Merge pull request #354 from petrkotas/add-interface

Add intesigation interface

Author: openshift-merge-bot[bot]

cadctl/cmd/investigate/investigate.go

@@ -22,9 +22,9 @@ import (
 	"path/filepath"
 
 	cmv1 "github.com/openshift-online/ocm-sdk-go/clustersmgmt/v1"
-	investigation "github.com/openshift/configuration-anomaly-detection/pkg/investigations"
+	investigations "github.com/openshift/configuration-anomaly-detection/pkg/investigations"
 	"github.com/openshift/configuration-anomaly-detection/pkg/investigations/ccam"
-	investigation_mapping "github.com/openshift/configuration-anomaly-detection/pkg/investigations/mapping"
+	investigation "github.com/openshift/configuration-anomaly-detection/pkg/investigations/investigation"
 	"github.com/openshift/configuration-anomaly-detection/pkg/logging"
 	"github.com/openshift/configuration-anomaly-detection/pkg/managedcloud"
 	"github.com/openshift/configuration-anomaly-detection/pkg/metrics"
@@ -71,7 +71,8 @@ func run(_ *cobra.Command, _ []string) error {
 
 	logging.Infof("Incident link: %s", pdClient.GetIncidentRef())
 
-	alertInvestigation := investigation_mapping.GetInvestigation(pdClient.GetTitle())
+	_, cadExperimentalEnabled := os.LookupEnv("CAD_EXPERIMENTAL_ENABLED")
+	alertInvestigation := investigations.GetInvestigation(pdClient.GetTitle(), cadExperimentalEnabled)
 
 	// Escalate all unsupported alerts
 	if alertInvestigation == nil {
@@ -82,7 +83,7 @@ func run(_ *cobra.Command, _ []string) error {
 		return nil
 	}
 
-	metrics.Inc(metrics.Alerts, alertInvestigation.Name)
+	metrics.Inc(metrics.Alerts, alertInvestigation.Name())
 
 	// clusterID can end up being either be the internal or external ID.
 	// We don't really care, as we only use this to initialize the cluster object,
@@ -121,16 +122,17 @@ func run(_ *cobra.Command, _ []string) error {
 	customerAwsClient, err := managedcloud.CreateCustomerAWSClient(cluster, ocmClient)
 	if err != nil {
 		ccamResources := &investigation.Resources{Name: "ccam", Cluster: cluster, ClusterDeployment: clusterDeployment, AwsClient: customerAwsClient, OcmClient: ocmClient, PdClient: pdClient, AdditionalResources: map[string]interface{}{"error": err}}
-		result, err := ccam.Investigate(ccamResources)
-		updateMetrics(alertInvestigation.Name, &result)
+		inv := ccam.CCAM{}
+		result, err := inv.Run(ccamResources)
+		updateMetrics(alertInvestigation.Name(), &result)
 		return err
 	}
 
-	investigationResources := &investigation.Resources{Name: alertInvestigation.Name, Cluster: cluster, ClusterDeployment: clusterDeployment, AwsClient: customerAwsClient, OcmClient: ocmClient, PdClient: pdClient}
+	investigationResources := &investigation.Resources{Name: alertInvestigation.Name(), Cluster: cluster, ClusterDeployment: clusterDeployment, AwsClient: customerAwsClient, OcmClient: ocmClient, PdClient: pdClient}
 
 	logging.Infof("Starting investigation for %s", alertInvestigation.Name)
 	result, err := alertInvestigation.Run(investigationResources)
-	updateMetrics(alertInvestigation.Name, &result)
+	updateMetrics(alertInvestigation.Name(), &result)
 	return err
 }
 

interceptor/pkg/interceptor/pdinterceptor.go

@@ -8,9 +8,10 @@ import (
 	"fmt"
 	"io"
 	"net/http"
+	"os"
 	"time"
 
-	investigation_mapping "github.com/openshift/configuration-anomaly-detection/pkg/investigations/mapping"
+	investigations "github.com/openshift/configuration-anomaly-detection/pkg/investigations"
 	"github.com/openshift/configuration-anomaly-detection/pkg/pagerduty"
 	triggersv1 "github.com/tektoncd/triggers/pkg/apis/triggers/v1beta1"
 	"github.com/tektoncd/triggers/pkg/interceptors"
@@ -106,7 +107,8 @@ func (pdi *PagerDutyInterceptor) Process(ctx context.Context, r *triggersv1.Inte
 		return interceptors.Failf(codes.InvalidArgument, "could not initialize pagerduty client: %v", err)
 	}
 
