CEP-45: Incremental Repair integration w/Mutation Tracking

Implement mutation tracking repair as a new repair task that replaces
Merkle tree validation and streaming for tracked keyspaces. Instead of
building hash trees and comparing data, MutationTrackingSyncCoordinator
sends MT_SYNC_REQ to all participants to collect their witnessed offsets,
then waits for background offset broadcasts to confirm all replicas have
reconciled to those target offsets.

Key changes:

- Add MutationTrackingIncrementalRepairTask that creates per-range
  MutationTrackingSyncCoordinator instances and blocks until all
  shards reach the target reconciled offsets or timeout.

- Add MT_SYNC_REQ/MT_SYNC_RSP verbs and MutationTrackingSyncRequest/
  MutationTrackingSyncResponse messages for the repair protocol to
  establish a happens-before relationship between repair start and
  offset collection.

- RepairCoordinator.create() factory method snapshots TCM state to
  decide whether to use mutation tracking repair and flips incremental
  to false (skipping anti-compaction) when MT is active without
  migration.

- Support mutation tracking migration: during untracked->tracked
  migration, run incremental repair first (for pre-migration data),
  then MT sync. KeyspaceMigrationInfo validates that repair ranges
  don't partially overlap pending migration ranges and routes
  streaming/SSTable finalization through the correct tracked vs
  untracked path.

- Temporarily route read repair mutations through the untracked write
  path during migration by adding isReadRepair flag to Mutation and
  bypassing migration routing checks in ReadRepairVerbHandler and
  CassandraKeyspaceWriteHandler. This is a stopgap; CASSANDRA-21252
  will roll back this approach and handle read repair properly.

- Add offset collection APIs to CoordinatorLog, Node2OffsetsMap, and
  Shard for computing union and intersection of witnessed offsets
  scoped to specific participant host IDs (supporting --force with
  dead node exclusion).

- Add configurable mutation_tracking_sync_timeout (default 2m) with
  JMX get/set on StorageServiceMBean.

- Fix ActiveRepairService to use tryFailure() instead of setFailure()
  to avoid double-completion exceptions during concurrent repair
  failures.

Co-Authored-By: Ariel Weisberg <aweisberg@apple.com>

src/java/org/apache/cassandra/config/Config.java

@@ -169,6 +169,8 @@ public class Config
     @Replaces(oldName = "repair_request_timeout_in_ms", converter = Converters.MILLIS_DURATION_LONG, deprecated = true)
     public volatile DurationSpec.LongMillisecondsBound repair_request_timeout = new DurationSpec.LongMillisecondsBound("120000ms");
 
+    public volatile DurationSpec.LongMillisecondsBound mutation_tracking_sync_timeout = new DurationSpec.LongMillisecondsBound("2m");
+
     public Integer streaming_connections_per_host = 1;
     @Replaces(oldName = "streaming_keep_alive_period_in_secs", converter = Converters.SECONDS_DURATION, deprecated = true)
     public DurationSpec.IntSecondsBound streaming_keep_alive_period = new DurationSpec.IntSecondsBound("300s");

src/java/org/apache/cassandra/config/DatabaseDescriptor.java

@@ -2592,6 +2592,16 @@ public class DatabaseDescriptor
         conf.repair_request_timeout = new DurationSpec.LongMillisecondsBound(timeOutInMillis);
     }
 
+    public static long getMutationTrackingSyncTimeout(TimeUnit unit)
+    {
+        return conf.mutation_tracking_sync_timeout.to(unit);
+    }
+
+    public static void setMutationTrackingSyncTimeout(long timeoutInMillis)
+    {
+        conf.mutation_tracking_sync_timeout = new DurationSpec.LongMillisecondsBound(timeoutInMillis);
+    }
+
     public static boolean hasCrossNodeTimeout()
     {
         return conf.internode_timeout;

src/java/org/apache/cassandra/db/AbstractMutationVerbHandler.java

@@ -202,6 +202,13 @@ public abstract class AbstractMutationVerbHandler<T extends IMutation> implement
      */
     protected ClusterMetadata checkReplicationMigration(ClusterMetadata metadata, Message<T> message, InetAddressAndPort respondTo)
     {
+        // Read repair mutations always bypass mutation tracking and use the untracked
+        // write path, so skip the replication migration routing check. The isReadRepair
+        // flag on the mutation hasn't been set yet at this point — it's set later in
+        // applyMutation() — so we check the handler type instead.
+        if (this instanceof ReadRepairVerbHandler)
+            return metadata;
+
         IMutation mutation = message.payload;
         MutationRouting expected = mutation.id().isNone() ? MutationRouting.UNTRACKED : MutationRouting.TRACKED;
         if (expected == MigrationRouter.getMutationRouting(metadata, mutation))

src/java/org/apache/cassandra/db/CassandraKeyspaceWriteHandler.java

@@ -47,7 +47,8 @@ public class CassandraKeyspaceWriteHandler implements KeyspaceWriteHandler
         {
             group = Keyspace.writeOrder.start();
 
-            MigrationRouter.validateUntrackedMutation(mutation);
+            if (!mutation.isReadRepair())
+                MigrationRouter.validateUntrackedMutation(mutation);
             // write the mutation to the commitlog and memtables
             CommitLogPosition position = null;
             if (makeDurable)

src/java/org/apache/cassandra/db/Keyspace.java

@@ -437,7 +437,7 @@ public class Keyspace
                       boolean updateIndexes,
                       boolean isDroppable)
     {
-        if (MigrationRouter.isFullyTracked(mutation))
+        if (MigrationRouter.isFullyTracked(mutation) && !mutation.isReadRepair())
             applyInternalTracked(mutation, null);
         else
             applyInternal(mutation, makeDurable, updateIndexes, isDroppable, false, null);
@@ -460,7 +460,7 @@ public class Keyspace
                                                boolean isDeferrable,
                                                Promise<?> future)
     {
-        Preconditions.checkState(!getMetadata().useMutationTracking() && mutation.id().isNone());
+        Preconditions.checkState((!getMetadata().useMutationTracking() || mutation.isReadRepair()) && mutation.id().isNone());
 
         if (TEST_FAIL_WRITES && getMetadata().name.equals(TEST_FAIL_WRITES_KS))
             throw new RuntimeException("Testing write failures");

src/java/org/apache/cassandra/db/Mutation.java

@@ -108,6 +108,21 @@ public class Mutation implements IMutation, Supplier<Mutation>, Commitable
     // because it is being applied by one or in a context where transaction conflicts don't occur
     private PotentialTxnConflicts potentialTxnConflicts;
 
+    // Transient: not serialized on the wire. Set by ReadRepairVerbHandler on the
+    // receiving side so downstream code (Keyspace.apply, write handlers) can route
+    // read repair mutations through the untracked write path during migration.
+    private transient boolean isReadRepair;
+
+    public void setReadRepair(boolean readRepair)
+    {
+        this.isReadRepair = readRepair;
+    }
+
+    public boolean isReadRepair()
+    {
+        return isReadRepair;
+    }
+
     public Mutation(MutationId id, PartitionUpdate update)
     {
         this(id, update.metadata().keyspace, update.partitionKey(), ImmutableMap.of(update.metadata().id, update), approxTime.now(), update.metadata().params.cdc, PotentialTxnConflicts.DISALLOW);
@@ -148,7 +163,9 @@ public class Mutation implements IMutation, Supplier<Mutation>, Commitable
     @Override
     public Mutation withMutationId(MutationId mutationId)
     {
-        return new Mutation(mutationId, keyspaceName, key, modifications, approxCreatedAtNanos, cdcEnabled, potentialTxnConflicts);
+        Mutation m = new Mutation(mutationId, keyspaceName, key, modifications, approxCreatedAtNanos, cdcEnabled, potentialTxnConflicts);
+        m.isReadRepair = this.isReadRepair;
+        return m;
     }
 
     private static boolean cdcEnabled(Iterable<PartitionUpdate> modifications)
@@ -182,7 +199,9 @@ public class Mutation implements IMutation, Supplier<Mutation>, Commitable
 
         Map<TableId, PartitionUpdate> updates = builder.build();
         checkState(!updates.isEmpty(), "Updates should not be empty");
-        return new Mutation(id, keyspaceName, key, builder.build(), approxCreatedAtNanos, potentialTxnConflicts);
+        Mutation result = new Mutation(id, keyspaceName, key, builder.build(), approxCreatedAtNanos, potentialTxnConflicts);
+        result.isReadRepair = this.isReadRepair;
+        return result;
     }
 
     public @Nullable Mutation without(TableId tableId)

src/java/org/apache/cassandra/db/ReadRepairVerbHandler.java

@@ -27,6 +27,7 @@ public class ReadRepairVerbHandler extends AbstractMutationVerbHandler<Mutation>
 
     public void applyMutation(Mutation mutation)
     {
+        mutation.setReadRepair(true);
         mutation.apply();
     }
 

src/java/org/apache/cassandra/db/lifecycle/Tracker.java

@@ -63,6 +63,7 @@ import org.apache.cassandra.notifications.TableDroppedNotification;
 import org.apache.cassandra.notifications.TablePreScrubNotification;
 import org.apache.cassandra.notifications.TruncationNotification;
 import org.apache.cassandra.replication.ImmutableCoordinatorLogOffsets;
+import org.apache.cassandra.tcm.ClusterMetadata;
 import org.apache.cassandra.utils.Pair;
 import org.apache.cassandra.utils.Throwables;
 import org.apache.cassandra.utils.TimeUUID;
@@ -272,7 +273,10 @@ public class Tracker
 
     public void addSSTables(Collection<SSTableReader> sstables)
     {
-        Preconditions.checkState(!cfstore.metadata().replicationType().isTracked());
+        // Tracked tables may legitimately use this path during migration from untracked to tracked,
+        // when incremental repair streams SSTables that were written before tracking was enabled.
+        Preconditions.checkState(!cfstore.metadata().replicationType().isTracked()
+                                 || ClusterMetadata.current().mutationTrackingMigrationState.isMigrating(cfstore.metadata().keyspace));
         addSSTablesInternal(sstables, false, true, true);
     }
 

src/java/org/apache/cassandra/db/streaming/CassandraStreamReceiver.java

@@ -53,6 +53,7 @@ import org.apache.cassandra.service.accord.AccordTopology;
 import org.apache.cassandra.service.accord.IAccordService;
 import org.apache.cassandra.service.accord.TimeOnlyRequestBookkeeping.LatencyRequestBookkeeping;
 import org.apache.cassandra.replication.MutationTrackingService;
+import org.apache.cassandra.service.replication.migration.KeyspaceMigrationInfo;
 import org.apache.cassandra.replication.PendingLocalTransfer;
 import org.apache.cassandra.streaming.IncomingStream;
 import org.apache.cassandra.streaming.StreamReceiver;
@@ -104,6 +105,19 @@ public class CassandraStreamReceiver implements StreamReceiver
         this.requiresWritePath = requiresWritePath(cfs);
     }
 
+    /**
+     * Whether this stream should use the tracked transfer path (pending until activation).
+     * Returns false during mutation tracking migration for ranges that are still pending,
+     * since migration repair uses the untracked streaming path for those ranges.
+     */
+    private boolean useTrackedTransferPath()
+    {
+        if (!cfs.metadata().replicationType().isTracked() || !session.streamOperation().isTrackable())
+            return false;
+
+        return KeyspaceMigrationInfo.shouldUseTrackedTransfers(ClusterMetadata.current(), cfs.getKeyspaceName(), cfs.metadata().id, ranges);
+    }
+
     public static CassandraStreamReceiver fromReceiver(StreamReceiver receiver)
     {
         Preconditions.checkArgument(receiver instanceof CassandraStreamReceiver);
@@ -135,7 +149,7 @@ public class CassandraStreamReceiver implements StreamReceiver
         sstables.addAll(finished);
         receivedEntireSSTable = file.isEntireSSTable();
 
-        if (cfs.metadata().replicationType().isTracked() && session.streamOperation().isTrackable())
+        if (useTrackedTransferPath())
         {
             PendingLocalTransfer transfer = new PendingLocalTransfer(cfs.metadata().id, session.planId(), sstables);
             MutationTrackingService.instance().received(transfer);
@@ -266,7 +280,7 @@ public class CassandraStreamReceiver implements StreamReceiver
                 logger.debug("[Stream #{}] Received {} sstables from {} ({})", session.planId(), readers.size(), session.peer, readers);
 
                 // SSTables involved in a coordinated transfer become live when the transfer is activated
-                if (cfs.metadata().replicationType().isTracked() && session.streamOperation().isTrackable())
+                if (useTrackedTransferPath())
                     return;
 
                 if (session.streamOperation() == StreamOperation.BOOTSTRAP)

src/java/org/apache/cassandra/dht/NormalizedRanges.java

@@ -134,6 +134,19 @@ public class NormalizedRanges<T extends RingPosition<T>> extends AbstractList<Ra
         return isIn;
     }
 
+    /**
+     * Check if any of these ranges intersect with the given range.
+     */
+    public boolean intersects(Range<T> range)
+    {
+        for (Range<T> r : this)
+        {
+            if (r.intersects(range))
+                return true;
+        }
+        return false;
+    }
+
     public NormalizedRanges<T> subtract(NormalizedRanges<T> b)
     {
         if (b.isEmpty())

src/java/org/apache/cassandra/io/sstable/format/SSTableWriter.java

@@ -62,6 +62,8 @@ import org.apache.cassandra.io.util.MmappedRegionsCache;
 import org.apache.cassandra.replication.ImmutableCoordinatorLogOffsets;
 import org.apache.cassandra.replication.MutationTrackingService;
 import org.apache.cassandra.service.ActiveRepairService;
+import org.apache.cassandra.service.replication.migration.KeyspaceMigrationInfo;
+import org.apache.cassandra.tcm.ClusterMetadata;
 import org.apache.cassandra.utils.Clock;
 import org.apache.cassandra.utils.Throwables;
 import org.apache.cassandra.utils.TimeUUID;
@@ -340,13 +342,25 @@ public abstract class SSTableWriter extends SSTable implements Transactional
         // Reconciliation should not occur before activation for coordinated transfer streams for tracked keyspaces.
         boolean reconcile = txn.opType() != OperationType.STREAM;
 
+        // During migration, incremental repair handles repair status for ranges still pending migration.
+        // Only apply mutation tracking reconciliation for ranges NOT in the migration pending set.
+        // For SSTables whose range falls within pending migration ranges, IR sets pendingRepair/repairedAt.
         if (metadata().replicationType().isTracked() && repairedAt == ActiveRepairService.UNREPAIRED_SSTABLE && reconcile)
         {
-            Preconditions.checkState(Objects.equals(pendingRepair, ActiveRepairService.NO_PENDING_REPAIR));
-            if (MutationTrackingService.instance().isDurablyReconciled(coordinatorLogOffsets))
+            KeyspaceMigrationInfo migrationInfo = ClusterMetadata.current().mutationTrackingMigrationState.getKeyspaceInfo(metadata().keyspace);
+            boolean inMigrationPendingRange = migrationInfo != null
+                                              && migrationInfo.isRangeInPendingMigration(metadata().id,
+                                                                                         first.getToken(),
+                                                                                         last.getToken());
+
+            if (!inMigrationPendingRange)
             {
-                repairedAt = Clock.Global.currentTimeMillis();
-                logger.debug("Marking SSTable {} as reconciled with repairedAt {}", descriptor, repairedAt);
+                Preconditions.checkState(Objects.equals(pendingRepair, ActiveRepairService.NO_PENDING_REPAIR));
+                if (MutationTrackingService.instance().isDurablyReconciled(coordinatorLogOffsets))
+                {
+                    repairedAt = Clock.Global.currentTimeMillis();
+                    logger.debug("Marking SSTable {} as reconciled with repairedAt {}", descriptor, repairedAt);
+                }
             }
         }
 

src/java/org/apache/cassandra/net/Verb.java

@@ -64,6 +64,8 @@ import org.apache.cassandra.repair.messages.FailSession;
 import org.apache.cassandra.repair.messages.FinalizeCommit;
 import org.apache.cassandra.repair.messages.FinalizePromise;
 import org.apache.cassandra.repair.messages.FinalizePropose;
+import org.apache.cassandra.repair.messages.MutationTrackingSyncRequest;
+import org.apache.cassandra.repair.messages.MutationTrackingSyncResponse;
 import org.apache.cassandra.repair.messages.PrepareConsistentRequest;
 import org.apache.cassandra.repair.messages.PrepareConsistentResponse;
 import org.apache.cassandra.repair.messages.PrepareMessage;
@@ -287,8 +289,10 @@ public enum Verb
     FINALIZE_PROMISE_MSG   (111, P1, repairWithBackoffTimeout,      ANTI_ENTROPY,      () -> FinalizePromise.serializer,           () -> RepairMessageVerbHandler.instance(),   REPAIR_RSP          ),
     FINALIZE_COMMIT_MSG    (112, P1, repairWithBackoffTimeout,      ANTI_ENTROPY,      () -> FinalizeCommit.serializer,            () -> RepairMessageVerbHandler.instance(),   REPAIR_RSP          ),
     FAILED_SESSION_MSG     (113, P1, repairWithBackoffTimeout,      ANTI_ENTROPY,      () -> FailSession.serializer,               () -> RepairMessageVerbHandler.instance(),   REPAIR_RSP          ),
-    STATUS_RSP             (115, P1, repairTimeout,   ANTI_ENTROPY,      () -> StatusResponse.serializer,            () -> RepairMessageVerbHandler.instance(),   REPAIR_RSP          ),
-    STATUS_REQ             (114, P1, repairTimeout,   ANTI_ENTROPY,      () -> StatusRequest.serializer,             () -> RepairMessageVerbHandler.instance(),   REPAIR_RSP          ),
+    STATUS_RSP             (115, P1, repairTimeout,                 ANTI_ENTROPY,      () -> StatusResponse.serializer,            () -> RepairMessageVerbHandler.instance(),   REPAIR_RSP          ),
+    STATUS_REQ             (114, P1, repairTimeout,                 ANTI_ENTROPY,      () -> StatusRequest.serializer,             () -> RepairMessageVerbHandler.instance(),   REPAIR_RSP          ),
+    MT_SYNC_RSP            (117, P1, repairWithBackoffTimeout,      REQUEST_RESPONSE,  () -> MutationTrackingSyncResponse.serializer, RESPONSE_HANDLER                                              ),
+    MT_SYNC_REQ            (116, P1, repairWithBackoffTimeout,      ANTI_ENTROPY,      () -> MutationTrackingSyncRequest.serializer, () -> RepairMessageVerbHandler.instance(), MT_SYNC_RSP         ),
 
