feat: support FTS indexes on JSON fields (#4219)

`FTS()` rejects JSON columns because validation only considers their
`LargeBinary` storage type. Validate the complete Arrow field in
embedded and remote creation so `lance.json` columns reach Lance's
existing JSON-aware inverted index, while ordinary binary fields remain
rejected.

JSON searches use path-qualified triples such as `meta.tag,str,nature`;
the API documentation describes this syntax and its field-scoping
behavior. Remote creation preserves the supplied FTS configuration.

Two integration limits remain: malformed triples can still panic in the
pinned Lance dependency (tracked by lance-format/lance#8812 and the
pending fix in lance-format/lance#8813), and the opt-in server test has
not run against a real deployment because no remote database is
configured. That test covers index completion, statistics, search
results, and index use; HTTP mock coverage only verifies the client
request.

Refs #3891.

python/python/lancedb/index.py

@@ -112,9 +112,42 @@ class Fm:
 class FTS:
     """Describe a FTS index configuration.
 
-    `FTS` is a full-text search index that can be used on `String` columns
+    `FTS` supports string columns, lists of strings, and JSON columns created
+    from PyArrow ``pa.json_()``. JSON columns use Lance's JSON-aware tokenizer
+    and are stored as ``LargeBinary`` with the ``lance.json`` logical type.
+    Arbitrary binary columns are not supported.
 
-    For example, it works with `title`, `description`, `content`, etc.
+    JSON queries use ``path,str,value`` triples. Paths are relative to the JSON
+    column and use dots for nested object keys. For example,
+    ``meta.tag,str,nature`` matches the string value at ``$.meta.tag``;
+    it does not search other paths or JSON key names. The value is processed
+    with the configured text tokenizer. Use ``str``, not ``string``.
+
+    Create and query a JSON index through the modern API::
+
+        import pyarrow as pa
+        from lancedb.index import FTS
+
+        table = db.create_table(
+            "documents",
+            pa.table({"doc": pa.array(
+                ['{"meta":{"tag":"nature"}}'], type=pa.json_()
+            )}),
+        )
+        table.create_index("doc", config=FTS())
+        table.search(
+            "meta.tag,str,nature", query_type="fts", fts_columns="doc"
+        ).to_list()
+
+    Create the index on the JSON column itself, not a JSON subpath such as
+    ``doc.meta.tag``. JSON-path scalar indexes do not provide FTS path scoping.
+    Plain text such as ``nature`` is not a valid JSON FTS query. The current
+    Lance tokenizer can panic on malformed triples; see
+    https://github.com/lance-format/lance/issues/8812.
+
+    Remote tables use the same ``FTS()`` configuration and require a server
+    that supports JSON FTS. Wait for ``table.wait_for_index(["doc_idx"])``
+    before searching a newly created remote index with its default name.
 
     Examples
     --------

python/python/tests/test_e2e_remote_db.py

@@ -2,13 +2,78 @@
 # SPDX-FileCopyrightText: Copyright The LanceDB Authors
 
 
+import os
+from datetime import timedelta
+from uuid import uuid4
+
+import lancedb
 import numpy as np
+import pyarrow as pa
 import pytest
 from lancedb import LanceDBConnection
+from lancedb.index import FTS
 
