Skip to content

KAFKA-19163: Avoid deleting groups with pending transactional offsets #7

New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

Sign up for GitHub

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Open
wants to merge 2 commits into
base: trunk
Choose a base branch
from
Open

KAFKA-19163: Avoid deleting groups with pending transactional offsets #7

Show file tree
Hide file tree
Changes from all commits
Commits
Show all changes
2 commits
Select commit Hold shift + click to select a range
d670f68
KAFKA-19164: Keep track of groups when deleting transactional offsets
squah-confluent Apr 16, 2025
3c0cbb0
KAFKA-19163: Avoid deleting groups with pending transactional offsets
squah-confluent Apr 16, 2025
File filter

Filter by extension

Filter by extension
Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
17 changes: 12 additions & 5 deletions ...p-coordinator/src/main/java/org/apache/kafka/coordinator/group/OffsetMetadataManager.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -198,12 +198,19 @@ public OffsetMetadataManager build() {
private final TimelineHashMap<String, TimelineHashSet<Long>> openTransactionsByGroup;

private class Offsets {
/**
* Whether to preserve empty entries for groups when removing offsets.
* We use this to keep track of the groups associated with pending transactions.
*/
private final boolean preserveGroups;

/**
* The offsets keyed by group id, topic name and partition id.
*/
private final TimelineHashMap<String, TimelineHashMap<String, TimelineHashMap<Integer, OffsetAndMetadata>>> offsetsByGroup;

private Offsets() {
private Offsets(boolean preserveGroups) {
this.preserveGroups = preserveGroups;
this.offsetsByGroup = new TimelineHashMap<>(snapshotRegistry, 0);
}

Expand Down Expand Up @@ -256,7 +263,7 @@ private OffsetAndMetadata remove(
if (partitionOffsets.isEmpty())
topicOffsets.remove(topic);

if (topicOffsets.isEmpty())
if (!preserveGroups && topicOffsets.isEmpty())
offsetsByGroup.remove(groupId);

return removedValue;
Expand All @@ -278,7 +285,7 @@ private OffsetAndMetadata remove(
this.groupMetadataManager = groupMetadataManager;
this.config = config;
this.metrics = metrics;
this.offsets = new Offsets();
this.offsets = new Offsets(false);
this.pendingTransactionalOffsets = new TimelineHashMap<>(snapshotRegistry, 0);
this.openTransactionsByGroup = new TimelineHashMap<>(snapshotRegistry, 0);
}
Expand Down Expand Up @@ -851,7 +858,7 @@ public boolean cleanupExpiredOffsets(String groupId, List<CoordinatorRecord> rec
TimelineHashMap<String, TimelineHashMap<Integer, OffsetAndMetadata>> offsetsByTopic =
offsets.offsetsByGroup.get(groupId);
if (offsetsByTopic == null) {
return true;
return !openTransactionsByGroup.containsKey(groupId);
}

// We expect the group to exist.
Expand Down Expand Up @@ -995,7 +1002,7 @@ public void replay(
// offsets store. Pending offsets there are moved to the main store when
// the transaction is committed; or removed when the transaction is aborted.
pendingTransactionalOffsets
.computeIfAbsent(producerId, __ -> new Offsets())
.computeIfAbsent(producerId, __ -> new Offsets(true))
.put(
groupId,
topic,
Expand Down
97 changes: 97 additions & 0 deletions ...ordinator/src/test/java/org/apache/kafka/coordinator/group/OffsetMetadataManagerTest.java
View file
Open in desktop
Original file line number Diff line number Diff line change
Expand Up @@ -2593,6 +2593,103 @@ public void testCleanupExpiredOffsetsWithPendingTransactionalOffsets() {
assertEquals(List.of(), records);
}

@Test
public void testCleanupExpiredOffsetsWithDeletedPendingTransactionalOffsets() {
GroupMetadataManager groupMetadataManager = mock(GroupMetadataManager.class);
Group group = mock(Group.class);

