khenaidoo | 59ce9dd | 2019-11-11 13:05:32 -0500 | [diff] [blame] | 1 | // Copyright 2014 The Prometheus Authors |
| 2 | // Licensed under the Apache License, Version 2.0 (the "License"); |
| 3 | // you may not use this file except in compliance with the License. |
| 4 | // You may obtain a copy of the License at |
| 5 | // |
| 6 | // http://www.apache.org/licenses/LICENSE-2.0 |
| 7 | // |
| 8 | // Unless required by applicable law or agreed to in writing, software |
| 9 | // distributed under the License is distributed on an "AS IS" BASIS, |
| 10 | // WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. |
| 11 | // See the License for the specific language governing permissions and |
| 12 | // limitations under the License. |
| 13 | |
| 14 | package expfmt |
| 15 | |
| 16 | import ( |
| 17 | "bufio" |
| 18 | "bytes" |
| 19 | "fmt" |
| 20 | "io" |
| 21 | "math" |
| 22 | "strconv" |
| 23 | "strings" |
| 24 | |
| 25 | dto "github.com/prometheus/client_model/go" |
| 26 | |
| 27 | "github.com/golang/protobuf/proto" |
| 28 | "github.com/prometheus/common/model" |
| 29 | ) |
| 30 | |
| 31 | // A stateFn is a function that represents a state in a state machine. By |
| 32 | // executing it, the state is progressed to the next state. The stateFn returns |
| 33 | // another stateFn, which represents the new state. The end state is represented |
| 34 | // by nil. |
| 35 | type stateFn func() stateFn |
| 36 | |
| 37 | // ParseError signals errors while parsing the simple and flat text-based |
| 38 | // exchange format. |
| 39 | type ParseError struct { |
| 40 | Line int |
| 41 | Msg string |
| 42 | } |
| 43 | |
| 44 | // Error implements the error interface. |
| 45 | func (e ParseError) Error() string { |
| 46 | return fmt.Sprintf("text format parsing error in line %d: %s", e.Line, e.Msg) |
| 47 | } |
| 48 | |
| 49 | // TextParser is used to parse the simple and flat text-based exchange format. Its |
| 50 | // zero value is ready to use. |
| 51 | type TextParser struct { |
| 52 | metricFamiliesByName map[string]*dto.MetricFamily |
| 53 | buf *bufio.Reader // Where the parsed input is read through. |
| 54 | err error // Most recent error. |
| 55 | lineCount int // Tracks the line count for error messages. |
| 56 | currentByte byte // The most recent byte read. |
| 57 | currentToken bytes.Buffer // Re-used each time a token has to be gathered from multiple bytes. |
| 58 | currentMF *dto.MetricFamily |
| 59 | currentMetric *dto.Metric |
| 60 | currentLabelPair *dto.LabelPair |
| 61 | |
| 62 | // The remaining member variables are only used for summaries/histograms. |
| 63 | currentLabels map[string]string // All labels including '__name__' but excluding 'quantile'/'le' |
| 64 | // Summary specific. |
| 65 | summaries map[uint64]*dto.Metric // Key is created with LabelsToSignature. |
| 66 | currentQuantile float64 |
| 67 | // Histogram specific. |
| 68 | histograms map[uint64]*dto.Metric // Key is created with LabelsToSignature. |
| 69 | currentBucket float64 |
| 70 | // These tell us if the currently processed line ends on '_count' or |
| 71 | // '_sum' respectively and belong to a summary/histogram, representing the sample |
| 72 | // count and sum of that summary/histogram. |
| 73 | currentIsSummaryCount, currentIsSummarySum bool |
| 74 | currentIsHistogramCount, currentIsHistogramSum bool |
| 75 | } |
| 76 | |
| 77 | // TextToMetricFamilies reads 'in' as the simple and flat text-based exchange |
| 78 | // format and creates MetricFamily proto messages. It returns the MetricFamily |
| 79 | // proto messages in a map where the metric names are the keys, along with any |
| 80 | // error encountered. |
| 81 | // |
| 82 | // If the input contains duplicate metrics (i.e. lines with the same metric name |
| 83 | // and exactly the same label set), the resulting MetricFamily will contain |
| 84 | // duplicate Metric proto messages. Similar is true for duplicate label |
| 85 | // names. Checks for duplicates have to be performed separately, if required. |
