[
https://issues.apache.org/jira/browse/CASSANDRA-14892?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16702917#comment-16702917
]
Branimir Lambov commented on CASSANDRA-14892:
---------------------------------------------
As requested by reviewer off-line, added extra tests for writing and reading
various input and compressed buffer sizes, as well as a unit test for
{{ByteBufferUtil.writeZeroes}}.
> Possible corruption in compressed files with uncompressed chunks
> ----------------------------------------------------------------
>
> Key: CASSANDRA-14892
> URL: https://issues.apache.org/jira/browse/CASSANDRA-14892
> Project: Cassandra
> Issue Type: Bug
> Components: Local Write-Read Paths
> Reporter: Branimir Lambov
> Assignee: Branimir Lambov
> Priority: Major
>
> When deciding to switch to writing a chunk uncompressed in a compressed file
> (see CASSANDRA-10520) and that chunk is smaller than the full chunk size
> (only currently happens in the last chunk of the file), it may very rarely
> happen that the chunk is:
> - bigger than the compression limit when compressed
> - smaller than the compression limit when left uncompressed
> If this happens the writer will write it uncompressed, but the reader will
> treat it as compressed and fail when attempting to read it.
> Such chunks should be padded with 0s to the minimum uncompressed size.
--
This message was sent by Atlassian JIRA
(v7.6.3#76005)
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
