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" //nolint:staticcheck // Ignore SA1019. Need to keep deprecated package for compatibility.
|
---|
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 | // Check for duplicate label names.
|
---|
303 | labels := make(map[string]struct{})
|
---|
304 | for _, l := range p.currentMetric.Label {
|
---|
305 | lName := l.GetName()
|
---|
306 | if _, exists := labels[lName]; !exists {
|
---|
307 | labels[lName] = struct{}{}
|
---|
308 | } else {
|
---|
309 | p.parseError(fmt.Sprintf("duplicate label names for metric %q", p.currentMF.GetName()))
|
---|
310 | return nil
|
---|
311 | }
|
---|
312 | }
|
---|
313 | return p.startLabelValue
|
---|
314 | }
|
---|
315 |
|
---|
316 | // startLabelValue represents the state where the next byte read from p.buf is
|
---|
317 | // the start of a (quoted) label value (or whitespace leading up to it).
|
---|
318 | func (p *TextParser) startLabelValue() stateFn {
|
---|
319 | if p.skipBlankTab(); p.err != nil {
|
---|
320 | return nil // Unexpected end of input.
|
---|
321 | }
|
---|
322 | if p.currentByte != '"' {
|
---|
323 | p.parseError(fmt.Sprintf("expected '\"' at start of label value, found %q", p.currentByte))
|
---|
324 | return nil
|
---|
325 | }
|
---|
326 | if p.readTokenAsLabelValue(); p.err != nil {
|
---|
327 | return nil
|
---|
328 | }
|
---|
329 | if !model.LabelValue(p.currentToken.String()).IsValid() {
|
---|
330 | p.parseError(fmt.Sprintf("invalid label value %q", p.currentToken.String()))
|
---|
331 | return nil
|
---|
332 | }
|
---|
333 | p.currentLabelPair.Value = proto.String(p.currentToken.String())
|
---|
334 | // Special treatment of summaries:
|
---|
335 | // - Quantile labels are special, will result in dto.Quantile later.
|
---|
336 | // - Other labels have to be added to currentLabels for signature calculation.
|
---|
337 | if p.currentMF.GetType() == dto.MetricType_SUMMARY {
|
---|
338 | if p.currentLabelPair.GetName() == model.QuantileLabel {
|
---|
339 | if p.currentQuantile, p.err = parseFloat(p.currentLabelPair.GetValue()); p.err != nil {
|
---|
340 | // Create a more helpful error message.
|
---|
341 | p.parseError(fmt.Sprintf("expected float as value for 'quantile' label, got %q", p.currentLabelPair.GetValue()))
|
---|
342 | return nil
|
---|
343 | }
|
---|
344 | } else {
|
---|
345 | p.currentLabels[p.currentLabelPair.GetName()] = p.currentLabelPair.GetValue()
|
---|
346 | }
|
---|
347 | }
|
---|
348 | // Similar special treatment of histograms.
|
---|
349 | if p.currentMF.GetType() == dto.MetricType_HISTOGRAM {
|
---|
350 | if p.currentLabelPair.GetName() == model.BucketLabel {
|
---|
351 | if p.currentBucket, p.err = parseFloat(p.currentLabelPair.GetValue()); p.err != nil {
|
---|
352 | // Create a more helpful error message.
|
---|
353 | p.parseError(fmt.Sprintf("expected float as value for 'le' label, got %q", p.currentLabelPair.GetValue()))
|
---|
354 | return nil
|
---|
355 | }
|
---|
356 | } else {
|
---|
357 | p.currentLabels[p.currentLabelPair.GetName()] = p.currentLabelPair.GetValue()
|
---|
358 | }
|
---|
359 | }
|
---|
360 | if p.skipBlankTab(); p.err != nil {
|
---|
361 | return nil // Unexpected end of input.
|
---|
362 | }
|
---|
363 | switch p.currentByte {
|
---|
364 | case ',':
|
---|
365 | return p.startLabelName
|
---|
366 |
|
---|
367 | case '}':
|
---|
368 | if p.skipBlankTab(); p.err != nil {
|
---|
369 | return nil // Unexpected end of input.
