Merge branch 'release-8.1' into cherry-pick-8264-to-release-8.1

client/client.go

@@ -1424,17 +1424,6 @@ func (c *client) scatterRegionsWithOptions(ctx context.Context, regionsID []uint
 	return resp, nil
 }
 
-// IsLeaderChange will determine whether there is a leader change.
-func IsLeaderChange(err error) bool {
-	if err == errs.ErrClientTSOStreamClosed {
-		return true
-	}
-	errMsg := err.Error()
-	return strings.Contains(errMsg, errs.NotLeaderErr) ||
-		strings.Contains(errMsg, errs.MismatchLeaderErr) ||
-		strings.Contains(errMsg, errs.NotServedErr)
-}
-
 const (
 	httpSchemePrefix  = "http://"
 	httpsSchemePrefix = "https://"

client/errs/errno.go

@@ -20,21 +20,20 @@ import (
 	"github.com/pingcap/errors"
 )
 
+// Note: keep the same as the ones defined on the server side to ensure the client can use them correctly.
 const (
+	// NoLeaderErr indicates there is no leader in the cluster currently.
+	NoLeaderErr = "no leader"
 	// NotLeaderErr indicates the non-leader member received the requests which should be received by leader.
-	// Note: keep the same as the ones defined on the server side, because the client side checks if an error message
-	// contains this string to judge whether the leader is changed.
-	NotLeaderErr = "is not leader"
+	NotLeaderErr = "not leader"
 	// MismatchLeaderErr indicates the non-leader member received the requests which should be received by leader.
-	// Note: keep the same as the ones defined on the server side, because the client side checks if an error message
-	// contains this string to judge whether the leader is changed.
 	MismatchLeaderErr = "mismatch leader id"
 	// NotServedErr indicates an tso node/pod received the requests for the keyspace groups which are not served by it.
-	// Note: keep the same as the ones defined on the server side, because the client side checks if an error message
-	// contains this string to judge whether the leader is changed.
 	NotServedErr = "is not served"
 	// RetryTimeoutErr indicates the server is busy.
 	RetryTimeoutErr = "retry timeout"
+	// NotPrimaryErr indicates the non-primary member received the requests which should be received by primary.
+	NotPrimaryErr = "not primary"
 )
 
 // client errors

client/errs/errs.go

@@ -15,11 +15,29 @@
 package errs
 
 import (
+	"strings"
+
 	"github.com/pingcap/errors"
 	"go.uber.org/zap"
 	"go.uber.org/zap/zapcore"
 )
 
+// IsLeaderChange will determine whether there is a leader/primary change.
+func IsLeaderChange(err error) bool {
+	if err == nil {
+		return false
+	}
+	if err == ErrClientTSOStreamClosed {
+		return true
+	}
+	errMsg := err.Error()
+	return strings.Contains(errMsg, NoLeaderErr) ||
+		strings.Contains(errMsg, NotLeaderErr) ||
+		strings.Contains(errMsg, MismatchLeaderErr) ||
+		strings.Contains(errMsg, NotServedErr) ||
+		strings.Contains(errMsg, NotPrimaryErr)
+}
+
 // ZapError is used to make the log output easier.
 func ZapError(err error, causeError ...error) zap.Field {
 	if err == nil {