     REPLICATION_DONE_RSP   (82,  P0, rpcTimeout,      MISC,              () -> NoPayload.serializer,                 RESPONSE_HANDLER                             ),
     REPLICATION_DONE_REQ   (22,  P0, rpcTimeout,      MISC,              () -> NoPayload.serializer,                 () -> ReplicationDoneVerbHandler.instance, REPLICATION_DONE_RSP),

src/java/org/apache/cassandra/repair/MutationTrackingIncrementalRepairTask.java

@@ -0,0 +1,229 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.cassandra.repair;
+
+import java.util.ArrayList;
+import java.util.Collection;
+import java.util.List;
+import java.util.concurrent.TimeUnit;
+
+import org.apache.cassandra.concurrent.ExecutorPlus;
+import org.apache.cassandra.config.DatabaseDescriptor;
+import org.apache.cassandra.dht.Range;
+import org.apache.cassandra.dht.Token;
+import org.apache.cassandra.replication.MutationTrackingSyncCoordinator;
+import org.apache.cassandra.schema.KeyspaceMetadata;
+import org.apache.cassandra.tcm.ClusterMetadata;
+import org.apache.cassandra.utils.Throwables;
+import org.apache.cassandra.utils.TimeUUID;
+import org.apache.cassandra.utils.concurrent.AsyncPromise;
+import org.apache.cassandra.utils.concurrent.Future;
+
+import static com.google.common.base.Preconditions.checkState;
+
+/** Repair task that syncs mutation tracking offsets across replicas */
+public class MutationTrackingIncrementalRepairTask extends AbstractRepairTask
+{
+
+    private final TimeUUID parentSession;
+    private final RepairCoordinator.NeighborsAndRanges neighborsAndRanges;
+    private final String[] cfnames;
+    private final ClusterMetadata metadata;
+
+    protected MutationTrackingIncrementalRepairTask(RepairCoordinator coordinator,
+                                                    TimeUUID parentSession,
+                                                    RepairCoordinator.NeighborsAndRanges neighborsAndRanges,
+                                                    String[] cfnames)
+    {
+        super(coordinator);
+        this.parentSession = parentSession;
+        this.neighborsAndRanges = neighborsAndRanges;
+        this.cfnames = cfnames;
+        this.metadata = coordinator.metadata;
+    }
+
+    @Override
+    public String name()
+    {
+        return "MutationTrackingRepair";
+    }
+
+    @Override
+    public Future<CoordinatedRepairResult> performUnsafe(ExecutorPlus executor, Scheduler validationScheduler)
+    {
+        List<CommonRange> allRanges = neighborsAndRanges.filterCommonRanges(keyspace, cfnames);
+        checkState(!allRanges.isEmpty(), "No ranges to repair");
+
+        List<MutationTrackingSyncCoordinator> syncCoordinators = new ArrayList<>();
+        List<Collection<Range<Token>>> rangeCollections = new ArrayList<>();
+
+        for (CommonRange commonRange : allRanges)
+        {
+            for (Range<Token> range : commonRange.ranges)
+            {
+                RepairJobDesc desc = new RepairJobDesc(parentSession, TimeUUID.Generator.nextTimeUUID(),
+                                                       keyspace, "Mutation Tracking Sync", List.of(range));
+                MutationTrackingSyncCoordinator syncCoordinator =
+                    new MutationTrackingSyncCoordinator(coordinator.ctx, desc, commonRange.endpoints, metadata);
+                syncCoordinator.start();
+                syncCoordinators.add(syncCoordinator);
+                rangeCollections.add(List.of(range));
+
+                logger.info("Started mutation tracking sync for range {}", range);
+            }
+        }
+
+        coordinator.notifyProgress("Started mutation tracking sync for " + syncCoordinators.size() + " ranges");
+
+        AsyncPromise<CoordinatedRepairResult> resultPromise = new AsyncPromise<>();
+
+        executor.execute(() -> {
+            try
+            {
+                waitForSyncCompletion(syncCoordinators, rangeCollections, resultPromise);
+            }
+            catch (InterruptedException e)
+            {
+                try
+                {
+                    resultPromise.tryFailure(new RuntimeException("Interrupted waiting for Mutation Tracking sync coordinators to finish", e));
+                }
+                finally
+                {
+                    Thread.currentThread().interrupt();
+                }
+            }
+            catch (Exception e)
+            {
+                logger.error("Error during mutation tracking repair", e);
+                resultPromise.tryFailure(e);
+            }
+        });
+
+        return resultPromise;
+    }
+
+    private void waitForSyncCompletion(List<MutationTrackingSyncCoordinator> syncCoordinators,
+                                       List<Collection<Range<Token>>> rangeCollections,
+                                       AsyncPromise<CoordinatedRepairResult> resultPromise) throws Exception
+    {
+        long deadlineNanos = coordinator.ctx.clock().nanoTime() + TimeUnit.MILLISECONDS.toNanos(
+            DatabaseDescriptor.getMutationTrackingSyncTimeout(TimeUnit.MILLISECONDS));
+        coordinator.ctx.scheduledTasks().schedule(() -> {
+            for (MutationTrackingSyncCoordinator syncCoordinator : syncCoordinators)
+            {
+                try
+                {
+                    syncCoordinator.timeout();
+                }
+                catch (Exception e)
+                {
+                    logger.error("Exception cancelling mutation tracking sync coordinator", e);
+                }
+            }
+        }, deadlineNanos - coordinator.ctx.clock().nanoTime(), TimeUnit.NANOSECONDS);
+
+        Exception error = null;
+        for (MutationTrackingSyncCoordinator syncCoordinator : syncCoordinators)
+        {
+            try
+            {
+                syncCoordinator.awaitCompletion();
+            }
+            catch (InterruptedException e)
+            {
+                if (error != null)
+                    e.addSuppressed(error);
+                for (MutationTrackingSyncCoordinator c : syncCoordinators)
+                {
+                    try
+                    {
+                        c.cancel();
+                    }
+                    catch (Exception e2)
+                    {
+                        e.addSuppressed(e2);
+                    }
+                }
+                throw e;
+            }
+            catch (Exception e)
+            {
+                error = Throwables.merge(error, e);
+            }
+        }
+
+        if (error != null)
+        {
+            logger.warn("Mutation tracking sync failed for keyspace {}", keyspace, error);
+            throw error;
+        }
+
+        coordinator.notifyProgress("Mutation tracking sync completed for all ranges");
+
+        List<RepairSessionResult> results = new ArrayList<>();
+        for (int i = 0; i < rangeCollections.size(); i++)
+        {
+            Collection<Range<Token>> ranges = rangeCollections.get(i);
+            results.add(new RepairSessionResult(parentSession, keyspace, ranges, List.of(), false));
+        }
+        resultPromise.trySuccess(CoordinatedRepairResult.create(rangeCollections, results));
+    }
+
+    /**
+     * Determines if this keyspace should use mutation tracking incremental repair.
+     * Returns true if:
+     * - Keyspace uses mutation tracking replication, OR
+     * - Keyspace is currently migrating (either direction)
+     *
+     * @param metadata the snapshotted cluster metadata to evaluate against
+     * @param keyspace the keyspace name to check
+     */
+    public static boolean shouldUseMutationTrackingRepair(ClusterMetadata metadata, String keyspace)
+    {
+        KeyspaceMetadata ksm = metadata.schema.maybeGetKeyspaceMetadata(keyspace).orElse(null);
+        if (ksm == null)
+            return false;
+
+        // Check if keyspace uses mutation tracking
+        if (ksm.useMutationTracking())
+            return true;
+
+        // For tracked→untracked migration (keyspace is currently untracked but migration is in progress),
+        // use regular incremental repair instead of MT repair. The MT sync step can't complete for this
+        // direction because streaming doesn't update mutation tracking offsets, and the keyspace is moving
+        // away from tracking. Regular incremental repair will sync the data and the RepairJob callback
+        // handler will still advance the migration state.
+        // TODO (desired): This is an over simplification in that depending on which ranges are migrated we might be able to just run MT sync but running IR should also be fine
+        return false;
+    }
+
+    /**
+     * Determines if a mutation tracking migration is in progress for this keyspace.
+     * Returns true during migration:
+     * - Migrating TO mutation tracking: need traditional repair to sync pre-migration data
+     * - Migrating FROM mutation tracking: need traditional repair for post-migration consistency
+     *
+     * @param metadata the snapshotted cluster metadata to evaluate against
+     * @param keyspace the keyspace name to check
+     */
+    public static boolean isMutationTrackingMigrationInProgress(ClusterMetadata metadata, String keyspace)
+    {
+        return metadata.mutationTrackingMigrationState.isMigrating(keyspace);
+    }
+}

src/java/org/apache/cassandra/repair/RepairCoordinator.java

@@ -53,6 +53,7 @@ import org.apache.cassandra.cql3.statements.SelectStatement;
 import org.apache.cassandra.db.ColumnFamilyStore;
 import org.apache.cassandra.db.ConsistencyLevel;
 import org.apache.cassandra.db.Keyspace;
+import org.apache.cassandra.dht.NormalizedRanges;
 import org.apache.cassandra.dht.Range;
 import org.apache.cassandra.dht.Token;
 import org.apache.cassandra.exceptions.RepairException;
@@ -67,12 +68,14 @@ import org.apache.cassandra.repair.messages.RepairMessage;
 import org.apache.cassandra.repair.messages.RepairOption;
 import org.apache.cassandra.repair.state.CoordinatorState;
 import org.apache.cassandra.repair.state.ParticipateState;
+import org.apache.cassandra.schema.KeyspaceMetadata;
 import org.apache.cassandra.schema.SchemaConstants;
 import org.apache.cassandra.schema.SystemDistributedKeyspace;
 import org.apache.cassandra.schema.TableMetadata;
 import org.apache.cassandra.service.ActiveRepairService;
 import org.apache.cassandra.service.ActiveRepairService.ParentRepairStatus;
 import org.apache.cassandra.service.ClientState;
+import org.apache.cassandra.service.replication.migration.KeyspaceMigrationInfo;
 import org.apache.cassandra.service.StorageService;
 import org.apache.cassandra.tcm.ClusterMetadata;
 import org.apache.cassandra.tcm.Epoch;
@@ -86,6 +89,7 @@ import org.apache.cassandra.utils.Pair;
 import org.apache.cassandra.utils.Throwables;
 import org.apache.cassandra.utils.TimeUUID;
 import org.apache.cassandra.utils.WrappedRunnable;
+import org.apache.cassandra.utils.concurrent.AsyncPromise;
 import org.apache.cassandra.utils.concurrent.Future;
 import org.apache.cassandra.utils.progress.ProgressEvent;
 import org.apache.cassandra.utils.progress.ProgressEventNotifier;
@@ -113,20 +117,66 @@ public class RepairCoordinator implements Runnable, ProgressEventNotifier, Repai
     final SharedContext ctx;
     final Scheduler validationScheduler;
 
+    // Mutation tracking decision, snapshotted once at creation time from TCM
+    final boolean useMutationTracking;
+    final boolean mutationTrackingMigrationInProgress;
+    final ClusterMetadata metadata;
+
     private TraceState traceState;
 
-    public RepairCoordinator(StorageService storageService, int cmd, RepairOption options, String keyspace, Epoch minEpoch)
+    /**
+     * Creates a RepairCoordinator, snapshotting TCM state to decide whether mutation tracking
+     * should be used. If mutation tracking is active for the keyspace (and no migration is in progress),
+     * the incremental flag in RepairOption is flipped to false to prevent anti-compaction.
+     */
+    public static RepairCoordinator create(StorageService storageService, int cmd, RepairOption options, String keyspace, Epoch minEpoch)
     {
-        this(SharedContext.Global.instance,
-             (ks, tables) -> storageService.getValidColumnFamilies(false, false, ks, tables),
-             storageService::getLocalReplicas,
-             cmd, options, keyspace, minEpoch);
+        ClusterMetadata metadata = ClusterMetadata.current();
+        boolean useMT = options.isIncremental()
+                         && MutationTrackingIncrementalRepairTask.shouldUseMutationTrackingRepair(metadata, keyspace);
+        KeyspaceMigrationInfo migrationInfo = metadata.mutationTrackingMigrationState.getKeyspaceInfo(keyspace);
+        boolean mtMigration = useMT && migrationInfo != null;
+
+        // If using mutation tracking without migration, flip incremental to false
+        // to prevent anti-compaction since mutation tracking manages marking tables repaired itself
+        if (useMT && !mtMigration)
+        {
+            logger.info("Keyspace {} uses mutation tracking; disabling incremental repair to skip anti-compaction", keyspace);
+            options = options.withIncremental(false);
+        }
+
+        // During migration, validate that repair ranges don't partially overlap with pending migration ranges.
+        // Ranges must be entirely inside or entirely outside the pending set so that a compatible repair
+        // behavior (MT vs normal IR) can be selected. The tables also need to have matching migration ranges.
+        if (mtMigration)
+        {
+            NormalizedRanges<Token> repairRanges = NormalizedRanges.normalizedRanges(options.getRanges());
+            KeyspaceMetadata ksm = metadata.schema.getKeyspaceMetadata(keyspace);
+            Collection<String> cfs = options.getColumnFamilies();
+            migrationInfo.assertRangesNotMixedMigration(repairRanges, ksm, cfs.isEmpty() ? null : cfs);
+        }
+
+        return new RepairCoordinator(SharedContext.Global.instance,
+                                     (ks, tables) -> storageService.getValidColumnFamilies(false, false, ks, tables),
+                                     storageService::getLocalReplicas,
+                                     cmd, options, keyspace, minEpoch,
+                                     useMT, mtMigration, metadata);
     }
 
     RepairCoordinator(SharedContext ctx,
                       BiFunction<String, String[], Iterable<ColumnFamilyStore>> validColumnFamilies,
                       Function<String, RangesAtEndpoint> getLocalReplicas,
                       int cmd, RepairOption options, String keyspace, Epoch minEpoch)
+    {
+        this(ctx, validColumnFamilies, getLocalReplicas, cmd, options, keyspace, minEpoch, false, false, null);
+    }
+
+    RepairCoordinator(SharedContext ctx,
+                      BiFunction<String, String[], Iterable<ColumnFamilyStore>> validColumnFamilies,
+                      Function<String, RangesAtEndpoint> getLocalReplicas,
+                      int cmd, RepairOption options, String keyspace, Epoch minEpoch,
+                      boolean useMutationTracking, boolean mutationTrackingMigrationInProgress,
+                      ClusterMetadata metadata)
     {
         this.ctx = ctx;
         this.minEpoch = minEpoch;
@@ -135,6 +185,9 @@ public class RepairCoordinator implements Runnable, ProgressEventNotifier, Repai
         this.tag = "repair:" + cmd;
         this.validColumnFamilies = validColumnFamilies;
         this.getLocalReplicas = getLocalReplicas;
+        this.useMutationTracking = useMutationTracking;
+        this.mutationTrackingMigrationInProgress = mutationTrackingMigrationInProgress;
+        this.metadata = metadata;
         ctx.repair().register(state);
     }
 
@@ -496,24 +549,57 @@ public class RepairCoordinator implements Runnable, ProgressEventNotifier, Repai
 
     private Future<Pair<CoordinatedRepairResult, Supplier<String>>> repair(String[] cfnames, NeighborsAndRanges neighborsAndRanges)
     {
-        RepairTask task;
+        ExecutorPlus executor = createExecutor();
+        state.phase.repairSubmitted();
+
         if (state.options.isPreview())
         {
-            task = new PreviewRepairTask(this, state.id, neighborsAndRanges.filterCommonRanges(state.keyspace, cfnames), neighborsAndRanges.shouldExcludeDeadParticipants, cfnames);
+            RepairTask task = new PreviewRepairTask(this, state.id, neighborsAndRanges.filterCommonRanges(state.keyspace, cfnames), neighborsAndRanges.shouldExcludeDeadParticipants, cfnames);
+            return submitRepairTask(task, executor);
+        }
+        else if (useMutationTracking)
+        {
+            RepairTask mtTask = new MutationTrackingIncrementalRepairTask(this, state.id, neighborsAndRanges, cfnames);
+            if (mutationTrackingMigrationInProgress)
+            {
+                // During migration, run incremental repair first, then mutation tracking sync.
+                // Propagate the IR result on success since it drives migration advancement.
+                RepairTask incrementalTask = new IncrementalRepairTask(this, state.id, neighborsAndRanges, cfnames);
+                AsyncPromise<Pair<CoordinatedRepairResult, Supplier<String>>> result = new AsyncPromise<>();
+                logger.info("Migration to mutation tracking in progress for {}; running incremental repair before MT sync", state.keyspace);
+                incrementalTask.perform(executor, validationScheduler).addCallback(
+                        irResult -> {
+                            logger.info("Incremental repair completed for migration keyspace {}: hasFailed={}", state.keyspace, irResult.hasFailed());
+                            Pair<CoordinatedRepairResult, Supplier<String>> irPair = Pair.create(irResult, incrementalTask::successMessage);
+                            mtTask.perform(executor, validationScheduler)
+                                    .addCallback(
+                                            mtResult -> result.trySuccess(irPair),
+                                            result::tryFailure);
+                        },
+                        failure -> {
+                            logger.warn("Incremental repair FAILED for migration keyspace {}", state.keyspace, failure);
+                            result.tryFailure(failure);
+                        }
+                );
+                return result.addCallback((s, f) -> executor.shutdown());
+            }
+            return submitRepairTask(mtTask, executor);
         }
         else if (state.options.isIncremental())
         {
-            task = new IncrementalRepairTask(this, state.id, neighborsAndRanges, cfnames);
+            RepairTask task = new IncrementalRepairTask(this, state.id, neighborsAndRanges, cfnames);
+            return submitRepairTask(task, executor);
         }
         else
         {
-            task = new NormalRepairTask(this, state.id, neighborsAndRanges.filterCommonRanges(state.keyspace, cfnames), neighborsAndRanges.shouldExcludeDeadParticipants, cfnames);
+            RepairTask task = new NormalRepairTask(this, state.id, neighborsAndRanges.filterCommonRanges(state.keyspace, cfnames), neighborsAndRanges.shouldExcludeDeadParticipants, cfnames);
+            return submitRepairTask(task, executor);
         }
+    }
 
-        ExecutorPlus executor = createExecutor();
-        state.phase.repairSubmitted();
+    private Future<Pair<CoordinatedRepairResult, Supplier<String>>> submitRepairTask(RepairTask task, ExecutorPlus executor)
+    {
         return task.perform(executor, validationScheduler)
-                   // after adding the callback java could no longer infer the type...
                    .<Pair<CoordinatedRepairResult, Supplier<String>>>map(r -> Pair.create(r, task::successMessage))
                    .addCallback((s, f) -> executor.shutdown());
     }

src/java/org/apache/cassandra/repair/RepairJob.java

@@ -57,6 +57,7 @@ import org.apache.cassandra.service.accord.IAccordService;
 import org.apache.cassandra.service.accord.repair.AccordRepair;
 import org.apache.cassandra.service.accord.repair.AccordRepair.AccordRepairResult;
 import org.apache.cassandra.service.consensus.migration.ConsensusMigrationRepairResult;
+import org.apache.cassandra.service.replication.migration.KeyspaceMigrationInfo;
 import org.apache.cassandra.service.replication.migration.MutationTrackingMigrationRepairResult;
 import org.apache.cassandra.service.paxos.cleanup.PaxosCleanup;
 import org.apache.cassandra.service.paxos.cleanup.PaxosUpdateLowBallot;
@@ -261,7 +262,7 @@ public class RepairJob extends AsyncFuture<RepairResult> implements Runnable
                                                                             .flatMap(this::executeTasks, taskExecutor);
 
             // For tracked keyspaces, we need to ensure sync'd data is present in the log
-            boolean isTracked = cfs.metadata().replicationType().isTracked();
+            boolean isTracked = useTrackedTransfers();
             if (isTracked)
                 syncResults = TransferTrackingService.instance().onRepairSyncCompletion(this, syncResults, taskExecutor);
         }
@@ -347,7 +348,7 @@ public class RepairJob extends AsyncFuture<RepairResult> implements Runnable
             else
                 syncTasks = createStandardSyncTasks(trees);
 
-            return ks.getMetadata().params.replicationType.isTracked() 
+            return useTrackedTransfers()
                    ? SyncTasks.tracked(ks, syncTasks)
                    : SyncTasks.untracked(syncTasks);
         }, taskExecutor);
@@ -369,6 +370,23 @@ public class RepairJob extends AsyncFuture<RepairResult> implements Runnable
         return desc.keyspace.equals(METADATA_KEYSPACE_NAME);
     }
 
+    /**
+     * Whether tracked repair transfers should be used for this repair job.
+     * Returns true only when the keyspace uses tracked replication AND the repair ranges
+     * for this table have completed migration (or no migration is in progress).
+     * During migration, ranges still in the pending set use the traditional untracked
+     * streaming path because:
+     * - The data being streamed is pre-migration data without mutation tracking offsets
+     * - TrackedRepairTransfer does not support --force (dead node exclusion)
+     */
+    private boolean useTrackedTransfers()
+    {
+        if (!cfs.metadata().replicationType().isTracked())
+            return false;
+
+        return KeyspaceMigrationInfo.shouldUseTrackedTransfers(ClusterMetadata.current(), desc.keyspace, cfs.metadata().id, desc.ranges);
+    }
+
     private boolean isTransient(InetAddressAndPort ep)
     {
         return session.state.commonRange.transEndpoints.contains(ep);
@@ -465,7 +483,7 @@ public class RepairJob extends AsyncFuture<RepairResult> implements Runnable
             if (!tasks.isEmpty())
                 state.phase.streamSubmitted();
 
-            if (cfs.metadata().replicationType().isTracked())
+            if (useTrackedTransfers())
                 TransferTrackingService.instance().onRepairSyncExecution(tasks);
 
             for (SyncTask task : tasks)

src/java/org/apache/cassandra/repair/RepairMessageVerbHandler.java

@@ -19,6 +19,7 @@ package org.apache.cassandra.repair;
 
 import java.util.ArrayList;
 import java.util.List;
+import java.util.Map;
 import java.util.concurrent.TimeoutException;
 import java.util.function.BiFunction;
 import java.util.function.Function;
@@ -29,11 +30,19 @@ import org.slf4j.LoggerFactory;
 
 import org.apache.cassandra.db.ColumnFamilyStore;
 import org.apache.cassandra.db.Keyspace;
+import org.apache.cassandra.dht.Range;
+import org.apache.cassandra.dht.Token;
 import org.apache.cassandra.locator.InetAddressAndPort;
 import org.apache.cassandra.net.IVerbHandler;
 import org.apache.cassandra.net.Message;
+
+import org.apache.cassandra.replication.CoordinatorLogId;
+import org.apache.cassandra.replication.MutationTrackingService;
+import org.apache.cassandra.replication.Offsets;
 import org.apache.cassandra.repair.messages.CleanupMessage;
 import org.apache.cassandra.repair.messages.FailSession;
+import org.apache.cassandra.repair.messages.MutationTrackingSyncRequest;
+import org.apache.cassandra.repair.messages.MutationTrackingSyncResponse;
 import org.apache.cassandra.repair.messages.PrepareMessage;
 import org.apache.cassandra.repair.messages.RepairMessage;
 import org.apache.cassandra.repair.messages.StatusRequest;
@@ -381,6 +390,10 @@ public class RepairMessageVerbHandler implements IVerbHandler<RepairMessage>
                     ctx.repair().consistent.local.handleStatusResponse(message.from(), (StatusResponse) message.payload);
                     break;
 
