Alerting: Clear alerting rule evaluation errors after intermittent failures (#42386)

* Alerting: Clear alerting rule evaluation errors after intermittent failures When an alert transitioned in a way that `alerting -> error -> (alerting|nodata)`, the error provided by the `error` state would never be cleared thus the API and UI would show the health as an error.
4 years ago · dd5a2e5128
parent 725dbf8d95
commit dd5a2e5128
2 changed files with 188 additions and 1 deletions
--- a/pkg/services/ngalert/state/manager_test.go
+++ b/pkg/services/ngalert/state/manager_test.go
@ -3,6 +3,7 @@ package state_test
 import (
 	"context"
 	"errors"
+	"fmt"
 	"testing"
 	"time"

@ -1254,6 +1255,187 @@ func TestProcessEvalResults(t *testing.T) {
 				},
 			},
 		},
+		{
+			desc: "normal -> alerting -> error -> alerting - it should clear the error",
+			alertRule: &models.AlertRule{
+				OrgID:           1,
+				Title:           "test_title",
+				UID:             "test_alert_rule_uid_2",
+				NamespaceUID:    "test_namespace_uid",
+				Annotations:     map[string]string{"annotation": "test"},
+				Labels:          map[string]string{"label": "test"},
+				IntervalSeconds: 10,
+				For:             30 * time.Second,
+			},
+			evalResults: []eval.Results{
+				{
+					eval.Result{
+						Instance:           data.Labels{"instance_label": "test"},
+						State:              eval.Normal,
+						EvaluatedAt:        evaluationTime,
+						EvaluationDuration: evaluationDuration,
+					},
+				},
+				{
+					eval.Result{
+						Instance:           data.Labels{"instance_label": "test"},
+						State:              eval.Alerting,
+						EvaluatedAt:        evaluationTime.Add(30 * time.Second),
+						EvaluationDuration: evaluationDuration,
+					},
+				},
+				{
+					eval.Result{
+						Instance:           data.Labels{"instance_label": "test"},
+						State:              eval.Error,
+						Error:              fmt.Errorf("Failed to query data"),
+						EvaluatedAt:        evaluationTime.Add(40 * time.Second),
+						EvaluationDuration: evaluationDuration,
+					},
+				},
+				{
+					eval.Result{
+						Instance:           data.Labels{"instance_label": "test"},
+						State:              eval.Alerting,
+						EvaluatedAt:        evaluationTime.Add(70 * time.Second),
+						EvaluationDuration: evaluationDuration,
+					},
+				},
+			},
+			expectedStates: map[string]*state.State{
+				`[["__alert_rule_namespace_uid__","test_namespace_uid"],["__alert_rule_uid__","test_alert_rule_uid_2"],["alertname","test_title"],["instance_label","test"],["label","test"]]`: {
+					AlertRuleUID: "test_alert_rule_uid_2",
+					OrgID:        1,
+					CacheId:      `[["__alert_rule_namespace_uid__","test_namespace_uid"],["__alert_rule_uid__","test_alert_rule_uid_2"],["alertname","test_title"],["instance_label","test"],["label","test"]]`,
+					Labels: data.Labels{
+						"__alert_rule_namespace_uid__": "test_namespace_uid",
+						"__alert_rule_uid__":           "test_alert_rule_uid_2",
+						"alertname":                    "test_title",
+						"label":                        "test",
+						"instance_label":               "test",
+					},
+					State: eval.Alerting,
+					Results: []state.Evaluation{
+						{
+							EvaluationTime:  evaluationTime,
+							EvaluationState: eval.Normal,
+							Values:          make(map[string]state.EvaluationValue),
+						},
+						{
+							EvaluationTime:  evaluationTime.Add(30 * time.Second),
+							EvaluationState: eval.Alerting,
+							Values:          make(map[string]state.EvaluationValue),
+						},
+						{
+							EvaluationTime:  evaluationTime.Add(40 * time.Second),
+							EvaluationState: eval.Error,
+							Values:          make(map[string]state.EvaluationValue),
+						},
+						{
+							EvaluationTime:  evaluationTime.Add(70 * time.Second),
+							EvaluationState: eval.Alerting,
+							Values:          make(map[string]state.EvaluationValue),
+						},
+					},
+					StartsAt:           evaluationTime.Add(70 * time.Second),
+					EndsAt:             evaluationTime.Add(70 * time.Second).Add(state.ResendDelay * 3),
+					LastEvaluationTime: evaluationTime.Add(70 * time.Second),
+					EvaluationDuration: evaluationDuration,
+					Annotations:        map[string]string{"annotation": "test"},
+				},
+			},
+		},
+		{
+			desc: "normal -> alerting -> error -> no data - it should clear the error",
+			alertRule: &models.AlertRule{
+				OrgID:           1,
+				Title:           "test_title",
