildID, err = getString(p.stringTable, &m.buildIDX, err) + mappings[m.ID] = m + } + + functions := make(map[uint64]*Function) + 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) + functions[f.ID] = f + } + + locations := make(map[uint64]*Location) + for _, l := range p.Location { + l.Mapping = mappings[l.mappingIDX] + l.mappingIDX = 0 + for i, ln := range l.Line { + if id := ln.functionIDX; id != 0 { + l.Line[i].Function = functions[id] + if l.Line[i].Function == nil { + return fmt.Errorf("Function ID %d not found", id) + } + l.Line[i].functionIDX = 0 + } + } + 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) + } + + for _, s := range p.Sample { + labels := make(map[string][]string) + numLabels := make(map[string][]int64) + 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 { + numLabels[key] = append(numLabels[key], l.numX) + } + } + if len(labels) > 0 { + s.Label = labels + } + if len(numLabels) > 0 { + s.NumLabel = numLabels + } + s.Location = nil + for _, lid := range s.locationIDX { + s.Location = append(s.Location, 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.stringTable = nil + return nil +} + +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) + for _, x := range p.Value { + encodeInt64(b, 2, x) + } + 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) +} + +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) }, +} + +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]) + } +} + +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 (p *Line) decoder() []decoder { + return lineDecoder +} + +func (p *Line) encode(b *buffer) { + encodeUint64Opt(b, 1, p.functionIDX) + encodeInt64Opt(b, 2, p.Line) +} + +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) }, +} + +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 +} diff --git a/src/internal/profile/filter.go b/src/internal/profile/filter.go new file mode 100644 index 0000000..9cad866 --- /dev/null +++ b/src/internal/profile/filter.go @@ -0,0 +1,158 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Implements methods to filter samples from profiles. + +package profile + +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 *regexp.Regexp) (fm, im, hm bool) { + 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 + } + } + } + + 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 +} + +// matchesName reports whether the function name or file in the +// location matches the regular expression. +func (loc *Location) matchesName(re *regexp.Regexp) bool { + for _, ln := range loc.Line { + if fn := ln.Function; fn != nil { + if re.MatchString(fn.Name) { + return true + } + if re.MatchString(fn.Filename) { + 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 { + var lines []Line + for _, ln := range loc.Line { + if fn := ln.Function; fn != nil { + if re.MatchString(fn.Name) { + continue + } + if 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(key, val string, nval int64) 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 := focusedSample(s, focus, ignore) + fm = fm || focused + im = im || ignored + if focused && !ignored { + samples = append(samples, s) + } + } + p.Sample = samples + return +} + +// focusedTag checks a sample against focus and ignore regexps. +// Returns whether the focus/ignore regexps match any tags +func focusedSample(s *Sample, focus, ignore TagMatch) (fm, im bool) { + fm = focus == nil + for key, vals := range s.Label { + for _, val := range vals { + if ignore != nil && ignore(key, val, 0) { + im = true + } + if !fm && focus(key, val, 0) { + fm = true + } + } + } + for key, vals := range s.NumLabel { + for _, val := range vals { + if ignore != nil && ignore(key, "", val) { + im = true + } + if !fm && focus(key, "", val) { + fm = true + } + } + } + return fm, im +} diff --git a/src/internal/profile/legacy_profile.go b/src/internal/profile/legacy_profile.go new file mode 100644 index 0000000..d69f8de --- /dev/null +++ b/src/internal/profile/legacy_profile.go @@ -0,0 +1,1266 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// 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(\w+) profile: total \d+\n\z`) + countRE = regexp.MustCompile(`\A(\d+) @(( 0x[0-9a-f]+)+)\n\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: ---`) + + procMapsRE = regexp.MustCompile(`([[:xdigit:]]+)-([[:xdigit:]]+)\s+([-rwxp]+)\s+([[:xdigit:]]+)\s+([[:xdigit:]]+):([[:xdigit:]]+)\s+([[:digit:]]+)\s*(\S+)?`) + + briefMapsRE = regexp.MustCompile(`\s*([[:xdigit:]]+)-([[:xdigit:]]+):\s*(\S+)(\s.*@)?([[:xdigit:]]+)?