+                case MT_SYNC_REQ:
+                    handleMutationTrackingSyncRequest(message);
+                    break;
+
                 default:
                     ctx.repair().handleMessage(message);
                     break;
@@ -489,4 +502,29 @@ public class RepairMessageVerbHandler implements IVerbHandler<RepairMessage>
                                      "validation request",
                                      from);
     }
+
+    @SuppressWarnings("unchecked")
+    private void handleMutationTrackingSyncRequest(Message<RepairMessage> message)
+    {
+        MutationTrackingSyncRequest request = (MutationTrackingSyncRequest) message.payload;
+        RepairJobDesc desc = request.desc;
+        logger.debug("Handling mutation tracking sync request {}", desc);
+
+        try
+        {
+            Map<Range<Token>, Map<CoordinatorLogId, Offsets.Immutable>> offsets =
+                MutationTrackingService.instance().collectWitnessedOffsetsForRanges(desc.keyspace, desc.ranges, request.liveHostIds);
+
+            MutationTrackingSyncResponse response = new MutationTrackingSyncResponse(
+                desc,
+                offsets);
+
+            ctx.messaging().send(message.responseWith(response), message.from());
+        }
+        catch (Exception e)
+        {
+            logger.error("Failed to handle mutation tracking sync request {}", desc, e);
+            sendFailureResponse(message);
+        }
+    }
 }

src/java/org/apache/cassandra/repair/RepairResult.java

@@ -22,6 +22,8 @@ import java.util.List;
 import org.apache.cassandra.service.consensus.migration.ConsensusMigrationRepairResult;
 import org.apache.cassandra.service.replication.migration.MutationTrackingMigrationRepairResult;
 
+import javax.annotation.Nullable;
+
 /**
  * RepairJob's result
  */
@@ -29,7 +31,9 @@ public class RepairResult
 {
     public final RepairJobDesc desc;
     public final List<SyncStat> stats;
+    @Nullable
     public final ConsensusMigrationRepairResult consensusMigrationRepairResult;
+    @Nullable
     public final MutationTrackingMigrationRepairResult mutationTrackingMigrationRepairResult;
 
     public RepairResult(RepairJobDesc desc, List<SyncStat> stats, ConsensusMigrationRepairResult consensusMigrationRepairResult, MutationTrackingMigrationRepairResult mutationTrackingMigrationRepairResult)

src/java/org/apache/cassandra/repair/autorepair/AutoRepairState.java

@@ -98,8 +98,8 @@ public abstract class AutoRepairState
 
     protected RepairCoordinator getRepairRunnable(String keyspace, RepairOption options)
     {
-        return new RepairCoordinator(StorageService.instance, StorageService.nextRepairCommand.incrementAndGet(),
-                                     options, keyspace, ClusterMetadata.current().epoch);
+        return RepairCoordinator.create(StorageService.instance, StorageService.nextRepairCommand.incrementAndGet(),
+                                        options, keyspace, ClusterMetadata.current().epoch);
     }
 
     public long getLastRepairTime()

src/java/org/apache/cassandra/repair/messages/MutationTrackingSyncRequest.java

@@ -0,0 +1,81 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.cassandra.repair.messages;
+
+import java.io.IOException;
+import java.util.Set;
+
+import org.apache.cassandra.io.IVersionedSerializer;
+import org.apache.cassandra.io.util.DataInputPlus;
+import org.apache.cassandra.io.util.DataOutputPlus;
+import org.apache.cassandra.repair.RepairJobDesc;
+
+import static org.apache.cassandra.utils.CollectionSerializers.nullableIntSetSerializer;
+
+/**
+ * Request sent from the mutation tracking repair coordinator to each participant to collect
+ * their current witnessed offsets. This establishes a happens-before relationship: the
+ * participant's response contains offsets captured after receiving this request, which is
+ * sent after the repair starts.
+ * <p>
+ * The liveHostIds set tells the responder which hosts are participating in this repair,
+ * so that the response only includes offsets witnessed by those hosts. This prevents the
+ * coordinator from setting sync targets that include offsets only known to down nodes.
+ */
+public class MutationTrackingSyncRequest extends RepairMessage
+{
+    /** The set of host IDs participating in this repair. Null means all replicas. */
+    public final Set<Integer> liveHostIds;
+
+    public MutationTrackingSyncRequest(RepairJobDesc desc, Set<Integer> liveHostIds)
+    {
+        super(desc);
+        this.liveHostIds = liveHostIds;
+    }
+
+    @Override
+    public String toString()
+    {
+        return "MutationTrackingSyncRequest{" +
+               "desc=" + desc +
+               ", liveHostIds=" + liveHostIds +
+               '}';
+    }
+
+    public static final IVersionedSerializer<MutationTrackingSyncRequest> serializer = new IVersionedSerializer<>()
+    {
+        public void serialize(MutationTrackingSyncRequest request, DataOutputPlus out, int version) throws IOException
+        {
+            RepairJobDesc.serializer.serialize(request.desc, out, version);
+            nullableIntSetSerializer.serialize(request.liveHostIds, out);
+        }
+
+        public MutationTrackingSyncRequest deserialize(DataInputPlus in, int version) throws IOException
+        {
+            RepairJobDesc desc = RepairJobDesc.serializer.deserialize(in, version);
+            Set<Integer> liveHostIds = nullableIntSetSerializer.deserialize(in);
+            return new MutationTrackingSyncRequest(desc, liveHostIds);
+        }
+
+        public long serializedSize(MutationTrackingSyncRequest request, int version)
+        {
+            return RepairJobDesc.serializer.serializedSize(request.desc, version)
+                   + nullableIntSetSerializer.serializedSize(request.liveHostIds);
+        }
+    };
+}

src/java/org/apache/cassandra/repair/messages/MutationTrackingSyncResponse.java

@@ -0,0 +1,110 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.cassandra.repair.messages;
+
+import java.io.IOException;
+import java.util.Map;
+import java.util.Objects;
+
+import org.apache.cassandra.dht.AbstractBounds;
+import org.apache.cassandra.dht.Range;
+import org.apache.cassandra.dht.Token;
+import org.apache.cassandra.io.IVersionedSerializer;
+import org.apache.cassandra.io.util.DataInputPlus;
+import org.apache.cassandra.io.util.DataOutputPlus;
+import org.apache.cassandra.replication.CoordinatorLogId;
+import org.apache.cassandra.replication.Offsets;
+import org.apache.cassandra.repair.RepairJobDesc;
+import org.apache.cassandra.utils.CollectionSerializers;
+
+/**
+ * Response from a participant to a {@link MutationTrackingSyncRequest}. Contains the
+ * participant's current witnessed offsets for each shard overlapping the requested ranges.
+ * These offsets are captured after the request is received, establishing a happens-before
+ * relationship with the repair start.
+ */
+public class MutationTrackingSyncResponse extends RepairMessage
+{
+    /** Per-shard witnessed offsets: shard range -> (logId -> offsets) */
+    public final Map<Range<Token>, Map<CoordinatorLogId, Offsets.Immutable>> offsetsByShard;
+
+    public MutationTrackingSyncResponse(RepairJobDesc desc,
+                                        Map<Range<Token>, Map<CoordinatorLogId, Offsets.Immutable>> offsetsByShard)
+    {
+        super(desc);
+        Objects.requireNonNull(offsetsByShard);
+        this.offsetsByShard = offsetsByShard;
+    }
+
+    @Override
+    public boolean equals(Object o)
+    {
+        if (!(o instanceof MutationTrackingSyncResponse))
+            return false;
+        MutationTrackingSyncResponse other = (MutationTrackingSyncResponse) o;
+        return Objects.equals(desc, other.desc)
+               && Objects.equals(offsetsByShard, other.offsetsByShard);
+    }
+
+    @Override
+    public int hashCode()
+    {
+        return Objects.hash(desc, offsetsByShard);
+    }
+
+    @Override
+    public String toString()
+    {
+        return "MutationTrackingSyncResponse{" +
+               "desc=" + desc +
+               ", shardCount=" + offsetsByShard.size() +
+               '}';
+    }
+
+    private static final IVersionedSerializer<Map<CoordinatorLogId, Offsets.Immutable>> offsetsMapSerializer =
+        CollectionSerializers.newMapSerializer(CoordinatorLogId.serializer, Offsets.serializer);
+
+    @SuppressWarnings("unchecked")
+    public static final IVersionedSerializer<MutationTrackingSyncResponse> serializer = new IVersionedSerializer<>()
+    {
+        public void serialize(MutationTrackingSyncResponse response, DataOutputPlus out, int version) throws IOException
+        {
+            RepairJobDesc.serializer.serialize(response.desc, out, version);
+            CollectionSerializers.serializeMap((Map<AbstractBounds<Token>, Map<CoordinatorLogId, Offsets.Immutable>>) (Map<?, ?>) response.offsetsByShard,
+                                              out, version, Range.tokenSerializer, offsetsMapSerializer);
+        }
+
+        public MutationTrackingSyncResponse deserialize(DataInputPlus in, int version) throws IOException
+        {
+            RepairJobDesc desc = RepairJobDesc.serializer.deserialize(in, version);
+            Map<AbstractBounds<Token>, Map<CoordinatorLogId, Offsets.Immutable>> raw =
+                CollectionSerializers.deserializeMap(in, version, Range.tokenSerializer, offsetsMapSerializer);
+            Map<Range<Token>, Map<CoordinatorLogId, Offsets.Immutable>> offsetsByShard =
+                (Map<Range<Token>, Map<CoordinatorLogId, Offsets.Immutable>>) (Map<?, ?>) raw;
+            return new MutationTrackingSyncResponse(desc, offsetsByShard);
+        }
+
+        public long serializedSize(MutationTrackingSyncResponse response, int version)
+        {
+            long size = RepairJobDesc.serializer.serializedSize(response.desc, version);
+            size += CollectionSerializers.serializedMapSize((Map<AbstractBounds<Token>, Map<CoordinatorLogId, Offsets.Immutable>>) (Map<?, ?>) response.offsetsByShard,
+                                                           version, Range.tokenSerializer, offsetsMapSerializer);
+            return size;
+        }
+    };
+}

src/java/org/apache/cassandra/repair/messages/RepairMessage.java

@@ -93,6 +93,8 @@ public abstract class RepairMessage
         map.put(Verb.FINALIZE_PROMISE_MSG, SUPPORTS_RETRY);
         map.put(Verb.FINALIZE_COMMIT_MSG, SUPPORTS_RETRY);
         map.put(Verb.FAILED_SESSION_MSG, SUPPORTS_RETRY);
+        // Mutation tracking messages
+        map.put(Verb.MT_SYNC_REQ, SUPPORTS_RETRY);
         VERB_TIMEOUT_VERSIONS = Collections.unmodifiableMap(map);
 
         EnumSet<Verb> allowsRetry = EnumSet.noneOf(Verb.class);
@@ -110,6 +112,8 @@ public abstract class RepairMessage
         allowsRetry.add(Verb.FINALIZE_PROMISE_MSG);
         allowsRetry.add(Verb.FINALIZE_COMMIT_MSG);
         allowsRetry.add(Verb.FAILED_SESSION_MSG);
+        // Mutation tracking messages
+        allowsRetry.add(Verb.MT_SYNC_REQ);
         ALLOWS_RETRY = Collections.unmodifiableSet(allowsRetry);
     }
 

src/java/org/apache/cassandra/repair/messages/RepairOption.java

@@ -494,6 +494,25 @@ public class RepairOption
                ')';
     }
 
+    /**
+     * Returns a new RepairOption with the incremental flag set to the given value.
+     * All other options are copied from this instance.
+     */
+    public RepairOption withIncremental(boolean incremental)
+    {
+        if (this.incremental == incremental)
+            return this;
+
+        RepairOption copy = new RepairOption(parallelism, primaryRange, incremental, trace, jobThreads,
+                                            ranges, pullRepair, forceRepair, previewKind, optimiseStreams,
+                                            ignoreUnreplicatedKeyspaces, repairData, repairPaxos,
+                                            dontPurgeTombstones, repairAccord);
+        copy.getColumnFamilies().addAll(columnFamilies);
+        copy.getDataCenters().addAll(dataCenters);
+        copy.getHosts().addAll(hosts);
+        return copy;
+    }
+
     public Map<String, String> asMap()
     {
         Map<String, String> options = new HashMap<>();

src/java/org/apache/cassandra/replication/CoordinatorLog.java

@@ -284,6 +284,76 @@ public abstract class CoordinatorLog
         }
     }
 
+    /**
+     * Returns the UNION of all witnessed offsets from all participants.
+     * This represents all offsets that ANY replica has witnessed.
+     */
+    Offsets.Immutable collectUnionOfWitnessedOffsets()
+    {
+        lock.readLock().lock();
+        try
+        {
+            return Offsets.Immutable.copy(witnessedOffsets.union());
+        }
+        finally
+        {
+            lock.readLock().unlock();
+        }
+    }
+
+    /**
+     * Returns the UNION of witnessed offsets scoped to only the specified host IDs.
+     */
+    Offsets.Immutable collectUnionOfWitnessedOffsets(Set<Integer> liveHostIds)
+    {
+        Offsets.Mutable union = new Offsets.Mutable(logId);
+        lock.readLock().lock();
+        try
+        {
+            for (int hostId : liveHostIds)
+            {
+                if (!participants.contains(hostId))
+                    continue;
+
+                Offsets.Mutable nodeOffsets = witnessedOffsets.get(hostId);
+                union.addAll(nodeOffsets);
+            }
+        }
+        finally
+        {
+            lock.readLock().unlock();
+        }
+        return Offsets.Immutable.copy(union);
+    }
+
+    /**
+     * Returns the intersection of witnessed offsets scoped to only the specified host IDs.
+     */
+    Offsets.Immutable collectReconciledOffsets(Set<Integer> liveHostIds)
+    {
+        lock.readLock().lock();
+        try
+        {
+            Offsets.Mutable intersection = null;
+            for (int hostId : liveHostIds)
+            {
+                if (!participants.contains(hostId))
+                    continue;
+
+                Offsets.Mutable nodeOffsets = witnessedOffsets.get(hostId);
+                if (intersection == null)
+                    intersection = Offsets.Mutable.copy(nodeOffsets);
+                else
+                    intersection = Offsets.Mutable.intersection(intersection, nodeOffsets);
+            }
+            return intersection == null ? new Offsets.Immutable(logId) : Offsets.Immutable.copy(intersection);
+        }
+        finally
+        {
+            lock.readLock().unlock();
+        }
+    }
+
     public long getUnreconciledCount()
     {
         lock.readLock().lock();

src/java/org/apache/cassandra/replication/MutationTrackingService.java

@@ -192,6 +192,8 @@ public class MutationTrackingService
     private final IncomingMutations incomingMutations = new IncomingMutations();
     private final OutgoingMutations outgoingMutations = new OutgoingMutations();
 
+    private final Map<String, Set<MutationTrackingSyncCoordinator>> syncCoordinatorsByKeyspace = new ConcurrentHashMap<>();
+
     private volatile boolean started = false;
 
     private MutationTrackingService()
@@ -400,6 +402,19 @@ public class MutationTrackingService
         {
             shardLock.readLock().unlock();
         }
+
+        // Notify any registered sync coordinators about the offset update
+        Set<MutationTrackingSyncCoordinator> coordinators = syncCoordinatorsByKeyspace.get(keyspace);
+        if (coordinators != null)
+        {
+            for (MutationTrackingSyncCoordinator coordinator : coordinators)
+            {
+                if (range.intersects(coordinator.getRange()))
+                {
+                    coordinator.onOffsetsReceived();
+                }
+            }
+        }
     }
 
     public void recordFullyReconciledOffsets(ReconciledLogSnapshot reconciledSnapshot)
@@ -457,6 +472,30 @@ public class MutationTrackingService
         return incomingMutations.subscribe(mutationId, callback);
     }
 
+    /**
+     * Register a sync coordinator to be notified when offset updates arrive.
+     */
+    public void registerSyncCoordinator(MutationTrackingSyncCoordinator coordinator)
+    {
+        syncCoordinatorsByKeyspace.computeIfAbsent(coordinator.getKeyspace(), k -> ConcurrentHashMap.newKeySet())
+                                  .add(coordinator);
+    }
+
+    /**
+     * Unregister a sync coordinator.
+     */
+    public void unregisterSyncCoordinator(MutationTrackingSyncCoordinator coordinator)
+    {
+        Set<MutationTrackingSyncCoordinator> coordinators = syncCoordinatorsByKeyspace.get(coordinator.getKeyspace());
+        if (coordinators != null)
+        {
+            coordinators.remove(coordinator);
+
+            if (coordinators.isEmpty())
+                syncCoordinatorsByKeyspace.remove(coordinator.getKeyspace(), coordinators);
+        }
+    }
+
     public void executeTransfers(String keyspace, Set<SSTableReader> sstables, ConsistencyLevel cl)
     {
         shardLock.readLock().lock();
@@ -614,6 +653,58 @@ public class MutationTrackingService
         return shards;
     }
 
+    public void forEachShardInKeyspace(String keyspace, Consumer<Shard> consumer)
+    {
+        shardLock.readLock().lock();
+        try
+        {
+            KeyspaceShards ksShards = keyspaceShards.get(keyspace);
+            if (ksShards != null)
+                ksShards.forEachShard(consumer);
+        }
+        finally
+        {
+            shardLock.readLock().unlock();
+        }
+    }
+
+    /**
+     * Collects the union of witnessed offsets for all shards in the given keyspace that overlap
+     * with the specified ranges. Used by the mutation tracking repair protocol to establish
+     * a happens-before relationship.
+     *
+     * @param keyspace the keyspace to collect offsets for
+     * @param ranges the token ranges to find overlapping shards for
+     * @return a map from shard range to the union of witnessed offsets per coordinator log
+     */
+    public Map<Range<Token>, Map<CoordinatorLogId, Offsets.Immutable>> collectWitnessedOffsetsForRanges(String keyspace, Collection<Range<Token>> ranges, Set<Integer> liveHostIds)
+    {
+        Map<Range<Token>, Map<CoordinatorLogId, Offsets.Immutable>> result = new HashMap<>();
+        shardLock.readLock().lock();
+        try
+        {
+            KeyspaceShards ksShards = keyspaceShards.get(keyspace);
+            if (ksShards != null)
+            {
+                ksShards.forEachShard(shard -> {
+                    for (Range<Token> range : ranges)
+                    {
+                        if (shard.range.intersects(range))
+                        {
+                            result.put(shard.range, shard.collectUnionOfWitnessedOffsetsPerLog(liveHostIds));
+                            break;
+                        }
+                    }
+                });
+            }
+        }
+        finally
+        {
+            shardLock.readLock().unlock();
+        }
+        return result;
+    }
+
     public void collectLocallyMissingMutations(MutationSummary remoteSummary, Log2OffsetsMap.Mutable into)
     {
         shardLock.readLock().lock();
@@ -983,7 +1074,7 @@ public class MutationTrackingService
                 }
 
                 if (prevKsm == null)
-                    return nextKsm.useMutationTracking() ? UpdateDecision.CREATE : UpdateDecision.NONE;
+                    return nextKsm.useMutationTracking() ? (hasExisting ? UpdateDecision.REPLICA_GROUP : UpdateDecision.CREATE) : UpdateDecision.NONE;
 
                 if (nextKsm == null)
                     return prevKsm.useMutationTracking() ? UpdateDecision.DROP : UpdateDecision.NONE;
@@ -1058,7 +1149,7 @@ public class MutationTrackingService
 
         static KeyspaceShards make(KeyspaceMetadata keyspace, ClusterMetadata cluster, LongSupplier logIdProvider, BiConsumer<Shard, CoordinatorLog> onNewLog)
         {
-            Preconditions.checkArgument(keyspace.params.replicationType.isTracked() || cluster.mutationTrackingMigrationState.getKeyspaceInfo(keyspace.name) != null);
+            Preconditions.checkArgument(keyspace.params.replicationType.isTracked() || cluster.mutationTrackingMigrationState.isMigrating(keyspace.name));
 
