KIP-534 — Reorganize checkpoint system in log cleaner to per partition
Accepted Kafka 3.1 Broker
Fixes log compaction to ensure that tombstone records and transaction abort markers are retained for at least `delete.retention.ms` milliseconds from the point the compacting segment was written, not just from the segment roll time. Under the prior implementation, tombstones could be physically removed before `delete.retention.ms` had elapsed from the last possible read of the preceding non-tombstone record, violating the semantic guarantee that consumers reading within the retention window will observe deletions.
Details
| Author | Richard Yu |
| Status | Accepted |
| Kafka Version | 3.1 |
| JIRA | KAFKA-8522 |
| Wiki | View on Apache Wiki |
| Created | 2019-09-02 |
| Last Modified | 2019-10-18 |
Explore how this KIP affects the Kafka protocol in the Protocol Explorer, or see the full KIP database.