|
| 1 | +// Copyright 2023 The Go Authors. All rights reserved. |
| 2 | +// Use of this source code is governed by a BSD-style |
| 3 | +// license that can be found in the LICENSE file. |
| 4 | + |
| 5 | +// This file contains data types that all implementations of the trace format |
| 6 | +// parser need to provide to the rest of the package. |
| 7 | + |
| 8 | +package trace |
| 9 | + |
| 10 | +import ( |
| 11 | + "fmt" |
| 12 | + "strings" |
| 13 | + |
| 14 | + "internal/trace/v2/event" |
| 15 | + "internal/trace/v2/event/go122" |
| 16 | + "internal/trace/v2/version" |
| 17 | +) |
| 18 | + |
| 19 | +// maxArgs is the maximum number of arguments for "plain" events, |
| 20 | +// i.e. anything that could reasonably be represented as a Base. |
| 21 | +const maxArgs = 5 |
| 22 | + |
| 23 | +// baseEvent is the basic unprocessed event. This serves as a common |
| 24 | +// fundamental data structure across. |
| 25 | +type baseEvent struct { |
| 26 | + typ event.Type |
| 27 | + time Time |
| 28 | + args [maxArgs - 1]uint64 |
| 29 | +} |
| 30 | + |
| 31 | +// extra returns a slice representing extra available space in args |
| 32 | +// that the parser can use to pass data up into Event. |
| 33 | +func (e *baseEvent) extra(v version.Version) []uint64 { |
| 34 | + switch v { |
| 35 | + case version.Go122: |
| 36 | + return e.args[len(go122.Specs()[e.typ].Args)-1:] |
| 37 | + } |
| 38 | + panic(fmt.Sprintf("unsupported version: go 1.%d", v)) |
| 39 | +} |
| 40 | + |
| 41 | +// evTable contains the per-generation data necessary to |
| 42 | +// interpret an individual event. |
| 43 | +type evTable struct { |
| 44 | + freq frequency |
| 45 | + strings dataTable[stringID, string] |
| 46 | + stacks dataTable[stackID, stack] |
| 47 | + |
| 48 | + // extraStrings are strings that get generated during |
| 49 | + // parsing but haven't come directly from the trace, so |
| 50 | + // they don't appear in strings. |
| 51 | + extraStrings []string |
| 52 | + extraStringIDs map[string]extraStringID |
| 53 | + nextExtra extraStringID |
| 54 | +} |
| 55 | + |
| 56 | +// addExtraString adds an extra string to the evTable and returns |
| 57 | +// a unique ID for the string in the table. |
| 58 | +func (t *evTable) addExtraString(s string) extraStringID { |
| 59 | + if s == "" { |
| 60 | + return 0 |
| 61 | + } |
| 62 | + if t.extraStringIDs == nil { |
| 63 | + t.extraStringIDs = make(map[string]extraStringID) |
| 64 | + } |
| 65 | + if id, ok := t.extraStringIDs[s]; ok { |
| 66 | + return id |
| 67 | + } |
| 68 | + t.nextExtra++ |
| 69 | + id := t.nextExtra |
| 70 | + t.extraStrings = append(t.extraStrings, s) |
| 71 | + t.extraStringIDs[s] = id |
| 72 | + return id |
| 73 | +} |
| 74 | + |
| 75 | +// getExtraString returns the extra string for the provided ID. |
| 76 | +// The ID must have been produced by addExtraString for this evTable. |
| 77 | +func (t *evTable) getExtraString(id extraStringID) string { |
| 78 | + if id == 0 { |
| 79 | + return "" |
| 80 | + } |
| 81 | + return t.extraStrings[id-1] |
| 82 | +} |
| 83 | + |
| 84 | +// dataTable is a mapping from EIs to Es. |
| 85 | +type dataTable[EI ~uint64, E any] struct { |
| 86 | + present []uint8 |
| 87 | + dense []E |
| 88 | + sparse map[EI]E |
| 89 | +} |
| 90 | + |
| 91 | +// insert tries to add a mapping from id to s. |
| 92 | +// |
| 93 | +// Returns an error if a mapping for id already exists, regardless |
| 94 | +// of whether or not s is the same in content. This should be used |
| 95 | +// for validation during parsing. |
| 96 | +func (d *dataTable[EI, E]) insert(id EI, data E) error { |
| 97 | + if d.sparse == nil { |
| 98 | + d.sparse = make(map[EI]E) |
| 99 | + } |
| 100 | + if existing, ok := d.get(id); ok { |
| 101 | + return fmt.Errorf("multiple %Ts with the same ID: id=%d, new=%v, existing=%v", data, id, data, existing) |
| 102 | + } |
| 103 | + d.sparse[id] = data |
| 104 | + return nil |
| 105 | +} |
| 106 | + |
| 107 | +// compactify attempts to compact sparse into dense. |
| 108 | +// |
| 109 | +// This is intended to be called only once after insertions are done. |
| 110 | +func (d *dataTable[EI, E]) compactify() { |
| 111 | + if d.sparse == nil || len(d.dense) != 0 { |
| 112 | + // Already compactified. |
| 113 | + return |
| 114 | + } |
| 115 | + // Find the range of IDs. |
| 116 | + maxID := EI(0) |
| 117 | + minID := ^EI(0) |
| 118 | + for id := range d.sparse { |
| 119 | + if id > maxID { |
| 120 | + maxID = id |
| 121 | + } |
| 122 | + if id < minID { |