 # TODO: setup integ test mark and script
 
 
+@pytest.mark.skipif(
+    not os.environ.get("LANCEDB_URI"), reason="Set LANCEDB_URI to a JSON FTS server"
+)
+def test_json_fts_against_server():
+    """Requires a writable database and capacity to run the distributed indexer.
+
+    Configure LANCEDB_URI, LANCEDB_API_KEY, and optionally LANCEDB_REGION and
+    LANCEDB_HOST_OVERRIDE before running this test.
+    """
+    uri = os.environ["LANCEDB_URI"]
+    assert uri.startswith("db://"), "LANCEDB_URI must identify a remote database"
+    conn = lancedb.connect(
+        uri,
+        api_key=os.environ.get("LANCEDB_API_KEY"),
+        region=os.environ.get("LANCEDB_REGION", "us-east-1"),
+        host_override=os.environ.get("LANCEDB_HOST_OVERRIDE"),
+    )
+    name = f"json_fts_{uuid4().hex}"
+    table = conn.create_table(
+        name,
+        pa.table(
+            {
+                "id": [0, 1, 2],
+                "doc": pa.array(
+                    [
+                        '{"meta":{"tag":"nature"},"author":"alice"}',
+                        '{"meta":{"tag":"city"},"author":"nature"}',
+                        '{"meta":{"tag":"nature"},"author":"bob"}',
+                    ],
+                    type=pa.json_(),
+                ),
+                "raw": pa.array([b"nature"] * 3, type=pa.large_binary()),
+            }
+        ),
+    )
+    try:
+        table.create_index("doc", config=FTS(), name="doc_fts")
+        table.wait_for_index(["doc_fts"], timeout=timedelta(minutes=5))
+        stats = table.index_stats("doc_fts")
+        assert stats.num_indexed_rows == 3
+        assert stats.num_unindexed_rows == 0
+        assert stats.index_type == "FTS"
+        for query, expected in [
+            ("meta.tag,str,nature", [0, 2]),
+            ("meta.tag,str,alice", []),
+            ("meta.tag,str,tag", []),
+        ]:
+            search = table.search(query, query_type="fts", fts_columns="doc")
+            assert sorted(row["id"] for row in search.to_list()) == expected
+            plan = search.explain_plan(True)
+            assert "MatchQuery" in plan, plan
+            assert "FlatMatchQuery" not in plan, plan
+        with pytest.raises(ValueError, match="FTS.*lance.json.*raw binary"):
+            table.create_index("raw", config=FTS())
+    finally:
+        conn.drop_table(name)
+
+
 @pytest.mark.skip(reason="Need to set up a local server")
 def test_against_local_server():
     conn = LanceDBConnection("lancedb+http://localhost:10024")

python/python/tests/test_fts.py

@@ -45,6 +45,75 @@ from utils import exception_output
 TEST_LANGUAGE_MODEL_HOME = Path(__file__).parent / "models"
 
 
+@pytest.mark.skipif(not hasattr(pa, "json_"), reason="requires PyArrow JSON type")
+@pytest.mark.parametrize("block_size", [128, 256])
+def test_json_fts(tmp_path, block_size):
+    db = ldb.connect(tmp_path)
+    data = pa.table(
+        {
+            "id": [0, 1, 2, 3],
+            "doc": pa.array(
+                [
+                    '{"title":"brown fox","meta":{"tag":"nature"},"author":"alice"}',
+                    '{"title":"blue car","meta":{"tag":"city"}}',
+                    '{"title":"brown bear","meta":{"tag":"nature"}}',
+                    '{"title":"red fox","author":"alice"}',
+                ],
+                type=pa.json_(),
+            ),
+        }
+    )
+    table = db.create_table("json_fts", data)
+    table.create_index("doc", config=FTS(block_size=block_size), name="doc_fts")
+    table = db.open_table("json_fts")
+    index = table.list_indices()[0]
+    assert index.name == "doc_fts"
+    assert index.index_type == "FTS"
+    assert index.columns == ["doc"]
+    stats = table.index_stats("doc_fts")
+    assert stats.num_indexed_rows == 4
+    assert stats.num_unindexed_rows == 0
+
+    for query, expected in [
+        ("title,str,brown", [0, 2]),
+        ("meta.tag,str,nature", [0, 2]),
+        ("author,str,alice", [0, 3]),
+        ("title,str,alice", []),
+        ("title,str,title", []),
+    ]:
+        search = table.search(query, query_type="fts", fts_columns="doc")
+        assert sorted(row["id"] for row in search.to_list()) == expected
+        plan = search.explain_plan(True)
+        assert "MatchQuery" in plan, plan
+        assert "FlatMatchQuery" not in plan, plan
+
+
+@pytest.mark.skipif(not hasattr(pa, "json_"), reason="requires PyArrow JSON type")
+@pytest.mark.asyncio
+async def test_json_fts_async(tmp_path):
+    db = await ldb.connect_async(tmp_path)
+    table = await db.create_table(
+        "json_fts",
+        pa.table({"doc": pa.array(['{"meta":{"tag":"nature"}}'], type=pa.json_())}),
+    )
+    await table.create_index("doc", config=FTS())
+    search = table.query().nearest_to_text(MatchQuery("meta.tag,str,nature", "doc"))
+    assert len(await search.to_list()) == 1
+    plan = await search.explain_plan(True)
+    assert "MatchQuery" in plan, plan
+    assert "FlatMatchQuery" not in plan, plan
+
+
+@pytest.mark.parametrize("dtype", [pa.binary(), pa.large_binary()])
+def test_fts_rejects_raw_binary(tmp_path, dtype):
+    table = ldb.connect(tmp_path).create_table(
+        "binary_fts", pa.table({"doc": pa.array([b'{"title":"brown"}'], type=dtype)})
+    )
+    with pytest.raises(ValueError, match="FTS.*lance.json.*raw binary"):
+        table.create_index("doc", config=FTS())
+    assert table.list_indices() == []
+
+
 @pytest.fixture
 def table(tmp_path) -> ldb.table.LanceTable:
     # Use local random state to avoid affecting other tests