             Map<Range<Token>, Shard> shards = new HashMap<>();
             Map<Range<Token>, VersionedEndpoints.ForRange> groups = new HashMap<>();

src/java/org/apache/cassandra/replication/MutationTrackingSyncCoordinator.java

@@ -0,0 +1,414 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package org.apache.cassandra.replication;
+
+import java.util.ArrayList;
+import java.util.List;
+import java.util.Map;
+import java.util.Set;
+import java.util.concurrent.ConcurrentHashMap;
+import java.util.concurrent.TimeoutException;
+import java.util.concurrent.atomic.AtomicBoolean;
+
+import org.slf4j.Logger;
+import org.slf4j.LoggerFactory;
+
+import com.google.common.collect.ImmutableMap;
+import com.google.common.collect.ImmutableSet;
+import com.google.common.collect.Iterables;
+
+import org.apache.cassandra.dht.Range;
+import org.apache.cassandra.dht.Token;
+import org.apache.cassandra.exceptions.RequestFailure;
+import org.apache.cassandra.locator.InetAddressAndPort;
+import org.apache.cassandra.net.Message;
+import org.apache.cassandra.net.RequestCallback;
+import org.apache.cassandra.net.Verb;
+import org.apache.cassandra.repair.RepairJobDesc;
+import org.apache.cassandra.repair.SharedContext;
+import org.apache.cassandra.repair.messages.MutationTrackingSyncRequest;
+import org.apache.cassandra.repair.messages.MutationTrackingSyncResponse;
+import org.apache.cassandra.repair.messages.RepairMessage;
+import org.apache.cassandra.tcm.ClusterMetadata;
+import org.apache.cassandra.utils.concurrent.AsyncPromise;
+
+import static com.google.common.base.Preconditions.checkState;
+
+public class MutationTrackingSyncCoordinator
+{
+    private static final Logger logger = LoggerFactory.getLogger(MutationTrackingSyncCoordinator.class);
+
+    private final SharedContext ctx;
+    private final RepairJobDesc desc;
+    private final String keyspace;
+    private final Range<Token> range;
+    private final Set<InetAddressAndPort> participants;
+    private final AsyncPromise<Void> completionFuture = new AsyncPromise<>();
+
+    // Per-shard state: tracks what each node has reported for that shard.
+    private ImmutableMap<Range<Token>, ShardSyncState> shardStates;
+
+    // Host IDs of participants for scoped offset collection/completion.
+    // Null means all shard participants (no filtering).
+    private final Set<Integer> liveHostIds;
+
+    private final AtomicBoolean started = new AtomicBoolean(false);
+
+    // Remote participants we are waiting for sync responses from. Completion is
+    // not possible until all responses have been received, since remote nodes may
+    // report targets that the local node doesn't know about yet.
+    private final Set<InetAddressAndPort> pendingSyncResponses = ConcurrentHashMap.newKeySet();
+
+    /**
+     * @param ctx shared context
+     * @param desc repair job descriptor
+     * @param participants the set of remote endpoints that should participate in this sync,
+     *                     as determined by the repair options (force, specific hosts).
+     *                     Only these endpoints will receive sync requests. If null,
+     *                     all remote replicas for overlapping shards will participate.
+     * @param metadata the snapshotted cluster metadata used to resolve endpoint-to-host-ID mappings
+     */
+    public MutationTrackingSyncCoordinator(SharedContext ctx, RepairJobDesc desc, Set<InetAddressAndPort> participants, ClusterMetadata metadata)
+    {
+        this.ctx = ctx;
+        this.desc = desc;
+        this.keyspace = desc.keyspace;
+        this.range = Iterables.getOnlyElement(desc.ranges);
+        this.participants = participants;
+
+        // Convert participant endpoints to host IDs for scoped completion checks.
+        // If participants is null (no filtering), all shard participants are live.
+        if (participants != null)
+        {
+            ImmutableSet.Builder<Integer> builder = ImmutableSet.builder();
+            for (InetAddressAndPort ep : participants)
+            {
+                builder.add(metadata.directory.peerId(ep).id());
+            }
+            // Always include the local node
+            builder.add(metadata.directory.peerId(ctx.broadcastAddressAndPort()).id());
+            liveHostIds = builder.build();
+        }
+        else
+        {
+            liveHostIds = null;
+        }
+    }
+
+    public void start()
+    {
+        if (!started.compareAndSet(false, true))
+            throw new IllegalStateException("Sync coordinator already started");
+
+        List<Shard> overlappingShards = new ArrayList<>();
+        MutationTrackingService.instance().forEachShardInKeyspace(keyspace, shard -> {
+            if (shard.range.intersects(range))
+                overlappingShards.add(shard);
+        });
+
+        checkState(!overlappingShards.isEmpty(), "No intersecting shards found for keyspace {} range {}", keyspace, range);
+
+        ImmutableMap.Builder<Range<Token>, ShardSyncState> builder = ImmutableMap.builder();
+        for (Shard shard : overlappingShards)
+        {
+            ShardSyncState state = new ShardSyncState(shard, liveHostIds);
+            builder.put(shard.range, state);
+        }
+        shardStates = builder.build();
+
+        // Register to receive offset updates
+        MutationTrackingService.instance().registerSyncCoordinator(this);
+
+        // Capture local targets
+        captureTargets();
+
+        logger.info("Sync coordinator started for keyspace {} range {}, tracking {} shards",
+                   keyspace, range, overlappingShards.size());
+
+        // Send sync requests to all remote participants
+        sendSyncRequests();
+
+        // Check if already complete (e.g. single node, no targets)
+        checkIfReadyToComplete();
+    }
+
+    private void complete()
+    {
+        if (completionFuture.trySuccess(null))
+            MutationTrackingService.instance().unregisterSyncCoordinator(this);
+    }
+
+    private void sendSyncRequests()
+    {
+        MutationTrackingSyncRequest request = new MutationTrackingSyncRequest(desc, liveHostIds);
+        // Collect remote replicas, filtering to only allowed participants if specified.
+        // This respects --force (which excludes dead nodes) and --hosts (which
+        // restricts to specific nodes).
+        Set<InetAddressAndPort> remoteParticipants = ConcurrentHashMap.newKeySet();
+        for (ShardSyncState state : shardStates.values())
+            remoteParticipants.addAll(state.shard.remoteReplicas());
+
+        if (participants != null)
+            remoteParticipants.retainAll(participants);
+
+        pendingSyncResponses.addAll(remoteParticipants);
+
+        for (InetAddressAndPort participant : remoteParticipants)
+        {
+            logger.debug("Sending mutation tracking sync request to {} for {}", participant, desc);
+
+            RepairMessage.sendMessageWithRetries(ctx,
+                                                 RepairMessage.notDone(completionFuture),
+                                                 request,
+                                                 Verb.MT_SYNC_REQ,
+                                                 participant,
+                                                 new RequestCallback<MutationTrackingSyncResponse>()
+                                                 {
+                                                     @Override
+                                                     public void onResponse(Message<MutationTrackingSyncResponse> msg)
+                                                     {
+                                                         onSyncResponse(msg.from(), msg.payload);
+                                                     }
+
+                                                     @Override
+                                                     public void onFailure(InetAddressAndPort from, RequestFailure failure)
+                                                     {
+                                                         fail(new RuntimeException(
+                                                             String.format("Mutation tracking sync failed: participant %s returned failure %s", from, failure.reason)));
+                                                     }
+
+                                                     @Override
+                                                     public boolean invokeOnFailure()
+                                                     {
+                                                         return true;
+                                                     }
+                                                 });
+        }
+    }
+
+    private void captureTargets()
+    {
+        checkState(!completionFuture.isDone());
+        checkForTopologyChange();
+
+        for (ShardSyncState state : shardStates.values())
+        {
+            state.captureTargets();
+        }
+    }
+
+    /**
+     * Checks if any of the shards we're tracking have changed due to topology updates.
+     * If a change is detected, fails the repair via {@link #fail(Throwable)}.
+     */
+    private void checkForTopologyChange()
+    {
+        for (ShardSyncState state : shardStates.values())
+        {
+            Shard currentShard = getCurrentShard(state.shard.range);
+            if (currentShard != state.shard)
+            {
+                fail(new RuntimeException("Repair failed: topology changed during sync"));
+                return;
+            }
+        }
+    }
+
+    private Shard getCurrentShard(Range<Token> shardRange)
+    {
+        Shard[] result = new Shard[1];
+        MutationTrackingService.instance().forEachShardInKeyspace(keyspace, shard -> {
+            if (shard.range.equals(shardRange))
+                result[0] = shard;
+        });
+        return result[0];
+    }
+
+    private void fail(Throwable cause)
+    {
+        if (completionFuture.tryFailure(cause))
+        {
+            MutationTrackingService.instance().unregisterSyncCoordinator(this);
+            logger.warn("Sync coordinator for keyspace {} range {} failed: {}",
+                        keyspace, range, cause.getMessage());
+        }
+    }
+
+    /**
+     * Check if all targets are reconciled across all shards.
+     */
+    private void checkIfReadyToComplete()
+    {
+        if (completionFuture.isDone())
+            return;
+        checkForTopologyChange();
+
+        if (checkIfComplete())
+        {
+            logger.info("Sync coordinator completed for keyspace {} range {}", keyspace, range);
+            complete();
+        }
+    }
+
+    private boolean checkIfComplete()
+    {
+        if (completionFuture.isDone())
+            return true;
+
+        if (!pendingSyncResponses.isEmpty())
+            return false;
+
+        for (ShardSyncState state : shardStates.values())
+        {
+            if (!state.isComplete())
+                return false;
+        }
+        return true;
+    }
+
+    /**
+     * Called when offset updates are received from a participant.
+     */
+    public void onOffsetsReceived()
+    {
+        if (completionFuture.isDone())
+            return;
+
+        checkIfReadyToComplete();
+    }
+
+    /**
+     * Called when a sync response is received from a participant in response to a
+     * MutationTrackingSyncRequest. Updates the shard targets with the offsets from the
+     * response, establishing a happens-before relationship with the repair start.
+     *
+     * @param from the participant that sent the response
+     * @param response the sync response from a participant
+     */
+    public void onSyncResponse(InetAddressAndPort from, MutationTrackingSyncResponse response)
+    {
+        if (completionFuture.isDone())
+            return;
+
+        // Deduplicate: retries of MT_SYNC_REQ can produce multiple responses from the
+        // same participant. Only process the first one.
+        if (!pendingSyncResponses.remove(from))
+            return;
+
+        // Update shard targets with the offsets received from the participant
+        for (Map.Entry<Range<Token>, Map<CoordinatorLogId, Offsets.Immutable>> entry : response.offsetsByShard.entrySet())
+        {
+            Range<Token> shardRange = entry.getKey();
+            ShardSyncState state = shardStates.get(shardRange);
+            if (state != null)
+            {
+                state.targets.putAll(entry.getValue());
+            }
+        }
+
+        logger.trace("Sync coordinator received sync response from {}", from);
+
+        checkIfReadyToComplete();
+    }
+
+    public String getKeyspace()
+    {
+        return keyspace;
+    }
+
+    public Range<Token> getRange()
+    {
+        return range;
+    }
+
+    public void awaitCompletion() throws Exception
+    {
+        completionFuture.get();
+    }
+
+    public void cancel()
+    {
+        if (completionFuture.tryFailure(new RuntimeException("Sync cancelled")))
+            MutationTrackingService.instance().unregisterSyncCoordinator(this);
+    }
+
+    public void timeout()
+    {
+        if (completionFuture.tryFailure(new TimeoutException("Mutation tracking sync timed out")))
+            MutationTrackingService.instance().unregisterSyncCoordinator(this);
+    }
+
+    /**
+     * Tracks sync state for a single shard.
+     * Completion is scoped to only the live participant host IDs when provided,
+     * so that dead/excluded nodes don't block sync completion.
+     */
+    private static class ShardSyncState
+    {
+        private final Shard shard;
+
+        // If non-null, only these host IDs are considered for union/intersection.
+        // If null, all shard participants are used (equivalent to no filtering).
+        private final Set<Integer> liveHostIds;
+
+        // Target offsets: LogId -> the offsets we're waiting for live nodes to have
+        private final Map<CoordinatorLogId, Offsets.Immutable> targets = new ConcurrentHashMap<>();
+
+        ShardSyncState(Shard shard, Set<Integer> liveHostIds)
+        {
+            this.shard = shard;
+            this.liveHostIds = liveHostIds;
+        }
+
+        void captureTargets()
+        {
+            Map<CoordinatorLogId, Offsets.Immutable> unionOffsets = shard.collectUnionOfWitnessedOffsetsPerLog(liveHostIds);
+            targets.putAll(unionOffsets);
+        }
+
+        boolean isComplete()
+        {
+            Map<CoordinatorLogId, Offsets.Immutable> currentReconciled = shard.collectReconciledOffsetsPerLog(liveHostIds);
+
+            for (Map.Entry<CoordinatorLogId, Offsets.Immutable> entry : targets.entrySet())
+            {
+                CoordinatorLogId logId = entry.getKey();
+                Offsets.Immutable target = entry.getValue();
+
+                Offsets.Immutable reconciled = currentReconciled.get(logId);
+                if (reconciled == null)
+                    return false;
+
+                if (!containsAll(reconciled, target))
+                    return false;
+            }
+            return true;
+        }
+
+        private boolean containsAll(Offsets reconciled, Offsets target)
+        {
+            for (ShortMutationId id : target)
+            {
+                if (!reconciled.contains(id.offset()))
+                    return false;
+            }
+            return true;
+        }
+    }
+}

src/java/org/apache/cassandra/replication/Node2OffsetsMap.java

@@ -73,6 +73,22 @@ public class Node2OffsetsMap
         return intersection;
     }
 
+    Offsets.Mutable union()
+    {
+        if (offsetsMap.isEmpty())
+            throw new IllegalStateException("Cannot compute union of empty offsets map");
+
+        Iterator<Offsets.Mutable> iter = offsetsMap.values().iterator();
+        if (offsetsMap.size() == 1)
+            return Offsets.Mutable.copy(iter.next());
+
+        Offsets.Mutable union = Offsets.Mutable.copy(iter.next());
+        while (iter.hasNext())
+            union.addAll(iter.next());
+
+        return union;
+    }
+
     public void add(int node, Offsets offsets)
     {
         Offsets.Mutable current = offsetsMap.get(node);

src/java/org/apache/cassandra/replication/Shard.java

@@ -20,7 +20,9 @@ package org.apache.cassandra.replication;
 import java.util.ArrayList;
 import java.util.Collections;
 import java.util.Comparator;
+import java.util.HashMap;
 import java.util.List;
+import java.util.Map;
 import java.util.Set;
 import java.util.SortedMap;
 import java.util.TreeMap;
@@ -407,6 +409,76 @@ public class Shard
         logs.values().forEach(log -> keyspaceBuilder.put(log.logId, log.collectReconciledOffsets(), range));
     }
 
+    /**
+     * Returns the reconciled offsets for each coordinator log in this shard.
+     * Reconciled offsets are the intersection of what all participants have.
+     */
+    public Map<CoordinatorLogId, Offsets.Immutable> collectReconciledOffsetsPerLog()
+    {
+        Map<CoordinatorLogId, Offsets.Immutable> result = new HashMap<>();
+        for (CoordinatorLog log : logs.values())
+        {
+            Offsets.Immutable reconciled = log.collectReconciledOffsets();
+            if (!reconciled.isEmpty())
+                result.put(log.logId, reconciled);
+        }
+        return result;
+    }
+
+    /**
+     * Returns the intersection of witnessed offsets scoped to only the specified participant host IDs.
+     * If liveHostIds is null, behaves the same as {@link #collectReconciledOffsetsPerLog()}.
+     */
+    public Map<CoordinatorLogId, Offsets.Immutable> collectReconciledOffsetsPerLog(Set<Integer> liveHostIds)
+    {
+        if (liveHostIds == null)
+            return collectReconciledOffsetsPerLog();
+
+        Map<CoordinatorLogId, Offsets.Immutable> result = new HashMap<>();
+        for (CoordinatorLog log : logs.values())
+        {
+            Offsets.Immutable reconciled = log.collectReconciledOffsets(liveHostIds);
+            if (!reconciled.isEmpty())
+                result.put(log.logId, reconciled);
+        }
+        return result;
+    }
+
+    /**
+     * Returns the UNION of witnessed offsets from all participants for each coordinator log.
+     * Union = all offsets that ANY replica has witnessed.
+     */
+    public Map<CoordinatorLogId, Offsets.Immutable> collectUnionOfWitnessedOffsetsPerLog()
+    {
+        Map<CoordinatorLogId, Offsets.Immutable> result = new HashMap<>();
+        for (CoordinatorLog log : logs.values())
+        {
+            Offsets.Immutable union = log.collectUnionOfWitnessedOffsets();
+            if (!union.isEmpty())
+                result.put(log.logId, union);
+        }
+        return result;
+    }
+
+    /**
+     * Returns the UNION of witnessed offsets scoped to only the specified participant host IDs.
+     * If liveHostIds is null, behaves the same as {@link #collectUnionOfWitnessedOffsetsPerLog()}.
+     */
+    public Map<CoordinatorLogId, Offsets.Immutable> collectUnionOfWitnessedOffsetsPerLog(Set<Integer> liveHostIds)
+    {
+        if (liveHostIds == null)
+            return collectUnionOfWitnessedOffsetsPerLog();
+
+        Map<CoordinatorLogId, Offsets.Immutable> result = new HashMap<>();
+        for (CoordinatorLog log : logs.values())
+        {
+            Offsets.Immutable union = log.collectUnionOfWitnessedOffsets(liveHostIds);
+            if (!union.isEmpty())
+                result.put(log.logId, union);
+        }
+        return result;
+    }
+
     @Override
     public String toString()
     {

src/java/org/apache/cassandra/service/ActiveRepairService.java

@@ -780,7 +780,7 @@ public class ActiveRepairService implements IEndpointStateChangeSubscriber, IFai
                 if (failure.reason == RequestFailureReason.TIMEOUT)
                 {
                     pending.set(-1);
-                    promise.setFailure(failRepairException(parentRepairSession, "Did not get replies from all endpoints."));
+                    promise.tryFailure(failRepairException(parentRepairSession, "Did not get replies from all endpoints."));
                 }
                 else
                 {
@@ -798,7 +798,7 @@ public class ActiveRepairService implements IEndpointStateChangeSubscriber, IFai
                     }
                     else
                     {
-                        promise.setFailure(failRepairException(parentRepairSession, "Got negative replies from endpoints " + failedNodes));
+                        promise.tryFailure(failRepairException(parentRepairSession, "Got negative replies from endpoints " + failedNodes));
                     }
                 }
             }

src/java/org/apache/cassandra/service/StorageService.java

@@ -3139,7 +3139,7 @@ public class StorageService extends NotificationBroadcasterSupport implements IE
             throw new IllegalArgumentException("data center(s) " + datacenters.toString() + " not found");
         }
 
-        RepairCoordinator task = new RepairCoordinator(this, cmd, options, keyspace, ClusterMetadata.current().epoch);
+        RepairCoordinator task = RepairCoordinator.create(this, cmd, options, keyspace, ClusterMetadata.current().epoch);
         task.addProgressListener(progressSupport);
         for (ProgressListener listener : listeners)
             task.addProgressListener(listener);
@@ -5464,6 +5464,19 @@ public class StorageService extends NotificationBroadcasterSupport implements IE
         DatabaseDescriptor.setRepairRpcTimeout(timeoutInMillis);
         logger.info("RepairRpcTimeout set to {}ms via JMX", timeoutInMillis);
     }
+
+    public long getMutationTrackingSyncTimeout()
+    {
+        return DatabaseDescriptor.getMutationTrackingSyncTimeout(MILLISECONDS);
+    }
+
+    public void setMutationTrackingSyncTimeout(long timeoutInMillis)
+    {
+        checkArgument(timeoutInMillis > 0);
+        DatabaseDescriptor.setMutationTrackingSyncTimeout(timeoutInMillis);
+        logger.info("MutationTrackingSyncTimeout set to {}ms via JMX", timeoutInMillis);
+    }
+
     public void evictHungRepairs()
     {
         logger.info("StorageService#clearPaxosRateLimiters called via jmx");

src/java/org/apache/cassandra/service/StorageServiceMBean.java

@@ -1334,6 +1334,9 @@ public interface StorageServiceMBean extends NotificationEmitter
     public Long getRepairRpcTimeout();
     public void setRepairRpcTimeout(Long timeoutInMillis);
 
