vector-store: allow syntax for local indexes

[ewienik]

ScylladDB currently supports syntax for creating local indexes. Vector-store needs to be able to read them and create local indexes using specified primary key. There will be several PRs to introduce full local indexes. This PR parses local indexes metadata and creates local indexes as global indexes.

The change introduces parsing and testing local indexes and refactoring of tests and types.

Fixes: vector-531

[ewienik]

Changelog for 1fd0207

• fix clippy warning

range diff

1:  cdc881af ! 1:  02bc1860 vector-store/tests: use DbIndexType in integration tests
    @@ crates/vector-store/tests/integration/usearch.rs: pub(crate) async fn setup_stor
              index_name: "ann".into(),
              target_column: "embedding".into(),
     +        index_type,
    -+        filtering_columns: columns.iter().map(|(name, _)| name.clone()).collect(),
    ++        filtering_columns: columns.keys().cloned().collect(),
              dimensions: dimension,
     -        index_type: DbIndexType::Global,
     -        filtering_columns: vec![],
2:  c0f1b763 = 2:  e3043edd vector-store: refactor AsRef trait for *Name types
3:  4b9e659d = 3:  d5e72d43 validator: refactor keyspace/table/index names
4:  2b226301 = 4:  8a07e0eb validator: refactor create_index with CreateIndexQuery
5:  deb33596 = 5:  e1ef607e validator: implement support for local indexes in CreateIndexQuery
6:  15dcaa77 = 6:  1fd0207b validator: implement index_create test case

[ewienik]

Changelog for 81bb2df

• provide local partition key's columns within DbIndexType::Local

range diff

1:  ec8a85f2 ! 1:  4a4dd003 vector-store: implement local indexes as global indexes
    @@ crates/vector-store/src/db.rs: impl Statements {
     +
     +        let mut columns = target_option.ck.into_iter();
     +        return Ok((
    -+            DbIndexType::Local,
    ++            DbIndexType::Local(target_option.pk.into_iter().map(ColumnName::from).collect()),
     +            columns.next().unwrap().into(),
     +            columns.map(ColumnName::from).collect(),
     +        ));
    @@ crates/vector-store/src/db.rs: impl Statements {
     +    let target_name = target_option.pk.remove(0);
     +    validate_target_type(&target_name)?;
     +    Ok((
    -+        DbIndexType::Local,
    ++        DbIndexType::Global,
     +        target_name.into(),
     +        target_option.ck.into_iter().map(ColumnName::from).collect(),
     +    ))
    @@ crates/vector-store/src/lib.rs: impl IndexMetadata {
          }
      }
      
    -+#[derive(Clone, Copy, Debug, PartialEq, Eq, Hash)]
    ++#[derive(Clone, Debug, PartialEq, Eq, Hash)]
     +pub enum DbIndexType {
     +    Global,
    -+    Local,
    ++    Local(Vec<ColumnName>),
     +}
     +
      #[derive(Debug)]
2:  02bc1860 = 2:  f4c9e7fa vector-store/tests: use DbIndexType in integration tests
3:  e3043edd = 3:  7a881921 vector-store: refactor AsRef trait for *Name types
4:  d5e72d43 = 4:  1bb77bd9 validator: refactor keyspace/table/index names
5:  8a07e0eb = 5:  923d2262 validator: refactor create_index with CreateIndexQuery
6:  e1ef607e = 6:  55ec8a11 validator: implement support for local indexes in CreateIndexQuery
7:  1fd0207b = 7:  81bb2df7 validator: implement index_create test case

[knowack1]

Look great. Please check some of my nit thoughts/comments. I would consider unit tests for a new code in db.rs - the logic looks quite complex there.

[knowack1]

Would be nice to have unit tests for this logic.

[ewienik]

It is impossible to unit test this as the function uses cql driver's Table struct which we cannot prepare manually (it needs working session with db)

[knowack1]

I would be good to have an jira issue for this. Otherwise we will forget.

[knowack1]

consider encapsulating this loop into a function - also in other places.