prometheus/model/textparse/openmetricsparse.go

531 lines
15 KiB
Go
Raw Permalink Normal View History

// Copyright 2018 The Prometheus Authors
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
//go:generate go get -u modernc.org/golex
//go:generate golex -o=openmetricslex.l.go openmetricslex.l
package textparse
import (
"errors"
"fmt"
"io"
"math"
"strings"
"unicode/utf8"
"github.com/prometheus/common/model"
"github.com/prometheus/prometheus/model/exemplar"
Style cleanup of all the changes in sparsehistogram so far A lot of this code was hacked together, literally during a hackathon. This commit intends not to change the code substantially, but just make the code obey the usual style practices. A (possibly incomplete) list of areas: * Generally address linter warnings. * The `pgk` directory is deprecated as per dev-summit. No new packages should be added to it. I moved the new `pkg/histogram` package to `model` anticipating what's proposed in #9478. * Make the naming of the Sparse Histogram more consistent. Including abbreviations, there were just too many names for it: SparseHistogram, Histogram, Histo, hist, his, shs, h. The idea is to call it "Histogram" in general. Only add "Sparse" if it is needed to avoid confusion with conventional Histograms (which is rare because the TSDB really has no notion of conventional Histograms). Use abbreviations only in local scope, and then really abbreviate (not just removing three out of seven letters like in "Histo"). This is in the spirit of https://github.com/golang/go/wiki/CodeReviewComments#variable-names * Several other minor name changes. * A lot of formatting of doc comments. For one, following https://github.com/golang/go/wiki/CodeReviewComments#comment-sentences , but also layout question, anticipating how things will look like when rendered by `godoc` (even where `godoc` doesn't render them right now because they are for unexported types or not a doc comment at all but just a normal code comment - consistency is queen!). * Re-enabled `TestQueryLog` and `TestEndopints` (they pass now, leaving them disabled was presumably an oversight). * Bucket iterator for histogram.Histogram is now created with a method. * HistogramChunk.iterator now allows iterator recycling. (I think @dieterbe only commented it out because he was confused by the question in the comment.) * HistogramAppender.Append panics now because we decided to treat staleness marker differently. Signed-off-by: beorn7 <beorn@grafana.com>
2021-10-09 06:57:07 -07:00
"github.com/prometheus/prometheus/model/histogram"
"github.com/prometheus/prometheus/model/labels"
"github.com/prometheus/prometheus/model/value"
)
type openMetricsLexer struct {
b []byte
i int
start int
err error
state int
}
// buf returns the buffer of the current token.
func (l *openMetricsLexer) buf() []byte {
return l.b[l.start:l.i]
}
// next advances the openMetricsLexer to the next character.
func (l *openMetricsLexer) next() byte {
l.i++
if l.i >= len(l.b) {
l.err = io.EOF
return byte(tEOF)
}
// Lex struggles with null bytes. If we are in a label value or help string, where
// they are allowed, consume them here immediately.
for l.b[l.i] == 0 && (l.state == sLValue || l.state == sMeta2 || l.state == sComment) {
l.i++
if l.i >= len(l.b) {
l.err = io.EOF
return byte(tEOF)
}
}
return l.b[l.i]
}
func (l *openMetricsLexer) Error(es string) {
l.err = errors.New(es)
}
// OpenMetricsParser parses samples from a byte slice of samples in the official
// OpenMetrics text exposition format.
// This is based on the working draft https://docs.google.com/document/u/1/d/1KwV0mAXwwbvvifBvDKH_LU1YjyXE_wxCkHNoCGq1GX0/edit
type OpenMetricsParser struct {
l *openMetricsLexer
builder labels.ScratchBuilder
series []byte
text []byte
mtype model.MetricType
val float64
ts int64
hasTS bool
start int
// offsets is a list of offsets into series that describe the positions
// of the metric name and label names and values for this series.
// p.offsets[0] is the start character of the metric name.