| 86 | // Also note that neither the metrics within each MetricFamily are sorted nor |
| 87 | // the label pairs within each Metric. Sorting is not required for the most |
| 88 | // frequent use of this method, which is sample ingestion in the Prometheus |
| 89 | // server. However, for presentation purposes, you might want to sort the |
| 90 | // metrics, and in some cases, you must sort the labels, e.g. for consumption by |
| 91 | // the metric family injection hook of the Prometheus registry. |
| 92 | // |
| 93 | // Summaries and histograms are rather special beasts. You would probably not |
| 94 | // use them in the simple text format anyway. This method can deal with |
| 95 | // summaries and histograms if they are presented in exactly the way the |
| 96 | // text.Create function creates them. |
| 97 | // |
| 98 | // This method must not be called concurrently. If you want to parse different |
| 99 | // input concurrently, instantiate a separate Parser for each goroutine. |
| 100 | func (p *TextParser) TextToMetricFamilies(in io.Reader) (map[string]*dto.MetricFamily, error) { |
| 101 | p.reset(in) |
| 102 | for nextState := p.startOfLine; nextState != nil; nextState = nextState() { |
| 103 | // Magic happens here... |
| 104 | } |
| 105 | // Get rid of empty metric families. |
| 106 | for k, mf := range p.metricFamiliesByName { |
| 107 | if len(mf.GetMetric()) == 0 { |
| 108 | delete(p.metricFamiliesByName, k) |
| 109 | } |
| 110 | } |
| 111 | // If p.err is io.EOF now, we have run into a premature end of the input |
| 112 | // stream. Turn this error into something nicer and more |
| 113 | // meaningful. (io.EOF is often used as a signal for the legitimate end |
| 114 | // of an input stream.) |
| 115 | if p.err == io.EOF { |
| 116 | p.parseError("unexpected end of input stream") |
| 117 | } |
| 118 | return p.metricFamiliesByName, p.err |
| 119 | } |
| 120 | |
| 121 | func (p *TextParser) reset(in io.Reader) { |
| 122 | p.metricFamiliesByName = map[string]*dto.MetricFamily{} |
| 123 | if p.buf == nil { |
| 124 | p.buf = bufio.NewReader(in) |
| 125 | } else { |
| 126 | p.buf.Reset(in) |
| 127 | } |
| 128 | p.err = nil |
| 129 | p.lineCount = 0 |
| 130 | if p.summaries == nil || len(p.summaries) > 0 { |
| 131 | p.summaries = map[uint64]*dto.Metric{} |
| 132 | } |
| 133 | if p.histograms == nil || len(p.histograms) > 0 { |
| 134 | p.histograms = map[uint64]*dto.Metric{} |
| 135 | } |
| 136 | p.currentQuantile = math.NaN() |
| 137 | p.currentBucket = math.NaN() |
| 138 | } |
| 139 | |
| 140 | // startOfLine represents the state where the next byte read from p.buf is the |
| 141 | // start of a line (or whitespace leading up to it). |
| 142 | func (p *TextParser) startOfLine() stateFn { |
| 143 | p.lineCount++ |
| 144 | if p.skipBlankTab(); p.err != nil { |
| 145 | // End of input reached. This is the only case where |
| 146 | // that is not an error but a signal that we are done. |
| 147 | p.err = nil |
| 148 | return nil |
| 149 | } |
| 150 | switch p.currentByte { |
| 151 | case '#': |
| 152 | return p.startComment |
| 153 | case '\n': |
| 154 | return p.startOfLine // Empty line, start the next one. |
| 155 | } |
| 156 | return p.readingMetricName |
| 157 | } |
| 158 | |
| 159 | // startComment represents the state where the next byte read from p.buf is the |
| 160 | // start of a comment (or whitespace leading up to it). |
| 161 | func (p *TextParser) startComment() stateFn { |
| 162 | if p.skipBlankTab(); p.err != nil { |
| 163 | return nil // Unexpected end of input. |
| 164 | } |
| 165 | if p.currentByte == '\n' { |
| 166 | return p.startOfLine |
| 167 | } |
| 168 | if p.readTokenUntilWhitespace(); p.err != nil { |
| 169 | return nil // Unexpected end of input. |
| 170 | } |
| 171 | // If we have hit the end of line already, there is nothing left |
| 172 | // to do. This is not considered a syntax error. |
| 173 | if p.currentByte == '\n' { |
