samueldlightfoot commented on code in PR #4606:
URL: https://github.com/apache/cassandra/pull/4606#discussion_r2917778986
##########
src/java/org/apache/cassandra/io/sstable/SSTableCursorReader.java:
##########
@@ -221,7 +227,7 @@ private SSTableCursorReader(SSTableReader reader,
TableMetadata metadata, Ref<SS
deserializationHelper = new DeserializationHelper(metadata,
version.correspondingMessagingVersion(), DeserializationHelper.Flag.LOCAL,
null);
serializationHeader = reader.header;
- dataReader = reader.openDataReader();
+ dataReader = reader.openDataReaderForScan(diskAccessMode);
Review Comment:
For reference, prior to the SSTableCursorReader changes on this PR, which
ensure the passed in compaction Scanners are closed before opening new cursor
readers, _openDataReader()_ was unintentionally (and fortunately) using scan
reads due to the already opened scanners.
Given we now close the old unused scanners before opening our cursor ones,
we are required to adopt openDataReaderForScan() to ensure behavioural equality.
For completeness, I also ran unthrottled compaction performance tests with
scan and non-scan:
| Config | Compaction Throughput | CQL p99 (ms) | CQL Mean (ms) |
|---|---|---|---|
| Scan + DIO | ~217 MiB/s | 33.03 | 2.44 |
| Scan + buffered | ~201 MiB/s | 35.91 | 2.15 |
| Non-scan + buffered | ~71 MiB/s | 4.75 | 0.68 |
| Non-scan + DIO | ~28 MiB/s | 1.47 | 0.48 |
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
To unsubscribe, e-mail: [email protected]
For queries about this service, please contact Infrastructure at:
[email protected]
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
