Merge pull request #1329 from hongkailiu/OTA-1813

OTA-1813: Populate risks from alerts

Author: openshift-merge-bot[bot]

.openshift-tests-extension/openshift_payload_cluster-version-operator.json

@@ -1,4 +1,18 @@
 [
+  {
+    "name": "[Jira:\"Cluster Version Operator\"] cluster-version-operator should work with risks from alerts",
+    "labels": {
+      "Local": {},
+      "OTA-1813": {},
+      "Serial": {}
+    },
+    "resources": {
+      "isolation": {}
+    },
+    "source": "openshift:payload:cluster-version-operator",
+    "lifecycle": "blocking",
+    "environmentSelector": {}
+  },
   {
     "name": "[Jira:\"Cluster Version Operator\"] cluster-version-operator should work with accept risks",
     "labels": {

cmd/cluster-version-operator-tests/main.go

@@ -24,7 +24,7 @@ func main() {
 		Name:    "openshift/cluster-version-operator/conformance/parallel",
 		Parents: []string{"openshift/conformance/parallel"},
 		Qualifiers: []string{
-			`!(name.contains("[Serial]") || "Serial" in labels || name.contains("[Slow]"))`,
+			`!(name.contains("[Serial]") || "Serial" in labels || name.contains("[Slow]") || "Local" in labels)`,
 		},
 	})
 
@@ -33,7 +33,7 @@ func main() {
 		Name:    "openshift/cluster-version-operator/conformance/serial",
 		Parents: []string{"openshift/conformance/serial"},
 		Qualifiers: []string{
-			`name.contains("[Serial]") || "Serial" in labels`,
+			`(name.contains("[Serial]") || "Serial" in labels) && !("Local" in labels)`,
 		},
 	})
 

go.mod

@@ -17,6 +17,7 @@ require (
 	github.com/operator-framework/api v0.17.1
 	github.com/operator-framework/operator-lifecycle-manager v0.22.0
 	github.com/pkg/errors v0.9.1
+	github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring v0.86.0
 	github.com/prometheus-operator/prometheus-operator/pkg/client v0.86.0
 	github.com/prometheus/client_golang v1.23.2
 	github.com/prometheus/client_model v0.6.2
@@ -72,7 +73,6 @@ require (
 	github.com/mwitkow/go-conntrack v0.0.0-20190716064945-2f068394615f // indirect
 	github.com/mxk/go-flowrate v0.0.0-20140419014527-cca7078d478f // indirect
 	github.com/pmezard/go-difflib v1.0.1-0.20181226105442-5d4384ee4fb2 // indirect
-	github.com/prometheus-operator/prometheus-operator/pkg/apis/monitoring v0.86.0 // indirect
 	github.com/prometheus/procfs v0.16.1 // indirect
 	github.com/robfig/cron v1.2.0 // indirect
 	github.com/sirupsen/logrus v1.9.3 // indirect

pkg/clusterconditions/promql/alerts.go

@@ -0,0 +1,89 @@
+package promql
+
+import (
+	"context"
+	"fmt"
+	"sync"
+
+	"github.com/prometheus/client_golang/api"
+	prometheusv1 "github.com/prometheus/client_golang/api/prometheus/v1"
+	"github.com/prometheus/common/config"
+
+	"k8s.io/klog/v2"
+
+	"github.com/openshift/cluster-version-operator/pkg/clusterconditions"
+)
+
+type Getter interface {
+	Get(ctx context.Context) prometheusv1.AlertsResult
+}
+
+func NewAlertGetter(promQLTarget clusterconditions.PromQLTarget) Getter {
+	p := NewPromQL(promQLTarget)
+	condition := p.Condition
+	v, ok := condition.(*PromQL)
+	if !ok {
+		panic("invalid condition type")
+	}
+	return &ocAlertGetter{promQL: v}
+}
+
+type ocAlertGetter struct {
+	promQL *PromQL
+
+	mutex  sync.Mutex
+	cached prometheusv1.AlertsResult
+}
+
+func (o *ocAlertGetter) Get(ctx context.Context) prometheusv1.AlertsResult {
+	if err := o.refresh(ctx); err != nil {
+		klog.Errorf("Failed to refresh alerts, using stale cache instead: %v", err)
+	}
+	return o.cached
+}
+
+func (o *ocAlertGetter) refresh(ctx context.Context) error {
+	o.mutex.Lock()
+	defer o.mutex.Unlock()
+
+	klog.Info("refresh alerts ...")
+	p := o.promQL
+	host, err := p.Host(ctx)
+	if err != nil {
+		return fmt.Errorf("failure determine thanos IP: %w", err)
+	}
+	p.url.Host = host
+	clientConfig := api.Config{Address: p.url.String()}
+
+	if roundTripper, err := config.NewRoundTripperFromConfig(p.HTTPClientConfig, "cluster-conditions"); err == nil {
+		clientConfig.RoundTripper = roundTripper
+	} else {
+		return fmt.Errorf("creating PromQL round-tripper: %w", err)
+	}
+
+	promqlClient, err := api.NewClient(clientConfig)
+	if err != nil {
+		return fmt.Errorf("creating PromQL client: %w", err)
+	}
+
+	client := &statusCodeNotImplementedForPostClient{
+		client: promqlClient,
+	}
+
+	v1api := prometheusv1.NewAPI(client)
+
+	queryContext := ctx
+	if p.QueryTimeout > 0 {
+		var cancel context.CancelFunc
+		queryContext, cancel = context.WithTimeout(ctx, p.QueryTimeout)
+		defer cancel()
+	}
+
+	r, err := v1api.Alerts(queryContext)
+	if err != nil {
+		return fmt.Errorf("failed to get alerts: %w", err)
+	}
+	o.cached = r
+	klog.Infof("refreshed: %d alerts", len(o.cached.Alerts))
+	return nil
+}