+    public long getMutationTrackingSyncTimeout();
+    public void setMutationTrackingSyncTimeout(long timeoutInMillis);
+
     public void evictHungRepairs();
     public void clearPaxosRepairs();
     public void setSkipPaxosRepairCompatibilityCheck(boolean v);

src/java/org/apache/cassandra/service/replication/migration/KeyspaceMigrationInfo.java

@@ -19,6 +19,7 @@
 package org.apache.cassandra.service.replication.migration;
 
 import java.io.IOException;
+import java.util.ArrayList;
 import java.util.Collection;
 import java.util.Collections;
 import java.util.List;
@@ -26,6 +27,7 @@ import java.util.Map;
 import java.util.Objects;
 import java.util.Set;
 import javax.annotation.Nonnull;
+import javax.annotation.Nullable;
 
 import com.google.common.collect.ImmutableMap;
 import com.google.common.collect.Maps;
@@ -36,11 +38,15 @@ import org.apache.cassandra.dht.Range;
 import org.apache.cassandra.dht.Token;
 import org.apache.cassandra.io.util.DataInputPlus;
 import org.apache.cassandra.io.util.DataOutputPlus;
+import org.apache.cassandra.schema.KeyspaceMetadata;
 import org.apache.cassandra.schema.TableId;
+import org.apache.cassandra.schema.TableMetadata;
+import org.apache.cassandra.tcm.ClusterMetadata;
 import org.apache.cassandra.tcm.Epoch;
 import org.apache.cassandra.tcm.serialization.MetadataSerializer;
 import org.apache.cassandra.tcm.serialization.Version;
 
+import static com.google.common.base.Preconditions.checkArgument;
 import static org.apache.cassandra.db.TypeSizes.sizeof;
 import static org.apache.cassandra.utils.CollectionSerializers.deserializeList;
 import static org.apache.cassandra.utils.CollectionSerializers.deserializeMap;
@@ -226,6 +232,26 @@ public class KeyspaceMigrationInfo
         return tableRanges.intersects(token);
     }
 
+    /**
+     * Check if a range intersects with any pending migration range for the given table.
+     */
+    public boolean isRangeInPendingMigration(TableId tableId, Range<Token> range)
+    {
+        NormalizedRanges<Token> tableRanges = pendingRangesPerTable.get(tableId);
+        if (tableRanges == null)
+            return false;
+        return tableRanges.intersects(range);
+    }
+
+    /**
+     * Check if the range defined by startToken and endToken intersects with any pending migration range
+     * for the given table.
+     */
+    public boolean isRangeInPendingMigration(TableId tableId, Token startToken, Token endToken)
+    {
+        return isRangeInPendingMigration(tableId, new Range<>(startToken, endToken));
+    }
+
     /**
      * Determine if read operations on a token should use tracked replication during migration.
      *
@@ -246,6 +272,107 @@ public class KeyspaceMigrationInfo
         return isTracked || isTokenInPendingRange(tableId, token);
     }
 
+    /**
+     * Asserts that the given ranges are either entirely inside or entirely outside the pending
+     * migration set for each specified table. During migration, reads for pending ranges continue
+     * to use the untracked read path with blocking read repair, so it is safe to exclude mutation
+     * tracking streaming for those ranges. However, partial overlap is not supported — repair must
+     * operate on ranges that are fully migrated or fully pending.
+     *
+     * @param ranges the normalized ranges to check
+     * @param tables the tables to check against
+     * @throws IllegalStateException if ranges partially overlap with pending ranges for any table
+     */
+    public void assertRangesNotMixedMigration(@Nonnull NormalizedRanges<Token> ranges,
+                                              @Nonnull Iterable<TableMetadata> tables)
+    {
+        for (TableMetadata table : tables)
+        {
+            NormalizedRanges<Token> pendingRanges = getPendingRangesForTable(table.id);
+            if (pendingRanges.isEmpty())
+                continue;
+
+            NormalizedRanges<Token> overlap = pendingRanges.intersection(ranges);
+            if (overlap.isEmpty())
+                continue;
+
+            // Some ranges overlap with pending — verify ALL ranges are pending for this table
+            NormalizedRanges<Token> outside = ranges.subtract(pendingRanges);
+            if (!outside.isEmpty())
+                throw new IllegalStateException(String.format(
+                    "Ranges for keyspace %s partially overlap with migration pending ranges for table %s. " +
+                    "Ranges must be entirely inside or entirely outside the pending set.",
+                    keyspace, table.name));
+        }
+    }
+
+    /**
+     * Convenience overload that resolves column family names to table metadata before checking.
+     *
+     * @param ranges the normalized ranges to check
+     * @param ksm the keyspace metadata for resolving table names
+     * @param columnFamilies specific table names to check, or null for all tables
+     * @throws IllegalStateException if ranges partially overlap with pending ranges for any table
+     */
+    public void assertRangesNotMixedMigration(@Nonnull NormalizedRanges<Token> ranges,
+                                              @Nonnull KeyspaceMetadata ksm,
+                                              @Nullable Collection<String> columnFamilies)
+    {
+        checkArgument(columnFamilies == null || !columnFamilies.isEmpty(), "columnFmilies must not be empty");
+        Iterable<TableMetadata> tables;
+        if (columnFamilies != null)
+        {
+            List<TableMetadata> tableList = new ArrayList<>(columnFamilies.size());
+            for (String cf : columnFamilies)
+            {
+                TableMetadata table = ksm.tables.getNullable(cf);
+                if (table != null)
+                    tableList.add(table);
+            }
+            tables = tableList;
+        }
+        else
+        {
+            tables = ksm.tables;
+        }
+        assertRangesNotMixedMigration(ranges, tables);
+    }
+
+    /**
+     * Determines whether the given ranges for a table should use the tracked transfer path
+     * (coordinated activation via TrackedRepairTransfer) or the untracked streaming path.
+     * <p>
+     * Returns true (use tracked) when:
+     * - No migration is in progress for this keyspace, OR
+     * - The ranges don't overlap with pending migration ranges for this table
+     * <p>
+     * Returns false (use untracked) when:
+     * - A migration is in progress AND the ranges overlap with pending ranges
+     *
+     * @param metadata cluster metadata snapshot
+     * @param keyspace keyspace name
+     * @param tableId table to check
+     * @param ranges the ranges being repaired or streamed
+     * @return true if tracked transfers should be used
+     */
+    public static boolean shouldUseTrackedTransfers(@Nonnull ClusterMetadata metadata,
+                                                    @Nonnull String keyspace,
+                                                    @Nonnull TableId tableId,
+                                                    @Nonnull Collection<Range<Token>> ranges)
+    {
+        KeyspaceMigrationInfo migrationInfo = metadata.mutationTrackingMigrationState.getKeyspaceInfo(keyspace);
+        if (migrationInfo == null)
+            return true;
+
+        NormalizedRanges<Token> pendingRanges = migrationInfo.getPendingRangesForTable(tableId);
+        if (pendingRanges.isEmpty())
+            return true;
+
+        NormalizedRanges<Token> normalizedRanges = NormalizedRanges.normalizedRanges(ranges);
+        NormalizedRanges<Token> overlap = pendingRanges.intersection(normalizedRanges);
+        return overlap.isEmpty();
+    }
+
     @Override
     public String toString()
     {

src/java/org/apache/cassandra/service/replication/migration/MutationTrackingMigrationState.java

@@ -256,6 +256,11 @@ public class MutationTrackingMigrationState implements MetadataValue<MutationTra
         return keyspaceInfo.get(keyspace);
     }
 
+    public boolean isMigrating(String keyspace)
+    {
+        return keyspaceInfo.containsKey(keyspace);
+    }
+
     public boolean hasMigratingKeyspaces()
     {
         return !keyspaceInfo.isEmpty();

src/java/org/apache/cassandra/utils/CollectionSerializers.java

@@ -77,6 +77,9 @@ public class CollectionSerializers
         }
     };
 
+    public static final UnversionedSerializer<Set<Integer>> intSetSerializer = newSetSerializer(Int32Serializer.serializer);
+    public static final UnversionedSerializer<Set<Integer>> nullableIntSetSerializer = NullableSerializer.wrap(intSetSerializer);
+
     public static <V> void serializeCollection(Collection<V> values, DataOutputPlus out, UnversionedSerializer<V> valueSerializer) throws IOException
     {
         out.writeUnsignedVInt32(values.size());
@@ -642,6 +645,30 @@ public class CollectionSerializers
         return result;
     }
 
+    public static <V> UnversionedSerializer<Set<V>> newSetSerializer(UnversionedSerializer<V> itemSerializer)
+    {
+        return new UnversionedSerializer<Set<V>>()
+        {
+            @Override
+            public void serialize(Set<V> set, DataOutputPlus out) throws IOException
+            {
+                serializeCollection(set, out, itemSerializer);
+            }
+
+            @Override
+            public Set<V> deserialize(DataInputPlus in) throws IOException
+            {
+                return deserializeSet(in, itemSerializer);
+            }
+
+            @Override
+            public long serializedSize(Set<V> t)
+            {
+                return serializedCollectionSize(t, itemSerializer);
+            }
+        };
+    }
+
     public static <V> UnversionedSerializer<List<V>> newListSerializer(UnversionedSerializer<V> itemSerializer)
     {
         return new UnversionedSerializer<List<V>>()
@@ -666,6 +693,30 @@ public class CollectionSerializers
         };
     }
 
+    public static <K, V> IVersionedSerializer<Map<K, V>> newMapSerializer(IVersionedSerializer<K> keySerializer, IVersionedSerializer<V> valueSerializer)
+    {
+        return new IVersionedSerializer<Map<K, V>>()
+        {
+            @Override
+            public void serialize(Map<K, V> map, DataOutputPlus out, int version) throws IOException
+            {
+                serializeMap(map, out, version, keySerializer, valueSerializer);
+            }
+
+            @Override
+            public Map<K, V> deserialize(DataInputPlus in, int version) throws IOException
+            {
+                return deserializeMap(in, version, keySerializer, valueSerializer);
+            }
+
+            @Override
+            public long serializedSize(Map<K, V> map, int version)
+            {
+                return serializedMapSize(map, version, keySerializer, valueSerializer);
+            }
+        };
+    }
+
     public static <V> IVersionedSerializer<List<V>> newListSerializer(IVersionedSerializer<V> itemSerializer)
     {
         return new IVersionedSerializer<List<V>>()

test/distributed/org/apache/cassandra/distributed/test/MutationTrackingMigrationTest.java

@@ -32,7 +32,6 @@ import org.apache.cassandra.distributed.api.Feature;
 import org.apache.cassandra.distributed.api.ICoordinator;
 import org.apache.cassandra.replication.MutationJournal;
 import org.apache.cassandra.schema.KeyspaceMetadata;
-import org.apache.cassandra.service.replication.migration.KeyspaceMigrationInfo;
 import org.apache.cassandra.service.replication.migration.MutationTrackingMigrationState;
 import org.apache.cassandra.tcm.ClusterMetadata;
 import org.apache.cassandra.tcm.ClusterMetadataService;
@@ -41,8 +40,7 @@ import org.apache.cassandra.utils.concurrent.UncheckedInterruptedException;
 
 import static java.lang.String.format;
 import static org.junit.Assert.assertEquals;
-import static org.junit.Assert.assertNotNull;
-import static org.junit.Assert.assertNull;
+import static org.junit.Assert.assertFalse;
 import static org.junit.Assert.assertTrue;
 
 /**
@@ -136,31 +134,31 @@ public class MutationTrackingMigrationTest extends TestBaseImpl
                 ClusterMetadata metadata = ClusterMetadata.current();
                 KeyspaceMetadata ksm = expectedState != ExpectedKeyspaceState.DROPPED ? metadata.schema.getKeyspaceMetadata(keyspace) : null;
                 MutationTrackingMigrationState migrationState = metadata.mutationTrackingMigrationState;
-                KeyspaceMigrationInfo migrationInfo = migrationState.getKeyspaceInfo(keyspace);
+                boolean migrating = migrationState.isMigrating(keyspace);
 
                 switch (expectedState)
                 {
                     case UNTRACKED:
                         assertTrue(!ksm.params.replicationType.isTracked());
-                        assertNull(migrationInfo);
+                        assertFalse(migrating);
                         break;
 
                     case MIGRATING_TO_TRACKED:
                         assertTrue(ksm.params.replicationType.isTracked());
-                        assertNotNull(migrationInfo);
+                        assertTrue(migrating);
                         break;
 
                     case MIGRATING_TO_UNTRACKED:
                         assertTrue(!ksm.params.replicationType.isTracked());
-                        assertNotNull(migrationInfo);
+                        assertTrue(migrating);
                         break;
 
                     case TRACKED:
                         assertTrue(ksm.params.replicationType.isTracked());
-                        assertNull(migrationInfo);
+                        assertFalse(migrating);
                         break;
                     case DROPPED:
-                        assertNull(migrationInfo);
+                        assertFalse(migrating);
                         break;
                     default:
                         throw new AssertionError("Unexpected state: " + expectedState);

test/distributed/org/apache/cassandra/distributed/test/repair/MutationTrackingIncrementalRepairTaskTest.java

@@ -0,0 +1,207 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.cassandra.distributed.test.repair;
+
+import java.io.IOException;
+import java.util.concurrent.atomic.AtomicInteger;
+
+import org.junit.AfterClass;
+import org.junit.BeforeClass;
+import org.junit.Test;
+
+import org.apache.cassandra.distributed.Cluster;
+import org.apache.cassandra.distributed.api.Feature;
+import org.apache.cassandra.distributed.test.TestBaseImpl;
+import org.apache.cassandra.repair.MutationTrackingIncrementalRepairTask;
+import org.apache.cassandra.tcm.ClusterMetadata;
+
+import static org.junit.Assert.*;
+
+/**
+ * Tests for MutationTrackingIncrementalRepairTask.
+ * Tests the decision logic for when to use mutation tracking repair.
+ *
+ * Uses a shared cluster across all tests to minimize overhead.
+ */
+public class MutationTrackingIncrementalRepairTaskTest extends TestBaseImpl
+{
+    private static Cluster CLUSTER;
+    private static final AtomicInteger ksCounter = new AtomicInteger();
+
+    @BeforeClass
+    public static void setupCluster() throws IOException
+    {
+        CLUSTER = Cluster.build()
+                         .withNodes(3)
+                         .withConfig(cfg -> cfg.with(Feature.NETWORK, Feature.GOSSIP))
+                         .start();
+    }
+
+    @AfterClass
+    public static void teardownCluster()
+    {
+        if (CLUSTER != null)
+            CLUSTER.close();
+    }
+
+    private static String nextKsName()
+    {
+        return "mtirt_ks" + ksCounter.incrementAndGet();
+    }
+
+    @Test
+    public void testShouldUseMutationTrackingRepairForTrackedKeyspace() throws Throwable
+    {
+        String ksName = nextKsName();
+        CLUSTER.schemaChange("CREATE KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                             "AND replication_type='tracked'");
+
+        Boolean shouldUse = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.shouldUseMutationTrackingRepair(metadata, ksName);
+        });
+
+        assertTrue("Tracked keyspace should use mutation tracking repair", shouldUse);
+    }
+
+    @Test
+    public void testShouldNotUseMutationTrackingRepairForUntrackedKeyspace() throws Throwable
+    {
+        String ksName = nextKsName();
+        CLUSTER.schemaChange("CREATE KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                             "AND replication_type='untracked'");
+
+        Boolean shouldUse = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.shouldUseMutationTrackingRepair(metadata, ksName);
+        });
+
+        assertFalse("Untracked keyspace should not use mutation tracking repair", shouldUse);
+    }
+
+    @Test
+    public void testRequiresTraditionalRepairReturnsFalseForNonMigratingKeyspace() throws Throwable
+    {
+        String ksName = nextKsName();
+        CLUSTER.schemaChange("CREATE KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                             "AND replication_type='tracked'");
+
+        Boolean migrationInProgress = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.isMutationTrackingMigrationInProgress(metadata, ksName);
+        });
+
+        assertFalse("Non-migrating keyspace should not have migration in progress", migrationInProgress);
+    }
+
+    @Test
+    public void testShouldUseMutationTrackingRepairForNonexistentKeyspace() throws Throwable
+    {
+        Boolean shouldUse = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.shouldUseMutationTrackingRepair(metadata, "nonexistent_ks_xyz");
+        });
+
+        assertFalse("Nonexistent keyspace should return false", shouldUse);
+    }
+
+    @Test
+    public void testMigrationFromUntrackedToTracked() throws Throwable
+    {
+        String ksName = nextKsName();
+        CLUSTER.schemaChange("CREATE KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                             "AND replication_type='untracked'");
+        CLUSTER.schemaChange("CREATE TABLE " + ksName + ".tbl (k int PRIMARY KEY, v int)");
+
+        // Verify initial state
+        Boolean shouldUseBefore = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.shouldUseMutationTrackingRepair(metadata, ksName);
+        });
+        assertFalse("Untracked keyspace should not use mutation tracking repair", shouldUseBefore);
+
+        Boolean migrationBefore = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.isMutationTrackingMigrationInProgress(metadata, ksName);
+        });
+        assertFalse("Non-migrating keyspace should not have migration in progress", migrationBefore);
+
+        // Trigger migration by altering to tracked
+        CLUSTER.schemaChange("ALTER KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                             "AND replication_type='tracked'");
+
+        // Verify migration state - both methods should now return true
+        Boolean shouldUseAfter = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.shouldUseMutationTrackingRepair(metadata, ksName);
+        });
+        assertTrue("Migrating keyspace should use mutation tracking repair", shouldUseAfter);
+
+        Boolean migrationAfter = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.isMutationTrackingMigrationInProgress(metadata, ksName);
+        });
+        assertTrue("Migrating keyspace should have migration in progress", migrationAfter);
+    }
+
+    @Test
+    public void testMigrationFromTrackedToUntracked() throws Throwable
+    {
+        String ksName = nextKsName();
+        CLUSTER.schemaChange("CREATE KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                             "AND replication_type='tracked'");
+        CLUSTER.schemaChange("CREATE TABLE " + ksName + ".tbl (k int PRIMARY KEY, v int)");
+
+        // Verify initial state
+        Boolean shouldUseBefore = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.shouldUseMutationTrackingRepair(metadata, ksName);
+        });
+        assertTrue("Tracked keyspace should use mutation tracking repair", shouldUseBefore);
+
+        Boolean migrationBefore = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.isMutationTrackingMigrationInProgress(metadata, ksName);
+        });
+        assertFalse("Non-migrating tracked keyspace should not have migration in progress", migrationBefore);
+
+        // Migrate back to untracked
+        CLUSTER.schemaChange("ALTER KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                             "AND replication_type='untracked'");
+
+        // During reverse migration, both should still apply
+        Boolean shouldUseAfter = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.shouldUseMutationTrackingRepair(metadata, ksName);
+        });
+        assertFalse("Keyspace migrating from tracked should not use mutation tracking repair", shouldUseAfter);
+
+        Boolean migrationAfter = CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.isMutationTrackingMigrationInProgress(metadata, ksName);
+        });
+        assertTrue("Keyspace migrating from tracked should have migration in progress", migrationAfter);
+    }
+}

test/distributed/org/apache/cassandra/distributed/test/repair/MutationTrackingRepairTest.java