OffsetMetadataManagerTestContext context = new OffsetMetadataManagerTestContext.Builder()
.withGroupMetadataManager(groupMetadataManager)
.withOffsetsRetentionMinutes(1)
.build();

long commitTimestamp = context.time.milliseconds();

context.commitOffset("group-id", "foo", 0, 100L, 0, commitTimestamp);
context.commitOffset(10L, "group-id", "foo", 1, 101L, 0, commitTimestamp + 500);

when(groupMetadataManager.group("group-id")).thenReturn(group);
when(group.offsetExpirationCondition()).thenReturn(Optional.of(
new OffsetExpirationConditionImpl(offsetAndMetadata -> offsetAndMetadata.commitTimestampMs)));
when(group.isSubscribedToTopic("foo")).thenReturn(false);

// Delete the pending transactional offset.
OffsetDeleteRequestData.OffsetDeleteRequestTopicCollection requestTopicCollection =
new OffsetDeleteRequestData.OffsetDeleteRequestTopicCollection(List.of(
new OffsetDeleteRequestData.OffsetDeleteRequestTopic()
.setName("foo")
.setPartitions(List.of(
new OffsetDeleteRequestData.OffsetDeleteRequestPartition().setPartitionIndex(1)
))
).iterator());
CoordinatorResult<OffsetDeleteResponseData, CoordinatorRecord> result = context.deleteOffsets(
new OffsetDeleteRequestData()
.setGroupId("group-id")
.setTopics(requestTopicCollection)
);
List<CoordinatorRecord> expectedRecords = List.of(
GroupCoordinatorRecordHelpers.newOffsetCommitTombstoneRecord("group-id", "foo", 1)
);
assertEquals(expectedRecords, result.records());

context.time.sleep(Duration.ofMinutes(1).toMillis());

// The group should not be deleted because it has a pending transaction.
expectedRecords = List.of(
GroupCoordinatorRecordHelpers.newOffsetCommitTombstoneRecord("group-id", "foo", 0)
);
List<CoordinatorRecord> records = new ArrayList<>();
assertFalse(context.cleanupExpiredOffsets("group-id", records));
assertEquals(expectedRecords, records);

// Commit the ongoing transaction.
context.replayEndTransactionMarker(10L, TransactionResult.COMMIT);

// The group should be deletable now.
context.commitOffset("group-id", "foo", 0, 100L, 0, commitTimestamp);
context.time.sleep(Duration.ofMinutes(1).toMillis());

records = new ArrayList<>();
assertTrue(context.cleanupExpiredOffsets("group-id", records));
assertEquals(expectedRecords, records);
}

@Test
public void testCleanupExpiredOffsetsWithPendingTransactionalOffsetsOnly() {
GroupMetadataManager groupMetadataManager = mock(GroupMetadataManager.class);
Group group = mock(Group.class);

OffsetMetadataManagerTestContext context = new OffsetMetadataManagerTestContext.Builder()
.withGroupMetadataManager(groupMetadataManager)
.withOffsetsRetentionMinutes(1)
.build();

long commitTimestamp = context.time.milliseconds();

context.commitOffset("group-id", "foo", 0, 100L, 0, commitTimestamp);
context.commitOffset(10L, "group-id", "foo", 1, 101L, 0, commitTimestamp + 500);

context.time.sleep(Duration.ofMinutes(1).toMillis());

when(groupMetadataManager.group("group-id")).thenReturn(group);
when(group.offsetExpirationCondition()).thenReturn(Optional.of(
new OffsetExpirationConditionImpl(offsetAndMetadata -> offsetAndMetadata.commitTimestampMs)));
when(group.isSubscribedToTopic("foo")).thenReturn(false);

// foo-0 is expired, but the group is not deleted beacuse it has pending transactional offset commits.
List<CoordinatorRecord> expectedRecords = List.of(
GroupCoordinatorRecordHelpers.newOffsetCommitTombstoneRecord("group-id", "foo", 0)
);
List<CoordinatorRecord> records = new ArrayList<>();
assertFalse(context.cleanupExpiredOffsets("group-id", records));
assertEquals(expectedRecords, records);

// No offsets are expired, and the group is still not deleted because it has pending transactional offset commits.
records = new ArrayList<>();
assertFalse(context.cleanupExpiredOffsets("group-id", records));
assertEquals(List.of(), records);
}

private static OffsetFetchResponseData.OffsetFetchResponsePartitions mkOffsetPartitionResponse(
int partition,
long offset,
Expand Down