This is an automated email from the ASF dual-hosted git repository.
brandonwilliams pushed a commit to branch cassandra-5.0
in repository https://gitbox.apache.org/repos/asf/cassandra.git
The following commit(s) were added to refs/heads/cassandra-5.0 by this push:
new 049160e70a Avoid reading of the same IndexInfo from disk many times
for a large partition when ShallowIndexedEntry is used
049160e70a is described below
commit 049160e70a22bfcc9c751a1fe3afec1e5c329f4c
Author: Brandon Williams <brandonwilli...@apache.org>
AuthorDate: Wed Apr 24 16:11:50 2024 -0700
Avoid reading of the same IndexInfo from disk many times for a large
partition when ShallowIndexedEntry is used
Patch by Dmitry Konstantinov; reviewed by bereng and brandonwilliams for
CASSANDRA-19557
---
CHANGES.txt | 1 +
.../apache/cassandra/io/sstable/format/big/IndexState.java | 13 ++++++++++++-
2 files changed, 13 insertions(+), 1 deletion(-)
diff --git a/CHANGES.txt b/CHANGES.txt
index 45500b642e..8af3d01ac9 100644
--- a/CHANGES.txt
+++ b/CHANGES.txt
@@ -1,4 +1,5 @@
5.0-beta2
+ * Avoid reading of the same IndexInfo from disk many times for a large
partition (CASSANDRA-19557)
* Resolve the oldest hints just from descriptors and current writer if
available (CASSANDRA-19600)
* Optionally fail writes when SAI refuses to index a term value exceeding
configured term max size (CASSANDRA-19493)
* Vector search can restrict on clustering keys when filtering isn't required
(CASSANDRA-19544)
diff --git
a/src/java/org/apache/cassandra/io/sstable/format/big/IndexState.java
b/src/java/org/apache/cassandra/io/sstable/format/big/IndexState.java
index f738f8ab7c..754b34b065 100644
--- a/src/java/org/apache/cassandra/io/sstable/format/big/IndexState.java
+++ b/src/java/org/apache/cassandra/io/sstable/format/big/IndexState.java
@@ -41,6 +41,9 @@ public class IndexState implements AutoCloseable
private int currentIndexIdx;
+ private int cachedIndexIdx = Integer.MIN_VALUE;
+ private IndexInfo cachedIndexInfo;
+
// Marks the beginning of the block corresponding to currentIndexIdx.
private DataPosition mark;
@@ -139,7 +142,15 @@ public class IndexState implements AutoCloseable
public IndexInfo index(int i) throws IOException
{
- return indexInfoRetriever.columnsIndex(i);
+ // during an iteration we retrieve the same IndexInfo many times
sequentially, for each row
+ // caching of the last retreived IndexInfo can save a lot of IO in
case of ShallowIndexedEntry
+ if (i == cachedIndexIdx)
+ {
+ return cachedIndexInfo;
+ }
+ cachedIndexInfo = indexInfoRetriever.columnsIndex(i);
+ cachedIndexIdx = i;
+ return cachedIndexInfo;
}
// Finds the index of the first block containing the provided bound,
starting at the provided index.
---------------------------------------------------------------------
To unsubscribe, e-mail: commits-unsubscr...@cassandra.apache.org
For additional commands, e-mail: commits-h...@cassandra.apache.org
