Check for a covering index using pg_catalog params, not ix name/string matching

[olirice]

Currently vecs uses string pattern matching on the index name to determine if an index exists and supports the correct vector ops

Logic is here

vecs/src/vecs/collection.py

Lines 623 to 677 in cc412ce

	def index(self) -> Optional[str]:
	"""
	PRIVATE
	Note:
	The `index` property is private and expected to undergo refactoring.
	Do not rely on it's output.
	Retrieves the SQL name of the collection's vector index, if it exists.
	Returns:
	Optional[str]: The name of the index, or None if no index exists.
	"""
	if self._index is None:
	query = text(
	"""
	select
	relname as table_name
	from
	pg_class pc
	where
	pc.relnamespace = 'vecs'::regnamespace
	and relname ilike 'ix_vector%'
	and pc.relkind = 'i'
	"""
	)
	with self.client.Session() as sess:
	ix_name = sess.execute(query).scalar()
	self._index = ix_name
	return self._index
	def is_indexed_for_measure(self, measure: IndexMeasure):
	"""
	Checks if the collection is indexed for a specific measure.
	Args:
	measure (IndexMeasure): The measure to check for.
	Returns:
	bool: True if the collection is indexed for the measure, False otherwise.
	"""
	index_name = self.index
	if index_name is None:
	return False
	ops = INDEX_MEASURE_TO_OPS.get(measure)
	if ops is None:
	return False
	if ops in index_name:
	return True
	return False

It would be preferable to lookup the supported ops in pg_catalog such that it is more intuitive

[KaranPradhan266]

Hey @olirice — I’d love to work on this. I recently contributed to the Supabase stripe sync engine and would be excited to take a shot at this as well. Let me know if it’s still available to work on!

[omwanere]

Hi 👋
I can take a look at this and update the index detection logic to use pg_catalog metadata instead of index name matching.
I’ll investigate and share a PR if I find a clean solution.