|
---|
370 | }
|
---|
371 | return p.readingValue
|
---|
372 | default:
|
---|
373 | p.parseError(fmt.Sprintf("unexpected end of label value %q", p.currentLabelPair.GetValue()))
|
---|
374 | return nil
|
---|
375 | }
|
---|
376 | }
|
---|
377 |
|
---|
378 | // readingValue represents the state where the last byte read (now in
|
---|
379 | // p.currentByte) is the first byte of the sample value (i.e. a float).
|
---|
380 | func (p *TextParser) readingValue() stateFn {
|
---|
381 | // When we are here, we have read all the labels, so for the
|
---|
382 | // special case of a summary/histogram, we can finally find out
|
---|
383 | // if the metric already exists.
|
---|
384 | if p.currentMF.GetType() == dto.MetricType_SUMMARY {
|
---|
385 | signature := model.LabelsToSignature(p.currentLabels)
|
---|
386 | if summary := p.summaries[signature]; summary != nil {
|
---|
387 | p.currentMetric = summary
|
---|
388 | } else {
|
---|
389 | p.summaries[signature] = p.currentMetric
|
---|
390 | p.currentMF.Metric = append(p.currentMF.Metric, p.currentMetric)
|
---|
391 | }
|
---|
392 | } else if p.currentMF.GetType() == dto.MetricType_HISTOGRAM {
|
---|
393 | signature := model.LabelsToSignature(p.currentLabels)
|
---|
394 | if histogram := p.histograms[signature]; histogram != nil {
|
---|
395 | p.currentMetric = histogram
|
---|
396 | } else {
|
---|
397 | p.histograms[signature] = p.currentMetric
|
---|
398 | p.currentMF.Metric = append(p.currentMF.Metric, p.currentMetric)
|
---|
399 | }
|
---|
400 | } else {
|
---|
401 | p.currentMF.Metric = append(p.currentMF.Metric, p.currentMetric)
|
---|
402 | }
|
---|
403 | if p.readTokenUntilWhitespace(); p.err != nil {
|
---|
404 | return nil // Unexpected end of input.
|
---|
405 | }
|
---|
406 | value, err := parseFloat(p.currentToken.String())
|
---|
407 | if err != nil {
|
---|
408 | // Create a more helpful error message.
|
---|
409 | p.parseError(fmt.Sprintf("expected float as value, got %q", p.currentToken.String()))
|
---|
410 | return nil
|
---|
411 | }
|
---|
412 | switch p.currentMF.GetType() {
|
---|
413 | case dto.MetricType_COUNTER:
|
---|
414 | p.currentMetric.Counter = &dto.Counter{Value: proto.Float64(value)}
|
---|
415 | case dto.MetricType_GAUGE:
|
---|
416 | p.currentMetric.Gauge = &dto.Gauge{Value: proto.Float64(value)}
|
---|
417 | case dto.MetricType_UNTYPED:
|
---|
418 | p.currentMetric.Untyped = &dto.Untyped{Value: proto.Float64(value)}
|
---|
419 | case dto.MetricType_SUMMARY:
|
---|
420 | // *sigh*
|
---|
421 | if p.currentMetric.Summary == nil {
|
---|
422 | p.currentMetric.Summary = &dto.Summary{}
|
---|
423 | }
|
---|
424 | switch {
|
---|
425 | case p.currentIsSummaryCount:
|
---|
426 | p.currentMetric.Summary.SampleCount = proto.Uint64(uint64(value))
|
---|
427 | case p.currentIsSummarySum:
|
---|
428 | p.currentMetric.Summary.SampleSum = proto.Float64(value)
|
---|
429 | case !math.IsNaN(p.currentQuantile):
|
---|
430 | p.currentMetric.Summary.Quantile = append(
|
---|
431 | p.currentMetric.Summary.Quantile,
|
---|
432 | &dto.Quantile{
|
---|
433 | Quantile: proto.Float64(p.currentQuantile),
|
---|
434 | Value: proto.Float64(value),
|
---|
435 | },
|
---|
436 | )
|
---|
437 | }
|
---|
438 | case dto.MetricType_HISTOGRAM:
|
---|
439 | // *sigh*
|
---|
440 | if p.currentMetric.Histogram == nil {
|
---|
441 | p.currentMetric.Histogram = &dto.Histogram{}
|
---|
442 | }
|
---|
443 | switch {
|
---|
444 | case p.currentIsHistogramCount:
|
---|
445 | p.currentMetric.Histogram.SampleCount = proto.Uint64(uint64(value))
|
---|
446 | case p.currentIsHistogramSum:
|
---|
447 | p.currentMetric.Histogram.SampleSum = proto.Float64(value)
|
---|
448 | case !math.IsNaN(p.currentBucket):
|
---|
449 | p.currentMetric.Histogram.Bucket = append(
|
---|
450 | p.currentMetric.Histogram.Bucket,
|
---|
451 | &dto.Bucket{
|
---|
452 | UpperBound: proto.Float64(p.currentBucket),
|
---|
453 | CumulativeCount: proto.Uint64(uint64(value)),
|
---|
454 | },
|
---|
455 | )
|
---|
456 | }
|
---|
457 | default:
|
---|
458 | p.err = fmt.Errorf("unexpected type for metric name %q", p.currentMF.GetName())
|
---|
459 | }
|
---|
460 | if p.currentByte == '\n' {
|
---|
461 | return p.startOfLine
|
---|
462 | }
|
---|
463 | return p.startTimestamp
|
---|
464 | }
|
---|
465 |
|
---|
466 | // startTimestamp represents the state where the next byte read from p.buf is
|
---|
467 | // the start of the timestamp (or whitespace leading up to it).
