Copilot commented on code in PR #24235:
URL: https://github.com/apache/kafka/pull/24235#discussion_r4641740643
##########
share-coordinator/src/test/java/org/apache/kafka/coordinator/share/ShareCoordinatorShardTest.java:
##########
@@ -1806,28 +1809,131 @@ public void testInitializeStateIdempotent() {
}
@Test
- public void testInitializeStateNotIdempotentWhenStartOffsetChanges() {
- InitializeShareGroupStateRequestData.PartitionData partitionData = new
InitializeShareGroupStateRequestData.PartitionData()
+ public void testInitializeStateRetryPreservesConsumptionProgress() {
+ // New partitions of an already-known topic are initialized at offset
zero.
+ InitializeShareGroupStateRequestData.PartitionData initializeRequest =
new InitializeShareGroupStateRequestData.PartitionData()
.setPartition(PARTITION)
- .setStartOffset(10)
+ .setStartOffset(0)
.setStateEpoch(5);
- CoordinatorResult<InitializeShareGroupStateResponseData,
CoordinatorRecord> result = shard.initializeState(GROUP_ID, TOPIC_ID,
partitionData);
- result.records().forEach(record -> shard.replay(0L, 0L, (short) 0,
record));
- assertEquals(1, result.records().size());
+ CoordinatorResult<InitializeShareGroupStateResponseData,
CoordinatorRecord> initializeResult =
+ shard.initializeState(GROUP_ID, TOPIC_ID, initializeRequest);
+
assertEquals(InitializeShareGroupStateResponse.toResponseData(TOPIC_ID,
PARTITION), initializeResult.response());
+ initializeResult.records().forEach(record -> shard.replay(0L, 0L,
(short) 0, record));
+
+ ReadShareGroupStateRequestData readRequest = new
ReadShareGroupStateRequestData()
+ .setGroupId(GROUP_ID)
+ .setTopics(List.of(new
ReadShareGroupStateRequestData.ReadStateData()
+ .setTopicId(TOPIC_ID)
+ .setPartitions(List.of(new
ReadShareGroupStateRequestData.PartitionData()
+ .setPartition(PARTITION)
+ .setLeaderEpoch(1)))));
+ CoordinatorResult<ReadShareGroupStateResponseData, CoordinatorRecord>
readResult =
+ shard.readStateAndMaybeUpdateLeaderEpoch(readRequest);
+ assertEquals(Errors.NONE.code(),
readResult.response().results().get(0).partitions().get(0).errorCode());
+ readResult.records().forEach(record -> shard.replay(1L, 0L, (short) 0,
record));
+
+ // Consumption advances the start offset and persists acknowledgements
without changing the state epoch.
+ WriteShareGroupStateRequestData writeRequest = new
WriteShareGroupStateRequestData()
+ .setGroupId(GROUP_ID)
+ .setTopics(List.of(new
WriteShareGroupStateRequestData.WriteStateData()
+ .setTopicId(TOPIC_ID)
+ .setPartitions(List.of(new
WriteShareGroupStateRequestData.PartitionData()
+ .setPartition(PARTITION)
+ .setStateEpoch(5)
+ .setLeaderEpoch(1)
+ .setStartOffset(20)
+ .setDeliveryCompleteCount(5)
+ .setStateBatches(List.of(new
WriteShareGroupStateRequestData.StateBatch()
+ .setFirstOffset(25)
+ .setLastOffset(29)
+ .setDeliveryState((byte) 2)
+ .setDeliveryCount((short) 1)))))));
+ CoordinatorResult<WriteShareGroupStateResponseData, CoordinatorRecord>
writeResult = shard.writeState(writeRequest);
+ assertEquals(WriteShareGroupStateResponse.toResponseData(TOPIC_ID,
PARTITION), writeResult.response());
+ writeResult.records().forEach(record -> shard.replay(2L, 0L, (short)
0, record));
+
+ ReadShareGroupStateResponseData.PartitionResult stateBeforeRetry =
+
shard.readStateAndMaybeUpdateLeaderEpoch(readRequest).response().results().get(0).partitions().get(0);
+ assertEquals(20, stateBeforeRetry.startOffset());
Review Comment:
`readStateAndMaybeUpdateLeaderEpoch` can emit records (as evidenced earlier
in this test where you replay `readResult.records()`). Here the result records
are ignored, which can make the test brittle if future changes cause this read
to produce records. Consider capturing the `CoordinatorResult` and replaying
its records (or using a read API that does not generate records) before
asserting on the returned state.
##########
share-coordinator/src/main/java/org/apache/kafka/coordinator/share/ShareCoordinatorShard.java:
##########
@@ -540,7 +540,9 @@ public CoordinatorResult<DeleteShareGroupStateResponseData,
CoordinatorRecord> d
}
/**
- * This method writes a share snapshot records corresponding to the
requested share partition.
+ * Initializes or resets the state of the requested share partition.
+ * If state already exists at the requested state epoch, it is preserved
regardless of the
+ * requested offset and no record is written. Resetting existing state
requires a higher state epoch.
* <p>
* This method as called by the ShareCoordinatorService will be provided
with
* the request data for a single share partition (group, topicId and
partition data).
Review Comment:
Correct grammar in Javadoc: change “This method as called” to “This method
is called”.
