extend cluster health to return errors for IAM, and Bucket metadata (#19995)

Bonus: make API freeze to be opt-in instead of default

cmd/bootstrap-peer-server.go

@@ -114,6 +114,9 @@ var skipEnvs = map[string]struct{}{
 	"MINIO_ROOT_PASSWORD":          {},
 	"MINIO_ACCESS_KEY":             {},
 	"MINIO_SECRET_KEY":             {},
+	"MINIO_OPERATOR_VERSION":       {},
+	"MINIO_VSPHERE_PLUGIN_VERSION": {},
+	"MINIO_CI_CD":                  {},
 }
 
 func getServerSystemCfg() *ServerSystemConfig {

cmd/common-main.go

@@ -834,7 +834,7 @@ func serverHandleEnvVars() {
 		}
 	}
 
-	globalDisableFreezeOnBoot = env.Get("_MINIO_DISABLE_API_FREEZE_ON_BOOT", "") == "true" || serverDebugLog
+	globalEnableSyncBoot = env.Get("MINIO_SYNC_BOOT", config.EnableOff) == config.EnableOn
 }
 
 func loadRootCredentials() {

cmd/globals.go

@@ -449,8 +449,8 @@ var (
 	// dynamic sleeper for multipart expiration routine
 	deleteMultipartCleanupSleeper = newDynamicSleeper(5, 25*time.Millisecond, false)
 
-	// Is _MINIO_DISABLE_API_FREEZE_ON_BOOT set?
-	globalDisableFreezeOnBoot bool
+	// Is MINIO_SYNC_BOOT set?
+	globalEnableSyncBoot bool
 
 	// Contains NIC interface name used for internode communication
 	globalInternodeInterface     string

cmd/healthcheck-handler.go

@@ -29,14 +29,35 @@ import (
 
 const unavailable = "offline"
 
-// ClusterCheckHandler returns if the server is ready for requests.
-func ClusterCheckHandler(w http.ResponseWriter, r *http.Request) {
-	ctx := newContext(r, w, "ClusterCheckHandler")
-
+func checkHealth(w http.ResponseWriter) ObjectLayer {
 	objLayer := newObjectLayerFn()
 	if objLayer == nil {
 		w.Header().Set(xhttp.MinIOServerStatus, unavailable)
 		writeResponse(w, http.StatusServiceUnavailable, nil, mimeNone)
+		return nil
+	}
+
+	if !globalBucketMetadataSys.Initialized() {
+		w.Header().Set(xhttp.MinIOServerStatus, "bucket-metadata-offline")
+		writeResponse(w, http.StatusServiceUnavailable, nil, mimeNone)
+		return nil
+	}
+
+	if !globalIAMSys.Initialized() {
+		w.Header().Set(xhttp.MinIOServerStatus, "iam-offline")
+		writeResponse(w, http.StatusServiceUnavailable, nil, mimeNone)
+		return nil
+	}
+
+	return objLayer
+}
+
+// ClusterCheckHandler returns if the server is ready for requests.
+func ClusterCheckHandler(w http.ResponseWriter, r *http.Request) {
+	ctx := newContext(r, w, "ClusterCheckHandler")
+
+	objLayer := checkHealth(w)
+	if objLayer == nil {
 		return
 	}
 
@@ -72,10 +93,8 @@ func ClusterCheckHandler(w http.ResponseWriter, r *http.Request) {
 func ClusterReadCheckHandler(w http.ResponseWriter, r *http.Request) {
 	ctx := newContext(r, w, "ClusterReadCheckHandler")
 
-	objLayer := newObjectLayerFn()
+	objLayer := checkHealth(w)
 	if objLayer == nil {
-		w.Header().Set(xhttp.MinIOServerStatus, unavailable)
-		writeResponse(w, http.StatusServiceUnavailable, nil, mimeNone)
 		return
 	}
 

cmd/server-main.go

@@ -897,7 +897,7 @@ func serverMain(ctx *cli.Context) {
 		})
 	}
 
-	if !globalDisableFreezeOnBoot {
+	if globalEnableSyncBoot {
 		// Freeze the services until the bucket notification subsystem gets initialized.
 		bootstrapTrace("freezeServices", freezeServices)
 	}
@@ -1000,10 +1000,11 @@ func serverMain(ctx *cli.Context) {
 	}()
 
 	go func() {
-		if !globalDisableFreezeOnBoot {
+		if globalEnableSyncBoot {
 			defer bootstrapTrace("unfreezeServices", unfreezeServices)
 			t := time.AfterFunc(5*time.Minute, func() {
-				warnings = append(warnings, color.YellowBold("- Initializing the config subsystem is taking longer than 5 minutes. Please set '_MINIO_DISABLE_API_FREEZE_ON_BOOT=true' to not freeze the APIs"))
+				warnings = append(warnings,
+					color.YellowBold("- Initializing the config subsystem is taking longer than 5 minutes. Please remove 'MINIO_SYNC_BOOT=on' to not freeze the APIs"))
 			})
 			defer t.Stop()
 		}