natas_hw
/
cassandra
mirror of https://github.com/apache/cassandra
0
0
Fork 0
Code Issues Packages Projects Releases Wiki Activity

Fix AssertionError in hasReplicaWithOngoingRepair when parallel_repair_count > 1 

patch by Patrick Lee; reviewed by Paulo Motta, Jaydeepkumar Chovatia for CASSANDRA-21426

Co-authored-by: Paulo Motta <paulo@apache.org>
This commit is contained in:
Patrick Lee 2026-06-03 14:50:06 -05:00 committed by Stefan Miklosovic
parent 4b821d1bc7
commit db631675df
No known key found for this signature in database
GPG Key ID: 32F35CB2F546D93E
3 changed files with 56 additions and 1 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

1
CHANGES.txt
View File

@ -1,4 +1,5 @@
5.0.9
* Fix AssertionError in hasReplicaWithOngoingRepair when parallel_repair_count > 1 (CASSANDRA-21426)
* putShortVolatile is not volatile in InMemoryTrie (CASSANDRA-21353)
* Fix RequestFailureReason serializer and nits in a few others (CASSANDRA-21437)
* Remove golang dependency in gen-doc and replace with python implementation (CASSANDRA-21432)

2
src/java/org/apache/cassandra/repair/autorepair/AutoRepairUtils.java
View File

@ -735,7 +735,7 @@ public class AutoRepairUtils
for (Map.Entry<AbstractReplicationStrategy, List<String>> entry : replicationStrategies.entrySet())
{
AbstractReplicationStrategy replicationStrategy = entry.getKey();
EndpointsByRange endpointsByRange = replicationStrategy.getRangeAddresses(StorageService.instance.getTokenMetadata());
EndpointsByRange endpointsByRange = replicationStrategy.getRangeAddresses(StorageService.instance.getTokenMetadata().cachedOnlyTokenMap());
// get ranges of the eligible address for the given replication strategy.
RangesAtEndpoint rangesAtEndpoint = StorageService.instance.getReplicas(replicationStrategy, eligibleBroadcastAddress);

54
test/unit/org/apache/cassandra/repair/autorepair/AutoRepairUtilsTest.java
View File

@ -19,6 +19,7 @@
package org.apache.cassandra.repair.autorepair;
import java.net.UnknownHostException;
import java.util.ArrayList;
import java.util.List;
import java.util.Set;
import java.util.TreeSet;
@ -44,8 +45,11 @@ import org.apache.cassandra.cql3.UntypedResultSet;
import org.apache.cassandra.db.Keyspace;
import org.apache.cassandra.db.marshal.UTF8Type;
import org.apache.cassandra.db.marshal.UUIDType;
import org.apache.cassandra.dht.Murmur3Partitioner;
import org.apache.cassandra.locator.IEndpointSnitch;
import org.apache.cassandra.locator.InetAddressAndPort;
import org.apache.cassandra.locator.TokenMetadata;
import org.apache.cassandra.service.StorageService;
import org.apache.cassandra.repair.autorepair.AutoRepairConfig.RepairType;
import org.apache.cassandra.repair.autorepair.AutoRepairUtils.AutoRepairHistory;
import org.apache.cassandra.repair.autorepair.AutoRepairUtils.CurrentRepairStatus;
@ -709,4 +713,54 @@ public class AutoRepairUtilsTest extends CQLTester
boolean result = AutoRepairUtils.hasNodesBelowMinimumVersion();
assertTrue(result);
}
/**
* Reproduces CASSANDRA-21426: when parallel_repair_count > 1 and another node has an ongoing repair,
* hasReplicaWithOngoingRepair calls getRangeAddresses with the live TokenMetadata singleton, which
* triggers the assertion in getTopology() that prevents calling it on the live instance.
*/
@Test
public void testGetMostEligibleHostToRepairWithOngoingParallelRepair() throws UnknownHostException
{
InetAddressAndPort otherEndpoint = InetAddressAndPort.getByName("127.0.0.2");
UUID otherHostId = UUID.randomUUID();
TokenMetadata tokenMetadata = StorageService.instance.getTokenMetadata();
try
{
// Register a second endpoint in TokenMetadata so it has tokens and can be resolved
tokenMetadata.updateNormalToken(Murmur3Partitioner.instance.getRandomToken(), otherEndpoint);
tokenMetadata.updateHostId(otherHostId, otherEndpoint);
long currentMillis = System.currentTimeMillis();
// Build repair histories: other node has ongoing repair, local node is finished
AutoRepairHistory otherHistory = new AutoRepairHistory(otherHostId, null, currentMillis, currentMillis - 100,
null, 0, false);
AutoRepairHistory myHistory = new AutoRepairHistory(hostId, null, currentMillis - 200, currentMillis - 100,
null, 0, false);
// Construct CurrentRepairStatus directly to avoid CQL reads that fail with the extra endpoint
List<AutoRepairHistory> allHistories = new ArrayList<>();
allHistories.add(otherHistory);
allHistories.add(myHistory);
CurrentRepairStatus currentRepairStatus = new CurrentRepairStatus(allHistories, null, hostId);
// Verify preconditions: otherHostId is repairing, local node is not
assertTrue(currentRepairStatus.hostIdsWithOnGoingRepair.contains(otherHostId));
assertFalse(currentRepairStatus.hostIdsWithOnGoingRepair.contains(hostId));
// This exercises the code path:
// getMostEligibleHostToRepair -> hasReplicaWithOngoingRepair
// -> getRangeAddresses(getTokenMetadata()) -> calculateNaturalReplicas -> getTopology()
// Without the fix, getTopology() asserts because it's called on the live TokenMetadata singleton.
AutoRepairHistory result = AutoRepairUtils.getMostEligibleHostToRepair(repairType, currentRepairStatus, hostId);
assertNotNull(result);
assertEquals(hostId, result.hostId);
}
finally
{
tokenMetadata.removeEndpoint(otherEndpoint);
}
}
}