-	investigation := investigation_mapping.GetInvestigation(pdClient.GetTitle())
+	_, cadExperimentalEnabled := os.LookupEnv("CAD_EXPERIMENTAL_ENABLED")
+	investigation := investigations.GetInvestigation(pdClient.GetTitle(), cadExperimentalEnabled)
 	// If the alert is not in the whitelist, return `Continue: false` as interceptor response
 	// and escalate the alert to SRE
 	if investigation == nil {

pkg/investigations/ccam/ccam.go

@@ -7,19 +7,21 @@ import (
 	"regexp"
 
 	cmv1 "github.com/openshift-online/ocm-sdk-go/clustersmgmt/v1"
-	investigation "github.com/openshift/configuration-anomaly-detection/pkg/investigations"
+	investigation "github.com/openshift/configuration-anomaly-detection/pkg/investigations/investigation"
 	"github.com/openshift/configuration-anomaly-detection/pkg/logging"
 	"github.com/openshift/configuration-anomaly-detection/pkg/ocm"
 )
 
+type CCAM struct{}
+
 var ccamLimitedSupport = &ocm.LimitedSupportReason{
 	Summary: "Restore missing cloud credentials",
 	Details: "Your cluster requires you to take action because Red Hat is not able to access the infrastructure with the provided credentials. Please restore the credentials and permissions provided during install",
 }
 
 // Evaluate estimates if the awsError is a cluster credentials are missing error. If it determines that it is,
 // the cluster is placed into limited support (if the cluster state allows it), otherwise an error is returned.
-func Investigate(r *investigation.Resources) (investigation.InvestigationResult, error) {
+func (c *CCAM) Run(r *investigation.Resources) (investigation.InvestigationResult, error) {
 	result := investigation.InvestigationResult{}
 	cluster := r.Cluster
 	ocmClient := r.OcmClient
@@ -62,6 +64,22 @@ func Investigate(r *investigation.Resources) (investigation.InvestigationResult,
 	}
 }
 
+func (c *CCAM) Name() string {
+	return "Cluster Credentials Are Missing (CCAM)"
+}
+
+func (c *CCAM) Description() string {
+	return "Detects missing cluster credentials"
+}
+
+func (c *CCAM) ShouldInvestigateAlert(alert string) bool {
+	return false
+}
+
+func (c *CCAM) IsExperimental() bool {
+	return false
+}
+
 // userCausedErrors contains the list of backplane returned error strings that we map to
 // customer modifications/role deletions.
 var userCausedErrors = []string{

pkg/investigations/ccam/ccam_test.go

@@ -4,7 +4,7 @@ import (
 	"errors"
 	"testing"
 
-	investigation "github.com/openshift/configuration-anomaly-detection/pkg/investigations"
+	investigation "github.com/openshift/configuration-anomaly-detection/pkg/investigations/investigation"
 )
 
 func TestEvaluateRandomError(t *testing.T) {
@@ -19,7 +19,10 @@ func TestEvaluateRandomError(t *testing.T) {
 			"error": errors.New("timeout"),
 		},
 	}
-	_, err := Investigate(&input)
+
+	inv := CCAM{}
+
+	_, err := inv.Run(&input)
 	if err.Error() != timeoutError.Error() {
 		t.Fatalf("Expected error %v, but got %v", timeoutError, err)
 	}

pkg/investigations/chgm/chgm.go

@@ -8,7 +8,7 @@ import (
 	"time"
 
 	"github.com/openshift/configuration-anomaly-detection/pkg/aws"
-	investigation "github.com/openshift/configuration-anomaly-detection/pkg/investigations"
+	investigation "github.com/openshift/configuration-anomaly-detection/pkg/investigations/investigation"
 	"github.com/openshift/configuration-anomaly-detection/pkg/logging"
 	"github.com/openshift/configuration-anomaly-detection/pkg/networkverifier"
 	"github.com/openshift/configuration-anomaly-detection/pkg/notewriter"
@@ -36,8 +36,10 @@ var (
 	}
 )
 
-// Investigate runs the investigation for a triggered chgm pagerduty event
-func Investigate(r *investigation.Resources) (investigation.InvestigationResult, error) {
+type CHGM struct{}
+
+// Run runs the investigation for a triggered chgm pagerduty event
+func (c *CHGM) Run(r *investigation.Resources) (investigation.InvestigationResult, error) {
 	result := investigation.InvestigationResult{}
 	notes := notewriter.New("CHGM", logging.RawLogger)
 
@@ -116,6 +118,22 @@ func Investigate(r *investigation.Resources) (investigation.InvestigationResult,
 	return result, r.PdClient.EscalateIncidentWithNote(notes.String())
 }
 
+func (c *CHGM) Name() string {
+	return "Cluster Has Gone Missing (CHGM)"
+}
+
+func (c *CHGM) Description() string {
+	return "Detects reason for clusters that have gone missing"
+}
+
+func (c *CHGM) ShouldInvestigateAlert(alert string) bool {
+	return strings.Contains(alert, "has gone missing")
+}
+
+func (c *CHGM) IsExperimental() bool {
+	return false
+}
+
 // investigateHibernation checks if the cluster was recently woken up from
 // hibernation. If clusters are hibernated for more than 30 days, the internal
 // certificates of the kubelets can expire and CSRs need to be approved