|
---|
468 | func (p *TextParser) startTimestamp() stateFn {
|
---|
469 | if p.skipBlankTab(); p.err != nil {
|
---|
470 | return nil // Unexpected end of input.
|
---|
471 | }
|
---|
472 | if p.readTokenUntilWhitespace(); p.err != nil {
|
---|
473 | return nil // Unexpected end of input.
|
---|
474 | }
|
---|
475 | timestamp, err := strconv.ParseInt(p.currentToken.String(), 10, 64)
|
---|
476 | if err != nil {
|
---|
477 | // Create a more helpful error message.
|
---|
478 | p.parseError(fmt.Sprintf("expected integer as timestamp, got %q", p.currentToken.String()))
|
---|
479 | return nil
|
---|
480 | }
|
---|
481 | p.currentMetric.TimestampMs = proto.Int64(timestamp)
|
---|
482 | if p.readTokenUntilNewline(false); p.err != nil {
|
---|
483 | return nil // Unexpected end of input.
|
---|
484 | }
|
---|
485 | if p.currentToken.Len() > 0 {
|
---|
486 | p.parseError(fmt.Sprintf("spurious string after timestamp: %q", p.currentToken.String()))
|
---|
487 | return nil
|
---|
488 | }
|
---|
489 | return p.startOfLine
|
---|
490 | }
|
---|
491 |
|
---|
492 | // readingHelp represents the state where the last byte read (now in
|
---|
493 | // p.currentByte) is the first byte of the docstring after 'HELP'.
|
---|
494 | func (p *TextParser) readingHelp() stateFn {
|
---|
495 | if p.currentMF.Help != nil {
|
---|
496 | p.parseError(fmt.Sprintf("second HELP line for metric name %q", p.currentMF.GetName()))
|
---|
497 | return nil
|
---|
498 | }
|
---|
499 | // Rest of line is the docstring.
|
---|
500 | if p.readTokenUntilNewline(true); p.err != nil {
|
---|
501 | return nil // Unexpected end of input.
|
---|
502 | }
|
---|
503 | p.currentMF.Help = proto.String(p.currentToken.String())
|
---|
504 | return p.startOfLine
|
---|
505 | }
|
---|
506 |
|
---|
507 | // readingType represents the state where the last byte read (now in
|
---|
508 | // p.currentByte) is the first byte of the type hint after 'HELP'.
|
---|
509 | func (p *TextParser) readingType() stateFn {
|
---|
510 | if p.currentMF.Type != nil {
|
---|
511 | p.parseError(fmt.Sprintf("second TYPE line for metric name %q, or TYPE reported after samples", p.currentMF.GetName()))
|
---|
512 | return nil
|
---|
513 | }
|
---|
514 | // Rest of line is the type.
|
---|
515 | if p.readTokenUntilNewline(false); p.err != nil {
|
---|
516 | return nil // Unexpected end of input.
