fix(python): expose FTS build memory limits

python/python/lancedb/index.py

@@ -163,6 +163,15 @@ class FTS:
         The number of documents per compressed posting block. Supported values
         are 128 and 256. A value of 256 uses the experimental FTS V3 format
         and may introduce breaking changes.
+    memory_limit : int, optional
+        The total memory limit in MiB for the local FTS build stage. The limit
+        is divided evenly among indexing workers. This build-only setting is
+        not persisted with the index and does not apply to remote tables.
+    num_workers : int, optional
+        The number of workers for a local FTS build. By default Lance uses
+        roughly half of the available CPU cores. The effective value is
+        limited by the available compute capacity. This build-only setting is
+        not persisted with the index and does not apply to remote tables.
 
     Notes
     -----
@@ -185,6 +194,8 @@ class FTS:
     prefix_only: bool = False
     block_size: int = 128
     custom_stop_words: Optional[List[str]] = None
+    memory_limit: Optional[int] = None
+    num_workers: Optional[int] = None
 
 
 @dataclass

python/python/tests/test_fts.py

@@ -245,6 +245,14 @@ def test_create_inverted_index_rejects_invalid_block_size(table):
         table.create_index("text", config=FTS(block_size=129))
 
 
+def test_create_inverted_index_respects_build_memory_limit(table):
+    with pytest.raises(ValueError, match="exceeds worker memory limit"):
+        table.create_index(
+            "text",
+            config=FTS(memory_limit=0, num_workers=1),
+        )
+
+
 def test_custom_stop_words_list(table):
     table.create_index(
         "text",

python/src/index.rs

@@ -42,7 +42,7 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
             "Fm" => Ok(LanceDbIndex::Fm(FmIndexBuilder::default())),
             "FTS" => {
                 let params = source.extract::<FtsParams>()?;
-                let inner_opts = FtsIndexBuilder::default()
+                let mut inner_opts = FtsIndexBuilder::default()
                     .base_tokenizer(params.base_tokenizer)
                     .language(&params.language)
                     .map_err(|_| {
@@ -61,6 +61,12 @@ pub fn extract_index_params(source: &Option<Bound<'_, PyAny>>) -> PyResult<Lance
                     .ngram_max_length(params.ngram_max_length)
                     .ngram_prefix_only(params.prefix_only)
                     .custom_stop_words(params.custom_stop_words);
+                if let Some(memory_limit) = params.memory_limit {
+                    inner_opts = inner_opts.memory_limit_mb(memory_limit);
+                }
+                if let Some(num_workers) = params.num_workers {
+                    inner_opts = inner_opts.num_workers(num_workers);
+                }
                 let inner_opts = inner_opts
                     .block_size(params.block_size)
                     .map_err(|err| PyValueError::new_err(err.to_string()))?;
@@ -213,6 +219,8 @@ struct FtsParams {
     ngram_max_length: u32,
     prefix_only: bool,
     block_size: usize,
+    memory_limit: Option<u64>,
+    num_workers: Option<usize>,
 }
 
 #[derive(FromPyObject)]