patch healthcheck sli metric so that we only have a binary value

Change-Id: I0d87e29715432f772309a0d4a7305fff358c6d48
This commit is contained in:
Han Kang 2022-09-27 12:43:42 -07:00
parent db13f51db9
commit 334be489d4
2 changed files with 8 additions and 43 deletions

View File

@ -18,8 +18,6 @@ package slis
import ( import (
"context" "context"
"errors"
k8smetrics "k8s.io/component-base/metrics" k8smetrics "k8s.io/component-base/metrics"
) )
@ -28,17 +26,10 @@ type HealthcheckStatus string
const ( const (
Success HealthcheckStatus = "success" Success HealthcheckStatus = "success"
Error HealthcheckStatus = "error" Error HealthcheckStatus = "error"
Pending HealthcheckStatus = "pending"
) )
type HealthcheckType string type HealthcheckType string
const (
Livez HealthcheckType = "livez"
Readyz HealthcheckType = "readyz"
Healthz HealthcheckType = "healthz"
)
var ( var (
// healthcheck is a Prometheus Gauge metrics used for recording the results of a k8s healthcheck. // healthcheck is a Prometheus Gauge metrics used for recording the results of a k8s healthcheck.
healthcheck = k8smetrics.NewGaugeVec( healthcheck = k8smetrics.NewGaugeVec(
@ -48,7 +39,7 @@ var (
Help: "This metric records the result of a single healthcheck.", Help: "This metric records the result of a single healthcheck.",
StabilityLevel: k8smetrics.ALPHA, StabilityLevel: k8smetrics.ALPHA,
}, },
[]string{"name", "type", "status"}, []string{"name", "type"},
) )
// healthchecksTotal is a Prometheus Counter metrics used for counting the results of a k8s healthcheck. // healthchecksTotal is a Prometheus Counter metrics used for counting the results of a k8s healthcheck.
@ -61,8 +52,6 @@ var (
}, },
[]string{"name", "type", "status"}, []string{"name", "type", "status"},
) )
statuses = []HealthcheckStatus{Success, Error, Pending}
statusSet = map[HealthcheckStatus]struct{}{Success: {}, Error: {}, Pending: {}}
) )
func Register(registry k8smetrics.KubeRegistry) { func Register(registry k8smetrics.KubeRegistry) {
@ -76,15 +65,12 @@ func ResetHealthMetrics() {
} }
func ObserveHealthcheck(ctx context.Context, name string, healthcheckType string, status HealthcheckStatus) error { func ObserveHealthcheck(ctx context.Context, name string, healthcheckType string, status HealthcheckStatus) error {
if _, ok := statusSet[status]; !ok { if status == Success {
return errors.New("not a valid healthcheck status") healthcheck.WithContext(ctx).WithLabelValues(name, healthcheckType).Set(1)
} } else {
for _, s := range statuses { healthcheck.WithContext(ctx).WithLabelValues(name, healthcheckType).Set(0)
if status != s {
healthcheck.WithContext(ctx).WithLabelValues(name, healthcheckType, string(s)).Set(0)
}
} }
healthchecksTotal.WithContext(ctx).WithLabelValues(name, healthcheckType, string(status)).Inc() healthchecksTotal.WithContext(ctx).WithLabelValues(name, healthcheckType, string(status)).Inc()
healthcheck.WithContext(ctx).WithLabelValues(name, healthcheckType, string(status)).Set(1)
return nil return nil
} }

View File

@ -39,9 +39,7 @@ func TestObserveHealthcheck(t *testing.T) {
initialOutput := ` initialOutput := `
# HELP kubernetes_healthcheck [ALPHA] This metric records the result of a single healthcheck. # HELP kubernetes_healthcheck [ALPHA] This metric records the result of a single healthcheck.
# TYPE kubernetes_healthcheck gauge # TYPE kubernetes_healthcheck gauge
kubernetes_healthcheck{name="healthcheck-a",status="error",type="healthz"} 1 kubernetes_healthcheck{name="healthcheck-a",type="healthz"} 0
kubernetes_healthcheck{name="healthcheck-a",status="pending",type="healthz"} 0
kubernetes_healthcheck{name="healthcheck-a",status="success",type="healthz"} 0
# HELP kubernetes_healthchecks_total [ALPHA] This metric records the results of all healthcheck. # HELP kubernetes_healthchecks_total [ALPHA] This metric records the results of all healthcheck.
# TYPE kubernetes_healthchecks_total counter # TYPE kubernetes_healthchecks_total counter
kubernetes_healthchecks_total{name="healthcheck-a",status="error",type="healthz"} 1 kubernetes_healthchecks_total{name="healthcheck-a",status="error",type="healthz"} 1
@ -53,23 +51,6 @@ func TestObserveHealthcheck(t *testing.T) {
hcStatus HealthcheckStatus hcStatus HealthcheckStatus
want string want string
}{ }{
{
desc: "test pending",
name: healthcheckName,
hcType: "healthz",
hcStatus: Pending,
want: `
# HELP kubernetes_healthcheck [ALPHA] This metric records the result of a single healthcheck.
# TYPE kubernetes_healthcheck gauge
kubernetes_healthcheck{name="healthcheck-a",status="error",type="healthz"} 0
kubernetes_healthcheck{name="healthcheck-a",status="pending",type="healthz"} 1
kubernetes_healthcheck{name="healthcheck-a",status="success",type="healthz"} 0
# HELP kubernetes_healthchecks_total [ALPHA] This metric records the results of all healthcheck.
# TYPE kubernetes_healthchecks_total counter
kubernetes_healthchecks_total{name="healthcheck-a",status="error",type="healthz"} 1
kubernetes_healthchecks_total{name="healthcheck-a",status="pending",type="healthz"} 1
`,
},
{ {
desc: "test success", desc: "test success",
name: healthcheckName, name: healthcheckName,
@ -78,9 +59,7 @@ func TestObserveHealthcheck(t *testing.T) {
want: ` want: `
# HELP kubernetes_healthcheck [ALPHA] This metric records the result of a single healthcheck. # HELP kubernetes_healthcheck [ALPHA] This metric records the result of a single healthcheck.
# TYPE kubernetes_healthcheck gauge # TYPE kubernetes_healthcheck gauge
kubernetes_healthcheck{name="healthcheck-a",status="error",type="healthz"} 0 kubernetes_healthcheck{name="healthcheck-a",type="healthz"} 1
kubernetes_healthcheck{name="healthcheck-a",status="pending",type="healthz"} 0
kubernetes_healthcheck{name="healthcheck-a",status="success",type="healthz"} 1
# HELP kubernetes_healthchecks_total [ALPHA] This metric records the results of all healthcheck. # HELP kubernetes_healthchecks_total [ALPHA] This metric records the results of all healthcheck.
# TYPE kubernetes_healthchecks_total counter # TYPE kubernetes_healthchecks_total counter
kubernetes_healthchecks_total{name="healthcheck-a",status="error",type="healthz"} 1 kubernetes_healthchecks_total{name="healthcheck-a",status="error",type="healthz"} 1