tso, server: add debug logs for TSO sync, closure, and forwarding paths (#10439)

YuhaoZhang00 · web-flow · commit b21a1839fd96 · 2026-04-09T02:12:51.000Z
ref #10329 Signed-off-by: Yuhao Zhang <yhzhang00@outlook.com>
diff --git a/metrics/grafana/pd.json b/metrics/grafana/pd.json
@@ -10967,7 +10967,7 @@
           "grid": {},
           "gridPos": {
             "h": 9,
-            "w": 12,
+            "w": 8,
             "x": 0,
             "y": 123
           },
@@ -11082,8 +11082,8 @@
           "grid": {},
           "gridPos": {
             "h": 9,
-            "w": 12,
-            "x": 12,
+            "w": 8,
+            "x": 8,
             "y": 123
           },
           "id": 1204,
@@ -11210,6 +11210,104 @@
             "alignLevel": null
           }
         },
+        {
+          "aliasColors": {},
+          "bars": false,
+          "dashLength": 10,
+          "dashes": false,
+          "datasource": "${DS_TEST-CLUSTER}",
+          "description": "The rate of TSO streams forwarded to the independent TSO service.",
+          "editable": true,
+          "error": false,
+          "fill": 1,
+          "grid": {},
+          "gridPos": {
+            "h": 9,
+            "w": 8,
+            "x": 16,
+            "y": 123
+          },
+          "id": 1631,
+          "legend": {
+            "alignAsTable": true,
+            "avg": false,
+            "current": true,
+            "hideEmpty": true,
+            "hideZero": true,
+            "max": true,
+            "min": false,
+            "rightSide": true,
+            "show": true,
+            "sideWidth": 300,
+            "total": false,
+            "values": true
+          },
+          "lines": true,
+          "linewidth": 1,
+          "links": [],
+          "nullPointMode": "null as zero",
+          "paceLength": 10,
+          "percentage": false,
+          "pointradius": 5,
+          "points": false,
+          "renderer": "flot",
+          "seriesOverrides": [],
+          "spaceLength": 10,
+          "stack": false,
+          "steppedLine": false,
+          "targets": [
+            {
+              "expr": "sum(rate(pd_server_tso_forward_stream_total{k8s_cluster=\"$k8s_cluster\", tidb_cluster=\"$tidb_cluster\", instance=~\"$instance\"}[1m])) by (instance)",
+              "format": "time_series",
+              "interval": "",
+              "intervalFactor": 2,
+              "legendFormat": "{{instance}}",
+              "refId": "A",
+              "step": 2
+            }
+          ],
+          "thresholds": [],
+          "timeFrom": null,
+          "timeRegions": [],
+          "timeShift": null,
+          "title": "Forward TSO streams count",
+          "tooltip": {
+            "msResolution": false,
+            "shared": true,
+            "sort": 0,
+            "value_type": "individual"
+          },
+          "type": "graph",
+          "xaxis": {
+            "buckets": null,
+            "mode": "time",
+            "name": null,
+            "show": true,
+            "values": []
+          },
+          "yaxes": [
+            {
+              "format": "ops",
+              "label": null,
+              "logBase": 1,
+              "max": null,
+              "min": null,
+              "show": true
+            },
+            {
+              "format": "short",
+              "label": null,
+              "logBase": 1,
+              "max": null,
+              "min": null,
+              "show": true
+            }
+          ],
+          "yaxis": {
+            "align": false,
+            "alignLevel": null
+          }
+        },
         {
           "aliasColors": {},
           "bars": false,
diff --git a/pkg/tso/allocator.go b/pkg/tso/allocator.go
@@ -131,7 +131,8 @@ func (a *Allocator) allocatorUpdater() {
 				continue
 			}
 			if err := a.UpdateTSO(); err != nil {
-				log.Warn("failed to update allocator's timestamp", append(a.logFields, errs.ZapError(err))...)
+				log.Warn("failed to update allocator's timestamp, resetting the TSO allocator with leadership resignation",
+					append(a.logFields, errs.ZapError(err))...)
 				a.Reset(true)
 				// To wait for the allocator to be re-initialized next time.
 				continue
diff --git a/pkg/tso/tso.go b/pkg/tso/tso.go
@@ -223,10 +223,13 @@ func (t *timestampOracle) syncTimestamp() error {
 	t.metrics.syncSaveDuration.Observe(time.Since(start).Seconds())
 
 	t.metrics.syncOKEvent.Inc()
+	// "last" is the etcd value, "last-saved" is the in-memory window, "save" is the
+	// newly persisted window, and "next" is the physical time loaded into memory
 	log.Info("sync and save timestamp",
 		logutil.CondUint32("keyspace-group-id", t.keyspaceGroupID, t.keyspaceGroupID > 0),
 		zap.Time("last", last), zap.Time("last-saved", lastSavedTime),
-		zap.Time("save", save), zap.Time("next", next))
+		zap.Time("save", save), zap.Time("next", next),
+		zap.String("member-name", t.member.Name()))
 	// save into memory
 	t.setTSOPhysical(next)
 	return nil
@@ -292,6 +295,11 @@ func (t *timestampOracle) resetUserTimestamp(tso uint64, ignoreSmaller, skipUppe
 		}
 		t.lastSavedTime.Store(save)
 		t.metrics.resetSaveDuration.Observe(time.Since(start).Seconds())
+		log.Info("persisted tso window to etcd (user-reset)",
+			logutil.CondUint32("keyspace-group-id", t.keyspaceGroupID, t.keyspaceGroupID > 0),
+			zap.Time("save", save), zap.Time("next", nextPhysical),
+			zap.String("member-name", t.member.Name()),
+		)
 	}
 	// save into memory only if nextPhysical or nextLogical is greater.
 	t.tsoMux.physical = nextPhysical
@@ -399,6 +407,11 @@ func (t *timestampOracle) updateTimestamp(purpose updatePurpose) (bool, error) {
 		}
 		t.lastSavedTime.Store(save)
 		t.metrics.updateSaveDuration.Observe(time.Since(start).Seconds())
+		log.Debug("persisted tso window to etcd (update)",
+			logutil.CondUint32("keyspace-group-id", t.keyspaceGroupID, t.keyspaceGroupID > 0),
+			zap.Time("save", save), zap.Time("next", next),
+			zap.String("member-name", t.member.Name()),
+		)
 	}
 	// If it's an IntervalUpdate, we don't need to check logical overflow, just update physical time directly.
 	// Otherwise, the caller met logical overflow, so it will allocate physical time to alloc more timestamp in concurrent.
diff --git a/server/cluster/cluster.go b/server/cluster/cluster.go
@@ -120,6 +120,12 @@ const (
 	syncRegionTaskRunner = "sync-region-async"
 )
 