@@ -0,0 +1,917 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.cassandra.distributed.test.repair;
+
+import java.io.IOException;
+import java.net.UnknownHostException;
+import java.util.ArrayList;
+import java.util.List;
+import java.util.concurrent.CountDownLatch;
+import java.util.concurrent.ExecutorService;
+import java.util.concurrent.Executors;
+import java.util.concurrent.Future;
+import java.util.concurrent.TimeUnit;
+import java.util.concurrent.atomic.AtomicInteger;
+import java.util.function.BiPredicate;
+
+import org.junit.After;
+import org.junit.AfterClass;
+import org.junit.Before;
+import org.junit.BeforeClass;
+import org.junit.Test;
+
+import org.apache.cassandra.distributed.Cluster;
+import org.apache.cassandra.distributed.api.ConsistencyLevel;
+import org.apache.cassandra.distributed.api.Feature;
+
+import org.apache.cassandra.distributed.api.IMessageFilters;
+import org.apache.cassandra.distributed.api.NodeToolResult;
+import org.apache.cassandra.distributed.test.TestBaseImpl;
+import org.apache.cassandra.gms.EndpointState;
+import org.apache.cassandra.gms.FailureDetector;
+import org.apache.cassandra.gms.Gossiper;
+import org.apache.cassandra.dht.Murmur3Partitioner;
+import org.apache.cassandra.dht.Range;
+import org.apache.cassandra.dht.Token;
+import org.apache.cassandra.locator.InetAddressAndPort;
+import org.apache.cassandra.net.Verb;
+import org.apache.cassandra.repair.MutationTrackingIncrementalRepairTask;
+import org.apache.cassandra.service.StorageService;
+import org.apache.cassandra.service.replication.migration.KeyspaceMigrationInfo;
+import org.apache.cassandra.tcm.ClusterMetadata;
+import org.apache.cassandra.utils.ByteBufferUtil;
+import org.apache.cassandra.utils.FBUtilities;
+
+import static java.util.function.Predicate.not;
+import static org.apache.cassandra.distributed.api.IMessageFilters.Matcher.of;
+import static org.junit.Assert.*;
+
+/**
+ * End-to-end tests for mutation tracking repair.
+ *
+ * Each test creates a unique keyspace. Keyspaces are not dropped between tests because
+ * dropping a tracked keyspace while background offset broadcasts are in flight causes
+ * NoSuchElementException in MutationTrackingService.getOrCreateShards (the broadcast
+ * tries to look up the dropped keyspace's metadata). The keyspaces are cleaned up when
+ * the cluster is closed at the end of the test class.
+ */
+public class MutationTrackingRepairTest extends TestBaseImpl
+{
+    private static final int NUM_NODES = 3;
+    private static final List<Integer> ALL_NODES = List.of(1, 2, 3);
+
+    private static Cluster CLUSTER;
+    private static ExecutorService executor;
+    private static final AtomicInteger ksCounter = new AtomicInteger();
+
+    private String ksName;
+
+    @BeforeClass
+    public static void setupCluster() throws IOException
+    {
+        executor = Executors.newCachedThreadPool();
+        CLUSTER = Cluster.build()
+                .withNodes(NUM_NODES)
+                .withConfig(cfg -> cfg.set("mutation_tracking_sync_timeout", "10s")
+                        .set("request_timeout", "1000ms")
+                        .set("repair.retries.max_attempts", 10)
+                        .set("repair.retries.base_sleep_time", "100ms")
+                        .set("repair.retries.max_sleep_time", "500ms")
+                        .with(Feature.GOSSIP, Feature.NETWORK))
+                .start();
+    }
+
+    @AfterClass
+    public static void teardownCluster()
+    {
+        executor.shutdownNow();
+        if (CLUSTER != null)
+            CLUSTER.close();
+    }
+
+    @Before
+    public void setUp()
+    {
+        ksName = "mt_repair_" + ksCounter.incrementAndGet();
+        CLUSTER.schemaChange("CREATE KEYSPACE " + ksName + " WITH replication = " +
+                "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                "AND replication_type='tracked'");
+        CLUSTER.schemaChange("CREATE TABLE " + ksName + ".tbl (k int PRIMARY KEY, v int)");
+    }
+
+    @After
+    public void tearDown()
+    {
+        CLUSTER.filters().reset();
+        for (int i = 1; i <= CLUSTER.size(); i++)
+        {
+            CLUSTER.get(i).runOnInstance(() -> {
+                Gossiper.runInGossipStageBlocking(() -> {
+                    for (var entry : Gossiper.instance.endpointStateMap.entrySet())
+                    {
+                        InetAddressAndPort ep = entry.getKey();
+                        EndpointState state = entry.getValue();
+                        if (!ep.equals(FBUtilities.getBroadcastAddressAndPort()) && !state.isAlive())
+                        {
+                            FailureDetector.instance.report(ep);
+                            Gossiper.instance.realMarkAlive(ep, state);
+                        }
+                    }
+                });
+            });
+        }
+    }
+
+    private void setupUntracked()
+    {
+        ksName = "mt_repair_" + ksCounter.incrementAndGet();
+        CLUSTER.schemaChange("CREATE KEYSPACE " + ksName + " WITH replication = " +
+                "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                "AND replication_type='untracked'");
+        CLUSTER.schemaChange("CREATE TABLE " + ksName + ".tbl (k int PRIMARY KEY, v int)");
+    }
+
+    private void createTable(String tableName)
+    {
+        CLUSTER.schemaChange("CREATE TABLE " + ksName + '.' + tableName + " (k int PRIMARY KEY, v int)");
+    }
+
+    private void alterKeyspaceToTracked()
+    {
+        CLUSTER.schemaChange("ALTER KEYSPACE " + ksName + " WITH replication = " +
+                "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                "AND replication_type='tracked'");
+    }
+
+    private void alterKeyspaceToUntracked()
+    {
+        CLUSTER.schemaChange("ALTER KEYSPACE " + ksName + " WITH replication = " +
+                "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                "AND replication_type='untracked'");
+    }
+
+    private void insertData(String tableName, int start, int count)
+    {
+        for (int i = start; i < start + count; i++)
+        {
+            CLUSTER.coordinator(1).execute(
+                    "INSERT INTO " + ksName + '.' + tableName + " (k, v) VALUES (?, ?)",
+                    ConsistencyLevel.ALL, i, i);
+        }
+    }
+
+    private void insertDataWithInconsistency(String tableName, int start, int count)
+    {
+        insertDataWithInconsistency(2, tableName, start, count);
+    }
+
+    private void insertDataWithInconsistency(int isolatedNode, String tableName, int start, int count)
+    {
+        // Isolate a node so background reconcilation has some work to do
+        CLUSTER.filters().allVerbs().to(isolatedNode).drop();
+        CLUSTER.filters().allVerbs().from(isolatedNode).drop();
+
+        for (int i = start; i < start + count; i++)
+        {
+            CLUSTER.coordinator(1).execute(
+                    "INSERT INTO " + ksName + '.' + tableName + " (k, v) VALUES (?, ?)",
+                    ConsistencyLevel.QUORUM, i, i);
+        }
+
+        // Verify the isolated node is actually missing the data we just wrote
+        Object[][] results = CLUSTER.get(isolatedNode).executeInternal(
+                "SELECT k FROM " + ksName + '.' + tableName + " WHERE k >= ? AND k < ? ALLOW FILTERING",
+                start, start + count);
+        assertEquals("Node " + isolatedNode + " should not have data written while isolated",
+                0, results.length);
+        CLUSTER.filters().reset();
+    }
+
+    private void assertDataOnAllNodes(String tableName, List<Integer> keys)
+    {
+        for (int node = 1; node <= CLUSTER.size(); node++)
+        {
+            for (int key : keys)
+            {
+                Object[][] results = CLUSTER.get(node).executeInternal(
+                        "SELECT k, v FROM " + ksName + '.' + tableName + " WHERE k = ?", key);
+                assertEquals("Node " + node + " missing row k=" + key, 1, results.length);
+                assertEquals(key, results[0][0]);
+                assertEquals(key, results[0][1]);
+            }
+        }
+    }
+
+    private void assertDataOnAllNodes(String tableName, int start, int count)
+    {
+        List<Integer> keys = new ArrayList<>(count);
+        for (int i = start; i < start + count; i++)
+            keys.add(i);
+        assertDataOnAllNodes(tableName, keys);
+    }
+
+    private NodeToolResult nodetoolRepair(int node, String... args)
+    {
+        String[] cmd = new String[args.length + 1];
+        cmd[0] = "repair";
+        System.arraycopy(args, 0, cmd, 1, args.length);
+        return CLUSTER.get(node).nodetoolResult(cmd);
+    }
+
+    private List<NodeToolResult> repairConcurrently(List<Integer> nodes, String... args)
+    {
+        List<Future<NodeToolResult>> futures = new ArrayList<>();
+        for (int node : nodes)
+        {
+            int n = node;
+            futures.add(executor.submit(() -> nodetoolRepair(n, args)));
+        }
+        List<NodeToolResult> results = new ArrayList<>();
+        for (Future<NodeToolResult> f : futures)
+        {
+            try
+            {
+                results.add(f.get(60, TimeUnit.SECONDS));
+            }
+            catch (Exception e)
+            {
+                throw new RuntimeException("Repair future failed", e);
+            }
+        }
+        return results;
+    }
+
+    private void assertAllSuccess(List<NodeToolResult> results)
+    {
+        for (NodeToolResult r : results)
+            r.asserts().success();
+    }
+
+    private void assertAllFailure(List<NodeToolResult> results)
+    {
+        for (NodeToolResult r : results)
+            r.asserts().failure();
+    }
+
+    private String[] withPR(String... args)
+    {
+        String[] result = new String[args.length + 1];
+        System.arraycopy(args, 0, result, 0, args.length);
+        result[args.length] = "-pr";
+        return result;
+    }
+
+    private void repairResolvingInconsistency(String... args) throws Exception
+    {
+        repairResolvingInconsistency(2, ALL_NODES, withPR(args));
+    }
+
+    private void repairResolvingInconsistency(int isolatedNode, List<Integer> nodes, String... args) throws Exception
+    {
+        // Dropping messages is to check that repair retries messages if needed
+        CLUSTER.filters().allVerbs().to(isolatedNode).drop();
+        CLUSTER.filters().allVerbs().from(isolatedNode).drop();
+
+        List<Future<NodeToolResult>> futures = new ArrayList<>();
+        for (int node : nodes)
+        {
+            int n = node;
+            futures.add(executor.submit(() -> nodetoolRepair(n, args)));
+        }
+
+        Thread.sleep(2000);
+        assertTrue("Repair should be blocked while node " + isolatedNode + " is isolated",
+                futures.stream().allMatch(not(Future::isDone)));
+
+        CLUSTER.filters().reset();
+
+        List<NodeToolResult> results = new ArrayList<>();
+        for (Future<NodeToolResult> f : futures)
+            results.add(f.get(30, TimeUnit.SECONDS));
+        assertAllSuccess(results);
+
+        // Run a second time to make sure repair can be run multiple times without failing
+        assertAllSuccess(repairConcurrently(nodes, args));
+    }
+
+    private void repairFromNodesSuccess(List<Integer> nodes, String... args)
+    {
+        String[] prArgs = withPR(args);
+        assertAllSuccess(repairConcurrently(nodes, prArgs));
+        assertAllSuccess(repairConcurrently(nodes, prArgs));
+    }
+
+    private boolean isMigrationInProgress()
+    {
+        String ks = ksName;
+        return CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return MutationTrackingIncrementalRepairTask.isMutationTrackingMigrationInProgress(metadata, ks);
+        });
+    }
+
+    private boolean isMigrationComplete()
+    {
+        String ks = ksName;
+        return CLUSTER.get(1).callOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            return !metadata.mutationTrackingMigrationState.isMigrating(ks);
+        });
+    }
+
+    /**
+     * Get the primary token range for a node as [start, end] token values.
+     * With SimpleStrategy RF=3 and 3 nodes, each node has exactly one primary range.
+     */
+    private long[] getPrimaryRangeTokens(int node)
+    {
+        String ks = ksName;
+        return CLUSTER.get(node).callOnInstance(() -> {
+            var ranges = StorageService.instance.getPrimaryRanges(ks);
+            assertEquals(1, ranges.size());
+            Range<Token> range = ranges.iterator().next();
+            return new long[]{
+                    ((Murmur3Partitioner.LongToken) range.left).token,
+                    ((Murmur3Partitioner.LongToken) range.right).token
+            };
+        });
+    }
+
+    /**
+     * Compute which integer keys from [start, start+count) hash into the given token range.
+     */
+    private List<Integer> keysInTokenRange(int start, int count, long rangeStart, long rangeEnd)
+    {
+        Range<Token> range = new Range<>(new Murmur3Partitioner.LongToken(rangeStart),
+                new Murmur3Partitioner.LongToken(rangeEnd));
+        List<Integer> keys = new ArrayList<>();
+        for (int i = start; i < start + count; i++)
+        {
+            Token token = Murmur3Partitioner.instance.getToken(ByteBufferUtil.bytes(i));
+            if (range.contains(token))
+                keys.add(i);
+        }
+        return keys;
+    }
+
+    private String getBroadcastAddress(int node)
+    {
+        return CLUSTER.get(node).callOnInstance(() -> FBUtilities.getBroadcastAddressAndPort().getHostAddressAndPort());
+    }
+
+    private void isolateNode(int nodeToIsolate, int... observerNodes)
+    {
+        CLUSTER.filters().allVerbs().from(nodeToIsolate).drop();
+        CLUSTER.filters().allVerbs().to(nodeToIsolate).drop();
+
+        String isolatedAddress = CLUSTER.get(nodeToIsolate).callOnInstance(
+                () -> FBUtilities.getBroadcastAddressAndPort().getHostAddressAndPort());
+        for (int observer : observerNodes)
+        {
+            CLUSTER.get(observer).runOnInstance(() -> {
+                try
+                {
+                    InetAddressAndPort neighbor = InetAddressAndPort.getByName(isolatedAddress);
+                    FailureDetector.instance.forceConviction(neighbor);
+                }
+                catch (UnknownHostException e)
+                {
+                    throw new RuntimeException(e);
+                }
+            });
+        }
+    }
+
+    @Test
+    public void testBasicRepairHappyPath() throws Exception
+    {
+        insertDataWithInconsistency("tbl", 0, 100);
+
+        repairResolvingInconsistency(ksName);
+
+        assertDataOnAllNodes("tbl", 0, 100);
+    }
+
+    @Test
+    public void testRepairSpecificTable() throws Exception
+    {
+        createTable("tbl1");
+        createTable("tbl2");
+
+        // Repair only tbl1
+        insertDataWithInconsistency("tbl1", 0, 50);
+        repairResolvingInconsistency(ksName, "tbl1");
+        assertDataOnAllNodes("tbl1", 0, 50);
+
+        // Repair only tbl2 while tbl1 already has repaired data
+        insertDataWithInconsistency("tbl2", 0, 50);
+        repairResolvingInconsistency(ksName, "tbl2");
+        assertDataOnAllNodes("tbl2", 0, 50);
+
+        // Repair both tables together
+        insertDataWithInconsistency("tbl1", 50, 50);
+        insertDataWithInconsistency("tbl2", 50, 50);
+        repairResolvingInconsistency(ksName, "tbl1", "tbl2");
+        assertDataOnAllNodes("tbl1", 0, 100);
+        assertDataOnAllNodes("tbl2", 0, 100);
+    }
+
+    @Test
+    public void testRepairAllTables() throws Exception
+    {
+        createTable("tbl1");
+        createTable("tbl2");
+        createTable("tbl3");
+
+        insertDataWithInconsistency("tbl1", 0, 30);
+        insertDataWithInconsistency("tbl2", 100, 30);
+        insertDataWithInconsistency("tbl3", 200, 30);
+
+        repairResolvingInconsistency(ksName);
+
+        assertDataOnAllNodes("tbl1", 0, 30);
+        assertDataOnAllNodes("tbl2", 100, 30);
+        assertDataOnAllNodes("tbl3", 200, 30);
+    }
+
+    @Test
+    public void testForceRepairWithNodeDown()
+    {
+        insertDataWithInconsistency(3, "tbl", 0, 50);
+
+        isolateNode(2, 1, 3);
+
+        List<Integer> liveNodes = List.of(1, 3);
+        assertAllFailure(repairConcurrently(liveNodes, withPR(ksName)));
+
+        repairFromNodesSuccess(liveNodes, ksName, "--force");
+
+        for (int node : liveNodes)
+        {
+            for (int i = 0; i < 50; i++)
+            {
+                Object[][] results = CLUSTER.get(node).executeInternal(
+                        "SELECT k, v FROM " + ksName + ".tbl WHERE k = ?", i);
+                assertEquals("Node " + node + " missing row k=" + i, 1, results.length);
+            }
+        }
+    }
+
+    @Test
+    public void testForceRepairWithAllNodesUp() throws Exception
+    {
+        insertDataWithInconsistency("tbl", 0, 50);
+
+        repairResolvingInconsistency(ksName, "--force");
+
+        assertDataOnAllNodes("tbl", 0, 50);
+    }
+
+    @Test
+    public void testRepairWithSpecificHosts()
+    {
+        String addr1 = getBroadcastAddress(1);
+        String addr3 = getBroadcastAddress(3);
+
+        insertDataWithInconsistency(3, "tbl", 0, 50);
+
+        // Node 2 is down, so normal repair should fail
+        isolateNode(2, 1, 3);
+
+        List<Integer> liveNodes = List.of(1, 3);
+        assertAllFailure(repairConcurrently(liveNodes, withPR(ksName)));
+
+        // Repair with --in-hosts scoped to only the live nodes should succeed
+        // Note: --in-hosts cannot be combined with -pr
+        String[] args = new String[]{ksName, "--in-hosts", addr1 + ',' + addr3};
+        assertAllSuccess(repairConcurrently(liveNodes, args));
+        assertAllSuccess(repairConcurrently(liveNodes, args));
+
+        for (int node : liveNodes)
+        {
+            for (int i = 0; i < 50; i++)
+            {
+                Object[][] results = CLUSTER.get(node).executeInternal(
+                        "SELECT k, v FROM " + ksName + ".tbl WHERE k = ?", i);
+                assertEquals("Node " + node + " missing row k=" + i, 1, results.length);
+            }
+        }
+    }
+
+    @Test
+    public void testMigrationUntrackedToTrackedCompletesViaRepair() throws Exception
+    {
+        setupUntracked();
+        insertDataWithInconsistency("tbl", 0, 100);
+
+        alterKeyspaceToTracked();
+        assertTrue("Migration should be in progress after ALTER", isMigrationInProgress());
+
+        repairResolvingInconsistency(ksName);
+        assertTrue("Migration should complete after repair", isMigrationComplete());
+
+        assertDataOnAllNodes("tbl", 0, 100);
+    }
+
+    @Test
+    public void testDataAccessibleDuringMigrationToTracked() throws Exception
+    {
+        setupUntracked();
+        dataAccessibleDuringMigration(() -> alterKeyspaceToTracked());
+    }
+
+    @Test
+    public void testDataAccessibleDuringMigrationToUntracked() throws Exception
+    {
+        dataAccessibleDuringMigration(() -> alterKeyspaceToUntracked());
+    }
+
+    private void dataAccessibleDuringMigration(Runnable alterKeyspace) throws Exception
+    {
+        insertDataWithInconsistency("tbl", 0, 50);
+
+        alterKeyspace.run();
+
+        Object[][] results = CLUSTER.coordinator(1).execute(
+                "SELECT k, v FROM " + ksName + ".tbl", ConsistencyLevel.ALL);
+        assertEquals("Pre-migration data should be readable", 50, results.length);
+
+        insertData("tbl", 50, 50);
+
+        results = CLUSTER.coordinator(1).execute(
+                "SELECT k, v FROM " + ksName + ".tbl", ConsistencyLevel.ALL);
+        assertEquals("All data should be readable during migration", 100, results.length);
+
+        repairResolvingInconsistency(ksName);
+        assertTrue("Migration should complete after repair", isMigrationComplete());
+
+        results = CLUSTER.coordinator(1).execute(
+                "SELECT k, v FROM " + ksName + ".tbl", ConsistencyLevel.ALL);
+        assertEquals("All data should be readable after migration", 100, results.length);
+
+        insertData("tbl", 100, 50);
+        results = CLUSTER.coordinator(1).execute(
+                "SELECT k, v FROM " + ksName + ".tbl", ConsistencyLevel.ALL);
+        assertEquals("All data including post-migration should be readable", 150, results.length);
+
+        assertDataOnAllNodes("tbl", 0, 150);
+    }
+
+    @Test
+    public void testMigrationTrackedToUntrackedCompletesViaRepair() throws Exception
+    {
+        insertDataWithInconsistency("tbl", 0, 100);
+
+        alterKeyspaceToUntracked();
+        assertTrue("Migration should be in progress after ALTER", isMigrationInProgress());
+
+        repairResolvingInconsistency(ksName);
+        assertTrue("Migration should complete after repair", isMigrationComplete());
+
+        assertDataOnAllNodes("tbl", 0, 100);
+    }
+
+    @Test
+    public void testForceRepairWithDeadNodeDoesNotAdvanceMigration()
+    {
+        repairWithDeadNodeDoesNotAdvanceMigration(withPR(ksName, "--force"));
+    }
+
+    @Test
+    public void testInHostsRepairWithDeadNodeDoesNotAdvanceMigration()
+    {
+        String addr1 = getBroadcastAddress(1);
+        String addr3 = getBroadcastAddress(3);
+        repairWithDeadNodeDoesNotAdvanceMigration(ksName, "--in-hosts", addr1 + ',' + addr3);
+    }
+
+    private void repairWithDeadNodeDoesNotAdvanceMigration(String... repairArgs)
+    {
+        setupUntracked();
+        insertDataWithInconsistency(3, "tbl", 0, 50);
+
+        alterKeyspaceToTracked();
+        assertTrue("Migration should be in progress", isMigrationInProgress());
+
+        isolateNode(2, 1, 3);
+
+        List<Integer> liveNodes = List.of(1, 3);
+        assertAllSuccess(repairConcurrently(liveNodes, repairArgs));
+
+        String ks = ksName;
+        assertTrue("Migration should not advance with dead nodes excluded",
+                CLUSTER.get(1).callOnInstance(() -> {
+                    ClusterMetadata metadata = ClusterMetadata.current();
+                    return metadata.mutationTrackingMigrationState.isMigrating(ks);
+                }));
+    }
+
+    @Test
+    public void testInHostsRepairSucceedsWhenSpecifiedHostIsNetworkBlocked()
+    {
+        String addr1 = getBroadcastAddress(1);
+        String addr3 = getBroadcastAddress(3);
+
+        insertDataWithInconsistency(3, "tbl", 0, 50);
+
+        // Block network to node 2 but do NOT mark it down in gossip
+        CLUSTER.filters().allVerbs().from(2).drop();
+        CLUSTER.filters().allVerbs().to(2).drop();
+
+        // Repair specifying only live hosts should succeed despite node 2 being blocked
+        List<Integer> liveNodes = List.of(1, 3);
+        String[] args = new String[]{ksName, "--in-hosts", addr1 + ',' + addr3};
+        assertAllSuccess(repairConcurrently(liveNodes, args));
+    }
+
+    @Test
+    public void testPreviewRepairDoesNotAdvanceMigration() throws Exception
+    {
+        setupUntracked();
+        insertDataWithInconsistency("tbl", 0, 50);
+
+        alterKeyspaceToTracked();
+        assertTrue("Migration should be in progress", isMigrationInProgress());
+
+        repairResolvingInconsistency(ksName, "--preview");
+
+        assertTrue("Migration should not advance with preview repair", isMigrationInProgress());
+    }
+
+    @Test
+    public void testSubrangeRepair() throws Exception
+    {
+        long[] primaryRange = getPrimaryRangeTokens(1);
+        String st = Long.toString(primaryRange[0]);
+        String et = Long.toString(primaryRange[1]);
+
+        insertDataWithInconsistency("tbl", 0, 100);
+
+        repairResolvingInconsistency(2, ALL_NODES, ksName, "-st", st, "-et", et);
+
+        List<Integer> keysInRange = keysInTokenRange(0, 100, primaryRange[0], primaryRange[1]);
+        assertFalse("Should have keys hashing into node 1's primary range", keysInRange.isEmpty());
+
+        assertDataOnAllNodes("tbl", keysInRange);
+    }
+
+    @Test
+    public void testSubrangeRepairAdvancesMigrationOnlyForSpecifiedRange() throws Exception
+    {
+        setupUntracked();
+        long[] primaryRange = getPrimaryRangeTokens(1);
+        String st = Long.toString(primaryRange[0]);
+        String et = Long.toString(primaryRange[1]);
+
+        insertDataWithInconsistency("tbl", 0, 100);
+
+        alterKeyspaceToTracked();
+        assertTrue("Full ring should be pending", isMigrationInProgress());
+
+        // During migration, subrange repair uses incremental repair. Running from all nodes
+        // on the same subrange causes anti-compaction conflicts, so repair from a single node.
+        repairResolvingInconsistency(2, List.of(1), ksName, "-st", st, "-et", et);
+
+        assertTrue("Migration should not be complete after subrange repair",
+                isMigrationInProgress());
+
+        // Verify the repaired range is no longer pending but other ranges still are
+        String ks = ksName;
+        long rangeStart = primaryRange[0];
+        long rangeEnd = primaryRange[1];
+        CLUSTER.get(1).runOnInstance(() -> {
+            ClusterMetadata metadata = ClusterMetadata.current();
+            KeyspaceMigrationInfo info = metadata.mutationTrackingMigrationState.getKeyspaceInfo(ks);
+            assertNotNull("Migration info should still exist", info);
+
+            Range<Token> repairedRange = new Range<>(new Murmur3Partitioner.LongToken(rangeStart),
+                    new Murmur3Partitioner.LongToken(rangeEnd));
+            for (var entry : info.pendingRangesPerTable.entrySet())
+            {
+                for (Range<Token> pending : entry.getValue())
+                {
+                    assertFalse("Repaired range should not overlap with pending ranges for table " + entry.getKey(),
+                            repairedRange.intersects(pending));
+                }
+            }
+        });
+
+        // Verify all keys in the repaired range are present on all nodes
+        List<Integer> keysInRange = keysInTokenRange(0, 100, primaryRange[0], primaryRange[1]);
+        assertFalse("Should have keys hashing into node 1's primary range", keysInRange.isEmpty());
+
+        assertDataOnAllNodes("tbl", keysInRange);
+    }
+
+    @Test
+    public void testRepairRejectsMixedMigratedAndPendingRanges()
+    {
+        setupUntracked();
+        insertData("tbl", 0, 50);
+
+        alterKeyspaceToTracked();
+        assertTrue("Migration should be in progress after ALTER", isMigrationInProgress());
+
+        long[] primaryRange = getPrimaryRangeTokens(1);
+        String st = Long.toString(primaryRange[0]);
+        String et = Long.toString(primaryRange[1]);
+
+        // Repair node 1's primary range to advance migration for that subrange only.
+        // Run from single node to avoid anti-compaction conflicts during migration IR.
+        nodetoolRepair(1, ksName, "-st", st, "-et", et).asserts().success();
+
+        // Now attempt a repair with a range that straddles the migrated/pending boundary.
+        // Node 1's primary range has been repaired (no longer pending), but the range
+        // immediately after is still pending. A range spanning both should be rejected.
+        String straddleSt = Long.toString(primaryRange[1] - 1);
+        String straddleEt = Long.toString(primaryRange[1] + 1000);
+        NodeToolResult result = nodetoolRepair(1, ksName, "-st", straddleSt, "-et", straddleEt);
+        result.asserts().failure();
+        assertTrue("Expected partial overlap error but got: " + result.getStderr(),
+                result.getStderr().contains("partially overlap with migration pending ranges"));
+    }
+
+    @Test
+    public void testRepairTimeout()
+    {
+        insertData("tbl", 0, 50);
+
+        CLUSTER.filters().verbs(Verb.MT_SYNC_REQ.id).to(2).drop();
+        CLUSTER.filters().verbs(Verb.MT_SYNC_REQ.id).from(2).drop();
+
+        List<NodeToolResult> results = repairConcurrently(ALL_NODES, withPR(ksName));
+        assertAllFailure(results);
+        for (NodeToolResult r : results)
+            assertTrue("Expected timeout error but got: " + r.getStderr(),
+                    r.getStderr().contains("Mutation tracking sync timed out"));
+    }
+
+    /**
+     * Exercises the onFailure callback in MutationTrackingSyncCoordinator.sendSyncRequests().
+     * Unlike testRepairTimeout (which drops MT_SYNC_REQ entirely so the request times out),
+     * this test makes the remote handler throw an exception, which sends a FAILURE_RSP back
+     * to the coordinator, triggering the onFailure -> fail() path.
+     */
+    @Test
+    public void testSyncFailureResponse()
+    {
+        insertData("tbl", 0, 50);
+
+        // The matcher throwing causes uncaught exceptions on the receiving nodes' stage threads.
+        // These are expected, so filter them out to avoid failing at cluster close.
+        CLUSTER.setUncaughtExceptionsFilter((nodeNum, throwable) ->
+                throwable.getMessage() != null && throwable.getMessage().contains("sync failure injected"));
+        try
+        {
+            CLUSTER.verbs(Verb.MT_SYNC_REQ).messagesMatching(of(m -> {
+                throw new RuntimeException("sync failure injected");
+            })).drop();
+
+            List<NodeToolResult> results = repairConcurrently(ALL_NODES, withPR(ksName));
+            assertAllFailure(results);
+            for (NodeToolResult r : results)
+                assertTrue("Expected sync failure error but got: " + r.getStderr(),
+                        r.getStderr().contains("Mutation tracking sync failed"));
+        }
+        finally
+        {
+            CLUSTER.setUncaughtExceptionsFilter((BiPredicate<Integer, Throwable>) null);
+        }
+    }
+
+    /**
+     * During migration from untracked to tracked, incremental repair runs anti-compaction
+     * on SSTables that were written before tracking was enabled. When an SSTable partially
+     * overlaps the repair range, anti-compaction must split it by rewriting through
+     * SSTableWriter. The "inside repair range" writer gets pendingRepair set to the session ID.
+     *
+     * SSTableWriter.finalizeMetadata() must tolerate pendingRepair being set on a tracked
+     * table during migration. This test uses a narrow subrange to force anti-compaction to
+     * split SSTables (rather than just mutating fully-contained ones in place).
+     */
+    @Test
+    public void testMigrationSubrangeRepairAntiCompactionSplitsSSTables() throws Exception
+    {
+        setupUntracked();
+
+        // Write data and flush so SSTables span the full token ring on each node.
+        insertData("tbl", 0, 500);
+        for (int i = 1; i <= NUM_NODES; i++)
+            CLUSTER.get(i).flush(ksName);
+
+        alterKeyspaceToTracked();
+        assertTrue("Migration should be in progress", isMigrationInProgress());
+
+        // Use a subrange that's well within one local range but wide enough to contain
+        // data. With 3 nodes, node 2's primary range is approximately
+        // (-3074457345618258603, 3074457345618258602]. A range of (0, 3000000000000000000]
+        // is fully contained in that range and covers ~16% of the ring, so ~80 of our 500
+        // rows should hash into it. SSTables from the flush span the entire ring, so they
+        // will NOT be fully contained in this narrow range. Anti-compaction must split them
+        // via SSTableWriter, exercising the pendingRepair code path in finalizeMetadata().
+        String st = "0";
+        String et = "3000000000000000000";
+
+        // Run repair from a single node to avoid anti-compaction conflicts.
+        // This should succeed: anti-compaction splits SSTables and the repair completes.
+        NodeToolResult result = nodetoolRepair(1, ksName, "-st", st, "-et", et);
+        result.asserts().success();
+    }
+
+    @Test
+    public void testRepairSyncTimeout()
+    {
+        insertDataWithInconsistency("tbl", 0, 50);
+
+        // Drop only offset broadcasts so MT_SYNC_REQ/RSP can succeed but
+        // reconciliation never completes, triggering mutation_tracking_sync_timeout
+        CLUSTER.filters().verbs(Verb.BROADCAST_LOG_OFFSETS.id).drop();
+
+        List<NodeToolResult> results = repairConcurrently(ALL_NODES, withPR(ksName));
+        assertAllFailure(results);
+        for (NodeToolResult r : results)
+            assertTrue("Expected sync timeout error but got: " + r.getStderr(),
+                    r.getStderr().contains("Mutation tracking sync timed out"));
+    }
+
+    /**
+     * Verifies that a topology change during an active mutation tracking sync causes
+     * the repair to fail with "topology changed during sync".
+     *
+     * The strategy:
+     * 1. Insert data so sync has work to do
+     * 2. Drop BROADCAST_LOG_OFFSETS so the sync coordinator stays alive waiting for
+     *    offset reconciliation
+     * 3. Start repair in a background thread
+     * 4. Wait until the sync request has been sent (confirming sync is active)
+     * 5. ALTER KEYSPACE to change RF (3 -> 2), which triggers REPLICA_GROUP ->
+     *    withUpdatedMetadata -> new Shard instances
+     * 6. Turn off the BROADCAST_LOG_OFFSETS filter so offset broadcasts resume,
+     *    triggering onOffsetsReceived -> recaptureTargets -> checkForTopologyChange
+     *    which detects the identity mismatch and fails the repair
+     * 7. Assert the repair failed with the expected topology change message
+     */
+    @Test
+    public void testRepairFailsOnTopologyChange() throws Exception
+    {
+        insertData("tbl", 0, 50);
+
+        // Block offset broadcasts so the sync coordinator stays alive waiting
+        IMessageFilters.Filter offsetFilter = CLUSTER.filters().verbs(Verb.BROADCAST_LOG_OFFSETS.id).drop();
+
+        // Use a latch to detect when the sync request has been sent, meaning
+        // the sync coordinator is active and tracking shard references
+        CountDownLatch syncStarted = new CountDownLatch(1);
+        IMessageFilters.Filter syncObserver = CLUSTER.verbs(Verb.MT_SYNC_REQ).messagesMatching(
+                (from, to, msg) -> {
+                    syncStarted.countDown();
+                    return false; // don't drop the message
+                }).drop();
+
+        // Start repair in background
+        Future<NodeToolResult> repairFuture = executor.submit(() -> nodetoolRepair(1, withPR(ksName)));
+
+        // Wait until sync is active. The latch fires when MT_SYNC_REQ is sent, which
+        // happens after shardStates is fully populated in start(), so no additional
+        // delay is needed.
+        assertTrue("Timed out waiting for sync to start",
+                   syncStarted.await(30, TimeUnit.SECONDS));
+
+        // ALTER KEYSPACE to change RF from 3 to 2 — this changes the participants for
+        // every range, triggering REPLICA_GROUP -> withUpdatedMetadata -> new Shard instances.
+        // The sync coordinator's shardStates still holds references to the old Shard objects.
+        CLUSTER.schemaChange("ALTER KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 2} " +
+                             "AND replication_type='tracked'");
+
+        // Remove the sync observer since it's no longer needed
+        syncObserver.off();
+
+        // Turn off the offset broadcast filter so broadcasts resume. When an offset
+        // broadcast arrives, it calls onOffsetsReceived -> recaptureTargets ->
+        // checkForTopologyChange, which will detect that the current Shard instances
+        // (new objects from withUpdatedMetadata) differ from the ones stored in
+        // shardStates (reference equality check), and fail the repair.
+        offsetFilter.off();
+
+        NodeToolResult result = repairFuture.get(30, TimeUnit.SECONDS);
+        result.asserts().failure();
+        assertTrue("Expected topology change error but got: " + result.getStderr(),
+                   result.getStderr().contains("topology changed during sync"));
+    }
+}

