maintainer,heartbeatpb: add drain target plumbing

Author: hongyunyan

heartbeatpb/heartbeat.pb.go

@@ -139,6 +139,14 @@ message MaintainerHeartbeat {
     repeated MaintainerStatus statuses = 1;
 }
 
+// DrainProgress is maintainer-observed progress for dispatcher drain on one target.
+message DrainProgress {
+    string target_node_id = 1;
+    uint64 target_epoch = 2;
+    uint32 target_dispatcher_count = 3;
+    uint32 target_inflight_drain_move_count = 4;
+}
+
 message MaintainerStatus {
     ChangefeedID changefeedID = 1;
     string feed_state = 2;
@@ -147,6 +155,9 @@ message MaintainerStatus {
     repeated RunningError err = 5;
     bool bootstrap_done = 6;
     uint64 lastSyncedTs = 7;  // last synced ts of all tables in the changefeed, used in /:changefeed_id/synced API
+    // drain_progress reports the active dispatcher drain target observed by this maintainer.
+    // Nil means no active dispatcher drain target.
+    DrainProgress drain_progress = 8;
 }
 
 // NodeLiveness is node-reported liveness.
@@ -168,6 +179,10 @@ enum NodeLiveness {
 message NodeHeartbeat {
     NodeLiveness liveness = 1;
     uint64 node_epoch = 2;
+    // dispatcher_drain_target_* reports the manager-level dispatcher drain target
+    // currently applied on this node. Empty target means the drain target is clear.
+    string dispatcher_drain_target_node_id = 3;
+    uint64 dispatcher_drain_target_epoch = 4;
 }
 
 // SetNodeLivenessRequest asks a node to transition its local liveness.
@@ -182,6 +197,13 @@ message SetNodeLivenessResponse {
     uint64 node_epoch = 2;
 }
 
+// SetDispatcherDrainTargetRequest asks a maintainer manager to apply the dispatcher drain target.
+// target_node_id can be empty to clear the current target.
+message SetDispatcherDrainTargetRequest {
+    string target_node_id = 1;
+    uint64 target_epoch = 2;
+}
+
 message CoordinatorBootstrapRequest {
     int64 version = 1;
 }

heartbeatpb/heartbeat.proto

@@ -383,6 +383,12 @@ func (m *Maintainer) GetMaintainerStatus() *heartbeatpb.MaintainerStatus {
 	return status
 }
 
+// SetDispatcherDrainTarget applies the newest drain target to this maintainer.
+func (m *Maintainer) SetDispatcherDrainTarget(target node.ID, epoch uint64) {
+	m.controller.SetDispatcherDrainTarget(target, epoch)
+	m.statusChanged.Store(true)
+}
+
 func (m *Maintainer) initialize() error {
 	start := time.Now()
 	log.Info("start to initialize changefeed maintainer",

maintainer/maintainer.go

@@ -21,14 +21,15 @@ import (
 	"github.com/pingcap/ticdc/heartbeatpb"
 	"github.com/pingcap/ticdc/maintainer/operator"
 	"github.com/pingcap/ticdc/maintainer/replica"
+	mscheduler "github.com/pingcap/ticdc/maintainer/scheduler"
 	"github.com/pingcap/ticdc/maintainer/span"
 	"github.com/pingcap/ticdc/maintainer/split"
 	"github.com/pingcap/ticdc/pkg/common"
 	appcontext "github.com/pingcap/ticdc/pkg/common/context"
 	"github.com/pingcap/ticdc/pkg/config"
 	"github.com/pingcap/ticdc/pkg/messaging"
 	"github.com/pingcap/ticdc/pkg/node"
-	"github.com/pingcap/ticdc/pkg/scheduler"
+	pkgscheduler "github.com/pingcap/ticdc/pkg/scheduler"
 	"github.com/pingcap/ticdc/pkg/util"
 	"github.com/pingcap/ticdc/server/watcher"
 	"github.com/pingcap/ticdc/utils/threadpool"
@@ -43,7 +44,7 @@ type Controller struct {
 	bootstrapped bool
 	startTs      uint64
 
-	schedulerController    *scheduler.Controller
+	schedulerController    *pkgscheduler.Controller
 	operatorController     *operator.Controller
 	redoOperatorController *operator.Controller
 	spanController         *span.Controller
@@ -70,6 +71,10 @@ type Controller struct {
 
 	keyspaceMeta common.KeyspaceMeta
 	enableRedo   bool
+
+	// drainState keeps the latest dispatcher drain target visible to this
+	// maintainer even before drain-aware schedulers are introduced.
+	drainState *mscheduler.DrainState
 }
 
 func NewController(changefeedID common.ChangeFeedID,
@@ -135,6 +140,7 @@ func NewController(changefeedID common.ChangeFeedID,
 		splitter:               splitter,
 		keyspaceMeta:           keyspaceMeta,
 		enableRedo:             enableRedo,
+		drainState:             mscheduler.NewDrainState(),
 	}
 }
 
@@ -261,3 +267,15 @@ func (c *Controller) GetMinRedoCheckpointTs(minCheckpointTs uint64) uint64 {
 	minCheckpointTsForSpan := c.redoSpanController.GetMinCheckpointTsForNonReplicatingSpans(minCheckpointTs)
 	return min(minCheckpointTsForOperator, minCheckpointTsForSpan)
 }
+
+// SetDispatcherDrainTarget applies the newest drain target visible to this
+// changefeed. Older epochs are ignored so local state does not regress.
+func (c *Controller) SetDispatcherDrainTarget(target node.ID, epoch uint64) {
+	c.drainState.SetDispatcherDrainTarget(target, epoch)
+}
+
+// getDispatcherDrainTarget returns the current drain target snapshot used by
+// status reporting and later drain-aware schedulers.
+func (c *Controller) getDispatcherDrainTarget() (node.ID, uint64) {
+	return c.drainState.DispatcherDrainTarget()
+}