// p.offsets[1] is the end of the metric name.
// Subsequently, p.offsets is a pair of pair of offsets for the positions
// of the label name and value start and end characters.
offsets []int
eOffsets []int
exemplar []byte
exemplarVal float64
exemplarTs int64
hasExemplarTs bool
}
// NewOpenMetricsParser returns a new parser of the byte slice.
func NewOpenMetricsParser(b []byte, st *labels.SymbolTable) Parser {
return &OpenMetricsParser{
l: &openMetricsLexer{b: b},
builder: labels.NewScratchBuilderWithSymbolTable(st, 16),
}
}
// Series returns the bytes of the series, the timestamp if set, and the value
// of the current sample.
func (p *OpenMetricsParser) Series() ([]byte, *int64, float64) {
if p.hasTS {
ts := p.ts
return p.series, &ts, p.val
}
return p.series, nil, p.val
}
// Histogram returns (nil, nil, nil, nil) for now because OpenMetrics does not
// support sparse histograms yet.
func (p *OpenMetricsParser) Histogram() ([]byte, *int64, *histogram.Histogram, *histogram.FloatHistogram) {
return nil, nil, nil, nil
Hacky implementation of protobuf parsing This "brings back" protobuf parsing, with the only goal to play with the new sparse histograms. The Prom-2.x style parser is highly adapted to the structure of the Prometheus text format (and later OpenMetrics). Some jumping through hoops is required to feed protobuf into it. This is not meant to be a model for the final implementation. It should just enable sparse histogram ingestion at a reasonable efficiency. Following known shortcomings and flaws: - No tests yet. - Summaries and legacy histograms, i.e. without sparse buckets, are ignored. - Staleness doesn't work (but this could be fixed in the appender, to be discussed). - No tricks have been tried that would be similar to the tricks the text parsers do (like direct pointers into the HTTP response body). That makes things weird here. Tricky optimizations only make sense once the final format is specified, which will almost certainly not be the old protobuf format. (Interestingly, I expect this implementation to be in fact much more efficient than the original protobuf ingestion in Prom-1.x.) - This is using a proto3 version of metrics.proto (mostly to be consistent with the other protobuf uses). However, proto3 sees no difference between an unset field. We depend on that to distinguish between an unset timestamp and the timestamp 0 (1970-01-01, 00:00:00 UTC). In this experimental code, we just assume that timestamp is never specified and therefore a timestamp of 0 always is interpreted as "not set". Signed-off-by: beorn7 <beorn@grafana.com>
2021-06-29 14:45:23 -07:00
}
// Help returns the metric name and help text in the current entry.
// Must only be called after Next returned a help entry.
// The returned byte slices become invalid after the next call to Next.
func (p *OpenMetricsParser) Help() ([]byte, []byte) {
m := p.l.b[p.offsets[0]:p.offsets[1]]
// Replacer causes allocations. Replace only when necessary.
if strings.IndexByte(yoloString(p.text), byte('\\')) >= 0 {
// OpenMetrics always uses the Prometheus format label value escaping.
return m, []byte(lvalReplacer.Replace(string(p.text)))
}
return m, p.text
}
// Type returns the metric name and type in the current entry.
// Must only be called after Next returned a type entry.
// The returned byte slices become invalid after the next call to Next.
func (p *OpenMetricsParser) Type() ([]byte, model.MetricType) {
return p.l.b[p.offsets[0]:p.offsets[1]], p.mtype
}
// Unit returns the metric name and unit in the current entry.
// Must only be called after Next returned a unit entry.
// The returned byte slices become invalid after the next call to Next.
func (p *OpenMetricsParser) Unit() ([]byte, []byte) {
return p.l.b[p.offsets[0]:p.offsets[1]], p.text
}
// Comment returns the text of the current comment.
// Must only be called after Next returned a comment entry.