| 123 | + minID = id |
| 124 | + } |
| 125 | + } |
| 126 | + // We're willing to waste at most 2x memory. |
| 127 | + if int(maxID-minID) > 2*len(d.sparse) { |
| 128 | + return |
| 129 | + } |
| 130 | + if int(minID) > len(d.sparse) { |
| 131 | + return |
| 132 | + } |
| 133 | + size := int(maxID) + 1 |
| 134 | + d.present = make([]uint8, (size+7)/8) |
| 135 | + d.dense = make([]E, size) |
| 136 | + for id, data := range d.sparse { |
| 137 | + d.dense[id] = data |
| 138 | + d.present[id/8] |= uint8(1) << (id % 8) |
| 139 | + } |
| 140 | + d.sparse = nil |
| 141 | +} |
| 142 | + |
| 143 | +// get returns the E for id or false if it doesn't |
| 144 | +// exist. This should be used for validation during parsing. |
| 145 | +func (d *dataTable[EI, E]) get(id EI) (E, bool) { |
| 146 | + if id == 0 { |
| 147 | + return *new(E), true |
| 148 | + } |
| 149 | + if int(id) < len(d.dense) { |
| 150 | + if d.present[id/8]&(uint8(1)<<(id%8)) != 0 { |
| 151 | + return d.dense[id], true |
| 152 | + } |
| 153 | + } else if d.sparse != nil { |
| 154 | + if data, ok := d.sparse[id]; ok { |
| 155 | + return data, true |
| 156 | + } |
| 157 | + } |
| 158 | + return *new(E), false |
| 159 | +} |
| 160 | + |
| 161 | +// forEach iterates over all ID/value pairs in the data table. |
| 162 | +func (d *dataTable[EI, E]) forEach(yield func(EI, E) bool) bool { |
| 163 | + for id, value := range d.dense { |
| 164 | + if d.present[id/8]&(uint8(1)<<(id%8)) == 0 { |
| 165 | + continue |
| 166 | + } |
| 167 | + if !yield(EI(id), value) { |
| 168 | + return false |
| 169 | + } |
| 170 | + } |
| 171 | + if d.sparse == nil { |
| 172 | + return true |
| 173 | + } |
| 174 | + for id, value := range d.sparse { |
| 175 | + if !yield(id, value) { |
| 176 | + return false |
| 177 | + } |
| 178 | + } |
| 179 | + return true |
| 180 | +} |
| 181 | + |
| 182 | +// mustGet returns the E for id or panics if it fails. |
| 183 | +// |
| 184 | +// This should only be used if id has already been validated. |
| 185 | +func (d *dataTable[EI, E]) mustGet(id EI) E { |
| 186 | + data, ok := d.get(id) |
| 187 | + if !ok { |
| 188 | + panic(fmt.Sprintf("expected id %d in %T table", id, data)) |
| 189 | + } |
| 190 | + return data |
| 191 | +} |
| 192 | + |
| 193 | +// frequency is nanoseconds per timestamp unit. |
| 194 | +type frequency float64 |
| 195 | + |
| 196 | +// mul multiplies an unprocessed to produce a time in nanoseconds. |
| 197 | +func (f frequency) mul(t timestamp) Time { |
| 198 | + return Time(float64(t) * float64(f)) |
| 199 | +} |
| 200 | + |
| 201 | +// stringID is an index into the string table for a generation. |
| 202 | +type stringID uint64 |
| 203 | + |
| 204 | +// extraStringID is an index into the extra string table for a generation. |
| 205 | +type extraStringID uint64 |
| 206 | + |
| 207 | +// stackID is an index into the stack table for a generation. |
| 208 | +type stackID uint64 |
| 209 | + |
| 210 | +// cpuSample represents a CPU profiling sample captured by the trace. |
| 211 | +type cpuSample struct { |
| 212 | + schedCtx |
| 213 | + time Time |
| 214 | + stack stackID |
| 215 | +} |
| 216 | + |
| 217 | +// asEvent produces a complete Event from a cpuSample. It needs |
| 218 | +// the evTable from the generation that created it. |
| 219 | +// |
| 220 | +// We don't just store it as an Event in generation to minimize |
| 221 | +// the amount of pointer data floating around. |
| 222 | +func (s cpuSample) asEvent(table *evTable) Event { |
| 223 | + // TODO(mknyszek): This is go122-specific, but shouldn't be. |
| 224 | + // Generalize this in the future. |
| 225 | + e := Event{ |
| 226 | + table: table, |
| 227 | + ctx: s.schedCtx, |
| 228 | + base: baseEvent{ |
| 229 | + typ: go122.EvCPUSample, |
| 230 | + time: s.time, |
| 231 | + }, |
| 232 | + } |
| 233 | + e.base.args[0] = uint64(s.stack) |
| 234 | + return e |
| 235 | +} |
| 236 | + |
| 237 | +// stack represents a goroutine stack sample. |
| 238 | +type stack struct { |
| 239 | + frames []frame |
| 240 | +} |
| 241 | + |
| 242 | +func (s stack) String() string { |
| 243 | + var sb strings.Builder |
| 244 | + for _, frame := range s.frames { |
| 245 | + fmt.Fprintf(&sb, "\t%#v\n", frame) |
| 246 | + } |
| 247 | + return sb.String() |
| 248 | +} |
| 249 | + |
| 250 | +// frame represents a single stack frame. |
| 251 | +type frame struct { |
| 252 | + pc uint64 |
| 253 | + funcID stringID |
| 254 | + fileID stringID |
| 255 | + line uint64 |
| 256 | +} |
0 commit comments