##########
share-coordinator/src/test/java/org/apache/kafka/coordinator/share/ShareCoordinatorShardTest.java:
##########
@@ -1806,28 +1809,131 @@ public void testInitializeStateIdempotent() {
}
@Test
- public void testInitializeStateNotIdempotentWhenStartOffsetChanges() {
- InitializeShareGroupStateRequestData.PartitionData partitionData = new
InitializeShareGroupStateRequestData.PartitionData()
+ public void testInitializeStateRetryPreservesConsumptionProgress() {
+ // New partitions of an already-known topic are initialized at offset
zero.
+ InitializeShareGroupStateRequestData.PartitionData initializeRequest =
new InitializeShareGroupStateRequestData.PartitionData()
.setPartition(PARTITION)
- .setStartOffset(10)
+ .setStartOffset(0)
.setStateEpoch(5);
- CoordinatorResult<InitializeShareGroupStateResponseData,
CoordinatorRecord> result = shard.initializeState(GROUP_ID, TOPIC_ID,
partitionData);
- result.records().forEach(record -> shard.replay(0L, 0L, (short) 0,
record));
- assertEquals(1, result.records().size());
+ CoordinatorResult<InitializeShareGroupStateResponseData,
CoordinatorRecord> initializeResult =
+ shard.initializeState(GROUP_ID, TOPIC_ID, initializeRequest);
+
assertEquals(InitializeShareGroupStateResponse.toResponseData(TOPIC_ID,
PARTITION), initializeResult.response());
+ initializeResult.records().forEach(record -> shard.replay(0L, 0L,
(short) 0, record));
+
+ ReadShareGroupStateRequestData readRequest = new
ReadShareGroupStateRequestData()
+ .setGroupId(GROUP_ID)
+ .setTopics(List.of(new
ReadShareGroupStateRequestData.ReadStateData()
+ .setTopicId(TOPIC_ID)
+ .setPartitions(List.of(new
ReadShareGroupStateRequestData.PartitionData()
+ .setPartition(PARTITION)
+ .setLeaderEpoch(1)))));
+ CoordinatorResult<ReadShareGroupStateResponseData, CoordinatorRecord>
readResult =
+ shard.readStateAndMaybeUpdateLeaderEpoch(readRequest);
+ assertEquals(Errors.NONE.code(),
readResult.response().results().get(0).partitions().get(0).errorCode());
+ readResult.records().forEach(record -> shard.replay(1L, 0L, (short) 0,
record));
+
+ // Consumption advances the start offset and persists acknowledgements
without changing the state epoch.
+ WriteShareGroupStateRequestData writeRequest = new
WriteShareGroupStateRequestData()
+ .setGroupId(GROUP_ID)
+ .setTopics(List.of(new
WriteShareGroupStateRequestData.WriteStateData()
+ .setTopicId(TOPIC_ID)
+ .setPartitions(List.of(new
WriteShareGroupStateRequestData.PartitionData()
+ .setPartition(PARTITION)
+ .setStateEpoch(5)
+ .setLeaderEpoch(1)
+ .setStartOffset(20)
+ .setDeliveryCompleteCount(5)
+ .setStateBatches(List.of(new
WriteShareGroupStateRequestData.StateBatch()
+ .setFirstOffset(25)
+ .setLastOffset(29)
+ .setDeliveryState((byte) 2)
+ .setDeliveryCount((short) 1)))))));
+ CoordinatorResult<WriteShareGroupStateResponseData, CoordinatorRecord>
writeResult = shard.writeState(writeRequest);
+ assertEquals(WriteShareGroupStateResponse.toResponseData(TOPIC_ID,
PARTITION), writeResult.response());
+ writeResult.records().forEach(record -> shard.replay(2L, 0L, (short)
0, record));
+
+ ReadShareGroupStateResponseData.PartitionResult stateBeforeRetry =
+
shard.readStateAndMaybeUpdateLeaderEpoch(readRequest).response().results().get(0).partitions().get(0);
+ assertEquals(20, stateBeforeRetry.startOffset());
+ assertEquals(1, stateBeforeRetry.stateBatches().size());
+
+ // A delayed retry from an overlapping initialization must preserve
the state already consumed.
+ CoordinatorResult<InitializeShareGroupStateResponseData,
CoordinatorRecord> retryResult =
+ shard.initializeState(GROUP_ID, TOPIC_ID, initializeRequest);
+
assertEquals(InitializeShareGroupStateResponse.toResponseData(TOPIC_ID,
PARTITION), retryResult.response());
+ retryResult.records().forEach(record -> shard.replay(3L, 0L, (short)
0, record));
+
+ ReadShareGroupStateResponseData.PartitionResult stateAfterRetry =
+
shard.readStateAndMaybeUpdateLeaderEpoch(readRequest).response().results().get(0).partitions().get(0);
+ assertAll(
+ () -> assertEquals(stateBeforeRetry.startOffset(),
stateAfterRetry.startOffset(), "Start offset must not regress"),
+ () -> assertEquals(stateBeforeRetry.stateBatches(),
stateAfterRetry.stateBatches(), "Persisted acknowledgements must be preserved")
Review Comment:
The regression description mentions that retries can also clear
`deliveryCompleteCount`, but the test only asserts `startOffset` and
`stateBatches`. Add an assertion that `deliveryCompleteCount` (and any other
relevant persisted fields you expect to be preserved in this scenario) is
unchanged after the retry to fully cover the reported failure mode.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]