client/http/client.go

@@ -120,28 +120,47 @@ func (ci *clientInner) requestWithRetry(
 	headerOpts ...HeaderOption,
 ) error {
 	var (
+		serverURL  string
+		isLeader   bool
 		statusCode int
 		err        error
+		logFields  = append(reqInfo.logFields(), zap.String("source", ci.source))
 	)
 	execFunc := func() error {
+		defer func() {
+			// If the status code is 503, it indicates that there may be PD leader/follower changes.
+			// If the error message contains the leader/primary change information, it indicates that there may be PD leader/primary change.
+			if statusCode == http.StatusServiceUnavailable || errs.IsLeaderChange(err) {
+				ci.sd.ScheduleCheckMemberChanged()
+			}
+			log.Debug("[pd] http request finished", append(logFields,
+				zap.String("server-url", serverURL),
+				zap.Bool("is-leader", isLeader),
+				zap.Int("status-code", statusCode),
+				zap.Error(err))...)
+		}()
 		// It will try to send the request to the PD leader first and then try to send the request to the other PD followers.
 		clients := ci.sd.GetAllServiceClients()
 		if len(clients) == 0 {
 			return errs.ErrClientNoAvailableMember
 		}
 		skipNum := 0
 		for _, cli := range clients {
-			url := cli.GetURL()
-			if reqInfo.targetURL != "" && reqInfo.targetURL != url {
+			serverURL = cli.GetURL()
+			isLeader = cli.IsConnectedToLeader()
+			if len(reqInfo.targetURL) > 0 && reqInfo.targetURL != serverURL {
 				skipNum++
 				continue
 			}
-			statusCode, err = ci.doRequest(ctx, url, reqInfo, headerOpts...)
+			statusCode, err = ci.doRequest(ctx, serverURL, reqInfo, headerOpts...)
 			if err == nil || noNeedRetry(statusCode) {
 				return err
 			}
-			log.Debug("[pd] request url failed",
-				zap.String("source", ci.source), zap.Bool("is-leader", cli.IsConnectedToLeader()), zap.String("url", url), zap.Error(err))
+			log.Debug("[pd] http request url failed", append(logFields,
+				zap.String("server-url", serverURL),
+				zap.Bool("is-leader", isLeader),
+				zap.Int("status-code", statusCode),
+				zap.Error(err))...)
 		}
 		if skipNum == len(clients) {
 			return errs.ErrClientNoTargetMember
@@ -168,26 +187,21 @@ func noNeedRetry(statusCode int) bool {
 
 func (ci *clientInner) doRequest(
 	ctx context.Context,
-	url string, reqInfo *requestInfo,
+	serverURL string, reqInfo *requestInfo,
 	headerOpts ...HeaderOption,
 ) (int, error) {
 	var (
-		source      = ci.source
 		callerID    = reqInfo.callerID
 		name        = reqInfo.name
 		method      = reqInfo.method
 		body        = reqInfo.body
 		res         = reqInfo.res
 		respHandler = reqInfo.respHandler
+		url         = reqInfo.getURL(serverURL)
+		logFields   = append(reqInfo.logFields(),
+			zap.String("source", ci.source),
+			zap.String("url", url))
 	)
-	url = reqInfo.getURL(url)
-	logFields := []zap.Field{
-		zap.String("source", source),
-		zap.String("name", name),
-		zap.String("url", url),
-		zap.String("method", method),
-		zap.String("caller-id", callerID),
-	}
 	log.Debug("[pd] request the http url", logFields...)
 	req, err := http.NewRequestWithContext(ctx, method, url, bytes.NewBuffer(body))
 	if err != nil {
@@ -228,11 +242,14 @@ func (ci *clientInner) doRequest(
 		if readErr != nil {
 			logFields = append(logFields, zap.NamedError("read-body-error", err))
 		} else {
+			// API server will return a JSON body containing the detailed error message
+			// when the status code is not `http.StatusOK` 200.
+			bs = bytes.TrimSpace(bs)
 			logFields = append(logFields, zap.ByteString("body", bs))
 		}
 
 		log.Error("[pd] request failed with a non-200 status", logFields...)
-		return resp.StatusCode, errors.Errorf("request pd http api failed with status: '%s'", resp.Status)
+		return resp.StatusCode, errors.Errorf("request pd http api failed with status: '%s', body: '%s'", resp.Status, bs)
 	}
 
 	if res == nil {

client/http/request_info.go

@@ -18,6 +18,7 @@ import (
 	"fmt"
 
 	"github.com/tikv/pd/client/retry"
+	"go.uber.org/zap"
 )
 
