Add proper contexts with timeouts for etcd operations (#7097)

This fixes an issue of perceived hang when incorrect
unreachable URLs are specified in MINIO_ETCD_ENDPOINTS
variable.

Fixes #7096

cmd/common-main.go

@@ -242,6 +242,7 @@ func handleCommonEnvVars() {
 					return &cert, terr
 				}
 			}
+
 			globalEtcdClient, err = etcd.New(etcd.Config{
 				Endpoints:         etcdEndpoints,
 				DialTimeout:       defaultDialTimeout,

cmd/config-common.go

@@ -20,6 +20,7 @@ import (
 	"bytes"
 	"context"
 	"errors"
+	"fmt"
 
 	etcd "github.com/coreos/etcd/clientv3"
 	"github.com/minio/minio/cmd/logger"
@@ -60,8 +61,15 @@ func deleteConfig(ctx context.Context, objAPI ObjectLayer, configFile string) er
 }
 
 func saveConfigEtcd(ctx context.Context, client *etcd.Client, configFile string, data []byte) error {
-	_, err := client.Put(ctx, configFile, string(data))
+	timeoutCtx, cancel := context.WithTimeout(ctx, defaultContextTimeout)
-	return err
+	defer cancel()
+	_, err := client.Put(timeoutCtx, configFile, string(data))
+	if err == context.DeadlineExceeded {
+		return fmt.Errorf("etcd setup is unreachable, please check your endpoints %s", client.Endpoints())
+	} else if err != nil {
+		return fmt.Errorf("unexpected error %s returned by etcd setup, please check your endpoints %s", err, client.Endpoints())
+	}
+	return nil
 }
 
 func saveConfig(ctx context.Context, objAPI ObjectLayer, configFile string, data []byte) error {
@@ -75,9 +83,14 @@ func saveConfig(ctx context.Context, objAPI ObjectLayer, configFile string, data
 }
 
 func readConfigEtcd(ctx context.Context, client *etcd.Client, configFile string) ([]byte, error) {
-	resp, err := client.Get(ctx, configFile)
+	timeoutCtx, cancel := context.WithTimeout(ctx, defaultContextTimeout)
+	defer cancel()
+	resp, err := client.Get(timeoutCtx, configFile)
 	if err != nil {
-		return nil, err
+		if err == context.DeadlineExceeded {
+			return nil, fmt.Errorf("etcd setup is unreachable, please check your endpoints %s", client.Endpoints())
+		}
+		return nil, fmt.Errorf("unexpected error %s returned by etcd setup, please check your endpoints %s", err, client.Endpoints())
 	}
 	if resp.Count == 0 {
 		return nil, errConfigNotFound
@@ -93,7 +106,9 @@ func readConfigEtcd(ctx context.Context, client *etcd.Client, configFile string)
 // watchConfig - watches for changes on `configFile` on etcd and loads them.
 func watchConfig(objAPI ObjectLayer, configFile string, loadCfgFn func(ObjectLayer) error) {
 	if globalEtcdClient != nil {
-		for watchResp := range globalEtcdClient.Watch(context.Background(), configFile) {
+		ctx, cancel := context.WithTimeout(context.Background(), defaultContextTimeout)
+		defer cancel()
+		for watchResp := range globalEtcdClient.Watch(ctx, configFile) {
 			for _, event := range watchResp.Events {
 				if event.IsModify() || event.IsCreate() {
 					loadCfgFn(objAPI)
@@ -104,9 +119,14 @@ func watchConfig(objAPI ObjectLayer, configFile string, loadCfgFn func(ObjectLay
 }
 
 func checkConfigEtcd(ctx context.Context, client *etcd.Client, configFile string) error {
-	resp, err := globalEtcdClient.Get(ctx, configFile)
+	timeoutCtx, cancel := context.WithTimeout(ctx, defaultContextTimeout)
+	defer cancel()
+	resp, err := client.Get(timeoutCtx, configFile)
 	if err != nil {
-		return err
+		if err == context.DeadlineExceeded {
+			return fmt.Errorf("etcd setup is unreachable, please check your endpoints %s", client.Endpoints())
+		}
+		return fmt.Errorf("unexpected error %s returned by etcd setup, please check your endpoints %s", err, client.Endpoints())
 	}
 	if resp.Count == 0 {
 		return errConfigNotFound

cmd/config.go

@@ -24,7 +24,6 @@ import (
 	"path"
 	"runtime"
 	"strings"
-	"time"
 
