Arjun Ashok created CASSANDRA-21358:
---------------------------------------
Summary: CursorBasedCompaction: Final index block width off by one
byte in SSTableCursorWriter#appendBIGIndex()
Key: CASSANDRA-21358
URL: https://issues.apache.org/jira/browse/CASSANDRA-21358
Project: Apache Cassandra
Issue Type: Bug
Components: Local/SSTable
Reporter: Arjun Ashok
The cursor compaction path computes the width of the final index block as
(partitionEnd - partitionStart - 1) - indexBlockStartOffset, which excludes the
end-of-partition marker byte.
The iterator path (BigFormatPartitionWriter#finish()) calls addIndexBlock()
after writing the end marker, so its width includes that byte.
The resulting Index.db entry has a width 1 byte smaller than the iterator
produces. When seeking into a large partition near the end using the index, the
reader's internal byte accounting is off by 1, causing misaligned reads.
Affects any partition large enough to span multiple index blocks (default
threshold: 64 KiB).
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