 // The following constants are the names of the requests.
@@ -156,3 +157,13 @@ func (ri *requestInfo) WithTargetURL(targetURL string) *requestInfo {
 func (ri *requestInfo) getURL(addr string) string {
 	return fmt.Sprintf("%s%s", addr, ri.uri)
 }
+
+func (ri *requestInfo) logFields() []zap.Field {
+	return []zap.Field{
+		zap.String("caller-id", ri.callerID),
+		zap.String("name", ri.name),
+		zap.String("uri", ri.uri),
+		zap.String("method", ri.method),
+		zap.String("target-url", ri.targetURL),
+	}
+}

client/pd_service_discovery_test.go

@@ -29,6 +29,7 @@ import (
 	"github.com/pingcap/kvproto/pkg/pdpb"
 	"github.com/stretchr/testify/require"
 	"github.com/stretchr/testify/suite"
+	"github.com/tikv/pd/client/errs"
 	"github.com/tikv/pd/client/grpcutil"
 	"github.com/tikv/pd/client/testutil"
 	"google.golang.org/grpc"
@@ -205,7 +206,7 @@ func (suite *serviceClientTestSuite) TestServiceClient() {
 	re.NotNil(leaderConn)
 
 	_, err := pb.NewGreeterClient(followerConn).SayHello(suite.ctx, &pb.HelloRequest{Name: "pd"})
-	re.ErrorContains(err, "not leader")
+	re.ErrorContains(err, errs.NotLeaderErr)
 	resp, err := pb.NewGreeterClient(leaderConn).SayHello(suite.ctx, &pb.HelloRequest{Name: "pd"})
 	re.NoError(err)
 	re.Equal("Hello pd", resp.GetMessage())

client/resource_manager_client.go

@@ -16,7 +16,6 @@ package pd
 
 import (
 	"context"
-	"strings"
 	"time"
 
 	"github.com/gogo/protobuf/proto"
@@ -35,10 +34,6 @@ const (
 	modify                     actionType = 1
 	groupSettingsPathPrefix               = "resource_group/settings"
 	controllerConfigPathPrefix            = "resource_group/controller"
-	// errNotPrimary is returned when the requested server is not primary.
-	errNotPrimary = "not primary"
-	// errNotLeader is returned when the requested server is not pd leader.
-	errNotLeader = "not leader"
 )
 
 // GroupSettingsPathPrefixBytes is used to watch or get resource groups.
@@ -83,7 +78,7 @@ func (c *client) resourceManagerClient() (rmpb.ResourceManagerClient, error) {
 
 // gRPCErrorHandler is used to handle the gRPC error returned by the resource manager service.
 func (c *client) gRPCErrorHandler(err error) {
-	if strings.Contains(err.Error(), errNotPrimary) || strings.Contains(err.Error(), errNotLeader) {
+	if errs.IsLeaderChange(err) {
 		c.pdSvcDiscovery.ScheduleCheckMemberChanged()
 	}
 }

client/tso_dispatcher.go

@@ -522,7 +522,7 @@ tsoBatchLoop:
 			cancel()
 			stream = nil
 			// Because ScheduleCheckMemberChanged is asynchronous, if the leader changes, we better call `updateMember` ASAP.
-			if IsLeaderChange(err) {
+			if errs.IsLeaderChange(err) {
 				if err := bo.Exec(dispatcherCtx, c.svcDiscovery.CheckMemberChanged); err != nil {
 					select {
 					case <-dispatcherCtx.Done():

errors.toml

@@ -16,11 +16,21 @@ error = '''
 redirect failed
 '''
 
+["PD:apiutil:ErrRedirectNoLeader"]
+error = '''
+redirect finds no leader
+'''
+
 ["PD:apiutil:ErrRedirectToNotLeader"]
 error = '''
 redirect to not leader
 '''
 
+["PD:apiutil:ErrRedirectToNotPrimary"]
+error = '''
+redirect to not primary
+'''
+
 ["PD:autoscaling:ErrEmptyMetricsResponse"]
 error = '''
 metrics response from Prometheus is empty

pkg/core/region.go

@@ -44,6 +44,8 @@ import (
 const (
 	randomRegionMaxRetry = 10
 	scanRegionLimit      = 1000
+	// CollectFactor is the factor of collected region heartbeat.
+	CollectFactor = 0.9
 )
 
 // errRegionIsStale is error info for region is stale.
@@ -1396,6 +1398,12 @@ func (r *RegionsInfo) GetNotFromStorageRegionsCntByStore(storeID uint64) int {
 	return r.getNotFromStorageRegionsCntByStoreLocked(storeID)
 }
 