| 174 | return p.startOfLine |
| 175 | } |
| 176 | keyword := p.currentToken.String() |
| 177 | if keyword != "HELP" && keyword != "TYPE" { |
| 178 | // Generic comment, ignore by fast forwarding to end of line. |
| 179 | for p.currentByte != '\n' { |
| 180 | if p.currentByte, p.err = p.buf.ReadByte(); p.err != nil { |
| 181 | return nil // Unexpected end of input. |
| 182 | } |
| 183 | } |
| 184 | return p.startOfLine |
| 185 | } |
| 186 | // There is something. Next has to be a metric name. |
| 187 | if p.skipBlankTab(); p.err != nil { |
| 188 | return nil // Unexpected end of input. |
| 189 | } |
| 190 | if p.readTokenAsMetricName(); p.err != nil { |
| 191 | return nil // Unexpected end of input. |
| 192 | } |
| 193 | if p.currentByte == '\n' { |
| 194 | // At the end of the line already. |
| 195 | // Again, this is not considered a syntax error. |
| 196 | return p.startOfLine |
| 197 | } |
| 198 | if !isBlankOrTab(p.currentByte) { |
| 199 | p.parseError("invalid metric name in comment") |
| 200 | return nil |
| 201 | } |
| 202 | p.setOrCreateCurrentMF() |
| 203 | if p.skipBlankTab(); p.err != nil { |
| 204 | return nil // Unexpected end of input. |
| 205 | } |
| 206 | if p.currentByte == '\n' { |
| 207 | // At the end of the line already. |
| 208 | // Again, this is not considered a syntax error. |
| 209 | return p.startOfLine |
| 210 | } |
| 211 | switch keyword { |
| 212 | case "HELP": |
| 213 | return p.readingHelp |
| 214 | case "TYPE": |
| 215 | return p.readingType |
| 216 | } |
| 217 | panic(fmt.Sprintf("code error: unexpected keyword %q", keyword)) |
| 218 | } |
| 219 | |
| 220 | // readingMetricName represents the state where the last byte read (now in |
| 221 | // p.currentByte) is the first byte of a metric name. |
| 222 | func (p *TextParser) readingMetricName() stateFn { |
| 223 | if p.readTokenAsMetricName(); p.err != nil { |
| 224 | return nil |
| 225 | } |
| 226 | if p.currentToken.Len() == 0 { |
| 227 | p.parseError("invalid metric name") |
| 228 | return nil |
| 229 | } |
| 230 | p.setOrCreateCurrentMF() |
| 231 | // Now is the time to fix the type if it hasn't happened yet. |
| 232 | if p.currentMF.Type == nil { |
| 233 | p.currentMF.Type = dto.MetricType_UNTYPED.Enum() |
| 234 | } |
| 235 | p.currentMetric = &dto.Metric{} |
| 236 | // Do not append the newly created currentMetric to |
| 237 | // currentMF.Metric right now. First wait if this is a summary, |
| 238 | // and the metric exists already, which we can only know after |
| 239 | // having read all the labels. |
| 240 | if p.skipBlankTabIfCurrentBlankTab(); p.err != nil { |
| 241 | return nil // Unexpected end of input. |
| 242 | } |
| 243 | return p.readingLabels |
| 244 | } |
| 245 | |
| 246 | // readingLabels represents the state where the last byte read (now in |
| 247 | // p.currentByte) is either the first byte of the label set (i.e. a '{'), or the |
| 248 | // first byte of the value (otherwise). |
| 249 | func (p *TextParser) readingLabels() stateFn { |
| 250 | // Summaries/histograms are special. We have to reset the |
| 251 | // currentLabels map, currentQuantile and currentBucket before starting to |
| 252 | // read labels. |
| 253 | if p.currentMF.GetType() == dto.MetricType_SUMMARY || p.currentMF.GetType() == dto.MetricType_HISTOGRAM { |
| 254 | p.currentLabels = map[string]string{} |
| 255 | p.currentLabels[string(model.MetricNameLabel)] = p.currentMF.GetName() |
| 256 | p.currentQuantile = math.NaN() |
| 257 | p.currentBucket = math.NaN() |
| 258 | } |
| 259 | if p.currentByte != '{' { |
| 260 | return p.readingValue |
| 261 | } |
| 262 | return p.startLabelName |
| 263 | } |
| 264 | |
| 265 | // startLabelName represents the state where the next byte read from p.buf is |
| 266 | // the start of a label name (or whitespace leading up to it). |
| 267 | func (p *TextParser) startLabelName() stateFn { |
| 268 | if p.skipBlankTab(); p.err != nil { |
| 269 | return nil // Unexpected end of input. |
| 270 | } |