test/distributed/org/apache/cassandra/distributed/test/replication/MutationTrackingSyncCoordinatorTest.java

@@ -0,0 +1,282 @@
+/*
+ * Licensed to the Apache Software Foundation (ASF) under one
+ * or more contributor license agreements.  See the NOTICE file
+ * distributed with this work for additional information
+ * regarding copyright ownership.  The ASF licenses this file
+ * to you under the Apache License, Version 2.0 (the
+ * "License"); you may not use this file except in compliance
+ * with the License.  You may obtain a copy of the License at
+ *
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+package org.apache.cassandra.distributed.test.replication;
+
+import java.time.Duration;
+import java.util.concurrent.CompletableFuture;
+
+import org.apache.cassandra.tcm.ClusterMetadata;
+import org.junit.Test;
+
+import org.apache.cassandra.distributed.Cluster;
+import org.apache.cassandra.distributed.api.ConsistencyLevel;
+import org.apache.cassandra.distributed.test.TestBaseImpl;
+import org.apache.cassandra.dht.Murmur3Partitioner;
+import org.apache.cassandra.dht.Range;
+import org.apache.cassandra.dht.Token;
+import org.apache.cassandra.repair.RepairJobDesc;
+import org.apache.cassandra.repair.SharedContext;
+import org.apache.cassandra.replication.MutationTrackingService;
+import org.apache.cassandra.replication.MutationTrackingSyncCoordinator;
+import org.apache.cassandra.utils.TimeUUID;
+import org.awaitility.Awaitility;
+
+import static org.junit.Assert.*;
+
+/**
+ * Distributed tests for MutationTrackingSyncCoordinator.
+ *
+ * Tests that the sync coordinator correctly waits for offset convergence
+ * across all nodes in a cluster.
+ */
+public class MutationTrackingSyncCoordinatorTest extends TestBaseImpl
+{
+    private static final String KS_NAME = "sync_test_ks";
+    private static final String TBL_NAME = "sync_test_tbl";
+
+    private void createTrackedKeyspace(Cluster cluster, String keyspaceSuffix)
+    {
+        String ksName = KS_NAME + keyspaceSuffix;
+        cluster.schemaChange("CREATE KEYSPACE " + ksName + " WITH replication = " +
+                             "{'class': 'SimpleStrategy', 'replication_factor': 3} " +
+                             "AND replication_type='tracked'");
+        cluster.schemaChange("CREATE TABLE " + ksName + '.' + TBL_NAME + " (k int PRIMARY KEY, v int)");
+    }
+
+    private String tableName(String suffix)
+    {
+        return KS_NAME + suffix + '.' + TBL_NAME;
+    }
+
+    private void pauseOffsetBroadcasts(Cluster cluster, boolean pause)
+    {
+        for (int i = 1; i <= cluster.size(); i++)
+            cluster.get(i).runOnInstance(() -> MutationTrackingService.instance().pauseOffsetBroadcast(pause));
+    }
+
+    private static Range<Token> fullTokenRange()
+    {
+        return new Range<>(
+            new Murmur3Partitioner.LongToken(Long.MIN_VALUE),
+            new Murmur3Partitioner.LongToken(Long.MAX_VALUE)
+        );
+    }
+
+    @Test
+    public void testSyncCoordinatorCompletesWhenNoShards() throws Throwable
+    {
+        try (Cluster cluster = builder().withNodes(3).start())
+        {
+            createTrackedKeyspace(cluster, "");
+
+            // Create a sync coordinator for a range that has no data
+            // It should complete immediately since there are no offsets to sync
+            Boolean completed = cluster.get(1).callOnInstance(() -> {
+                Range<Token> range = fullTokenRange();
+                RepairJobDesc desc = new RepairJobDesc(TimeUUID.Generator.nextTimeUUID(),
+                                                       TimeUUID.Generator.nextTimeUUID(),
+                                                       KS_NAME, "", java.util.List.of(range));
+                MutationTrackingSyncCoordinator coordinator = new MutationTrackingSyncCoordinator(
+                    SharedContext.Global.instance, desc, null, ClusterMetadata.current());
+                coordinator.start();
+
+                try
+                {
+                    coordinator.awaitCompletion();
+                    return true;
+                }
+                catch (Exception e)
+                {
+                    return false;
+                }
+            });
+
+            assertTrue("Sync coordinator should complete when there are no pending offsets", completed);
+        }
+    }
+
+    @Test
+    public void testSyncCoordinatorWaitsForAllReplicasMutations() throws Throwable
+    {
+        // Enable repair retries with a short request timeout so that the dropped MT_SYNC_RSP
+        // from node 1 causes a quick timeout and retry rather than a 2-minute wait and failure.
+        // After the message filter is reset, the retried MT_SYNC_REQ will get a response.
+        try (Cluster cluster = builder().withNodes(3)
+                                        .withConfig(config -> config.set("request_timeout", "1000ms")
+                                                                    .set("repair.retries.max_attempts", 10)
+                                                                    .set("repair.retries.base_sleep_time", "100ms")
+                                                                    .set("repair.retries.max_sleep_time", "500ms"))
+                                        .start())
+        {
+            createTrackedKeyspace(cluster, "3");
+
+            // Block all messages FROM node 1 to prevent write replication
+            // and also to drop MT_SYNC_RSP from node 1 back to the coordinator.
+            // This ensures that write only succeeds locally on node 1 and the
+            // sync coordinator can't get node 1's sync response.
+            cluster.filters().allVerbs().from(1).drop();
+
+            cluster.coordinator(1).execute(
+                "INSERT INTO " + tableName("3") + " (k, v) VALUES (1, 1)",
+                ConsistencyLevel.ONE
+            );
+
+            // Start MutationTrackingSyncCoordinator on node 2 in a separate thread.
+            // It should wait for offsets to sync since node 1's sync response is being dropped.
+            // The coordinator sends MT_SYNC_REQ to nodes 1 and 3. Node 3's response comes back
+            // but node 1's response is dropped. The coordinator stays blocked because
+            // pendingSyncResponses still contains node 1. After the filter is reset, the
+            // retried MT_SYNC_REQ will succeed and the coordinator can proceed.
+            CompletableFuture<Boolean> coordinatorFuture = CompletableFuture.supplyAsync(() -> cluster.get(2).callOnInstance(() -> {
+                Range<Token> range = fullTokenRange();
+                RepairJobDesc desc = new RepairJobDesc(TimeUUID.Generator.nextTimeUUID(),
+                                                       TimeUUID.Generator.nextTimeUUID(),
+                                                       KS_NAME + '3', "", java.util.List.of(range));
+                MutationTrackingSyncCoordinator coordinator = new MutationTrackingSyncCoordinator(
+                    SharedContext.Global.instance, desc, null, ClusterMetadata.current());
+                coordinator.start();
+
+                try
+                {
+                    coordinator.awaitCompletion();
+                    return true;
+                }
+                catch (Exception e)
+                {
+                    return false;
+                }
+            }));
+
+            // Wait until node 1 has the data
+            Awaitility.await()
+                      .atMost(Duration.ofSeconds(5))
+                      .pollInterval(Duration.ofMillis(100))
+                      .untilAsserted(() -> {
+                          Object[][] results = cluster.get(1).executeInternal(
+                          "SELECT k, v FROM " + tableName("3") + " WHERE k = 1");
+                          assertEquals("Node 1 should have the data", 1, results.length);
+                      });
+
+            // Verify other nodes shouldn't have the data yet since we have blocked messages
+            for (int i = 2; i <= 3; i++)
+            {
+                Object[][] results = cluster.get(i).executeInternal(
+                    "SELECT k, v FROM " + tableName("3") + " WHERE k = 1"
+                );
+                assertEquals("Node " + i + " should not have data yet", 0, results.length);
+            }
+
+            // Verify coordinator stays blocked for at least 2 seconds while node 1's
+            // sync response is being dropped. The coordinator can't complete because
+            // pendingSyncResponses still contains node 1.
+            Awaitility.await()
+                      .during(Duration.ofSeconds(2))
+                      .atMost(Duration.ofSeconds(3))
+                      .until(() -> !coordinatorFuture.isDone());
+
+            // Reset filter so that retried MT_SYNC_REQ to node 1 can get a response,
+            // and offset broadcasts from node 1 can reach other nodes.
+            cluster.filters().reset();
+
+            // Force offset broadcasts on all nodes to drive reconciliation.
+            // After the sync response from node 1 establishes targets, the coordinator
+            // needs to see that all replicas have caught up via offset broadcasts.
+            for (int i = 1; i <= 3; i++)
+                cluster.get(i).runOnInstance(() -> MutationTrackingService.instance().broadcastOffsetsForTesting());
+
+            // Wait for coordinator to complete
+            Awaitility.await()
+                      .atMost(Duration.ofSeconds(30))
+                      .pollInterval(Duration.ofMillis(200))
+                      .until(coordinatorFuture::isDone);
+
+            assertTrue("Coordinator should complete successfully", coordinatorFuture.get());
+
+            // Verify data propagated to all replicas
+            for (int i = 1; i <= 3; i++)
+            {
+                final int nodeId = i;
+                Awaitility.await()
+                          .atMost(Duration.ofSeconds(10))
+                          .pollInterval(Duration.ofMillis(100))
+                          .untilAsserted(() -> {
+                              Object[][] results = cluster.get(nodeId).executeInternal(
+                              "SELECT k, v FROM " + tableName("3") + " WHERE k = 1");
+                              assertEquals("Node " + nodeId + " should have the data", 1, results.length);
+                              assertEquals(1, results[0][0]);
+                              assertEquals(1, results[0][1]);
+                          });
+            }
+        }
+    }
+
+    @Test
+    public void testSyncCoordinatorCancel() throws Throwable
+    {
+        try (Cluster cluster = builder().withNodes(3).start())
+        {
+            createTrackedKeyspace(cluster, "4");
+
+            // Pause offset broadcasts on all nodes to prevent sync from completing
+            pauseOffsetBroadcasts(cluster, true);
+
+            for (int i = 0; i < 100; i++)
+            {
+                cluster.coordinator(1).execute(
+                    "INSERT INTO " + tableName("4") + " (k, v) VALUES (?, ?)",
+                    ConsistencyLevel.ONE, i, i);
+            }
+
+            // Start coordinator - it will be stuck waiting for offsets
+            Boolean wasCancelled = cluster.get(1).callOnInstance(() -> {
+                Range<Token> range = fullTokenRange();
+                RepairJobDesc desc = new RepairJobDesc(TimeUUID.Generator.nextTimeUUID(),
+                                                       TimeUUID.Generator.nextTimeUUID(),
+                                                       KS_NAME + '4', "", java.util.List.of(range));
+                MutationTrackingSyncCoordinator coordinator = new MutationTrackingSyncCoordinator(
+                    SharedContext.Global.instance, desc, null, ClusterMetadata.current());
+                coordinator.start();
+
+                try
+                {
+                    Thread.sleep(100);
+                }
+                catch (InterruptedException e)
+                {
+                    Thread.currentThread().interrupt();
+                    return false;
+                }
+
+                coordinator.cancel(); // Cancel it
+
+                // Verify it was cancelled
+                try
+                {
+                    coordinator.awaitCompletion();
+                    return false; // Should have thrown
+                }
+                catch (Exception e)
+                {
+                    Throwable cause = e.getCause() != null ? e.getCause() : e;
+                    return cause.getMessage() != null && cause.getMessage().contains("cancelled");
+                }
+            });
+            assertTrue("Sync coordinator should be cancelled", wasCancelled);
+        }
+    }
+}