+// IsStorePrepared checks if a store is prepared.
+// For each store, the number of active regions should be more than total region of the store * CollectFactor
+func (r *RegionsInfo) IsStorePrepared(storeID uint64) bool {
+	return float64(r.GetNotFromStorageRegionsCntByStore(storeID)) >= float64(r.GetStoreRegionCount(storeID))*CollectFactor
+}
+
 // getNotFromStorageRegionsCntByStoreLocked gets the `NotFromStorageRegionsCnt` count of a store's leader, follower and learner by storeID.
 func (r *RegionsInfo) getNotFromStorageRegionsCntByStoreLocked(storeID uint64) int {
 	return r.leaders[storeID].notFromStorageRegionsCount() + r.followers[storeID].notFromStorageRegionsCount() + r.learners[storeID].notFromStorageRegionsCount()

pkg/errs/errno.go

@@ -195,10 +195,11 @@ var (
 
 // apiutil errors
 var (
-	ErrRedirect       = errors.Normalize("redirect failed", errors.RFCCodeText("PD:apiutil:ErrRedirect"))
-	ErrOptionNotExist = errors.Normalize("the option %s does not exist", errors.RFCCodeText("PD:apiutil:ErrOptionNotExist"))
-	// ErrRedirectToNotLeader is the error message for redirect to not leader.
-	ErrRedirectToNotLeader = errors.Normalize("redirect to not leader", errors.RFCCodeText("PD:apiutil:ErrRedirectToNotLeader"))
+	ErrRedirect             = errors.Normalize("redirect failed", errors.RFCCodeText("PD:apiutil:ErrRedirect"))
+	ErrOptionNotExist       = errors.Normalize("the option %s does not exist", errors.RFCCodeText("PD:apiutil:ErrOptionNotExist"))
+	ErrRedirectNoLeader     = errors.Normalize("redirect finds no leader", errors.RFCCodeText("PD:apiutil:ErrRedirectNoLeader"))
+	ErrRedirectToNotLeader  = errors.Normalize("redirect to not leader", errors.RFCCodeText("PD:apiutil:ErrRedirectToNotLeader"))
+	ErrRedirectToNotPrimary = errors.Normalize("redirect to not primary", errors.RFCCodeText("PD:apiutil:ErrRedirectToNotPrimary"))
 )
 
 // grpcutil errors

pkg/keyspace/keyspace.go

@@ -32,6 +32,7 @@ import (
 	"github.com/tikv/pd/pkg/storage/endpoint"
 	"github.com/tikv/pd/pkg/storage/kv"
 	"github.com/tikv/pd/pkg/utils/etcdutil"
+	"github.com/tikv/pd/pkg/utils/logutil"
 	"github.com/tikv/pd/pkg/utils/syncutil"
 	"go.uber.org/zap"
 )
@@ -328,7 +329,7 @@ func (manager *Manager) splitKeyspaceRegion(id uint32, waitRegionSplit bool) (er
 	if waitRegionSplit {
 		ranges := keyspaceRule.Data.([]*labeler.KeyRangeRule)
 		if len(ranges) < 2 {
-			log.Warn("[keyspace] failed to split keyspace region with insufficient range", zap.Any("label-rule", keyspaceRule))
+			log.Warn("[keyspace] failed to split keyspace region with insufficient range", logutil.ZapRedactString("label-rule", keyspaceRule.String()))
 			return ErrRegionSplitFailed
 		}
 		rawLeftBound, rawRightBound := ranges[0].StartKey, ranges[0].EndKey
@@ -377,7 +378,7 @@ func (manager *Manager) splitKeyspaceRegion(id uint32, waitRegionSplit bool) (er
 
 	log.Info("[keyspace] added region label for keyspace",
 		zap.Uint32("keyspace-id", id),
-		zap.Any("label-rule", keyspaceRule),
+		logutil.ZapRedactString("label-rule", keyspaceRule.String()),
 		zap.Duration("takes", time.Since(start)),
 	)
 	return