|
---|
517 | }
|
---|
518 | metricType, ok := dto.MetricType_value[strings.ToUpper(p.currentToken.String())]
|
---|
519 | if !ok {
|
---|
520 | p.parseError(fmt.Sprintf("unknown metric type %q", p.currentToken.String()))
|
---|
521 | return nil
|
---|
522 | }
|
---|
523 | p.currentMF.Type = dto.MetricType(metricType).Enum()
|
---|
524 | return p.startOfLine
|
---|
525 | }
|
---|
526 |
|
---|
527 | // parseError sets p.err to a ParseError at the current line with the given
|
---|
528 | // message.
|
---|
529 | func (p *TextParser) parseError(msg string) {
|
---|
530 | p.err = ParseError{
|
---|
531 | Line: p.lineCount,
|
---|
532 | Msg: msg,
|
---|
533 | }
|
---|
534 | }
|
---|
535 |
|
---|
536 | // skipBlankTab reads (and discards) bytes from p.buf until it encounters a byte
|
---|
537 | // that is neither ' ' nor '\t'. That byte is left in p.currentByte.
|
---|
538 | func (p *TextParser) skipBlankTab() {
|
---|
539 | for {
|
---|
540 | if p.currentByte, p.err = p.buf.ReadByte(); p.err != nil || !isBlankOrTab(p.currentByte) {
|
---|
541 | return
|
---|
542 | }
|
---|
543 | }
|
---|
544 | }
|
---|
545 |
|
---|
546 | // skipBlankTabIfCurrentBlankTab works exactly as skipBlankTab but doesn't do
|
---|
547 | // anything if p.currentByte is neither ' ' nor '\t'.
|
---|
548 | func (p *TextParser) skipBlankTabIfCurrentBlankTab() {
|
---|
549 | if isBlankOrTab(p.currentByte) {
|
---|
550 | p.skipBlankTab()
|
---|
551 | }
|
---|
552 | }
|
---|
553 |
|
---|
554 | // readTokenUntilWhitespace copies bytes from p.buf into p.currentToken. The
|
---|
555 | // first byte considered is the byte already read (now in p.currentByte). The
|
---|
556 | // first whitespace byte encountered is still copied into p.currentByte, but not
|
---|
557 | // into p.currentToken.
|
---|
558 | func (p *TextParser) readTokenUntilWhitespace() {
|
---|
559 | p.currentToken.Reset()
|
---|
560 | for p.err == nil && !isBlankOrTab(p.currentByte) && p.currentByte != '\n' {
|
---|
561 | p.currentToken.WriteByte(p.currentByte)
|
---|
562 | p.currentByte, p.err = p.buf.ReadByte()
|
---|
563 | }
|
---|
564 | }
|
---|
565 |
|
---|
566 | // readTokenUntilNewline copies bytes from p.buf into p.currentToken. The first
|
---|
567 | // byte considered is the byte already read (now in p.currentByte). The first
|
---|
568 | // newline byte encountered is still copied into p.currentByte, but not into
|
---|
569 | // p.currentToken. If recognizeEscapeSequence is true, two escape sequences are
|
---|
570 | // recognized: '\\' translates into '\', and '\n' into a line-feed character.
|
---|
571 | // All other escape sequences are invalid and cause an error.
|
---|
572 | func (p *TextParser) readTokenUntilNewline(recognizeEscapeSequence bool) {
|
---|
573 | p.currentToken.Reset()
|
---|
574 | escaped := false
|
---|
575 | for p.err == nil {
|
---|
576 | if recognizeEscapeSequence && escaped {
|
---|
577 | switch p.currentByte {
|
---|
578 | case '\\':
|
---|
579 | p.currentToken.WriteByte(p.currentByte)
|
---|
580 | case 'n':
|
---|
581 | p.currentToken.WriteByte('\n')
|
---|
582 | default:
|
---|
583 | p.parseError(fmt.Sprintf("invalid escape sequence '\\%c'", p.currentByte))
|
---|
584 | return
|
---|
585 | }
|
---|
586 | escaped = false
|
---|
587 | } else {
|
---|
588 | switch p.currentByte {
|
---|
589 | case '\n':
|
---|
590 | return
|
---|
591 | case '\\':
|
---|
592 | escaped = true
|
---|
593 | default:
|
---|
594 | p.currentToken.WriteByte(p.currentByte)
|
---|
595 | }
|
---|
596 | }
|
---|
597 | p.currentByte, p.err = p.buf.ReadByte()
|
---|
598 | }
|
---|
599 | }
|
---|
600 |
|
---|
601 | // readTokenAsMetricName copies a metric name from p.buf into p.currentToken.