test/distributed/org/apache/cassandra/distributed/test/sai/PartialUpdateHandlingTest.java

@@ -236,7 +236,11 @@ public class PartialUpdateHandlingTest extends TestBaseImpl
                 CLUSTER.coordinator(1).execute(insert.toString(), ConsistencyLevel.ALL);
             }
 
-            CLUSTER.get(1).nodetoolResult("repair", specification.keyspaceName()).asserts().success();
+            // Background reconciliation doesn't exist/work so incremental repair just hangs waiting for reconciliation that never occurs
+            if (specification.replicationType.isTracked())
+                CLUSTER.get(1).nodetoolResult("repair", "-full", specification.keyspaceName()).asserts().success();
+            else
+                CLUSTER.get(1).nodetoolResult("repair", specification.keyspaceName()).asserts().success();
         }
 
         public void writeUnrepairedRows()

test/distributed/org/apache/cassandra/distributed/test/tracking/MutationTrackingBounceTest.java

@@ -31,6 +31,7 @@ import org.apache.cassandra.harry.execution.InJvmDTestVisitExecutor;
 import org.apache.cassandra.harry.gen.Generator;
 import org.apache.cassandra.harry.gen.SchemaGenerators;
 import org.apache.cassandra.replication.MutationJournal;
+import org.junit.Ignore;
 import org.junit.Test;
 
 
@@ -45,48 +46,76 @@ public class MutationTrackingBounceTest extends FuzzTestBase
     {
         try (Cluster cluster = builder().withNodes(1).start())
         {
-            int tables = 10;
-            int writesPerKey = 2;
-            int pks = 100;
-            withRandom(rng -> {
-                cluster.schemaChange(String.format("CREATE KEYSPACE %s WITH replication = {'class': 'SimpleStrategy', 'replication_factor': 1} " +
-                                                   "AND replication_type='tracked'",
-                                                   KEYSPACE));
+            bounceTest(cluster, 1, 1);
+        }
+    }
 
-                List<HistoryBuilder> builders = new ArrayList<>();
-                for (int i = 0; i < tables; i++)
-                {
-                    Generator<SchemaSpec> schemaGen = SchemaGenerators.trivialSchema(KEYSPACE, () -> "mutation_tracking_bounce_" + (builders.size() + 1), POPULATION,
-                                                                                     SchemaSpec.optionsBuilder());
+    @Test
+    public void bounceTestMultiNode() throws Throwable
+    {
+        try (Cluster cluster = builder().withNodes(3).start())
+        {
+            bounceTest(cluster, 3, 1);
+        }
+    }
 
-                    SchemaSpec schema = schemaGen.generate(rng);
-                    cluster.schemaChange(schema.compile());
-                    builders.add(new ReplayingHistoryBuilder(schema.valueGenerators,
-                                                             hb -> InJvmDTestVisitExecutor.builder()
-                                                                                          .consistencyLevel(ConsistencyLevel.QUORUM)
-                                                                                          .build(schema, hb, cluster)));
-                }
+    @Ignore("https://issues.apache.org/jira/browse/CASSANDRA-21256")
+    @Test
+    public void doubleBounceTestMultiNode() throws Throwable
+    {
+        try (Cluster cluster = builder().withNodes(3).start())
+        {
+            bounceTest(cluster, 3, 2);
+        }
+    }
 
-                int counter = 0;
-                for (int pk = 0; pk < pks; pk++) {
-                    for (HistoryBuilder history : builders)
-                        for (int i = 0; i < writesPerKey; i++)
-                            history.insert(pk);
+    private void bounceTest(Cluster cluster, int rf, int bounces) throws Throwable
+    {
+        int tables = 10;
+        int writesPerKey = 2;
+        int pks = 100;
+        withRandom(rng -> {
+            cluster.schemaChange(String.format("CREATE KEYSPACE %s WITH replication = {'class': 'SimpleStrategy', 'replication_factor': %d} " +
+                                               "AND replication_type='tracked'",
+                                               KEYSPACE, rf));
+
+            List<HistoryBuilder> builders = new ArrayList<>();
+            for (int i = 0; i < tables; i++)
+            {
+                Generator<SchemaSpec> schemaGen = SchemaGenerators.trivialSchema(KEYSPACE, () -> "mutation_tracking_bounce_" + (builders.size() + 1), POPULATION,
+                                                                                 SchemaSpec.optionsBuilder());
+
+                SchemaSpec schema = schemaGen.generate(rng);
+                cluster.schemaChange(schema.compile());
+                builders.add(new ReplayingHistoryBuilder(schema.valueGenerators,
+                                                         hb -> InJvmDTestVisitExecutor.builder()
+                                                                                      .consistencyLevel(ConsistencyLevel.QUORUM)
+                                                                                      .build(schema, hb, cluster)));
+            }
+
+            int counter = 0;
+            for (int pk = 0; pk < pks; pk++)
+            {
+                for (HistoryBuilder history : builders)
+                    for (int i = 0; i < writesPerKey; i++)
+                        history.insert(pk);
 
                     if (++counter % 10 == 0)
                         cluster.get(1).runOnInstance(() -> MutationJournal.instance().closeCurrentSegmentForTestingIfNonEmpty());
                 }
 
+            for (int bounce = 0; bounce < bounces; bounce++)
+            {
                 ClusterUtils.stopUnchecked(cluster.get(1));
                 cluster.get(1).startup();
+            }
 
-                for (int pk = 0; pk < pks; pk++)
-                    for (HistoryBuilder history : builders)
-                        for (int i = 0; i < 10; i++)
-                            history.selectPartition(pk);
+            for (int pk = 0; pk < pks; pk++)
+                for (HistoryBuilder history : builders)
+                    for (int i = 0; i < 10; i++)
+                        history.selectPartition(pk);
 
-                cluster.get(1).runOnInstance(new MutationTrackingBounce_ValidateRunnable(tables * pks * writesPerKey));
-            });
-        }
+            cluster.get(1).runOnInstance(new MutationTrackingBounce_ValidateRunnable(tables * pks * writesPerKey));
+        });
     }
 }

test/distributed/org/apache/cassandra/distributed/test/tracking/MutationTrackingTest.java

@@ -418,19 +418,22 @@ public class MutationTrackingTest extends TestBaseImpl
                 Assert.assertEquals(0, summary.get(logId).reconciled.offsetCount());
             });
 
-            // resume the reconciler
+            // resume the reconciler and spin until reconciliation completes.
+            // The reconciler retries with PUSH_MUTATION_REQ whose response inherits the
+            // request's expiry (write_request_timeout). Under load the response can arrive
+            // after that expiry and be silently dropped by InboundMessageHandler, requiring
+            // a retry cycle. Spinning accommodates multiple retry rounds.
             cluster.get(1).runOnInstance(() -> MutationTrackingService.instance().resumeActiveReconciler());
-            Thread.sleep(1000); // wait for reconiciler to do its job
 
-            cluster.get(1).runOnInstance(() ->
-            {
-                TableMetadata table = Schema.instance.getTableMetadata(keyspaceName, "tbl");
-                DecoratedKey dk = Murmur3Partitioner.instance.decorateKey(ByteBufferUtil.bytes(1));
-                MutationSummary summary = MutationTrackingService.instance().createSummaryForKey(dk, table.id, false);
-                CoordinatorLogId logId = getOnlyLogId(summary);
-                Assert.assertEquals(0, summary.get(logId).unreconciled.offsetCount());
-                Assert.assertEquals(1, summary.get(logId).reconciled.offsetCount());
-            });
+            Util.spinUntilTrue(() ->
+                cluster.get(1).callOnInstance(() -> {
+                    TableMetadata table = Schema.instance.getTableMetadata(keyspaceName, "tbl");
+                    DecoratedKey dk = Murmur3Partitioner.instance.decorateKey(ByteBufferUtil.bytes(1));
+                    MutationSummary summary = MutationTrackingService.instance().createSummaryForKey(dk, table.id, false);
+                    CoordinatorLogId logId = getOnlyLogId(summary);
+                    return summary.get(logId).unreconciled.offsetCount() == 0
+                           && summary.get(logId).reconciled.offsetCount() == 1;
+                }), 10);
         }
     }
 }

test/unit/org/apache/cassandra/db/MutationTrackingMutationVerbHandlerMigrationTest.java

@@ -24,7 +24,6 @@ import org.junit.BeforeClass;
 import org.junit.Test;
 
 import org.apache.cassandra.ServerTestUtils;
-import org.apache.cassandra.config.DatabaseDescriptor;
 import org.apache.cassandra.db.ReadCommand.PotentialTxnConflicts;
 import org.apache.cassandra.db.marshal.Int32Type;
 import org.apache.cassandra.db.partitions.PartitionUpdate;

test/unit/org/apache/cassandra/service/replication/migration/MutationTrackingMigrationStateTest.java

@@ -153,7 +153,7 @@ public class MutationTrackingMigrationStateTest
         assertNotSame(state, completed);
 
         assertFalse(completed.hasMigratingKeyspaces());
-        assertNull(completed.getKeyspaceInfo("test_ks"));
+        assertFalse(completed.isMigrating("test_ks"));
     }
 
     @Test
@@ -169,7 +169,7 @@ public class MutationTrackingMigrationStateTest
         MutationTrackingMigrationState removed = state.dropKeyspaces(epoch, Collections.singleton("test_ks"));
 
         assertFalse(removed.hasMigratingKeyspaces());
-        assertNull(removed.getKeyspaceInfo("test_ks"));
+        assertFalse(removed.isMigrating("test_ks"));
     }
 
     @Test
@@ -253,7 +253,7 @@ public class MutationTrackingMigrationStateTest
         state = state.withRangesRepairedForTable("ks1", testTableId, Collections.singleton(fullRing), epoch);
 
         assertEquals(1, state.keyspaceInfo.size());
-        assertNull(state.getKeyspaceInfo("ks1"));
+        assertFalse(state.isMigrating("ks1"));
 
         // ks2 should still have full ring pending
         KeyspaceMigrationInfo expectedKs2AfterKs1Complete = createExpectedKeyspaceMigrationInfo(

test/unit/org/apache/cassandra/tcm/transformations/AdvanceMutationTrackingMigrationTest.java

@@ -139,7 +139,7 @@ public class AdvanceMutationTrackingMigrationTest
 
         // Verify migration was auto-completed (keyspace removed from state)
         assertFalse(updated.mutationTrackingMigrationState.hasMigratingKeyspaces());
-        assertNull(updated.mutationTrackingMigrationState.getKeyspaceInfo("test_ks"));
+        assertFalse(updated.mutationTrackingMigrationState.isMigrating("test_ks"));
     }
 
     @Test
@@ -240,7 +240,7 @@ public class AdvanceMutationTrackingMigrationTest
         assertTrue(result2.isSuccess());
         ClusterMetadata afterComplete = result2.success().metadata;
 
-        assertNull(afterComplete.mutationTrackingMigrationState.getKeyspaceInfo("test_ks"));
+        assertFalse(afterComplete.mutationTrackingMigrationState.isMigrating("test_ks"));
         assertFalse(afterComplete.mutationTrackingMigrationState.hasMigratingKeyspaces());
     }
 
@@ -285,7 +285,7 @@ public class AdvanceMutationTrackingMigrationTest
         assertTrue(result2.isSuccess());
         ClusterMetadata afterTable2 = result2.success().metadata;
 
-        assertNull(afterTable2.mutationTrackingMigrationState.getKeyspaceInfo("test_ks"));
+        assertFalse(afterTable2.mutationTrackingMigrationState.isMigrating("test_ks"));
         assertFalse(afterTable2.mutationTrackingMigrationState.hasMigratingKeyspaces());
     }
 

test/unit/org/apache/cassandra/tcm/transformations/AlterSchemaMutationTrackingTest.java

@@ -42,8 +42,7 @@ import org.apache.cassandra.tcm.Epoch;
 import static org.apache.cassandra.cql3.CQLTester.schemaChange;
 import static org.junit.Assert.assertEquals;
 import static org.junit.Assert.assertFalse;
-import static org.junit.Assert.assertNotNull;
-import static org.junit.Assert.assertNull;
+import static org.junit.Assert.assertTrue;
 
 /**
  * Tests for AlterSchema auto-starting mutation tracking migration when replication type changes.
@@ -80,7 +79,7 @@ public class AlterSchemaMutationTrackingTest
         schemaChange(String.format("CREATE TABLE %s.tbl (pk int PRIMARY KEY, val int)", ksName));
 
         ClusterMetadata metadata = ClusterMetadata.current();
-        assertNull(metadata.mutationTrackingMigrationState.getKeyspaceInfo(ksName));
+        assertFalse(metadata.mutationTrackingMigrationState.isMigrating(ksName));
 
         // Alter tracked replication
         schemaChange(String.format("ALTER KEYSPACE %s WITH replication_type = 'tracked'", ksName));
@@ -116,7 +115,7 @@ public class AlterSchemaMutationTrackingTest
         schemaChange(String.format("CREATE TABLE %s.tbl (pk int PRIMARY KEY, val int)", ksName));
 
         ClusterMetadata metadata = ClusterMetadata.current();
-        assertNull(metadata.mutationTrackingMigrationState.getKeyspaceInfo(ksName));
+        assertFalse(metadata.mutationTrackingMigrationState.isMigrating(ksName));
 
         // Alter keyspace to untracked
         schemaChange(String.format("ALTER KEYSPACE %s WITH replication_type = 'untracked'", ksName));
@@ -149,7 +148,7 @@ public class AlterSchemaMutationTrackingTest
             "AND replication_type = 'untracked'"
         );
         schemaChange(String.format("CREATE TABLE %s.tbl (pk int PRIMARY KEY, val int)", ksName));
-        assertNull(ClusterMetadata.current().mutationTrackingMigrationState.getKeyspaceInfo(ksName));
+        assertFalse(ClusterMetadata.current().mutationTrackingMigrationState.isMigrating(ksName));
 
         // Alter keyspace without changing replication type
         schemaChange(String.format(
@@ -158,7 +157,7 @@ public class AlterSchemaMutationTrackingTest
         ));
 
         // confirm no migrations were started
-        assertNull(ClusterMetadata.current().mutationTrackingMigrationState.getKeyspaceInfo(ksName));
+        assertFalse(ClusterMetadata.current().mutationTrackingMigrationState.isMigrating(ksName));
     }
 
     @Test
@@ -221,8 +220,8 @@ public class AlterSchemaMutationTrackingTest
         schemaChange(String.format("CREATE TABLE %s.tbl (pk int PRIMARY KEY, val int)", ksName));
 
         ClusterMetadata metadata = ClusterMetadata.current();
-        assertNull("Should have no migration before first alter for " + ksName,
-                   metadata.mutationTrackingMigrationState.getKeyspaceInfo(ksName));
+        assertFalse("Should have no migration before first alter for " + ksName,
+                    metadata.mutationTrackingMigrationState.isMigrating(ksName));
 
         // Alter to tracked (untracked → tracked)
         schemaChange(String.format("ALTER KEYSPACE %s WITH replication_type = 'tracked'", ksName));
@@ -249,7 +248,7 @@ public class AlterSchemaMutationTrackingTest
 
         // this should auto-complete the migration, since none of the ranges from the initial alter completed migration
         metadata = ClusterMetadata.current();
-        assertNull(metadata.mutationTrackingMigrationState.getKeyspaceInfo(ksName));
+        assertFalse(metadata.mutationTrackingMigrationState.isMigrating(ksName));
 
         // Alter back to tracked again
         schemaChange(String.format("ALTER KEYSPACE %s WITH replication_type = 'tracked'", ksName));
@@ -283,13 +282,13 @@ public class AlterSchemaMutationTrackingTest
         schemaChange(String.format("ALTER KEYSPACE %s WITH replication_type = 'tracked'", ksName));
 
         ClusterMetadata metadata = ClusterMetadata.current();
-        assertNotNull(metadata.mutationTrackingMigrationState.getKeyspaceInfo(ksName));
+        assertTrue(metadata.mutationTrackingMigrationState.isMigrating(ksName));
 
         // Drop the keyspace & confirm migration is also removed
         schemaChange(String.format("DROP KEYSPACE %s", ksName));
 
         ClusterMetadata afterDrop = ClusterMetadata.current();
-        assertNull(afterDrop.mutationTrackingMigrationState.getKeyspaceInfo(ksName));
+        assertFalse(afterDrop.mutationTrackingMigrationState.isMigrating(ksName));
     }
 
     @Test