add a negative cache to regexp decoder

Often, regexp decoder can run repeatedly on the same input and skip them
with regexp filter. An common example is matching cgroup path in a chain like so:

```
- name: cgroup
- name: regexp
  regexps:
  - ^.*(system.slice).*$
```

Anything that is not in system.slice will be skipped. In some cases
where we filter only a small subset of inputs, the overhead of regexp matching
can often be noticable.

We add a cache here to test for input that would produce ErrSkipLabelSet and
skip regex matching on them to reduce the work done on regexp matching.
The cache size is customizable with `skip_cache_size` configuration.

Signed-off-by: Daniel Dao <[email protected]>

Author: dqminh

.vscode/config-schema.yaml

@@ -145,7 +145,10 @@ definitions:
                 type: object
               allow_unknown:
                 type: boolean
+              skip_cache_size:
+                type: number
               regexps:
                 type: array
                 items:
                   type: string
+

config/config.go

@@ -73,6 +73,8 @@ type Decoder struct {
 	StaticMap    map[string]string `yaml:"static_map"`
 	Regexps      []string          `yaml:"regexps"`
 	AllowUnknown bool              `yaml:"allow_unknown"`
+	// SkipCacheSize is size of a cache that stores label value matches ErrSkipLabelSet
+	SkipCacheSize uint `yaml:"skip_cache_size"`
 }
 
 // HistogramBucketType is an enum to define how to interpret histogram

decoder/regexp.go

@@ -6,22 +6,32 @@ import (
 	"regexp"
 
 	"github.com/cloudflare/ebpf_exporter/v2/config"
+	lru "github.com/hashicorp/golang-lru/v2"
 )
 
 // Regexp is a decoder that only allows inputs matching regexp
 type Regexp struct {
-	cache map[string]*regexp.Regexp
+	cache     map[string]*regexp.Regexp
+	skipCache *lru.Cache[string, struct{}]
 }
 
 // Decode only allows inputs matching regexp
 func (r *Regexp) Decode(in []byte, conf config.Decoder) ([]byte, error) {
 	if conf.Regexps == nil {
 		return nil, errors.New("no regexps defined in config")
 	}
+	inputStr := string(in)
 
 	if r.cache == nil {
 		r.cache = map[string]*regexp.Regexp{}
 	}
+	if conf.SkipCacheSize > 0 && r.skipCache == nil {
+		skipCache, err := lru.New[string, struct{}](int(conf.SkipCacheSize))
+		if err != nil {
+			return nil, err
+		}
+		r.skipCache = skipCache
+	}
 
 	for _, expr := range conf.Regexps {
 		if _, ok := r.cache[expr]; !ok {
@@ -33,6 +43,12 @@ func (r *Regexp) Decode(in []byte, conf config.Decoder) ([]byte, error) {
 			r.cache[expr] = compiled
 		}
 
+		if r.skipCache != nil {
+			if _, ok := r.skipCache.Get(inputStr); ok {
+				return nil, ErrSkipLabelSet
+			}
+		}
+
 		matches := r.cache[expr].FindSubmatch(in)
 
 		// First sub-match if present
@@ -46,5 +62,9 @@ func (r *Regexp) Decode(in []byte, conf config.Decoder) ([]byte, error) {
 		}
 	}
 
+	if r.skipCache != nil {
+		r.skipCache.Add(inputStr, struct{}{})
+	}
+
 	return nil, ErrSkipLabelSet
 }

decoder/regexp_test.go

@@ -54,3 +54,15 @@ func TestRegexpDecoder(t *testing.T) {
 		}
 	}
 }
+
+func TestRegexpDecoderWithSkipCache(t *testing.T) {
+	d := &Regexp{}
+	input := []byte("whatever")
+	_, err := d.Decode(input, config.Decoder{Regexps: []string{"^(systemd).*$", "^syslog-ng$"}, SkipCacheSize: 100})
+	if !errors.Is(err, ErrSkipLabelSet) {
+		t.Errorf("Error decoding %s: %v", input, err)
+	}
+	if !d.skipCache.Contains("whatever") {
+		t.Errorf("failed to add to skipcache %s: kets=%v", input, d.skipCache.Keys())
+	}
+}

examples/regexp.yaml

@@ -8,6 +8,7 @@ metrics:
           decoders:
             - name: string
             - name: regexp
+              skip_cache_size: 100
               regexps:
                 - ^(kswapd).*$ # anything matching kswapd prefix, mapping to one kswapd bucket
                 - ^(systemd.*)$ # anything matching systemd prefix, mapping to one systemd bucket

go.mod

@@ -7,6 +7,7 @@ require (
 	github.com/aquasecurity/libbpfgo v0.8.0-libbpf-1.5
 	github.com/coreos/go-systemd v0.0.0-20191104093116-d3cd4ed1dbcf
 	github.com/elastic/go-perf v0.0.0-20191212140718-9c656876f595
+	github.com/hashicorp/golang-lru/v2 v2.0.7
 	github.com/iovisor/gobpf v0.2.0
 	github.com/jaypipes/pcidb v1.0.1
 	github.com/mdlayher/sdnotify v1.0.0

go.sum

@@ -30,6 +30,8 @@ github.com/google/uuid v1.6.0 h1:NIvaJDMOsjHA8n1jAhLSgzrAzy1Hgr+hNrb57e+94F0=
 github.com/google/uuid v1.6.0/go.mod h1:TIyPZe4MgqvfeYDBFedMoGGpEw/LqOeaOT+nhxU+yHo=
 github.com/grpc-ecosystem/grpc-gateway/v2 v2.26.1 h1:e9Rjr40Z98/clHv5Yg79Is0NtosR5LXRvdr7o/6NwbA=
 github.com/grpc-ecosystem/grpc-gateway/v2 v2.26.1/go.mod h1:tIxuGz/9mpox++sgp9fJjHO0+q1X9/UOWd798aAm22M=
+github.com/hashicorp/golang-lru/v2 v2.0.7 h1:a+bsQ5rvGLjzHuww6tVxozPZFVghXaHOwFs4luLUK2k=
+github.com/hashicorp/golang-lru/v2 v2.0.7/go.mod h1:QeFd9opnmA6QUJc5vARoKUSoFhyfM2/ZepoAG6RGpeM=
 github.com/iovisor/gobpf v0.2.0 h1:34xkQxft+35GagXBk3n23eqhm0v7q0ejeVirb8sqEOQ=
 github.com/iovisor/gobpf v0.2.0/go.mod h1:WSY9Jj5RhdgC3ci1QaacvbFdQ8cbrEjrpiZbLHLt2s4=
 github.com/jaypipes/pcidb v1.0.1 h1:WB2zh27T3nwg8AE8ei81sNRb9yWBii3JGNJtT7K9Oic=