mirror of https://github.com/apache/kafka.git
ff5025a21c
* The `ShareCoordinatorShard` maintains the the record offset information for `SharePartitionKey`s in the `ShareCoordinatorOffsetsManager` class. * Replay of `ShareSnapshot`s in the shards are reflected in the offsets manager including records created due to delete state. * However, if the share partition delete is due to topic delete, no record will ever be written for the same `SharePartitionKey` post the delete tombstone (as topic id will not repeat). As a result the offset manager will always consider the deleted share partition's offset as the last redundant one. * The fix is to make the offset manager aware of the tombstone records and remove them from the redundant offset calculation. * Unit tests have been updated for the same. Reviewers: Andrew Schofield <aschofield@confluent.io>, Apoorv Mittal <apoorvmittal10@gmail.com> |
||
|---|---|---|
| .. | ||
| main | ||
| test/java/org/apache/kafka/coordinator/share | ||