|
---|
602 | // The first byte considered is the byte already read (now in p.currentByte).
|
---|
603 | // The first byte not part of a metric name is still copied into p.currentByte,
|
---|
604 | // but not into p.currentToken.
|
---|
605 | func (p *TextParser) readTokenAsMetricName() {
|
---|
606 | p.currentToken.Reset()
|
---|
607 | if !isValidMetricNameStart(p.currentByte) {
|
---|
608 | return
|
---|
609 | }
|
---|
610 | for {
|
---|
611 | p.currentToken.WriteByte(p.currentByte)
|
---|
612 | p.currentByte, p.err = p.buf.ReadByte()
|
---|
613 | if p.err != nil || !isValidMetricNameContinuation(p.currentByte) {
|
---|
614 | return
|
---|
615 | }
|
---|
616 | }
|
---|
617 | }
|
---|
618 |
|
---|
619 | // readTokenAsLabelName copies a label name from p.buf into p.currentToken.
|
---|
620 | // The first byte considered is the byte already read (now in p.currentByte).
|
---|
621 | // The first byte not part of a label name is still copied into p.currentByte,
|
---|
622 | // but not into p.currentToken.
|
---|
623 | func (p *TextParser) readTokenAsLabelName() {
|
---|
624 | p.currentToken.Reset()
|
---|
625 | if !isValidLabelNameStart(p.currentByte) {
|
---|
626 | return
|
---|
627 | }
|
---|
628 | for {
|
---|
629 | p.currentToken.WriteByte(p.currentByte)
|
---|
630 | p.currentByte, p.err = p.buf.ReadByte()
|
---|
631 | if p.err != nil || !isValidLabelNameContinuation(p.currentByte) {
|
---|
632 | return
|
---|
633 | }
|
---|
634 | }
|
---|
635 | }
|
---|
636 |
|
---|
637 | // readTokenAsLabelValue copies a label value from p.buf into p.currentToken.
|
---|
638 | // In contrast to the other 'readTokenAs...' functions, which start with the
|
---|
639 | // last read byte in p.currentByte, this method ignores p.currentByte and starts
|
---|
640 | // with reading a new byte from p.buf. The first byte not part of a label value
|
---|
641 | // is still copied into p.currentByte, but not into p.currentToken.
|
---|
642 | func (p *TextParser) readTokenAsLabelValue() {
|
---|
643 | p.currentToken.Reset()
|
---|
644 | escaped := false
|
---|
645 | for {
|
---|
646 | if p.currentByte, p.err = p.buf.ReadByte(); p.err != nil {
|
---|
647 | return
|
---|
648 | }
|
---|
649 | if escaped {
|
---|
650 | switch p.currentByte {
|
---|
651 | case '"', '\\':
|
---|
652 | p.currentToken.WriteByte(p.currentByte)
|
---|
653 | case 'n':
|
---|
654 | p.currentToken.WriteByte('\n')
|
---|
655 | default:
|
---|
656 | p.parseError(fmt.Sprintf("invalid escape sequence '\\%c'", p.currentByte))
|
---|
657 | return
|
---|
658 | }
|
---|
659 | escaped = false
|
---|
660 | continue
|
---|
661 | }
|
---|
662 | switch p.currentByte {
|
---|
663 | case '"':
|
---|
664 | return
|
---|
665 | case '\n':
|
---|
666 | p.parseError(fmt.Sprintf("label value %q contains unescaped new-line", p.currentToken.String()))
|
---|
667 | return
|
---|
668 | case '\\':
|
---|
669 | escaped = true
|
---|
670 | default:
|
---|
671 | p.currentToken.WriteByte(p.currentByte)
|
---|
672 | }
|
---|
673 | }
|
---|
674 | }
|
---|
675 |
|
---|
676 | func (p *TextParser) setOrCreateCurrentMF() {
|
---|
677 | p.currentIsSummaryCount = false
|
---|
678 | p.currentIsSummarySum = false
|
---|
679 | p.currentIsHistogramCount = false
|
---|
680 | p.currentIsHistogramSum = false
|
---|
681 | name := p.currentToken.String()
|
---|
682 | if p.currentMF = p.metricFamiliesByName[name]; p.currentMF != nil {
|
---|
683 | return
|
---|
684 | }
|
---|
685 | // Try out if this is a _sum or _count for a summary/histogram.
