Merge pull request #129 from cloudflare/speedup-fingerprints

Speed up alert fingerprint generation

Makefile

@@ -88,7 +88,7 @@ endif
 
 .PHONY: test
 test: lint bindata_assetfs.go
-	go test -cover `go list ./... | grep -v /vendor/`
+	go test -bench=. -cover `go list ./... | grep -v /vendor/`
 
 .build/vendor.ok:
 	go get -u github.com/kardianos/govendor

alertmanager/models.go

@@ -168,6 +168,8 @@ func (am *Alertmanager) pullAlerts(version string) error {
 			for k, v := range alert.Labels {
 				transform.ColorLabel(colors, k, v)
 			}
+
+			alert.UpdateFingerprints()
 			alerts = append(alerts, alert)
 
 			// update internal metrics

models/alert.go

@@ -45,30 +45,41 @@ type Alert struct {
 	Alertmanager []AlertmanagerInstance `json:"alertmanager"`
 	Receiver     string                 `json:"receiver"`
 	Links        map[string]string      `json:"links"`
+	// fingerprints are precomputed for speed
+	labelsFP  string `hash:"-"`
+	contentFP string `hash:"-"`
+}
+
+// UpdateFingerprints will generate a new set of fingerprints for this alert
+// it should be called after modifying any field that isn't tagged with hash:"-"
+func (a *Alert) UpdateFingerprints() {
+	a.labelsFP = fmt.Sprintf("%x", structhash.Sha1(a.Labels, 1))
+	a.contentFP = fmt.Sprintf("%x", structhash.Sha1(a, 1))
 }
 
 // LabelsFingerprint is a checksum computed only from labels which should be
 // unique for every alert
-func (a Alert) LabelsFingerprint() string {
-	return fmt.Sprintf("%x", structhash.Sha1(a.Labels, 1))
+func (a *Alert) LabelsFingerprint() string {
+	return a.labelsFP
 }
 
 // ContentFingerprint is a checksum computed from entire alert object
-func (a Alert) ContentFingerprint() string {
-	return fmt.Sprintf("%x", structhash.Sha1(a, 1))
+// except some blacklisted fields tagged with hash:"-"
+func (a *Alert) ContentFingerprint() string {
+	return a.contentFP
 }
 
 // IsSilenced will return true if alert should be considered silenced
-func (a Alert) IsSilenced() bool {
+func (a *Alert) IsSilenced() bool {
 	return (a.State == AlertStateSuppressed && len(a.SilencedBy) > 0)
 }
 
 // IsInhibited will return true if alert should be considered silenced
-func (a Alert) IsInhibited() bool {
+func (a *Alert) IsInhibited() bool {
 	return (a.State == AlertStateSuppressed && len(a.InhibitedBy) > 0)
 }
 
 // IsActive will return true if alert is not suppressed in any way
-func (a Alert) IsActive() bool {
+func (a *Alert) IsActive() bool {
 	return (a.State == AlertStateActive)
 }

models/alert_test.go

@@ -2,6 +2,7 @@ package models_test
 
 import (
 	"testing"
+	"time"
 
 	"github.com/cloudflare/unsee/models"
 )
@@ -52,3 +53,49 @@ func TestAlertState(t *testing.T) {
 		}
 	}
 }
+
+func BenchmarkLabelsFingerprint(b *testing.B) {
+	alert := models.Alert{
+		Labels: map[string]string{
+			"foo1":        "bar1",
+			"foo1bar1":    "545jjjssd",
+			"foo1xxxx":    "bdjjs88ff",
+			"agdfdfd":     "bar1",
+			"fossdsf3o1":  "bar11111",
+			"fdfdgfdgoo1": "bar1",
+		},
+	}
+	for n := 0; n < b.N; n++ {
+		alert.LabelsFingerprint()
+	}
+}
+
+func BenchmarkLabelsContent(b *testing.B) {
+	alert := models.Alert{
+		Annotations: map[string]string{
+			"foo": "bar",
+			"abc": "Neque porro quisquam est qui dolorem ipsum quia dolor sit amet, consectetur, adipisci velit...",
+		},
+		Labels: map[string]string{
+			"foo1":        "bar1",
+			"foo1bar1":    "545jjjssd",
+			"foo1xxxx":    "bdjjs88ff",
+			"agdfdfd":     "bar1",
+			"fossdsf3o1":  "bar11111",
+			"fdfdgfdgoo1": "bar1",
+		},
+		State:    models.AlertStateActive,
+		StartsAt: time.Date(2015, time.March, 10, 0, 0, 0, 0, time.UTC),
+		Alertmanager: []models.AlertmanagerInstance{
+			models.AlertmanagerInstance{
+				Name:  "default",
+				URI:   "http://localhost",
+				State: models.AlertStateActive,
+			},
+		},
+	}
+	alert.UpdateFingerprints()
+	for n := 0; n < b.N; n++ {
+		alert.LabelsFingerprint()
+	}
+}

models/alertgroup_test.go

@@ -47,6 +47,7 @@ var alertListSortTests = []alertListSortTest{
 func TestUnseeAlertListSort(t *testing.T) {
 	al := models.AlertList{}
 	for _, testCase := range alertListSortTests {
+		testCase.alert.UpdateFingerprints()
 		al = append(al, testCase.alert)
 	}
 
@@ -56,6 +57,7 @@ func TestUnseeAlertListSort(t *testing.T) {
 	for i := 1; i <= iterations; i++ {
 		sort.Sort(al)
 		for _, testCase := range alertListSortTests {
+			testCase.alert.UpdateFingerprints()
 			if al[testCase.position].ContentFingerprint() != testCase.alert.ContentFingerprint() {
 				failures++
 			}
@@ -120,6 +122,13 @@ var agFPTests = []agFPTest{
 
 func TestAlertGroupContentFingerprint(t *testing.T) {
 	for _, testCase := range agFPTests {
+		alerts := models.AlertList{}
+		for _, alert := range testCase.ag.Alerts {
+			alert.UpdateFingerprints()
+			alerts = append(alerts, alert)
+		}
+		sort.Sort(alerts)
+		testCase.ag.Alerts = alerts
 		if testCase.ag.ContentFingerprint() != testCase.fingerprint {
 			t.Errorf("Invalid AlertGroup ContentFingerprint(), expected '%s', got '%s', AlertGroup: %v",
 				testCase.fingerprint, testCase.ag.ContentFingerprint(), testCase.ag)

views.go

@@ -133,6 +133,11 @@ func alerts(c *gin.Context) {
 			}
 			if !validFilters || (slices.BoolInSlice(results, true) && !slices.BoolInSlice(results, false)) {
 				matches++
+				// we need to update fingerprints since we've modified some fields in dedup
+				// and agCopy.ContentFingerprint() depends on per alert fingerprint
+				// we update it here rather than in dedup since here we can apply it
+				// only for alerts left after filtering
+				alert.UpdateFingerprints()
 				agCopy.Alerts = append(agCopy.Alerts, alert)
 
 				countLabel(counters, "@state", alert.State)