+const (
+	tsoDynamicSwitchingStateUnknown int32 = iota
+	tsoDynamicSwitchingStateEnabled
+	tsoDynamicSwitchingStateDisabled
+)
+
 // Server is the interface for cluster.
 type Server interface {
 	GetAllocator() id.Allocator
@@ -161,6 +167,7 @@ type RaftCluster struct {
 
 	running                bool
 	isKeyspaceGroupEnabled bool
+	tsoDynamicSwitchingState atomic.Int32
 	meta                   *metapb.Cluster
 	storage                storage.Storage
 	minResolvedTS          atomic.Value // Store as uint64
@@ -457,6 +464,10 @@ func (c *RaftCluster) checkSchedulingService() {
 func (c *RaftCluster) checkTSOService() {
 	if c.isKeyspaceGroupEnabled {
 		if c.opt.GetMicroserviceConfig().IsTSODynamicSwitchingEnabled() {
+			prev := c.tsoDynamicSwitchingState.Swap(tsoDynamicSwitchingStateEnabled)
+			if prev == tsoDynamicSwitchingStateDisabled {
+				log.Info("TSO dynamic switching is enabled, resuming TSO service checks")
+			}
 			servers, err := discovery.Discover(c.etcdClient, constant.TSOServiceName)
 			if err != nil || len(servers) == 0 {
 				if err := c.startTSOJobsIfNeeded(); err != nil {
@@ -474,6 +485,11 @@ func (c *RaftCluster) checkTSOService() {
 					c.SetServiceIndependent(constant.TSOServiceName)
 				}
 			}
+		} else {
+			prev := c.tsoDynamicSwitchingState.Swap(tsoDynamicSwitchingStateDisabled)
+			if prev != tsoDynamicSwitchingStateDisabled {
+				log.Info("TSO dynamic switching is disabled by config, skipping TSO service checks")
+			}
 		}
 		return
 	}
@@ -541,7 +557,7 @@ func (c *RaftCluster) stopTSOJobsIfNeeded() {
 	if !c.tsoAllocator.IsInitialize() {
 		return
 	}
-	log.Info("closing the TSO allocator")
+	log.Info("closing the embedded TSO allocator")
 	c.tsoAllocator.Reset(false)
 	failpoint.Inject("updateAfterResetTSO", func() {
 		if err := c.tsoAllocator.UpdateTSO(); !errorspkg.Is(err, errs.ErrUpdateTimestamp) {
diff --git a/server/grpc_service.go b/server/grpc_service.go
@@ -485,6 +485,7 @@ func (s *GrpcServer) Tso(stream pdpb.PD_TsoServer) error {
 		defer done()
 	}
 	if s.IsServiceIndependent(constant.TSOServiceName) {
+		tsoForwardStreamCounter.Inc()
 		return s.forwardToTSOService(stream)
 	}
 
diff --git a/server/metrics.go b/server/metrics.go
@@ -95,6 +95,13 @@ var (
 			Name:      "tso_proxy_forward_timeout_total",
 			Help:      "Counter of timeouts when tso proxy forwarding tso requests to tso service.",
 		})
+	tsoForwardStreamCounter = prometheus.NewCounter(
+		prometheus.CounterOpts{
+			Namespace: "pd",
+			Subsystem: "server",
+			Name:      "tso_forward_stream_total",
+			Help:      "Counter of TSO streams forwarded to the independent TSO service.",
+		})
 
 	tsoHandleDuration = prometheus.NewHistogram(
 		prometheus.HistogramOpts{
@@ -221,6 +228,7 @@ func init() {
 	prometheus.MustRegister(tsoProxyHandleDuration)
 	prometheus.MustRegister(tsoProxyBatchSize)
 	prometheus.MustRegister(tsoProxyForwardTimeoutCounter)
+	prometheus.MustRegister(tsoForwardStreamCounter)
 	prometheus.MustRegister(tsoHandleDuration)
 	prometheus.MustRegister(tsoBatchSize)
 	prometheus.MustRegister(queryRegionDuration)

Original file line number	Diff line number	Diff line change
`@@ -485,6 +485,7 @@ func (s *GrpcServer) Tso(stream pdpb.PD_TsoServer) error {`
`485`	`485`	`defer done()`
`486`	`486`	`}`
`487`	`487`	`if s.IsServiceIndependent(constant.TSOServiceName) {`
	`488`	`+ tsoForwardStreamCounter.Inc()`
`488`	`489`	`return s.forwardToTSOService(stream)`
`489`	`490`	`}`
`490`	`491`