Jun Rao created KAFKA-21000:
-------------------------------
Summary: Potential data corruption on clean broker restart
Key: KAFKA-21000
URL: https://issues.apache.org/jira/browse/KAFKA-21000
Project: Kafka
Issue Type: Improvement
Components: core
Reporter: Jun Rao
A similar issue to https://issues.apache.org/jira/browse/KAFKA-20979 could be
exposed in the .log file when preallocation is used.
When the active segment is closed, FileRecords.close() calls the following.
{code:java}
flush();
trim();{code}
When preallocation is used, trim() resizes the file length to what's actually
being used, but without flushing the length metadata. The broker cleanly shuts
down and crashes before the metadata is flushed. When the broker is restarted,
the on-disk file length still has the preallocation length but won't go through
recovery (since the broker was shutdown cleanly). When the broker tries to
calculate the next offset, it will hit the uninitialized bytes and cause an
exception.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