| 271 | if p.currentByte == '}' { |
| 272 | if p.skipBlankTab(); p.err != nil { |
| 273 | return nil // Unexpected end of input. |
| 274 | } |
| 275 | return p.readingValue |
| 276 | } |
| 277 | if p.readTokenAsLabelName(); p.err != nil { |
| 278 | return nil // Unexpected end of input. |
| 279 | } |
| 280 | if p.currentToken.Len() == 0 { |
| 281 | p.parseError(fmt.Sprintf("invalid label name for metric %q", p.currentMF.GetName())) |
| 282 | return nil |
| 283 | } |
| 284 | p.currentLabelPair = &dto.LabelPair{Name: proto.String(p.currentToken.String())} |
| 285 | if p.currentLabelPair.GetName() == string(model.MetricNameLabel) { |
| 286 | p.parseError(fmt.Sprintf("label name %q is reserved", model.MetricNameLabel)) |
| 287 | return nil |
| 288 | } |
| 289 | // Special summary/histogram treatment. Don't add 'quantile' and 'le' |
| 290 | // labels to 'real' labels. |
| 291 | if !(p.currentMF.GetType() == dto.MetricType_SUMMARY && p.currentLabelPair.GetName() == model.QuantileLabel) && |
| 292 | !(p.currentMF.GetType() == dto.MetricType_HISTOGRAM && p.currentLabelPair.GetName() == model.BucketLabel) { |
| 293 | p.currentMetric.Label = append(p.currentMetric.Label, p.currentLabelPair) |
| 294 | } |
| 295 | if p.skipBlankTabIfCurrentBlankTab(); p.err != nil { |
| 296 | return nil // Unexpected end of input. |
| 297 | } |
| 298 | if p.currentByte != '=' { |
| 299 | p.parseError(fmt.Sprintf("expected '=' after label name, found %q", p.currentByte)) |
| 300 | return nil |
| 301 | } |
| 302 | return p.startLabelValue |
| 303 | } |
| 304 | |
| 305 | // startLabelValue represents the state where the next byte read from p.buf is |
| 306 | // the start of a (quoted) label value (or whitespace leading up to it). |
| 307 | func (p *TextParser) startLabelValue() stateFn { |
| 308 | if p.skipBlankTab(); p.err != nil { |
| 309 | return nil // Unexpected end of input. |
| 310 | } |
| 311 | if p.currentByte != '"' { |
| 312 | p.parseError(fmt.Sprintf("expected '\"' at start of label value, found %q", p.currentByte)) |
| 313 | return nil |
| 314 | } |
| 315 | if p.readTokenAsLabelValue(); p.err != nil { |
| 316 | return nil |
| 317 | } |
| 318 | if !model.LabelValue(p.currentToken.String()).IsValid() { |
| 319 | p.parseError(fmt.Sprintf("invalid label value %q", p.currentToken.String())) |
| 320 | return nil |
| 321 | } |
| 322 | p.currentLabelPair.Value = proto.String(p.currentToken.String()) |
| 323 | // Special treatment of summaries: |
| 324 | // - Quantile labels are special, will result in dto.Quantile later. |
| 325 | // - Other labels have to be added to currentLabels for signature calculation. |
| 326 | if p.currentMF.GetType() == dto.MetricType_SUMMARY { |
| 327 | if p.currentLabelPair.GetName() == model.QuantileLabel { |
| 328 | if p.currentQuantile, p.err = strconv.ParseFloat(p.currentLabelPair.GetValue(), 64); p.err != nil { |
| 329 | // Create a more helpful error message. |
| 330 | p.parseError(fmt.Sprintf("expected float as value for 'quantile' label, got %q", p.currentLabelPair.GetValue())) |
| 331 | return nil |
| 332 | } |
| 333 | } else { |
| 334 | p.currentLabels[p.currentLabelPair.GetName()] = p.currentLabelPair.GetValue() |
| 335 | } |
| 336 | } |
| 337 | // Similar special treatment of histograms. |
| 338 | if p.currentMF.GetType() == dto.MetricType_HISTOGRAM { |
| 339 | if p.currentLabelPair.GetName() == model.BucketLabel { |
| 340 | if p.currentBucket, p.err = strconv.ParseFloat(p.currentLabelPair.GetValue(), 64); p.err != nil { |
| 341 | // Create a more helpful error message. |
| 342 | p.parseError(fmt.Sprintf("expected float as value for 'le' label, got %q", p.currentLabelPair.GetValue())) |
| 343 | return nil |
| 344 | } |
| 345 | } else { |
| 346 | p.currentLabels[p.currentLabelPair.GetName()] = p.currentLabelPair.GetValue() |
| 347 | } |
| 348 | } |
| 349 | if p.skipBlankTab(); p.err != nil { |
| 350 | return nil // Unexpected end of input. |
| 351 | } |
| 352 | switch p.currentByte { |