python/python/tests/test_remote_db.py

@@ -1410,6 +1410,72 @@ def test_remote_job_wait_raises_on_failure():
             job.wait()
 
 
+@pytest.mark.parametrize("server_version", ["0.2.0", "0.6.0"])
+def test_remote_json_fts_config(server_version):
+    from lancedb.index import FTS
+
+    received_requests = []
+
+    def handler(request):
+        request.send_response(200)
+        request.send_header("Content-Type", "application/json")
+        request.send_header("phalanx-version", server_version)
+        request.end_headers()
+        if request.path == "/v1/table/test/describe/":
+            fields = [
+                {
+                    "name": "doc",
+                    "type": {"type": "large_binary"},
+                    "nullable": True,
+                    "metadata": {"ARROW:extension:name": "lance.json"},
+                },
+                {
+                    "name": "raw",
+                    "type": {"type": "large_binary"},
+                    "nullable": True,
+                },
+            ]
+            request.wfile.write(
+                json.dumps({"version": 1, "schema": {"fields": fields}}).encode()
+            )
+        elif request.path == "/v1/table/test/create_index/":
+            received_requests.append(read_json_body(request))
+            request.wfile.write(b"{}")
+        else:
+            raise AssertionError(f"Unexpected request: {request.path}")
+
+    with mock_lancedb_connection(handler) as db:
+        table = db.open_table("test")
+        table.create_index(
+            "doc",
+            config=FTS(
+                with_position=True,
+                base_tokenizer="whitespace",
+                stem=False,
+                remove_stop_words=False,
+                lower_case=False,
+                block_size=256,
+            ),
+            name="doc_fts",
+            replace=False,
+        )
+        with pytest.raises(ValueError, match="FTS.*lance.json.*raw binary"):
+            table.create_index("raw", config=FTS())
+
+    assert len(received_requests) == 1
+    request = received_requests[0]
+    assert request["column"] == "doc"
+    assert request["index_type"] == "FTS"
+    assert request["name"] == "doc_fts"
+    assert request["replace"] is False
+    assert request["with_position"] is True
+    assert request["base_tokenizer"] == "whitespace"
+    assert request["stem"] is False
+    assert request["remove_stop_words"] is False
+    assert request["lower_case"] is False
+    assert request["block_size"] == 256
+
+
 def test_remote_create_index_new_api(caplog):
     received_requests = []