+				UID:             "test_alert_rule_uid_2",
+				NamespaceUID:    "test_namespace_uid",
+				Annotations:     map[string]string{"annotation": "test"},
+				Labels:          map[string]string{"label": "test"},
+				IntervalSeconds: 10,
+				For:             30 * time.Second,
+				NoDataState:     models.NoData,
+			},
+			evalResults: []eval.Results{
+				{
+					eval.Result{
+						Instance:           data.Labels{"instance_label": "test"},
+						State:              eval.Normal,
+						EvaluatedAt:        evaluationTime,
+						EvaluationDuration: evaluationDuration,
+					},
+				},
+				{
+					eval.Result{
+						Instance:           data.Labels{"instance_label": "test"},
+						State:              eval.Alerting,
+						EvaluatedAt:        evaluationTime.Add(30 * time.Second),
+						EvaluationDuration: evaluationDuration,
+					},
+				},
+				{
+					eval.Result{
+						Instance:           data.Labels{"instance_label": "test"},
+						State:              eval.Error,
+						Error:              fmt.Errorf("Failed to query data"),
+						EvaluatedAt:        evaluationTime.Add(40 * time.Second),
+						EvaluationDuration: evaluationDuration,
+					},
+				},
+				{
+					eval.Result{
+						Instance:           data.Labels{"instance_label": "test"},
+						State:              eval.NoData,
+						EvaluatedAt:        evaluationTime.Add(50 * time.Second),
+						EvaluationDuration: evaluationDuration,
+					},
+				},
+			},
+			expectedStates: map[string]*state.State{
+				`[["__alert_rule_namespace_uid__","test_namespace_uid"],["__alert_rule_uid__","test_alert_rule_uid_2"],["alertname","test_title"],["instance_label","test"],["label","test"]]`: {
+					AlertRuleUID: "test_alert_rule_uid_2",
+					OrgID:        1,
+					CacheId:      `[["__alert_rule_namespace_uid__","test_namespace_uid"],["__alert_rule_uid__","test_alert_rule_uid_2"],["alertname","test_title"],["instance_label","test"],["label","test"]]`,
+					Labels: data.Labels{
+						"__alert_rule_namespace_uid__": "test_namespace_uid",
+						"__alert_rule_uid__":           "test_alert_rule_uid_2",
+						"alertname":                    "test_title",
+						"label":                        "test",
+						"instance_label":               "test",
+					},
+					State: eval.NoData,
+					Results: []state.Evaluation{
+						{
+							EvaluationTime:  evaluationTime,
+							EvaluationState: eval.Normal,
+							Values:          make(map[string]state.EvaluationValue),
+						},
+						{
+							EvaluationTime:  evaluationTime.Add(30 * time.Second),
+							EvaluationState: eval.Alerting,
+							Values:          make(map[string]state.EvaluationValue),
+						},
+						{
+							EvaluationTime:  evaluationTime.Add(40 * time.Second),
+							EvaluationState: eval.Error,
+							Values:          make(map[string]state.EvaluationValue),
+						},
+						{
+							EvaluationTime:  evaluationTime.Add(50 * time.Second),
+							EvaluationState: eval.NoData,
+							Values:          make(map[string]state.EvaluationValue),
+						},
+					},
+					StartsAt:           evaluationTime.Add(30 * time.Second),
+					EndsAt:             evaluationTime.Add(50 * time.Second).Add(state.ResendDelay * 3),
+					LastEvaluationTime: evaluationTime.Add(50 * time.Second),
+					EvaluationDuration: evaluationDuration,
+					Annotations:        map[string]string{"annotation": "test"},
+				},
+			},
+		},
 		{
 			desc: "template is correctly expanded",
 			alertRule: &models.AlertRule{
--- a/pkg/services/ngalert/state/state.go
+++ b/pkg/services/ngalert/state/state.go
@ -57,15 +57,18 @@ func NewEvaluationValues(m map[string]eval.NumberValueCapture) map[string]Evalua
 }

 func (a *State) resultNormal(alertRule *ngModels.AlertRule, result eval.Result) {
+	a.Error = result.Error // should be nil since state is not error
+
 	if a.State != eval.Normal {
 		a.EndsAt = result.EvaluatedAt
 		a.StartsAt = result.EvaluatedAt
 	}
-	a.Error = result.Error // should be nil since state is not error
 	a.State = eval.Normal
 }

 func (a *State) resultAlerting(alertRule *ngModels.AlertRule, result eval.Result) {
+	a.Error = result.Error // should be nil since the state is not an error
+
 	switch a.State {
 	case eval.Alerting:
 		a.setEndsAt(alertRule, result)
@ -118,6 +121,8 @@ func (a *State) resultError(alertRule *ngModels.AlertRule, result eval.Result) {
 }

 func (a *State) resultNoData(alertRule *ngModels.AlertRule, result eval.Result) {
+	a.Error = result.Error
+
 	if a.StartsAt.IsZero() {
 		a.StartsAt = result.EvaluatedAt
 	}