// The returned byte slice becomes invalid after the next call to Next.
func (p *OpenMetricsParser) Comment() []byte {
return p.text
}
// Metric writes the labels of the current sample into the passed labels.
// It returns the string from which the metric was parsed.
func (p *OpenMetricsParser) Metric(l *labels.Labels) string {
// Copy the buffer to a string: this is only necessary for the return value.
s := string(p.series)
p.builder.Reset()
metricName := unreplace(s[p.offsets[0]-p.start : p.offsets[1]-p.start])
p.builder.Add(labels.MetricName, metricName)
for i := 2; i < len(p.offsets); i += 4 {
a := p.offsets[i] - p.start
b := p.offsets[i+1] - p.start
label := unreplace(s[a:b])
c := p.offsets[i+2] - p.start
d := p.offsets[i+3] - p.start
value := unreplace(s[c:d])
p.builder.Add(label, value)
}
p.builder.Sort()
*l = p.builder.Labels()
return s
}
// Exemplar writes the exemplar of the current sample into the passed exemplar.
// It returns whether an exemplar exists. As OpenMetrics only ever has one
// exemplar per sample, every call after the first (for the same sample) will
// always return false.
func (p *OpenMetricsParser) Exemplar(e *exemplar.Exemplar) bool {
if len(p.exemplar) == 0 {
return false
}
// Allocate the full immutable string immediately, so we just
// have to create references on it below.
s := string(p.exemplar)
e.Value = p.exemplarVal
if p.hasExemplarTs {
e.HasTs = true
e.Ts = p.exemplarTs
}
p.builder.Reset()
for i := 0; i < len(p.eOffsets); i += 4 {
a := p.eOffsets[i] - p.start
b := p.eOffsets[i+1] - p.start
c := p.eOffsets[i+2] - p.start
d := p.eOffsets[i+3] - p.start
p.builder.Add(s[a:b], s[c:d])
}
p.builder.Sort()
e.Labels = p.builder.Labels()
// Wipe exemplar so that future calls return false.
p.exemplar = p.exemplar[:0]
return true
}
// CreatedTimestamp returns nil as it's not implemented yet.
// TODO(bwplotka): https://github.com/prometheus/prometheus/issues/12980
func (p *OpenMetricsParser) CreatedTimestamp() *int64 {
return nil
}
// nextToken returns the next token from the openMetricsLexer.
func (p *OpenMetricsParser) nextToken() token {
tok := p.l.Lex()
return tok
}
func (p *OpenMetricsParser) parseError(exp string, got token) error {
e := p.l.i + 1
if len(p.l.b) < e {
e = len(p.l.b)
}
return fmt.Errorf("%s, got %q (%q) while parsing: %q", exp, p.l.b[p.l.start:e], got, p.l.b[p.start:e])
}
// Next advances the parser to the next sample.
// It returns (EntryInvalid, io.EOF) if no samples were read.
