[KAFKA-13831] Kafka retention can use old value of retention.ms

• Type: Bug
• Status: Open
• Priority: Major
• Resolution: Unresolved
• Affects Version/s: 2.8.0
• Component/s: core
• Assignee: Unassigned
• Reporter: Maciej Bryński
• Created: 15/Apr/22 09:08
• Updated: 15/Apr/22 11:58

Description

Hi,
I think I have found a bug in Kafka retention.
I'm using Confluent Platform 6.2.2 (Kafka 2.8.0).
I changed retention.ms for topic twice:
1. From 432000000ms to 180000ms (to clean the topic)
2. Back to 432000000ms.

After second change retention thread is still using 180000ms value.
Only broker restart fixes this issue.

Logs:

server.log.2022-04-15-03:[2022-04-15 03:29:08,445] INFO [Log partition=pm.hwe.lte.lcell.inc.intrarat.ho.x2.raw-0, dir=/data/kafka] Deleting segment LogSegment(baseOffset=1029819055, size=22996644, lastModifiedTime=1650007299179, largestRecordTimestamp=Some(1650007299178)) due to retention time 180000ms breach based on the largest record timestamp in the segment (kafka.log.Log)

Topic description:

kafka-topics --bootstrap-server localhost:9092 --describe --topic pm.hwe.lte.lcell.inc.intrarat.ho.x2.raw
Topic: pm.hwe.lte.lcell.inc.intrarat.ho.x2.raw TopicId: svLdGbOaRXmdkHGsdlaPUQ PartitionCount: 1 ReplicationFactor: 3 Configs: min.insync.replicas=2,segment.bytes=1073741824,retention.ms=432000000,segment.ms=86400000

EDIT: the scale of this problem is around 0,1% of topics where I did a change (a few topics from a few thousands)