maedhroz commented on code in PR #4353:
URL: https://github.com/apache/cassandra/pull/4353#discussion_r2678357959
##########
src/java/org/apache/cassandra/index/sai/plan/StorageAttachedIndexSearcher.java:
##########
@@ -468,79 +501,321 @@ private UnfilteredRowIterator
queryStorageAndFilter(List<PrimaryKey> keys)
}
}
- private UnfilteredRowIterator filterPartition(List<PrimaryKey> keys,
UnfilteredRowIterator partition, FilterTree tree)
+ @Override
+ public TableMetadata metadata()
{
- Row staticRow = partition.staticRow();
- DecoratedKey partitionKey = partition.partitionKey();
- List<Unfiltered> matches = new ArrayList<>();
- boolean hasMatch = false;
- Set<PrimaryKey> keysToShadow = topK ? new HashSet<>(keys) :
Collections.emptySet();
+ return queryController.metadata();
+ }
- while (partition.hasNext())
- {
- Unfiltered unfiltered = partition.next();
+ @Override
+ public void close()
+ {
+ FileUtils.closeQuietly(resultKeyIterator);
+ if (tableQueryMetrics != null)
tableQueryMetrics.record(queryContext);
+ }
+ }
- if (unfiltered.isRow())
- {
- queryContext.rowsFiltered++;
+ private static UnfilteredRowIterator filterPartition(UnfilteredRowIterator
partition, FilterTree tree, QueryContext context)
+ {
+ Row staticRow = partition.staticRow();
+ DecoratedKey partitionKey = partition.partitionKey();
+ List<Unfiltered> matches = new ArrayList<>();
+ boolean hasMatch = false;
- if (tree.isSatisfiedBy(partitionKey, (Row) unfiltered,
staticRow))
- {
- matches.add(unfiltered);
- hasMatch = true;
+ // todo static keys??
Review Comment:
Static rows have always been an extremely annoying thing to test with SAI
queries, and I don't know if that will be much different here. If we do allow
indexing static columns of the `vector` type, there are a couple things we need
to think through on the testing front:
1.) What happens when we index a vector that is a static column and (as is
usually the case), there are many rows in the associated partition? If that
vector is in the top K, do we give the user all the rows in the partition? Do
we fail the assertion `readAndValidatePartition()`?
2.) What happens if we have a non-static column vector indexed, but there's
also a filter on at least one non-vector SAI-indexed column? How do we resolve
the STATIC and WIDE (?) `PrimaryKey` instances together?
To be honest, I would have no issue at all with simply not allowing SAI
indexes on static vector columns. It might not save us from #2 above, but there
are so many corner cases with static columns, I don't know if it's worth
supporting. (i.e. What data model have we seen where someone really needs to do
this and they don't simply have a simple partition key-only primary key?)
--
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]