`) + + // LegacyHeapAllocated instructs the heapz parsers to use the + // allocated memory stats instead of the default in-use memory. Note + // that tcmalloc doesn't provide all allocated memory, only in-use + // stats. + LegacyHeapAllocated bool +) + +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) { + r := bytes.NewBuffer(b) + + var line string + var err error + for { + // Skip past comments and empty lines seeking a real header. + line, err = r.ReadString('\n') + if err != nil { + return nil, err + } + if !isSpaceOrComment(line) { + break + } + } + + m := countStartRE.FindStringSubmatch(line) + 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 { + line, err = r.ReadString('\n') + if err != nil { + if err == io.EOF { + break + } + return nil, err + } + 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 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 = parseAdditionalSections(strings.TrimSpace(line), r, 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() { + if len(p.Mapping) == 0 { + return + } + + // Some profile handlers will incorrectly set regions for the main + // executable if its section is remapped. Fix them through heuristics. + + // 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. + const expectedStart = 0x400000 + if m := p.Mapping[0]; m.Start-m.Offset == expectedStart { + m.Start = expectedStart + m.Offset = 0 + } + + for _, l := range p.Location { + if a := l.Address; a != 0 { + for _, m := range p.Mapping { + if m.Start <= a && a < m.Limit { + l.Mapping = m + break + } + } + } + } + + // 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:] +} + +// ParseTracebacks parses a set of tracebacks and returns a newly +// populated profile. It will accept any text file and generate a +// Profile out of it with any hex addresses it can identify, including +// a process map if it can recognize one. Each sample will include a +// tag "source" with the addresses recognized in string format. +func ParseTracebacks(b []byte) (*Profile, error) { + r := bytes.NewBuffer(b) + + p := &Profile{ + PeriodType: &ValueType{Type: "trace", Unit: "count"}, + Period: 1, + SampleType: []*ValueType{ + {Type: "trace", Unit: "count"}, + }, + } + + var sources []string + var sloc []*Location + + locs := make(map[uint64]*Location) + for { + l, err := r.ReadString('\n') + if err != nil { + if err != io.EOF { + return nil, err + } + if l == "" { + break + } + } + if sectionTrigger(l) == memoryMapSection { + break + } + if s, addrs := extractHexAddresses(l); len(s) > 0 { + 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) + } + + sources = append(sources, s...) + } else { + if len(sources) > 0 || len(sloc) > 0 { + addTracebackSample(sloc, sources, p) + sloc, sources = nil, nil + } + } + } + + // Add final sample to save any leftover data. + if len(sources) > 0 || len(sloc) > 0 { + addTracebackSample(sloc, sources, p) + } + + if err := p.ParseMemoryMap(r); err != nil { + return nil, err + } + return p, nil +} + +func addTracebackSample(l []*Location, s []string, p *Profile) { + p.Sample = append(p.Sample, + &Sample{ + Value: []int64{1}, + Location: l, + Label: map[string][]string{"source": s}, + }) +} + +// 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) + } + } + 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 all 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. + if len(p.Sample) > 1 && len(p.Sample[0].Location) > 1 { + allSame := true + id1 := p.Sample[0].Location[1].Address + for _, s := range p.Sample { + if len(s.Location) < 2 || id1 != s.Location[1].Address { + allSame = false + break + } + } + if allSame { + for _, s := range p.Sample { + s.Location = append(s.Location[:1], s.Location[2:]...) + } + } + } + + if err := p.ParseMemoryMap(bytes.NewBuffer(b)); err != nil { + return nil, err + } + return p, nil +} + +// 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) { + r := bytes.NewBuffer(b) + l, err := r.ReadString('\n') + if err != nil { + return nil, errUnrecognized + } + + sampling := "" + + if header := heapHeaderRE.FindStringSubmatch(l); header != nil { + p = &Profile{ + SampleType: []*ValueType{ + {Type: "objects", Unit: "count"}, + {Type: "space", Unit: "bytes"}, + }, + PeriodType: &ValueType{Type: "objects", Unit: "bytes"}, + } + + var period int64 + if len(header[6]) > 0 { + if period, err = strconv.ParseInt(header[6], 10, 64); err != nil { + return nil, errUnrecognized + } + } + + switch header[5] { + case "heapz_v2", "heap_v2": + sampling, p.Period = "v2", period + case "heapprofile": + sampling, p.Period = "", 1 + case "heap": + sampling, p.Period = "v2", period/2 + default: + return nil, errUnrecognized + } + } else if header = growthHeaderRE.FindStringSubmatch(l); header != nil { + p = &Profile{ + SampleType: []*ValueType{ + {Type: "objects", Unit: "count"}, + {Type: "space", Unit: "bytes"}, + }, + PeriodType: &ValueType{Type: "heapgrowth", Unit: "count"}, + Period: 1, + } + } else if header = fragmentationHeaderRE.FindStringSubmatch(l); header != nil { + p = &Profile{ + SampleType: []*ValueType{ + {Type: "objects", Unit: "count"}, + {Type: "space", Unit: "bytes"}, + }, + PeriodType: &ValueType{Type: "allocations", Unit: "count"}, + Period: 1, + } + } else { + return nil, errUnrecognized + } + + if LegacyHeapAllocated { + for _, st := range p.SampleType { + st.Type = "alloc_" + st.Type + } + } else { + for _, st := range p.SampleType { + st.Type = "inuse_" + st.Type + } + } + + locs := make(map[uint64]*Location) + for { + l, err = r.ReadString('\n') + if err != nil { + if err != io.EOF { + return nil, err + } + + if l == "" { + break + } + } + + if isSpaceOrComment(l) { + continue + } + l = strings.TrimSpace(l) + + if sectionTrigger(l) != unrecognizedSection { + break + } + + value, blocksize, addrs, err := parseHeapSample(l, p.Period, sampling) + 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 = parseAdditionalSections(l, r, p); err != nil { + return nil, err + } + return p, nil +} + +// parseHeapSample parses a single row from a heap profile into a new Sample. +func parseHeapSample(line string, rate int64, sampling string) (value []int64, blocksize int64, addrs []uint64, err error) { + sampleData := heapSampleRE.FindStringSubmatch(line) + if len(sampleData) != 6 { + return value, blocksize, addrs, fmt.Errorf("unexpected number of sample values: got %d, want 6", len(sampleData)) + } + + // Use first two values by default; tcmalloc sampling generates the + // same value for both, only the older heap-profile collect separate + // stats for in-use and allocated objects. + valueIndex := 1 + if LegacyHeapAllocated { + valueIndex = 3 + } + + var v1, v2 int64 + if v1, err = strconv.ParseInt(sampleData[valueIndex], 10, 64); err != nil { + return value, blocksize, addrs, fmt.Errorf("malformed sample: %s: %v", line, err) + } + if v2, err = strconv.ParseInt(sampleData[valueIndex+1], 10, 64); err != nil { + return value, blocksize, addrs, fmt.Errorf("malformed sample: %s: %v", line, err) + } + + if v1 == 0 { + if v2 != 0 { + return value, blocksize, addrs, fmt.Errorf("allocation count was 0 but allocation bytes was %d", v2) + } + } else { + blocksize = v2 / v1 + if sampling == "v2" { + v1, v2 = scaleHeapSample(v1, v2, rate) + } + } + + value = []int64{v1, v2} + addrs = parseHexAddresses(sampleData[5]) + + return value, blocksize, addrs, nil +} + +// extractHexAddresses extracts hex numbers from a string and returns +// them, together with their numeric value, in a slice. +func extractHexAddresses(s string) ([]string, []uint64) { + hexStrings := hexNumberRE.FindAllString(s, -1) + var ids []uint64 + for _, s := range hexStrings { + if id, err := strconv.ParseUint(s, 0, 64); err == nil { + ids = append(ids, id) + } else { + // Do not expect any parsing failures due to the regexp matching. + panic("failed to parse hex value:" + s) + } + } + return hexStrings, ids +} + +// parseHexAddresses parses hex numbers from a string and returns them +// in a slice. +func parseHexAddresses(s string) []uint64 { + _, ids := extractHexAddresses(s) + return ids +} + +// 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. +// This code converts the text output from runtime into a *Profile. (In the future +// the runtime might write a serialized Profile directly making this unnecessary.) +func parseContention(b []byte) (*Profile, error) { + r := bytes.NewBuffer(b) + var l string + var err error + for { + // Skip past comments and empty lines seeking a real header. + l, err = r.ReadString('\n') + if err != nil { + return nil, err + } + if !isSpaceOrComment(l) { + break + } + } + + if strings.HasPrefix(l, "--- contentionz ") { + return parseCppContention(r) + } else if strings.HasPrefix(l, "--- mutex:") { + return parseCppContention(r) + } else if strings.HasPrefix(l, "--- contention:") { + return parseCppContention(r) + } + return nil, errUnrecognized +} + +// parseCppContention parses the output from synchronization_profiling.cc +// for backward compatibility, and the compatible (non-debug) block profile +// output from the Go runtime. +func parseCppContention(r *bytes.Buffer) (*Profile, error) { + p := &Profile{ + PeriodType: &ValueType{Type: "contentions", Unit: "count"}, + Period: 1, + SampleType: []*ValueType{ + {Type: "contentions", Unit: "count"}, + {Type: "delay", Unit: "nanoseconds"}, + }, + } + + var cpuHz int64 + var l string + var err error + // Parse text of the form "attribute = value" before the samples. + const delimiter = "=" + for { + l, err = r.ReadString('\n') + if err != nil { + if err != io.EOF { + return nil, err + } + + if l == "" { + break + } + } + if isSpaceOrComment(l) { + continue + } + + if l = strings.TrimSpace(l); l == "" { + continue + } + + if strings.HasPrefix(l, "---") { + break + } + + attr := strings.SplitN(l, 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 + } + } + + locs := make(map[uint64]*Location) + for { + if !isSpaceOrComment(l) { + if l = strings.TrimSpace(l); strings.HasPrefix(l, "---") { + break + } + value, addrs, err := parseContentionSample(l, 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 l, err = r.ReadString('\n'); err != nil { + if err != io.EOF { + return nil, err + } + if l == "" { + break + } + } + } + + if err = parseAdditionalSections(l, r, 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 value, addrs, errUnrecognized + } + + v1, err := strconv.ParseInt(sampleData[1], 10, 64) + if err != nil { + return value, addrs, fmt.Errorf("malformed sample: %s: %v", line, err) + } + v2, err := strconv.ParseInt(sampleData[2], 10, 64) + if err != nil { + return value, addrs, 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 = parseHexAddresses(sampleData[3]) + + return value, addrs, nil +} + +// parseThread parses a Threadz profile and returns a new Profile. +func parseThread(b []byte) (*Profile, error) { + r := bytes.NewBuffer(b) + + var line string + var err error + for { + // Skip past comments and empty lines seeking a real header. + line, err = r.ReadString('\n') + if err != nil { + return nil, err + } + if !isSpaceOrComment(line) { + break + } + } + + if m := threadzStartRE.FindStringSubmatch(line); m != nil { + // Advance over initial comments until first stack trace. + for { + line, err = r.ReadString('\n') + if err != nil { + if err != io.EOF { + return nil, err + } + + if line == "" { + break + } + } + if sectionTrigger(line) != unrecognizedSection || line[0] == '-' { + 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 sectionTrigger(line) == unrecognizedSection { + if strings.HasPrefix(line, "---- no stack trace for") { + line = "" + break + } + if t := threadStartRE.FindStringSubmatch(line); len(t) != 4 { + return nil, errUnrecognized + } + + var addrs []uint64 + line, addrs, err = parseThreadSample(r) + if err != nil { + return nil, errUnrecognized + } + 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 _, 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: []int64{1}, + Location: sloc, + }) + } + + if err = parseAdditionalSections(line, r, p); err != nil { + return nil, err + } + + 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(b *bytes.Buffer) (nextl string, addrs []uint64, err error) { + var l string + sameAsPrevious := false + for { + if l, err = b.ReadString('\n'); err != nil { + if err != io.EOF { + return "", nil, err + } + if l == "" { + break + } + } + if l = strings.TrimSpace(l); l == "" { + continue + } + + if strings.HasPrefix(l, "---") { + break + } + if strings.Contains(l, "same as previous thread") { + sameAsPrevious = true + continue + } + + addrs = append(addrs, parseHexAddresses(l)...) + } + + if sameAsPrevious { + return l, nil, nil + } + return l, addrs, nil +} + +// parseAdditionalSections parses any additional sections in the +// profile, ignoring any unrecognized sections. +func parseAdditionalSections(l string, b *bytes.Buffer, p *Profile) (err error) { + for { + if sectionTrigger(l) == memoryMapSection { + break + } + // Ignore any unrecognized sections. + if l, err := b.ReadString('\n'); err != nil { + if err != io.EOF { + return err + } + if l == "" { + break + } + } + } + return p.ParseMemoryMap(b) +} + +// 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 { + b := bufio.NewReader(rd) + + var attrs []string + var r *strings.Replacer + const delimiter = "=" + for { + l, err := b.ReadString('\n') + if err != nil { + if err != io.EOF { + return err + } + if l == "" { + break + } + } + if l = strings.TrimSpace(l); l == "" { + continue + } + + if r != nil { + l = r.Replace(l) + } + m, err := parseMappingEntry(l) + if err != nil { + if err == errUnrecognized { + // Recognize assignments of the form: attr=value, and replace + // $attr with value on subsequent mappings. + if attr := strings.SplitN(l, delimiter, 2); len(attr) == 2 { + attrs = append(attrs, "$"+strings.TrimSpace(attr[0]), strings.TrimSpace(attr[1])) + r = strings.NewReplacer(attrs...) + } + // Ignore any unrecognized entries + continue + } + return err + } + if m == nil || (m.File == "" && len(p.Mapping) != 0) { + // In some cases the first entry may include the address range + // but not the name of the file. It should be followed by + // another entry with the name. + continue + } + if len(p.Mapping) == 1 && p.Mapping[0].File == "" { + // Update the name if this is the entry following that empty one. + p.Mapping[0].File = m.File + continue + } + p.Mapping = append(p.Mapping, m) + } + p.remapLocationIDs() + p.remapFunctionIDs() + p.remapMappingIDs() + return nil +} + +func parseMappingEntry(l string) (*Mapping, error) { + mapping := &Mapping{} + var err error + if me := procMapsRE.FindStringSubmatch(l); len(me) == 9 { + if !strings.Contains(me[3], "x") { + // Skip non-executable entries. + return nil, nil + } + if mapping.Start, err = strconv.ParseUint(me[1], 16, 64); err != nil { + return nil, errUnrecognized + } + if mapping.Limit, err = strconv.ParseUint(me[2], 16, 64); err != nil { + return nil, errUnrecognized + } + if me[4] != "" { + if mapping.Offset, err = strconv.ParseUint(me[4], 16, 64); err != nil { + return nil, errUnrecognized + } + } + mapping.File = me[8] + return mapping, nil + } + + if me := briefMapsRE.FindStringSubmatch(l); len(me) == 6 { + if mapping.Start, err = strconv.ParseUint(me[1], 16, 64); err != nil { + return nil, errUnrecognized + } + if mapping.Limit, err = strconv.ParseUint(me[2], 16, 64); err != nil { + return nil, errUnrecognized + } + mapping.File = me[3] + if me[5] != "" { + if mapping.Offset, err = strconv.ParseUint(me[5], 16, 64); err != nil { + return nil, errUnrecognized + } + } + return mapping, nil + } + + return nil, errUnrecognized +} + +type sectionType int + +const ( + unrecognizedSection sectionType = iota + memoryMapSection +) + +var memoryMapTriggers = []string{ + "--- Memory map: ---", + "MAPPED_LIBRARIES:", +} + +func sectionTrigger(line string) sectionType { + for _, trigger := range memoryMapTriggers { + if strings.Contains(line, trigger) { + return memoryMapSection + } + } + return unrecognizedSection +} + +func (p *Profile) addLegacyFrameInfo() { + switch { + case isProfileType(p, heapzSampleTypes) || + isProfileType(p, heapzInUseSampleTypes) || + isProfileType(p, heapzAllocSampleTypes): + 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 +var heapzInUseSampleTypes = []string{"inuse_objects", "inuse_space"} +var heapzAllocSampleTypes = []string{"alloc_objects", "alloc_space"} +var contentionzSampleTypes = []string{"contentions", "delay"} + +func isProfileType(p *Profile, t []string) bool { + st := p.SampleType + if len(st) != len(t) { + return false + } + + for i := range st { + if st[i].Type != t[i] { + return false + } + } + return true +} + +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.*`, + `(Mutex::)?AwaitCommon.*`, + `(Mutex::)?Unlock.*`, + `(Mutex::)?UnlockSlow.*`, + `(Mutex::)?ReaderUnlock.*`, + `(MutexLock::)?~MutexLock.*`, + `(SpinLock::)?Unlock.*`, + `(SpinLock::)?SlowUnlock.*`, + `(SpinLockHolder::)?~SpinLockHolder.*`, +}, `|`) diff --git a/src/internal/profile/merge.go b/src/internal/profile/merge.go new file mode 100644 index 0000000..3ea7d4c --- /dev/null +++ b/src/internal/profile/merge.go @@ -0,0 +1,461 @@ +// Copyright 2019 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package profile + +import ( + "fmt" + "sort" + "strconv" + "strings" +) + +// 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. +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 = make(map[uint64]*Location, 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) + } + } + } + + for _, s := range p.Sample { + if isZeroSample(s) { + // 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 map[uint64]*Location + 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 { + 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 + } + // Check memoization table. Must be done on the remapped location to + // account for the remapped mapping. Add current values to the + // existing sample. + k := s.key() + if ss, ok := pm.samples[k]; ok { + for i, v := range src.Value { + ss.Value[i] += v + } + return ss + } + copy(s.Value, src.Value) + pm.samples[k] = s + pm.p.Sample = append(pm.p.Sample, s) + return s +} + +// key generates sampleKey to be used as a key for maps. +func (sample *Sample) key() sampleKey { + ids := make([]string, len(sample.Location)) + for i, l := range sample.Location { + ids[i] = strconv.FormatUint(l.ID, 16) + } + + labels := make([]string, 0, len(sample.Label)) + for k, v := range sample.Label { + labels = append(labels, fmt.Sprintf("%q%q", k, v)) + } + sort.Strings(labels) + + numlabels := make([]string, 0, len(sample.NumLabel)) + for k, v := range sample.NumLabel { + numlabels = append(numlabels, fmt.Sprintf("%q%x%x", k, v, sample.NumUnit[k])) + } + sort.Strings(numlabels) + + return sampleKey{ + strings.Join(ids, "|"), + strings.Join(labels, ""), + strings.Join(numlabels, ""), + } +} + +type sampleKey struct { + locations string + labels string + numlabels string +} + +func (pm *profileMerger) mapLocation(src *Location) *Location { + if src == nil { + return nil + } + + if l, ok := pm.locationsByID[src.ID]; ok { + pm.locationsByID[src.ID] = l + 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[src.ID] = ll + return ll + } + pm.locationsByID[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)*2) + 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) + } + 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, + 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, + } + 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 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 + } + } + + 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, + } + 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 +} diff --git a/src/internal/profile/profile.go b/src/internal/profile/profile.go new file mode 100644 index 0000000..29568aa --- /dev/null +++ b/src/internal/profile/profile.go @@ -0,0 +1,613 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Package profile provides a representation of +// github.com/google/pprof/proto/profile.proto and +// methods to encode/decode/merge profiles in this format. +package profile + +import ( + "bytes" + "compress/gzip" + "fmt" + "io" + "regexp" + "strings" + "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 + + DropFrames string + KeepFrames string + + TimeNanos int64 + DurationNanos int64 + PeriodType *ValueType + Period int64 + + commentX []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 map[string][]string + NumLabel map[string][]int64 + 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 +} + +// 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 +} + +// 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 + + 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) { + orig, err := io.ReadAll(r) + if err != nil { + return nil, err + } + + var p *Profile + if len(orig) >= 2 && orig[0] == 0x1f && orig[1] == 0x8b { + gz, err := gzip.NewReader(bytes.NewBuffer(orig)) + if err != nil { + return nil, fmt.Errorf("decompressing profile: %v", err) + } + data, err := io.ReadAll(gz) + if err != nil { + return nil, fmt.Errorf("decompressing profile: %v", err) + } + orig = data + } + if p, err = parseUncompressed(orig); err != nil { + if p, err = parseLegacy(orig); 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") + +func parseLegacy(data []byte) (*Profile, error) { + parsers := []func([]byte) (*Profile, error){ + parseCPU, + parseHeap, + parseGoCount, // goroutine, threadcreate + parseThread, + parseContention, + } + + for _, parser := range parsers { + p, err := parser(data) + if err == nil { + p.setMain() + p.addLegacyFrameInfo() + return p, nil + } + if err != errUnrecognized { + return nil, err + } + } + return nil, errUnrecognized +} + +func parseUncompressed(data []byte) (*Profile, error) { + 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]+)`) + +// setMain scans Mapping entries and guesses which entry is main +// because legacy profiles don't obey the convention of putting main +// first. +func (p *Profile) setMain() { + for i := 0; i < len(p.Mapping); i++ { + file := strings.TrimSpace(strings.ReplaceAll(p.Mapping[i].File, "(deleted)", "")) + if len(file) == 0 { + continue + } + if len(libRx.FindStringSubmatch(file)) > 0 { + continue + } + if strings.HasPrefix(file, "[") { + continue + } + // Swap what we guess is main to position 0. + p.Mapping[i], p.Mapping[0] = p.Mapping[0], p.Mapping[i] + break + } +} + +// Write writes the profile as a gzip-compressed marshaled protobuf. +func (p *Profile) Write(w io.Writer) error { + p.preEncode() + b := marshal(p) + zw := gzip.NewWriter(w) + defer zw.Close() + _, err := zw.Write(b) + 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 len(s.Value) != sampleLen { + return fmt.Errorf("mismatch: sample has: %d values vs. %d types", len(s.Value), len(p.SampleType)) + } + } + + // 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.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.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.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 { + if f := ln.Function; f != nil { + 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, 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 { + 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 + } + } + if !address { + l.Address = 0 + } + } + } + + return p.CheckValid() +} + +// Print dumps a text representation of a profile. Intended mainly +// for debugging purposes. +func (p *Profile) String() string { + + ss := make([]string, 0, len(p.Sample)+len(p.Mapping)+len(p.Location)) + 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: %v", time.Duration(p.DurationNanos))) + } + + ss = append(ss, "Samples:") + var sh1 string + for _, s := range p.SampleType { + sh1 = sh1 + fmt.Sprintf("%s/%s ", s.Type, s.Unit) + } + ss = append(ss, strings.TrimSpace(sh1)) + for _, s := range p.Sample { + 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 { + ls := labelHeader + for k, v := range s.Label { + ls = ls + fmt.Sprintf("%s:%v ", k, v) + } + ss = append(ss, ls) + } + if len(s.NumLabel) > 0 { + ls := labelHeader + for k, v := range s.NumLabel { + ls = ls + fmt.Sprintf("%s:%v ", k, v) + } + ss = append(ss, ls) + } + } + + ss = append(ss, "Locations") + for _, l := range p.Location { + locStr := fmt.Sprintf("%6d: %#x ", l.ID, l.Address) + if m := l.Mapping; m != nil { + locStr = locStr + fmt.Sprintf("M=%d ", m.ID) + } + 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 s=%d", + fn.Name, + fn.Filename, + l.Line[li].Line, + 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 = " " + } + } + + ss = append(ss, "Mappings") + for _, m := range p.Mapping { + bits := "" + if m.HasFunctions { + bits += "[FN]" + } + if m.HasFilenames { + bits += "[FL]" + } + if m.HasLineNumbers { + bits += "[LN]" + } + if m.HasInlineFrames { + bits += "[IN]" + } + ss = append(ss, fmt.Sprintf("%d: %#x/%#x/%#x %s %s %s", + m.ID, + m.Start, m.Limit, m.Offset, + m.File, + m.BuildID, + bits)) + } + + return strings.Join(ss, "\n") + "\n" +} + +// Merge adds profile p adjusted by ratio r into profile p. Profiles +// must be compatible (same Type and SampleType). +// TODO(rsilvera): consider normalizing the profiles based on the +// total samples collected. +func (p *Profile) Merge(pb *Profile, r float64) error { + if err := p.Compatible(pb); err != nil { + return err + } + + pb = pb.Copy() + + // Keep the largest of the two periods. + if pb.Period > p.Period { + p.Period = pb.Period + } + + p.DurationNanos += pb.DurationNanos + + p.Mapping = append(p.Mapping, pb.Mapping...) + for i, m := range p.Mapping { + m.ID = uint64(i + 1) + } + p.Location = append(p.Location, pb.Location...) + for i, l := range p.Location { + l.ID = uint64(i + 1) + } + p.Function = append(p.Function, pb.Function...) + for i, f := range p.Function { + f.ID = uint64(i + 1) + } + + if r != 1.0 { + for _, s := range pb.Sample { + for i, v := range s.Value { + s.Value[i] = int64((float64(v) * r)) + } + } + } + p.Sample = append(p.Sample, pb.Sample...) + return p.CheckValid() +} + +// 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 !compatibleValueTypes(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 !compatibleValueTypes(p.SampleType[i], pb.SampleType[i]) { + return fmt.Errorf("incompatible sample types %v and %v", p.SampleType, pb.SampleType) + } + } + + 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 +} + +func compatibleValueTypes(v1, v2 *ValueType) bool { + if v1 == nil || v2 == nil { + return true // No grounds to disqualify. + } + return v1.Type == v2.Type && v1.Unit == v2.Unit +} + +// Copy makes a fully independent copy of a profile. +func (p *Profile) Copy() *Profile { + p.preEncode() + b := marshal(p) + + pp := &Profile{} + if err := unmarshal(b, pp); err != nil { + panic(err) + } + if err := pp.postDecode(); err != nil { + panic(err) + } + + return pp +} + +// Demangler maps symbol names to a human-readable form. This may +// include C++ demangling and additional simplification. Names that +// are not demangled may be missing from the resulting map. +type Demangler func(name []string) (map[string]string, error) + +// Demangle attempts to demangle and optionally simplify any function +// names referenced in the profile. It works on a best-effort basis: +// it will silently preserve the original names in case of any errors. +func (p *Profile) Demangle(d Demangler) error { + // Collect names to demangle. + var names []string + for _, fn := range p.Function { + names = append(names, fn.SystemName) + } + + // Update profile with demangled names. + demangled, err := d(names) + if err != nil { + return err + } + for _, fn := range p.Function { + if dd, ok := demangled[fn.SystemName]; ok { + fn.Name = dd + } + } + return nil +} + +// Empty reports whether the profile contains no samples. +func (p *Profile) Empty() bool { + return len(p.Sample) == 0 +} + +// Scale multiplies all sample values in a profile by a constant. +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. +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 + } + for _, s := range p.Sample { + for i, v := range s.Value { + if ratios[i] != 1 { + s.Value[i] = int64(float64(v) * ratios[i]) + } + } + } + return nil +} diff --git a/src/internal/profile/profile_test.go b/src/internal/profile/profile_test.go new file mode 100644 index 0000000..e1963f3 --- /dev/null +++ b/src/internal/profile/profile_test.go @@ -0,0 +1,79 @@ +// Copyright 2015 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +package profile + +import ( + "bytes" + "testing" +) + +func TestEmptyProfile(t *testing.T) { + var buf bytes.Buffer + p, err := Parse(&buf) + if err != nil { + t.Error("Want no error, got", err) + } + if p == nil { + t.Fatal("Want a valid profile, got <nil>") + } + if !p.Empty() { + t.Errorf("Profile should be empty, got %#v", p) + } +} + +func TestParseContention(t *testing.T) { + tests := []struct { + name string + in string + wantErr bool + }{ + { + name: "valid", + in: `--- mutex: +cycles/second=3491920901 +sampling period=1 +43227965305 1659640 @ 0x45e851 0x45f764 0x4a2be1 0x44ea31 +34035731690 15760 @ 0x45e851 0x45f764 0x4a2b17 0x44ea31 +`, + }, + { + name: "valid with comment", + in: `--- mutex: +cycles/second=3491920901 +sampling period=1 +43227965305 1659640 @ 0x45e851 0x45f764 0x4a2be1 0x44ea31 +# 0x45e850 sync.