| 353 | case ',': |
| 354 | return p.startLabelName |
| 355 | |
| 356 | case '}': |
| 357 | if p.skipBlankTab(); p.err != nil { |
| 358 | return nil // Unexpected end of input. |
| 359 | } |
| 360 | return p.readingValue |
| 361 | default: |
| 362 | p.parseError(fmt.Sprintf("unexpected end of label value %q", p.currentLabelPair.GetValue())) |
| 363 | return nil |
| 364 | } |
| 365 | } |
| 366 | |
| 367 | // readingValue represents the state where the last byte read (now in |
| 368 | // p.currentByte) is the first byte of the sample value (i.e. a float). |
| 369 | func (p *TextParser) readingValue() stateFn { |
| 370 | // When we are here, we have read all the labels, so for the |
| 371 | // special case of a summary/histogram, we can finally find out |
| 372 | // if the metric already exists. |
| 373 | if p.currentMF.GetType() == dto.MetricType_SUMMARY { |
| 374 | signature := model.LabelsToSignature(p.currentLabels) |
| 375 | if summary := p.summaries[signature]; summary != nil { |
| 376 | p.currentMetric = summary |
| 377 | } else { |
| 378 | p.summaries[signature] = p.currentMetric |
| 379 | p.currentMF.Metric = append(p.currentMF.Metric, p.currentMetric) |
| 380 | } |
| 381 | } else if p.currentMF.GetType() == dto.MetricType_HISTOGRAM { |
| 382 | signature := model.LabelsToSignature(p.currentLabels) |
| 383 | if histogram := p.histograms[signature]; histogram != nil { |
| 384 | p.currentMetric = histogram |
| 385 | } else { |
| 386 | p.histograms[signature] = p.currentMetric |
| 387 | p.currentMF.Metric = append(p.currentMF.Metric, p.currentMetric) |
| 388 | } |
| 389 | } else { |
| 390 | p.currentMF.Metric = append(p.currentMF.Metric, p.currentMetric) |
| 391 | } |
| 392 | if p.readTokenUntilWhitespace(); p.err != nil { |
| 393 | return nil // Unexpected end of input. |
| 394 | } |
| 395 | value, err := strconv.ParseFloat(p.currentToken.String(), 64) |
| 396 | if err != nil { |
| 397 | // Create a more helpful error message. |
| 398 | p.parseError(fmt.Sprintf("expected float as value, got %q", p.currentToken.String())) |
| 399 | return nil |
| 400 | } |
| 401 | switch p.currentMF.GetType() { |
| 402 | case dto.MetricType_COUNTER: |
| 403 | p.currentMetric.Counter = &dto.Counter{Value: proto.Float64(value)} |
| 404 | case dto.MetricType_GAUGE: |
| 405 | p.currentMetric.Gauge = &dto.Gauge{Value: proto.Float64(value)} |
| 406 | case dto.MetricType_UNTYPED: |
| 407 | p.currentMetric.Untyped = &dto.Untyped{Value: proto.Float64(value)} |
| 408 | case dto.MetricType_SUMMARY: |
| 409 | // *sigh* |
| 410 | if p.currentMetric.Summary == nil { |
| 411 | p.currentMetric.Summary = &dto.Summary{} |
| 412 | } |
| 413 | switch { |
| 414 | case p.currentIsSummaryCount: |
| 415 | p.currentMetric.Summary.SampleCount = proto.Uint64(uint64(value)) |
| 416 | case p.currentIsSummarySum: |
| 417 | p.currentMetric.Summary.SampleSum = proto.Float64(value) |
| 418 | case !math.IsNaN(p.currentQuantile): |
| 419 | p.currentMetric.Summary.Quantile = append( |
| 420 | p.currentMetric.Summary.Quantile, |
| 421 | &dto.Quantile{ |
| 422 | Quantile: proto.Float64(p.currentQuantile), |
| 423 | Value: proto.Float64(value), |
| 424 | }, |
| 425 | ) |
| 426 | } |
| 427 | case dto.MetricType_HISTOGRAM: |
| 428 | // *sigh* |
| 429 | if p.currentMetric.Histogram == nil { |
| 430 | p.currentMetric.Histogram = &dto.Histogram{} |
| 431 | } |
| 432 | switch { |
| 433 | case p.currentIsHistogramCount: |
| 434 | p.currentMetric.Histogram.SampleCount = proto.Uint64(uint64(value)) |
| 435 | case p.currentIsHistogramSum: |
| 436 | p.currentMetric.Histogram.SampleSum = proto.Float64(value) |
| 437 | case !math.IsNaN(p.currentBucket): |
| 438 | p.currentMetric.Histogram.Bucket = append( |
| 439 | p.currentMetric.Histogram.Bucket, |
| 440 | &dto.Bucket{ |
| 441 | UpperBound: proto.Float64(p.currentBucket), |
| 442 | CumulativeCount: proto.Uint64(uint64(value)), |
| 443 | }, |
| 444 | ) |
| 445 | } |
| 446 | default: |
