Merge pull request #13399 from ahrtr/serializable_health_check

Enhance health check endpoint to support serializable request
etcd-io · Nov 15, 2021 · d357f9b · d357f9b
2 parents 29c3b0f + 09ff051
commit d357f9b
Show file tree

Hide file tree

Showing 3 changed files with 24 additions and 12 deletions.
diff --git a/CHANGELOG-3.6.md b/CHANGELOG-3.6.md
@@ -34,6 +34,7 @@ See [code changes](https://github.com/etcd-io/etcd/compare/v3.5.0...v3.6.0).
 - Add [`etcd --log-format`](https://github.com/etcd-io/etcd/pull/13339) flag to support log format.
 - Fix [non mutating requests pass through quotaKVServer when NOSPACE](https://github.com/etcd-io/etcd/pull/13435)
 - Fix [exclude the same alarm type activated by multiple peers](https://github.com/etcd-io/etcd/pull/13467).
+- Fix [Provide a better liveness probe for when etcd runs as a Kubernetes pod](https://github.com/etcd-io/etcd/pull/13399)
 
 ### tools/benchmark
 

diff --git a/server/etcdserver/api/etcdhttp/metrics.go b/server/etcdserver/api/etcdhttp/metrics.go
@@ -40,14 +40,16 @@ const (
 // HandleMetricsHealth registers metrics and health handlers.
 func HandleMetricsHealth(lg *zap.Logger, mux *http.ServeMux, srv etcdserver.ServerV2) {
 	mux.Handle(PathMetrics, promhttp.Handler())
-	mux.Handle(PathHealth, NewHealthHandler(lg, func(excludedAlarms AlarmSet) Health { return checkV2Health(lg, srv, excludedAlarms) }))
+	mux.Handle(PathHealth, NewHealthHandler(lg, func(excludedAlarms AlarmSet, serializable bool) Health { return checkV2Health(lg, srv, excludedAlarms) }))
 }
 
 // HandleMetricsHealthForV3 registers metrics and health handlers. it checks health by using v3 range request
 // and its corresponding timeout.
 func HandleMetricsHealthForV3(lg *zap.Logger, mux *http.ServeMux, srv *etcdserver.EtcdServer) {
 	mux.Handle(PathMetrics, promhttp.Handler())
-	mux.Handle(PathHealth, NewHealthHandler(lg, func(excludedAlarms AlarmSet) Health { return checkV3Health(lg, srv, excludedAlarms) }))
+	mux.Handle(PathHealth, NewHealthHandler(lg, func(excludedAlarms AlarmSet, serializable bool) Health {
+		return checkV3Health(lg, srv, excludedAlarms, serializable)
+	}))
 }
 
 // HandlePrometheus registers prometheus handler on '/metrics'.
@@ -56,7 +58,7 @@ func HandlePrometheus(mux *http.ServeMux) {
 }
 
 // NewHealthHandler handles '/health' requests.
-func NewHealthHandler(lg *zap.Logger, hfunc func(excludedAlarms AlarmSet) Health) http.HandlerFunc {
+func NewHealthHandler(lg *zap.Logger, hfunc func(excludedAlarms AlarmSet, Serializable bool) Health) http.HandlerFunc {
 	return func(w http.ResponseWriter, r *http.Request) {
 		if r.Method != http.MethodGet {
 			w.Header().Set("Allow", http.MethodGet)
@@ -65,7 +67,12 @@ func NewHealthHandler(lg *zap.Logger, hfunc func(excludedAlarms AlarmSet) Health
 			return
 		}
 		excludedAlarms := getExcludedAlarms(r)
-		h := hfunc(excludedAlarms)
+		// Passing the query parameter "serializable=true" ensures that the
+		// health of the local etcd is checked vs the health of the cluster.
+		// This is useful for probes attempting to validate the liveness of
+		// the etcd process vs readiness of the cluster to serve requests.
+		serializableFlag := getSerializableFlag(r)
+		h := hfunc(excludedAlarms, serializableFlag)
 		defer func() {
 			if h.Health == "true" {
 				healthSuccess.Inc()
@@ -128,9 +135,13 @@ func getExcludedAlarms(r *http.Request) (alarms AlarmSet) {
 	return alarms
 }
 