(*Mutex).Unlock+0x80 /go/src/sync/mutex.go:126 +# 0x45f763 sync.(*RWMutex).Unlock+0x83 /go/src/sync/rwmutex.go:125 +# 0x4a2be0 main.main.func3+0x70 /go/src/internal/pprof/profile/a_binary.go:58 + +34035731690 15760 @ 0x45e851 0x45f764 0x4a2b17 0x44ea31 +# 0x45e850 sync.(*Mutex).Unlock+0x80 /go/src/sync/mutex.go:126 +# 0x45f763 sync.(*RWMutex).Unlock+0x83 /go/src/sync/rwmutex.go:125 +# 0x4a2b16 main.main.func2+0xd6 /go/src/internal/pprof/profile/a_binary.go:48 +`, + }, + { + name: "empty", + in: `--- mutex:`, + wantErr: true, + }, + { + name: "invalid header", + in: `--- channel: +43227965305 1659640 @ 0x45e851 0x45f764 0x4a2be1 0x44ea31`, + wantErr: true, + }, + } + for _, tc := range tests { + _, err := parseContention([]byte(tc.in)) + if tc.wantErr && err == nil { + t.Errorf("parseContention(%q) succeeded unexpectedly", tc.name) + } + if !tc.wantErr && err != nil { + t.Errorf("parseContention(%q) failed unexpectedly: %v", tc.name, err) + } + } + +} diff --git a/src/internal/profile/proto.go b/src/internal/profile/proto.go new file mode 100644 index 0000000..52cf1ef --- /dev/null +++ b/src/internal/profile/proto.go @@ -0,0 +1,363 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// This file is a simple protocol buffer encoder and decoder. +// +// 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 + typ int + u64 uint64 + data []byte + tmp [16]byte +} + +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|0) + 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 encodeInt64Opt(b *buffer, tag int, x int64) { + if x == 0 { + return + } + encodeInt64(b, tag, x) +} + +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 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 encodeStringOpt(b *buffer, tag int, x string) { + if x == "" { + return + } + encodeString(b, tag, x) +} + +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 == false { + return + } + 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 i int + 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 +} diff --git a/src/internal/profile/proto_test.go b/src/internal/profile/proto_test.go new file mode 100644 index 0000000..c2613fc --- /dev/null +++ b/src/internal/profile/proto_test.go @@ -0,0 +1,67 @@ +package profile + +import ( + "reflect" + "testing" +) + +func TestPackedEncoding(t *testing.T) { + + type testcase struct { + uint64s []uint64 + int64s []int64 + encoded []byte + } + for i, tc := range []testcase{ + { + []uint64{0, 1, 10, 100, 1000, 10000}, + []int64{1000, 0, 1000}, + []byte{10, 8, 0, 1, 10, 100, 232, 7, 144, 78, 18, 5, 232, 7, 0, 232, 7}, + }, + { + []uint64{10000}, + nil, + []byte{8, 144, 78}, + }, + { + nil, + []int64{-10000}, + []byte{16, 240, 177, 255, 255, 255, 255, 255, 255, 255, 1}, + }, + } { + source := &packedInts{tc.uint64s, tc.int64s} + if got, want := marshal(source), tc.encoded; !reflect.DeepEqual(got, want) { + t.Errorf("failed encode %d, got %v, want %v", i, got, want) + } + + dest := new(packedInts) + if err := unmarshal(tc.encoded, dest); err != nil { + t.Errorf("failed decode %d: %v", i, err) + continue + } + if got, want := dest.uint64s, tc.uint64s; !reflect.DeepEqual(got, want) { + t.Errorf("failed decode uint64s %d, got %v, want %v", i, got, want) + } + if got, want := dest.int64s, tc.int64s; !reflect.DeepEqual(got, want) { + t.Errorf("failed decode int64s %d, got %v, want %v", i, got, want) + } + } +} + +type packedInts struct { + uint64s []uint64 + int64s []int64 +} + +func (u *packedInts) decoder() []decoder { + return []decoder{ + nil, + func(b *buffer, m message) error { return decodeUint64s(b, &m.(*packedInts).uint64s) }, + func(b *buffer, m message) error { return decodeInt64s(b, &m.(*packedInts).int64s) }, + } +} + +func (u *packedInts) encode(b *buffer) { + encodeUint64s(b, 1, u.uint64s) + encodeInt64s(b, 2, u.int64s) +} diff --git a/src/internal/profile/prune.go b/src/internal/profile/prune.go new file mode 100644 index 0000000..1924fad --- /dev/null +++ b/src/internal/profile/prune.go @@ -0,0 +1,97 @@ +// Copyright 2014 The Go Authors. All rights reserved. +// Use of this source code is governed by a BSD-style +// license that can be found in the LICENSE file. + +// Implements methods to remove frames from profiles. + +package profile + +import ( + "fmt" + "regexp" +) + +// 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) + + 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 != "" { + funcName := fn.Name + // Account for leading '.' on the PPC ELF v1 ABI. + if funcName[0] == '.' { + funcName = funcName[1:] + } + if dropRx.MatchString(funcName) { + if keepRx == nil || !keepRx.MatchString(funcName) { + 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 +} |