| 447 | p.err = fmt.Errorf("unexpected type for metric name %q", p.currentMF.GetName()) |
| 448 | } |
| 449 | if p.currentByte == '\n' { |
| 450 | return p.startOfLine |
| 451 | } |
| 452 | return p.startTimestamp |
| 453 | } |
| 454 | |
| 455 | // startTimestamp represents the state where the next byte read from p.buf is |
| 456 | // the start of the timestamp (or whitespace leading up to it). |
| 457 | func (p *TextParser) startTimestamp() stateFn { |
| 458 | if p.skipBlankTab(); p.err != nil { |
| 459 | return nil // Unexpected end of input. |
| 460 | } |
| 461 | if p.readTokenUntilWhitespace(); p.err != nil { |
| 462 | return nil // Unexpected end of input. |
| 463 | } |
| 464 | timestamp, err := strconv.ParseInt(p.currentToken.String(), 10, 64) |
| 465 | if err != nil { |
| 466 | // Create a more helpful error message. |
| 467 | p.parseError(fmt.Sprintf("expected integer as timestamp, got %q", p.currentToken.String())) |
| 468 | return nil |
| 469 | } |
| 470 | p.currentMetric.TimestampMs = proto.Int64(timestamp) |
| 471 | if p.readTokenUntilNewline(false); p.err != nil { |
| 472 | return nil // Unexpected end of input. |
| 473 | } |
| 474 | if p.currentToken.Len() > 0 { |
| 475 | p.parseError(fmt.Sprintf("spurious string after timestamp: %q", p.currentToken.String())) |
| 476 | return nil |
| 477 | } |
| 478 | return p.startOfLine |
| 479 | } |
| 480 | |
| 481 | // readingHelp represents the state where the last byte read (now in |
| 482 | // p.currentByte) is the first byte of the docstring after 'HELP'. |
| 483 | func (p *TextParser) readingHelp() stateFn { |
| 484 | if p.currentMF.Help != nil { |
| 485 | p.parseError(fmt.Sprintf("second HELP line for metric name %q", p.currentMF.GetName())) |
| 486 | return nil |
| 487 | } |
| 488 | // Rest of line is the docstring. |
| 489 | if p.readTokenUntilNewline(true); p.err != nil { |
| 490 | return nil // Unexpected end of input. |
| 491 | } |
| 492 | p.currentMF.Help = proto.String(p.currentToken.String()) |
| 493 | return p.startOfLine |
| 494 | } |
| 495 | |
| 496 | // readingType represents the state where the last byte read (now in |
| 497 | // p.currentByte) is the first byte of the type hint after 'HELP'. |
| 498 | func (p *TextParser) readingType() stateFn { |
| 499 | if p.currentMF.Type != nil { |
| 500 | p.parseError(fmt.Sprintf("second TYPE line for metric name %q, or TYPE reported after samples", p.currentMF.GetName())) |
| 501 | return nil |
| 502 | } |
| 503 | // Rest of line is the type. |
| 504 | if p.readTokenUntilNewline(false); p.err != nil { |
| 505 | return nil // Unexpected end of input. |
| 506 | } |
| 507 | metricType, ok := dto.MetricType_value[strings.ToUpper(p.currentToken.String())] |
| 508 | if !ok { |
| 509 | p.parseError(fmt.Sprintf("unknown metric type %q", p.currentToken.String())) |
| 510 | return nil |
| 511 | } |
| 512 | p.currentMF.Type = dto.MetricType(metricType).Enum() |
| 513 | return p.startOfLine |
| 514 | } |
| 515 | |
| 516 | // parseError sets p.err to a ParseError at the current line with the given |
| 517 | // message. |
| 518 | func (p *TextParser) parseError(msg string) { |
| 519 | p.err = ParseError{ |
| 520 | Line: p.lineCount, |
| 521 | Msg: msg, |
| 522 | } |
| 523 | } |
| 524 | |
| 525 | // skipBlankTab reads (and discards) bytes from p.buf until it encounters a byte |
| 526 | // that is neither ' ' nor '\t'. That byte is left in p.currentByte. |
| 527 | func (p *TextParser) skipBlankTab() { |
| 528 | for { |
| 529 | if p.currentByte, p.err = p.buf.ReadByte(); p.err != nil || !isBlankOrTab(p.currentByte) { |
| 530 | return |
| 531 | } |
| 532 | } |
| 533 | } |
| 534 | |
| 535 | // skipBlankTabIfCurrentBlankTab works exactly as skipBlankTab but doesn't do |
| 536 | // anything if p.currentByte is neither ' ' nor '\t'. |
| 537 | func (p *TextParser) skipBlankTabIfCurrentBlankTab() { |
| 538 | if isBlankOrTab(p.currentByte) { |
| 539 | p.skipBlankTab() |
| 540 | } |
| 541 | } |
| 542 | |
