Add API Server Reachability Check (#2959)

* get api server ping code * add emit gauge for healthz ping code --------- Co-authored-by: Nicolas Ontiveros <[email protected]>
Azure · Jul 5, 2023 · 629913b · 629913b
1 parent d82d4f5
commit 629913b
Show file tree

Hide file tree

Showing 2 changed files with 23 additions and 7 deletions.
diff --git a/pkg/monitor/cluster/cluster.go b/pkg/monitor/cluster/cluster.go
@@ -137,18 +137,18 @@ func (mon *Monitor) Monitor(ctx context.Context) (errs []error) {
 		})
 	}
 
-	// If API is not returning 200, don't need to run the next checks
-	statusCode, err := mon.emitAPIServerHealthzCode(ctx)
-	if err != nil {
+	// If API Server is unreachable, don't need to run the next checks
+	statusCode, err := mon.getAPIServerPingCode(ctx)
+	if err != nil || statusCode != http.StatusOK {
 		errs = append(errs, err)
-		friendlyFuncName := steps.FriendlyName(mon.emitAPIServerHealthzCode)
+		friendlyFuncName := steps.FriendlyName(mon.getAPIServerPingCode)
 		mon.log.Printf("%s: %s", friendlyFuncName, err)
 		mon.emitGauge("monitor.clustererrors", 1, map[string]string{"monitor": friendlyFuncName})
-	}
-	if statusCode != http.StatusOK {
 		return
 	}
+
 	for _, f := range []func(context.Context) error{
+		mon.emitAPIServerHealthzCode,
 		mon.emitAroOperatorHeartbeat,
 		mon.emitAroOperatorConditions,
 		mon.emitNSGReconciliation,

diff --git a/pkg/monitor/cluster/healthz.go b/pkg/monitor/cluster/healthz.go
@@ -8,7 +8,7 @@ import (
 	"strconv"
 )
 
-func (mon *Monitor) emitAPIServerHealthzCode(ctx context.Context) (int, error) {
+func (mon *Monitor) emitAPIServerHealthzCode(ctx context.Context) error {
 	var statusCode int
 	err := mon.cli.Discovery().RESTClient().
 		Get().
@@ -21,5 +21,21 @@ func (mon *Monitor) emitAPIServerHealthzCode(ctx context.Context) (int, error) {
 		"code": strconv.FormatInt(int64(statusCode), 10),
 	})
 
+	return err
+}
+
+func (mon *Monitor) getAPIServerPingCode(ctx context.Context) (int, error) {
+	var statusCode int
+	err := mon.cli.Discovery().RESTClient().
+		Get().
+		AbsPath("/healthz/ping").
+		Do(ctx).
+		StatusCode(&statusCode).
+		Error()
+
+	mon.emitGauge("apiserver.healthz.ping.code", 1, map[string]string{
+		"code": strconv.FormatInt(int64(statusCode), 10),
+	})
+
 	return statusCode, err
 }