 	"github.com/minio/minio/cmd/logger"
 	"github.com/minio/minio/pkg/quick"
@@ -52,10 +51,7 @@ func saveServerConfig(ctx context.Context, objAPI ObjectLayer, config *serverCon
 
 	configFile := path.Join(minioConfigPrefix, minioConfigFile)
 	if globalEtcdClient != nil {
-		timeoutCtx, cancel := context.WithTimeout(ctx, 5*time.Minute)
+		return saveConfigEtcd(ctx, globalEtcdClient, configFile, data)
-		_, err = globalEtcdClient.Put(timeoutCtx, configFile, string(data))
-		defer cancel()
-		return err
 	}
 
 	// Create a backup of the current config
@@ -163,18 +159,16 @@ func initConfig(objAPI ObjectLayer) error {
 	}
 
 	if globalEtcdClient != nil {
-		ctx, cancel := context.WithTimeout(context.Background(), 20*time.Second)
+		if err := checkConfigEtcd(context.Background(), globalEtcdClient, getConfigFile()); err != nil {
-		resp, err := globalEtcdClient.Get(ctx, getConfigFile())
+			return err
-		cancel()
+		}
-		if err == nil && resp.Count > 0 {
+		// Migrates all configs at old location.
-			if err = migrateConfig(); err != nil {
+		if err := migrateConfig(); err != nil {
-				return err
+			return err
-			}
+		}
-
+		// Migrates etcd ${HOME}/.minio/config.json to '/config/config.json'
-			// Migrates etcd ${HOME}/.minio/config.json to '/config/config.json'
+		if err := migrateConfigToMinioSys(objAPI); err != nil {
-			if err := migrateConfigToMinioSys(objAPI); err != nil {
+			return err
-				return err
-			}
 		}
 	} else {
 		if isFile(getConfigFile()) {

cmd/iam.go

@@ -464,13 +464,13 @@ func (sys *IAMSys) IsAllowed(args iampolicy.Args) bool {
 	return args.IsOwner
 }
 
-var defaultContextTimeout = 5 * time.Minute
+var defaultContextTimeout = 30 * time.Second
 
 // Similar to reloadUsers but updates users, policies maps from etcd server,
 func reloadEtcdUsers(prefix string, usersMap map[string]auth.Credentials, policyMap map[string]string) error {
 	ctx, cancel := context.WithTimeout(context.Background(), defaultContextTimeout)
-	r, err := globalEtcdClient.Get(ctx, prefix, etcd.WithPrefix(), etcd.WithKeysOnly())
 	defer cancel()
+	r, err := globalEtcdClient.Get(ctx, prefix, etcd.WithPrefix(), etcd.WithKeysOnly())
 	if err != nil {
 		return err
 	}
@@ -536,8 +536,8 @@ func reloadEtcdUsers(prefix string, usersMap map[string]auth.Credentials, policy
 
 func reloadEtcdPolicies(prefix string, cannedPolicyMap map[string]iampolicy.Policy) error {
 	ctx, cancel := context.WithTimeout(context.Background(), defaultContextTimeout)
-	r, err := globalEtcdClient.Get(ctx, prefix, etcd.WithPrefix(), etcd.WithKeysOnly())
 	defer cancel()
+	r, err := globalEtcdClient.Get(ctx, prefix, etcd.WithPrefix(), etcd.WithKeysOnly())
 	if err != nil {
 		return err
 	}

pkg/quick/encoding.go

@@ -138,18 +138,26 @@ func saveFileConfigEtcd(filename string, clnt *etcd.Client, v interface{}) error
 		dataBytes = []byte(strings.Replace(string(dataBytes), "\n", "\r\n", -1))
 	}
 
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
-	_, err = clnt.Put(ctx, filename, string(dataBytes))
 	defer cancel()
-	return err
+	_, err = clnt.Put(ctx, filename, string(dataBytes))
+	if err == context.DeadlineExceeded {
+		return fmt.Errorf("etcd setup is unreachable, please check your endpoints %s", clnt.Endpoints())
+	} else if err != nil {
+		return fmt.Errorf("unexpected error %s returned by etcd setup, please check your endpoints %s", err, clnt.Endpoints())
+	}
+	return nil
 }
 
 func loadFileConfigEtcd(filename string, clnt *etcd.Client, v interface{}) error {
-	ctx, cancel := context.WithTimeout(context.Background(), 5*time.Minute)
+	ctx, cancel := context.WithTimeout(context.Background(), 30*time.Second)
-	resp, err := clnt.Get(ctx, filename)
 	defer cancel()
+	resp, err := clnt.Get(ctx, filename)
 	if err != nil {
-		return err
+		if err == context.DeadlineExceeded {
+			return fmt.Errorf("etcd setup is unreachable, please check your endpoints %s", clnt.Endpoints())
+		}
+		return fmt.Errorf("unexpected error %s returned by etcd setup, please check your endpoints %s", err, clnt.Endpoints())
 	}
 	if resp.Count == 0 {
 		return dns.ErrNoEntriesFound