func (p *OpenMetricsParser) Next() (Entry, error) {
var err error
p.start = p.l.i
p.offsets = p.offsets[:0]
p.eOffsets = p.eOffsets[:0]
p.exemplar = p.exemplar[:0]
p.exemplarVal = 0
p.hasExemplarTs = false
switch t := p.nextToken(); t {
case tEOFWord:
if t := p.nextToken(); t != tEOF {
return EntryInvalid, errors.New("unexpected data after # EOF")
}
return EntryInvalid, io.EOF
case tEOF:
return EntryInvalid, errors.New("data does not end with # EOF")
case tHelp, tType, tUnit:
switch t2 := p.nextToken(); t2 {
case tMName:
mStart := p.l.start
mEnd := p.l.i
if p.l.b[mStart] == '"' && p.l.b[mEnd-1] == '"' {
mStart++
mEnd--
}
p.offsets = append(p.offsets, mStart, mEnd)
default:
return EntryInvalid, p.parseError("expected metric name after "+t.String(), t2)
}
switch t2 := p.nextToken(); t2 {
case tText:
if len(p.l.buf()) > 1 {
p.text = p.l.buf()[1 : len(p.l.buf())-1]
} else {
p.text = []byte{}
}
default:
return EntryInvalid, fmt.Errorf("expected text in %s", t.String())
}
switch t {
case tType:
switch s := yoloString(p.text); s {
case "counter":
p.mtype = model.MetricTypeCounter
case "gauge":
p.mtype = model.MetricTypeGauge
case "histogram":
p.mtype = model.MetricTypeHistogram
case "gaugehistogram":
p.mtype = model.MetricTypeGaugeHistogram
case "summary":
p.mtype = model.MetricTypeSummary
case "info":
p.mtype = model.MetricTypeInfo
case "stateset":
p.mtype = model.MetricTypeStateset
case "unknown":
p.mtype = model.MetricTypeUnknown
default:
return EntryInvalid, fmt.Errorf("invalid metric type %q", s)
}
case tHelp:
if !utf8.Valid(p.text) {
return EntryInvalid, fmt.Errorf("help text %q is not a valid utf8 string", p.text)
}
}
switch t {
case tHelp:
return EntryHelp, nil
case tType:
return EntryType, nil
case tUnit:
m := yoloString(p.l.b[p.offsets[0]:p.offsets[1]])
u := yoloString(p.text)
if len(u) > 0 {
if !strings.HasSuffix(m, u) || len(m) < len(u)+1 || p.l.b[p.offsets[1]-len(u)-1] != '_' {
return EntryInvalid, fmt.Errorf("unit %q not a suffix of metric %q", u, m)
}
}
return EntryUnit, nil
}
case tBraceOpen:
// We found a brace, so make room for the eventual metric name. If these
// values aren't updated, then the metric name was not set inside the
// braces and we can return an error.
if len(p.offsets) == 0 {
p.offsets = []int{-1, -1}
}
if p.offsets, err = p.parseLVals(p.offsets, false); err != nil {
return EntryInvalid, err
}
p.series = p.l.b[p.start:p.l.i]
return p.parseMetricSuffix(p.nextToken())
case tMName:
p.offsets = append(p.offsets, p.start, p.l.i)
p.series = p.l.b[p.start:p.l.i]
t2 := p.nextToken()
if t2 == tBraceOpen {
p.offsets, err = p.parseLVals(p.offsets, false)
if err != nil {
return EntryInvalid, err
}
p.series = p.l.b[p.start:p.l.i]
t2 = p.nextToken()
}
return p.parseMetricSuffix(t2)
default:
err = p.parseError("expected a valid start token", t)
}
return EntryInvalid, err
}
func (p *OpenMetricsParser) parseComment() error {
var err error
// Parse the labels.
p.eOffsets, err = p.parseLVals(p.eOffsets, true)
if err != nil {
return err
}
p.exemplar = p.l.b[p.start:p.l.i]
// Get the value.
p.exemplarVal, err = p.getFloatValue(p.nextToken(), "exemplar labels")
if err != nil {
return err
}
// Read the optional timestamp.
p.hasExemplarTs = false
switch t2 := p.nextToken(); t2 {
case tEOF:
return errors.New("data does not end with # EOF")
case tLinebreak:
break
case tTimestamp:
p.hasExemplarTs = true
var ts float64
// A float is enough to hold what we need for millisecond resolution.
if ts, err = parseFloat(yoloString(p.l.buf()[1:])); err != nil {
return fmt.Errorf("%w while parsing: %q", err, p.l.b[p.start:p.l.i])
}
if math.IsNaN(ts) || math.IsInf(ts, 0) {
return fmt.Errorf("invalid exemplar timestamp %f", ts)
}
p.exemplarTs = int64(ts * 1000)
switch t3 := p.nextToken(); t3 {
case tLinebreak:
default:
return p.parseError("expected next entry after exemplar timestamp", t3)
}
default:
return p.parseError("expected timestamp or comment", t2)
}
return nil
}
func (p *OpenMetricsParser) parseLVals(offsets []int, isExemplar bool) ([]int, error) {
t := p.nextToken()
for {
curTStart := p.l.start
curTI := p.l.i
switch t {
case tBraceClose:
return offsets, nil
case tLName:
case tQString:
default:
return nil, p.parseError("expected label name", t)
}
t = p.nextToken()
// A quoted string followed by a comma or brace is a metric name. Set the
// offsets and continue processing. If this is an exemplar, this format
// is not allowed.