| 543 | // readTokenUntilWhitespace copies bytes from p.buf into p.currentToken. The |
| 544 | // first byte considered is the byte already read (now in p.currentByte). The |
| 545 | // first whitespace byte encountered is still copied into p.currentByte, but not |
| 546 | // into p.currentToken. |
| 547 | func (p *TextParser) readTokenUntilWhitespace() { |
| 548 | p.currentToken.Reset() |
| 549 | for p.err == nil && !isBlankOrTab(p.currentByte) && p.currentByte != '\n' { |
| 550 | p.currentToken.WriteByte(p.currentByte) |
| 551 | p.currentByte, p.err = p.buf.ReadByte() |
| 552 | } |
| 553 | } |
| 554 | |
| 555 | // readTokenUntilNewline copies bytes from p.buf into p.currentToken. The first |
| 556 | // byte considered is the byte already read (now in p.currentByte). The first |
| 557 | // newline byte encountered is still copied into p.currentByte, but not into |
| 558 | // p.currentToken. If recognizeEscapeSequence is true, two escape sequences are |
| 559 | // recognized: '\\' translates into '\', and '\n' into a line-feed character. |
| 560 | // All other escape sequences are invalid and cause an error. |
| 561 | func (p *TextParser) readTokenUntilNewline(recognizeEscapeSequence bool) { |
| 562 | p.currentToken.Reset() |
| 563 | escaped := false |
| 564 | for p.err == nil { |
| 565 | if recognizeEscapeSequence && escaped { |
| 566 | switch p.currentByte { |
| 567 | case '\\': |
| 568 | p.currentToken.WriteByte(p.currentByte) |
| 569 | case 'n': |
| 570 | p.currentToken.WriteByte('\n') |
| 571 | default: |
| 572 | p.parseError(fmt.Sprintf("invalid escape sequence '\\%c'", p.currentByte)) |
| 573 | return |
| 574 | } |
| 575 | escaped = false |
| 576 | } else { |
| 577 | switch p.currentByte { |
| 578 | case '\n': |
| 579 | return |
| 580 | case '\\': |
| 581 | escaped = true |
| 582 | default: |
| 583 | p.currentToken.WriteByte(p.currentByte) |
| 584 | } |
| 585 | } |
| 586 | p.currentByte, p.err = p.buf.ReadByte() |
| 587 | } |
| 588 | } |
| 589 | |
| 590 | // readTokenAsMetricName copies a metric name from p.buf into p.currentToken. |
| 591 | // The first byte considered is the byte already read (now in p.currentByte). |
| 592 | // The first byte not part of a metric name is still copied into p.currentByte, |
| 593 | // but not into p.currentToken. |
| 594 | func (p *TextParser) readTokenAsMetricName() { |
| 595 | p.currentToken.Reset() |
| 596 | if !isValidMetricNameStart(p.currentByte) { |
| 597 | return |
| 598 | } |
| 599 | for { |
| 600 | p.currentToken.WriteByte(p.currentByte) |
| 601 | p.currentByte, p.err = p.buf.ReadByte() |
| 602 | if p.err != nil || !isValidMetricNameContinuation(p.currentByte) { |
| 603 | return |
| 604 | } |
| 605 | } |
| 606 | } |
| 607 | |
| 608 | // readTokenAsLabelName copies a label name from p.buf into p.currentToken. |
| 609 | // The first byte considered is the byte already read (now in p.currentByte). |
| 610 | // The first byte not part of a label name is still copied into p.currentByte, |
| 611 | // but not into p.currentToken. |
| 612 | func (p *TextParser) readTokenAsLabelName() { |
| 613 | p.currentToken.Reset() |
| 614 | if !isValidLabelNameStart(p.currentByte) { |
| 615 | return |
| 616 | } |
| 617 | for { |
| 618 | p.currentToken.WriteByte(p.currentByte) |
| 619 | p.currentByte, p.err = p.buf.ReadByte() |
| 620 | if p.err != nil || !isValidLabelNameContinuation(p.currentByte) { |
| 621 | return |
| 622 | } |
| 623 | } |
| 624 | } |
| 625 | |
| 626 | // readTokenAsLabelValue copies a label value from p.buf into p.currentToken. |
| 627 | // In contrast to the other 'readTokenAs...' functions, which start with the |
| 628 | // last read byte in p.currentByte, this method ignores p.currentByte and starts |
| 629 | // with reading a new byte from p.buf. The first byte not part of a label value |
| 630 | // is still copied into p.currentByte, but not into p.currentToken. |
| 631 | func (p *TextParser) readTokenAsLabelValue() { |
| 632 | p.currentToken.Reset() |
