Merge branch 'improvesetmatches' into regexpopti

This commit is contained in:
Cyril Tovena 2021-10-06 16:36:29 +02:00
commit a1f6c747ed
No known key found for this signature in database
GPG key ID: FD8F768F9D633FB6
4 changed files with 43 additions and 22 deletions

View file

@ -113,6 +113,7 @@ func (m *Matcher) Inverse() (*Matcher, error) {
// SetMatches returns a set of equality matchers for the current regex matchers if possible.
// For examples the regexp `a(b|f)` will returns "ab" and "af".
// Returns nil if we can't replace the regexp by only equality matchers.
func (m *Matcher) SetMatches() []string {
if m.re == nil {
return nil

View file

@ -56,7 +56,7 @@ func NewFastRegexMatcher(v string) (*FastRegexMatcher, error) {
// findSetMatches extract equality matches from a regexp.
// Returns nil if we can't replace the regexp by only equality matchers.
func findSetMatches(re *syntax.Regexp, base string) []string {
// Matches are not case sensitive, if we find a case insensitive regexp.
// Matches are case sensitive, if we find a case insensitive regexp.
// We have to abort.
if isCaseInsensitive(re) {
return nil
@ -69,25 +69,19 @@ func findSetMatches(re *syntax.Regexp, base string) []string {
return []string{base}
}
case syntax.OpAlternate:
found := findSetMatchesFromAlternate(re, base)
if found != nil {
return found
}
return findSetMatchesFromAlternate(re, base)
case syntax.OpCapture:
clearCapture(re)
return findSetMatches(re, base)
case syntax.OpConcat:
found := findSetMatchesFromConcat(re, base)
if found != nil {
return found
}
return findSetMatchesFromConcat(re, base)
case syntax.OpCharClass:
if len(re.Rune) == 1 {
return []string{base + string(re.Rune)}
if len(re.Rune)%2 != 0 {
return nil
}
var matches []string
var totalSet int
for i := 0; i < len(re.Rune); i = i + 2 {
for i := 0; i+1 < len(re.Rune); i = i + 2 {
totalSet += int(re.Rune[i+1]-re.Rune[i]) + 1
}
// limits the total characters that can be used to create matches.
@ -96,7 +90,7 @@ func findSetMatches(re *syntax.Regexp, base string) []string {
if totalSet > maxSetMatches {
return nil
}
for i := 0; i < len(re.Rune); i = i + 2 {
for i := 0; i+1 < len(re.Rune); i = i + 2 {
lo, hi := re.Rune[i], re.Rune[i+1]
for c := lo; c <= hi; c++ {
matches = append(matches, base+string(c))
@ -111,19 +105,14 @@ func findSetMatches(re *syntax.Regexp, base string) []string {
}
func findSetMatchesFromConcat(re *syntax.Regexp, base string) []string {
if isCaseInsensitive(re) {
return nil
}
if len(re.Sub) == 0 {
return nil
}
clearBeginEndText(re)
clearCapture(re.Sub...)
matches := findSetMatches(re.Sub[0], base)
if matches == nil {
return nil
}
matches := []string{base}
for i := 1; i < len(re.Sub); i++ {
for i := 0; i < len(re.Sub); i++ {
var newMatches []string
for _, b := range matches {
m := findSetMatches(re.Sub[i], b)
@ -165,6 +154,25 @@ func clearCapture(regs ...*syntax.Regexp) {
}
}
// clearBeginEndText removes the begin and end text from the regexp. Prometheus regexp are anchored to the beginning and end of the string.
func clearBeginEndText(re *syntax.Regexp) {
if len(re.Sub) == 0 {
return
}
if len(re.Sub) == 1 {
if re.Sub[0].Op == syntax.OpBeginText || re.Sub[0].Op == syntax.OpEndText {
re.Sub = nil
return
}
}
if re.Sub[0].Op == syntax.OpBeginText {
re.Sub = re.Sub[1:]
}
if re.Sub[len(re.Sub)-1].Op == syntax.OpEndText {
re.Sub = re.Sub[:len(re.Sub)-1]
}
}
// isCaseInsensitive tells if a regexp is case insensitive.
// The flag should be check at each level of the syntax tree.
func isCaseInsensitive(reg *syntax.Regexp) bool {