|
---|
686 | summaryName := summaryMetricName(name)
|
---|
687 | if p.currentMF = p.metricFamiliesByName[summaryName]; p.currentMF != nil {
|
---|
688 | if p.currentMF.GetType() == dto.MetricType_SUMMARY {
|
---|
689 | if isCount(name) {
|
---|
690 | p.currentIsSummaryCount = true
|
---|
691 | }
|
---|
692 | if isSum(name) {
|
---|
693 | p.currentIsSummarySum = true
|
---|
694 | }
|
---|
695 | return
|
---|
696 | }
|
---|
697 | }
|
---|
698 | histogramName := histogramMetricName(name)
|
---|
699 | if p.currentMF = p.metricFamiliesByName[histogramName]; p.currentMF != nil {
|
---|
700 | if p.currentMF.GetType() == dto.MetricType_HISTOGRAM {
|
---|
701 | if isCount(name) {
|
---|
702 | p.currentIsHistogramCount = true
|
---|
703 | }
|
---|
704 | if isSum(name) {
|
---|
705 | p.currentIsHistogramSum = true
|
---|
706 | }
|
---|
707 | return
|
---|
708 | }
|
---|
709 | }
|
---|
710 | p.currentMF = &dto.MetricFamily{Name: proto.String(name)}
|
---|
711 | p.metricFamiliesByName[name] = p.currentMF
|
---|
712 | }
|
---|
713 |
|
---|
714 | func isValidLabelNameStart(b byte) bool {
|
---|
715 | return (b >= 'a' && b <= 'z') || (b >= 'A' && b <= 'Z') || b == '_'
|
---|
716 | }
|
---|
717 |
|
---|
718 | func isValidLabelNameContinuation(b byte) bool {
|
---|
719 | return isValidLabelNameStart(b) || (b >= '0' && b <= '9')
|
---|
720 | }
|
---|
721 |
|
---|
722 | func isValidMetricNameStart(b byte) bool {
|
---|
723 | return isValidLabelNameStart(b) || b == ':'
|
---|
724 | }
|
---|
725 |
|
---|
726 | func isValidMetricNameContinuation(b byte) bool {
|
---|
727 | return isValidLabelNameContinuation(b) || b == ':'
|
---|
728 | }
|
---|
729 |
|
---|
730 | func isBlankOrTab(b byte) bool {
|
---|
731 | return b == ' ' || b == '\t'
|
---|
732 | }
|
---|
733 |
|
---|
734 | func isCount(name string) bool {
|
---|
735 | return len(name) > 6 && name[len(name)-6:] == "_count"
|
---|
736 | }
|
---|
737 |
|
---|
738 | func isSum(name string) bool {
|
---|
739 | return len(name) > 4 && name[len(name)-4:] == "_sum"
|
---|
740 | }
|
---|
741 |
|
---|
742 | func isBucket(name string) bool {
|
---|
743 | return len(name) > 7 && name[len(name)-7:] == "_bucket"
|
---|
744 | }
|
---|
745 |
|
---|
746 | func summaryMetricName(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 | default:
|
---|
753 | return name
|
---|
754 | }
|
---|
755 | }
|
---|
756 |
|
---|
757 | func histogramMetricName(name string) string {
|
---|
758 | switch {
|
---|
759 | case isCount(name):
|
---|
760 | return name[:len(name)-6]
|
---|
761 | case isSum(name):
|
---|
762 | return name[:len(name)-4]
|
---|
763 | case isBucket(name):
|
---|
764 | return name[:len(name)-7]
|
---|
765 | default:
|
---|
766 | return name
|
---|
767 | }
|
---|
768 | }
|
---|
769 |
|
---|
770 | func parseFloat(s string) (float64, error) {
|
---|
771 | if strings.ContainsAny(s, "pP_") {
|
---|
772 | return 0, fmt.Errorf("unsupported character in float")
|
---|
773 | }
|
---|
774 | return strconv.ParseFloat(s, 64)
|
---|
775 | }
|
---|