if t == tComma || t == tBraceClose {
if isExemplar {
return nil, p.parseError("expected label name", t)
}
if offsets[0] != -1 || offsets[1] != -1 {
return nil, fmt.Errorf("metric name already set while parsing: %q", p.l.b[p.start:p.l.i])
}
offsets[0] = curTStart + 1
offsets[1] = curTI - 1
if t == tBraceClose {
return offsets, nil
}
t = p.nextToken()
continue
}
// We have a label name, and it might be quoted.
if p.l.b[curTStart] == '"' {
curTStart++
curTI--
}
offsets = append(offsets, curTStart, curTI)
if t != tEqual {
return nil, p.parseError("expected equal", t)
}
if t := p.nextToken(); t != tLValue {
return nil, p.parseError("expected label value", t)
}
if !utf8.Valid(p.l.buf()) {
return nil, fmt.Errorf("invalid UTF-8 label value: %q", p.l.buf())
}
// The openMetricsLexer ensures the value string is quoted. Strip first
// and last character.
offsets = append(offsets, p.l.start+1, p.l.i-1)
// Free trailing commas are allowed.
t = p.nextToken()
if t == tComma {
t = p.nextToken()
} else if t != tBraceClose {
return nil, p.parseError("expected comma or brace close", t)
}
}
}
// parseMetricSuffix parses the end of the line after the metric name and
// labels. It starts parsing with the provided token.
func (p *OpenMetricsParser) parseMetricSuffix(t token) (Entry, error) {
if p.offsets[0] == -1 {
return EntryInvalid, fmt.Errorf("metric name not set while parsing: %q", p.l.b[p.start:p.l.i])
}
var err error
p.val, err = p.getFloatValue(t, "metric")
if err != nil {
return EntryInvalid, err
}
p.hasTS = false
switch t2 := p.nextToken(); t2 {
case tEOF:
return EntryInvalid, errors.New("data does not end with # EOF")
case tLinebreak:
break
case tComment:
if err := p.parseComment(); err != nil {
return EntryInvalid, err
}
case tTimestamp:
p.hasTS = true
var ts float64
// A float is enough to hold what we need for millisecond resolution.
if ts, err = parseFloat(yoloString(p.l.buf()[1:])); err != nil {
return EntryInvalid, fmt.Errorf("%w while parsing: %q", err, p.l.b[p.start:p.l.i])
}
if math.IsNaN(ts) || math.IsInf(ts, 0) {
return EntryInvalid, fmt.Errorf("invalid timestamp %f", ts)
}
p.ts = int64(ts * 1000)
switch t3 := p.nextToken(); t3 {
case tLinebreak:
case tComment:
if err := p.parseComment(); err != nil {
return EntryInvalid, err
}
default:
return EntryInvalid, p.parseError("expected next entry after timestamp", t3)
}
}
return EntrySeries, nil
}
func (p *OpenMetricsParser) getFloatValue(t token, after string) (float64, error) {
if t != tValue {
return 0, p.parseError(fmt.Sprintf("expected value after %v", after), t)
}
val, err := parseFloat(yoloString(p.l.buf()[1:]))
if err != nil {
return 0, fmt.Errorf("%w while parsing: %q", err, p.l.b[p.start:p.l.i])
}
// Ensure canonical NaN value.
if math.IsNaN(p.exemplarVal) {
val = math.Float64frombits(value.NormalNaN)
}
return val, nil
}