View file

@ -122,6 +122,8 @@ func TestFindSetMatches(t *testing.T) {
}{
// Single value, coming from a `bar=~"foo"` selector.
{"foo", []string{"foo"}},
{"^foo", []string{"foo"}},
{"^foo$", []string{"foo"}},
// Simple sets alternates.
{"foo|bar|zz", []string{"foo", "bar", "zz"}},
// Simple sets alternate and concat (bar|baz is parsed as ba(r|z)).
@ -131,6 +133,8 @@ func TestFindSetMatches(t *testing.T) {
// Simple sets alternate and concat and alternates with empty matches
// parsed as b(ar|(?:)|uzz) where b(?:) means literal b.
{"bar|b|buzz", []string{"bar", "b", "buzz"}},
// Skip anchors it's enforced anyway at the root.
{"(^bar$)|(b$)|(^buzz)", []string{"bar", "b", "buzz"}},
// Simple sets containing escaped characters.
{"fo\\.o|bar\\?|\\^baz", []string{"fo.o", "bar?", "^baz"}},
// using charclass
@ -142,6 +146,9 @@ func TestFindSetMatches(t *testing.T) {
// triple concat with multiple alternates
{"(api|rpc)_(v1|prom)_push", []string{"api_v1_push", "api_prom_push", "rpc_v1_push", "rpc_prom_push"}},
{"(api|rpc)_(v1|prom)_(push|query)", []string{"api_v1_push", "api_v1_query", "api_prom_push", "api_prom_query", "rpc_v1_push", "rpc_v1_query", "rpc_prom_push", "rpc_prom_query"}},
// class starting with "-"
{"[-1-2][a-c]", []string{"-a", "-b", "-c", "1a", "1b", "1c", "2a", "2b", "2c"}},
{"[1^3]", []string{"1", "3", "^"}},
// OpPlus with concat
{"(.+)/(foo|bar)", nil},
// Simple sets containing special characters without escaping.

View file

@ -108,7 +108,9 @@ func benchmarkPostingsForMatchers(b *testing.B, ir IndexReader) {
iNot2 := labels.MustNewMatcher(labels.MatchNotEqual, "n", "2"+postingsBenchSuffix)
iNot2Star := labels.MustNewMatcher(labels.MatchNotRegexp, "i", "^2.*$")
iNotStar2Star := labels.MustNewMatcher(labels.MatchNotRegexp, "i", "^.*2.*$")
jFooBar := labels.MustNewMatcher(labels.MatchRegexp, "j", "foo|bar")
iCharSet := labels.MustNewMatcher(labels.MatchRegexp, "i", "1[0-9]")
iAlternate := labels.MustNewMatcher(labels.MatchRegexp, "i", "(1|2|3|4|5|6|20|55)")
cases := []struct {
name string
matchers []*labels.Matcher
@ -117,6 +119,9 @@ func benchmarkPostingsForMatchers(b *testing.B, ir IndexReader) {
{`n="1",j="foo"`, []*labels.Matcher{n1, jFoo}},
{`j="foo",n="1"`, []*labels.Matcher{jFoo, n1}},
{`n="1",j!="foo"`, []*labels.Matcher{n1, jNotFoo}},
{`i=~"1[0-9]",j=~"foo|bar"`, []*labels.Matcher{iCharSet, jFooBar}},
{`j=~"foo|bar"`, []*labels.Matcher{jFooBar}},
{`i=~"(1|2|3|4|5|6|20|55)"`, []*labels.Matcher{iAlternate}},
{`i=~".*"`, []*labels.Matcher{iStar}},
{`i=~"1.*"`, []*labels.Matcher{i1Star}},
{`i=~".*1"`, []*labels.Matcher{iStar1}},