+func getSerializableFlag(r *http.Request) bool {
+	return r.URL.Query().Get("serializable") == "true"
+}
+
 // TODO: etcdserver.ErrNoLeader in health API
 
-func checkHealth(lg *zap.Logger, srv etcdserver.ServerV2, excludedAlarms AlarmSet) Health {
+func checkHealth(lg *zap.Logger, srv etcdserver.ServerV2, excludedAlarms AlarmSet, serializable bool) Health {
 	h := Health{}
 	h.Health = "true"
 	as := srv.Alarms()
@@ -156,7 +167,7 @@ func checkHealth(lg *zap.Logger, srv etcdserver.ServerV2, excludedAlarms AlarmSe
 		}
 	}
 
-	if uint64(srv.Leader()) == raft.None {
+	if !serializable && (uint64(srv.Leader()) == raft.None) {
 		h.Health = "false"
 		h.Reason = "RAFT NO LEADER"
 		lg.Warn("serving /health false; no leader")
@@ -166,7 +177,7 @@ func checkHealth(lg *zap.Logger, srv etcdserver.ServerV2, excludedAlarms AlarmSe
 }
 
 func checkV2Health(lg *zap.Logger, srv etcdserver.ServerV2, excludedAlarms AlarmSet) (h Health) {
-	if h = checkHealth(lg, srv, excludedAlarms); h.Health != "true" {
+	if h = checkHealth(lg, srv, excludedAlarms, false); h.Health != "true" {
 		return
 	}
 	ctx, cancel := context.WithTimeout(context.Background(), time.Second)
@@ -182,12 +193,12 @@ func checkV2Health(lg *zap.Logger, srv etcdserver.ServerV2, excludedAlarms Alarm
 	return
 }
 
-func checkV3Health(lg *zap.Logger, srv *etcdserver.EtcdServer, excludedAlarms AlarmSet) (h Health) {
-	if h = checkHealth(lg, srv, excludedAlarms); h.Health != "true" {
+func checkV3Health(lg *zap.Logger, srv *etcdserver.EtcdServer, excludedAlarms AlarmSet, serializable bool) (h Health) {
+	if h = checkHealth(lg, srv, excludedAlarms, serializable); h.Health != "true" {
 		return
 	}
 	ctx, cancel := context.WithTimeout(context.Background(), srv.Cfg.ReqTimeout())
-	_, err := srv.Range(ctx, &etcdserverpb.RangeRequest{KeysOnly: true, Limit: 1})
+	_, err := srv.Range(ctx, &etcdserverpb.RangeRequest{KeysOnly: true, Limit: 1, Serializable: serializable})
 	cancel()
 	if err != nil && err != auth.ErrUserEmpty && err != auth.ErrPermissionDenied {
 		h.Health = "false"

diff --git a/server/proxy/grpcproxy/health.go b/server/proxy/grpcproxy/health.go
@@ -31,15 +31,15 @@ func HandleHealth(lg *zap.Logger, mux *http.ServeMux, c *clientv3.Client) {
 	if lg == nil {
 		lg = zap.NewNop()
 	}
-	mux.Handle(etcdhttp.PathHealth, etcdhttp.NewHealthHandler(lg, func(excludedAlarms etcdhttp.AlarmSet) etcdhttp.Health { return checkHealth(c) }))
+	mux.Handle(etcdhttp.PathHealth, etcdhttp.NewHealthHandler(lg, func(excludedAlarms etcdhttp.AlarmSet, serializable bool) etcdhttp.Health { return checkHealth(c) }))
 }
 
 // HandleProxyHealth registers health handler on '/proxy/health'.
 func HandleProxyHealth(lg *zap.Logger, mux *http.ServeMux, c *clientv3.Client) {
 	if lg == nil {
 		lg = zap.NewNop()
 	}
-	mux.Handle(etcdhttp.PathProxyHealth, etcdhttp.NewHealthHandler(lg, func(excludedAlarms etcdhttp.AlarmSet) etcdhttp.Health { return checkProxyHealth(c) }))
+	mux.Handle(etcdhttp.PathProxyHealth, etcdhttp.NewHealthHandler(lg, func(excludedAlarms etcdhttp.AlarmSet, serializable bool) etcdhttp.Health { return checkProxyHealth(c) }))
 }
 
 func checkHealth(c *clientv3.Client) etcdhttp.Health {