| 633 | escaped := false |
| 634 | for { |
| 635 | if p.currentByte, p.err = p.buf.ReadByte(); p.err != nil { |
| 636 | return |
| 637 | } |
| 638 | if escaped { |
| 639 | switch p.currentByte { |
| 640 | case '"', '\\': |
| 641 | p.currentToken.WriteByte(p.currentByte) |
| 642 | case 'n': |
| 643 | p.currentToken.WriteByte('\n') |
| 644 | default: |
| 645 | p.parseError(fmt.Sprintf("invalid escape sequence '\\%c'", p.currentByte)) |
| 646 | return |
| 647 | } |
| 648 | escaped = false |
| 649 | continue |
| 650 | } |
| 651 | switch p.currentByte { |
| 652 | case '"': |
| 653 | return |
| 654 | case '\n': |
| 655 | p.parseError(fmt.Sprintf("label value %q contains unescaped new-line", p.currentToken.String())) |
| 656 | return |
| 657 | case '\\': |
| 658 | escaped = true |
| 659 | default: |
| 660 | p.currentToken.WriteByte(p.currentByte) |
| 661 | } |
| 662 | } |
| 663 | } |
| 664 | |
| 665 | func (p *TextParser) setOrCreateCurrentMF() { |
| 666 | p.currentIsSummaryCount = false |
| 667 | p.currentIsSummarySum = false |
| 668 | p.currentIsHistogramCount = false |
| 669 | p.currentIsHistogramSum = false |
| 670 | name := p.currentToken.String() |
| 671 | if p.currentMF = p.metricFamiliesByName[name]; p.currentMF != nil { |
| 672 | return |
| 673 | } |
| 674 | // Try out if this is a _sum or _count for a summary/histogram. |
| 675 | summaryName := summaryMetricName(name) |
| 676 | if p.currentMF = p.metricFamiliesByName[summaryName]; p.currentMF != nil { |
| 677 | if p.currentMF.GetType() == dto.MetricType_SUMMARY { |
| 678 | if isCount(name) { |
| 679 | p.currentIsSummaryCount = true |
| 680 | } |
| 681 | if isSum(name) { |
| 682 | p.currentIsSummarySum = true |
| 683 | } |
| 684 | return |
| 685 | } |
| 686 | } |
| 687 | histogramName := histogramMetricName(name) |
| 688 | if p.currentMF = p.metricFamiliesByName[histogramName]; p.currentMF != nil { |
| 689 | if p.currentMF.GetType() == dto.MetricType_HISTOGRAM { |
| 690 | if isCount(name) { |
| 691 | p.currentIsHistogramCount = true |
| 692 | } |
| 693 | if isSum(name) { |
| 694 | p.currentIsHistogramSum = true |
| 695 | } |
| 696 | return |
| 697 | } |
| 698 | } |
| 699 | p.currentMF = &dto.MetricFamily{Name: proto.String(name)} |
| 700 | p.metricFamiliesByName[name] = p.currentMF |
| 701 | } |
| 702 | |
| 703 | func isValidLabelNameStart(b byte) bool { |
| 704 | return (b >= 'a' && b <= 'z') || (b >= 'A' && b <= 'Z') || b == '_' |
| 705 | } |
| 706 | |
| 707 | func isValidLabelNameContinuation(b byte) bool { |
| 708 | return isValidLabelNameStart(b) || (b >= '0' && b <= '9') |
| 709 | } |
| 710 | |
| 711 | func isValidMetricNameStart(b byte) bool { |
| 712 | return isValidLabelNameStart(b) || b == ':' |
| 713 | } |
| 714 | |
| 715 | func isValidMetricNameContinuation(b byte) bool { |
| 716 | return isValidLabelNameContinuation(b) || b == ':' |
| 717 | } |
| 718 | |
| 719 | func isBlankOrTab(b byte) bool { |
| 720 | return b == ' ' || b == '\t' |
| 721 | } |
| 722 | |
| 723 | func isCount(name string) bool { |
| 724 | return len(name) > 6 && name[len(name)-6:] == "_count" |
| 725 | } |
| 726 | |
| 727 | func isSum(name string) bool { |
| 728 | return len(name) > 4 && name[len(name)-4:] == "_sum" |
| 729 | } |
| 730 | |
| 731 | func isBucket(name string) bool { |
| 732 | return len(name) > 7 && name[len(name)-7:] == "_bucket" |
| 733 | } |
| 734 | |
| 735 | func summaryMetricName(name string) string { |
| 736 | switch { |
| 737 | case isCount(name): |
| 738 | return name[:len(name)-6] |
| 739 | case isSum(name): |
| 740 | return name[:len(name)-4] |
| 741 | default: |
| 742 | return name |
| 743 | } |
| 744 | } |
| 745 | |
| 746 | func histogramMetricName(name string) string { |
| 747 | switch { |
| 748 | case isCount(name): |
| 749 | return name[:len(name)-6] |
| 750 | case isSum(name): |
| 751 | return name[:len(name)-4] |
| 752 | case isBucket(name): |
| 753 | return name[:len(name)-7] |
| 754 | default: |
| 755 | return name |
| 756 | } |
| 757 | } |