Michael Semb Wever created CASSANDRA-21613:
----------------------------------------------
Summary: Report both checksums when a compressed chunk fails its
checksum
Key: CASSANDRA-21613
URL: https://issues.apache.org/jira/browse/CASSANDRA-21613
Project: Apache Cassandra
Issue Type: Improvement
Components: Feature/Compression
Reporter: Michael Semb Wever
Assignee: Michael Semb Wever
{{CorruptBlockException}} names the file, the chunk offset and the chunk length:
{code}
(/path/to/sstable-Data.db): corruption detected, chunk at 65536 of length 4096.
{code}
The message does not name the checksums, so an operator cannot tell a stored
checksum that was never written, and so reads as zero, from a genuine mismatch.
The two need different responses.
The patch adds the stored and the calculated checksum to the message, at the
five sites that compare them: {{CompressedSequentialWriter.resetAndTruncate}}
and the four chunk readers.
One of those sites also folded a short read of the checksum into the comparison:
{code:java}
if (channel.read(scratch, chunk.offset + chunk.length) != Integer.BYTES
|| scratch.getInt(0) != checksum)
throw new CorruptBlockException(channel.filePath(), chunk);
{code}
A truncated file and a mismatched checksum reported as one fault. The patch
separates them.
Patch:
[mck/upstream/corrupt-block-checksums/trunk|https://github.com/thelastpickle/cassandra/tree/mck/upstream/corrupt-block-checksums/trunk]
Provenance:
[1cce224756|https://github.com/datastax/cassandra/commit/1cce224756d11b9520e1a091bc9af1ae0fe50c9f]
by Michael Marshall
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
