[KAFKA-5630] Consumer poll loop over the same record after a CorruptRecordException

• Type: Bug
• Status: Resolved
• Priority: Critical
• Resolution: Fixed
• Affects Version/s: 0.11.0.0
• Fix Version/s: 0.11.0.1, 1.0.0
• Component/s: consumer
• Labels: regression reliability
• Assignee: Jiangjie Qin
• Reporter: Vincent Maurin
• Created: 24/Jul/17 09:09
• Updated: 12/Jun/18 19:10
• Resolved: 01/Aug/17 16:04

Description

Hello

While consuming a topic with log compaction enabled, I am getting an infinite consumption loop of the same record, i.e, each call to poll is returning to me 500 times one record (500 is my max.poll.records). I am using the java client 0.11.0.0.

Running the code with the debugger, the initial problem come from `Fetcher.PartitionRecords,fetchRecords()`.
Here I get a `org.apache.kafka.common.errors.CorruptRecordException: Record size is less than the minimum record overhead (14)`
Then the boolean `hasExceptionInLastFetch` is set to true, resulting the test block in `Fetcher.PartitionRecords.nextFetchedRecord()` to always return the last record.

I guess the corruption problem is similar too https://issues.apache.org/jira/browse/KAFKA-5582 but this behavior of the client is probably not the expected one

Issue Links

• links to: GitHub Pull Request #3573