redis/redis-py · error · DataError

Cannot set 'sortable' or 'no_index' in Vector fields.

Error message

Cannot set 'sortable' or 'no_index' in Vector fields.

What it means

Raised by VectorField.__init__ when kwargs contain sortable=True or no_index=True. Vector fields are inherently always indexed and never sortable in RediSearch — these flags have no meaning for vectors and the constructor rejects them rather than silently dropping them. Note: raises DataError.

Source

Thrown at redis/commands/search/field.py:195

    def __init__(self, name: str, algorithm: str, attributes: dict, **kwargs):
        """
        Create Vector Field. Notice that Vector cannot have sortable or no_index tag,
        although it's also a Field.

        ``name`` is the name of the field.

        ``algorithm`` can be "FLAT", "HNSW", or "SVS-VAMANA".

        ``attributes`` each algorithm can have specific attributes. Some of them
        are mandatory and some of them are optional. See
        https://oss.redis.com/redisearch/master/Vectors/#specific_creation_attributes_per_algorithm
        for more information.
        """
        sort = kwargs.get("sortable", False)
        noindex = kwargs.get("no_index", False)

        if sort or noindex:
            raise DataError("Cannot set 'sortable' or 'no_index' in Vector fields.")

        if algorithm.upper() not in ["FLAT", "HNSW", "SVS-VAMANA"]:
            raise DataError(
                "Realtime vector indexing supporting 3 Indexing Methods:"
                "'FLAT', 'HNSW', and 'SVS-VAMANA'."
            )

        attr_li = []

        for key, value in attributes.items():
            attr_li.extend([key, value])

        Field.__init__(
            self, name, args=[Field.VECTOR, algorithm, len(attr_li), *attr_li], **kwargs
        )

View on GitHub (pinned to da03cdc7e8)

Solutions

1. Drop sortable and no_index from VectorField construction — vectors are always indexed and non-sortable.
2. Filter kwargs before passing: kwargs.pop('sortable', None); kwargs.pop('no_index', None).
3. Build vector fields separately from text/numeric fields to avoid shared flag templates.

Example fix

// before
VectorField('vec', 'FLAT', {'TYPE': 'FLOAT32', 'DIM': 128, 'DISTANCE_METRIC': 'L2'}, sortable=True)
// after
VectorField('vec', 'FLAT', {'TYPE': 'FLOAT32', 'DIM': 128, 'DISTANCE_METRIC': 'L2'})

def make_vector_field(name, algorithm, attributes, **kw):
    kw.pop("sortable", None)
    kw.pop("no_index", None)
    if kw:
        raise ValueError(f"Unexpected vector kwargs: {list(kw)}")
    from redis.commands.search.field import VectorField
    return VectorField(name, algorithm, attributes)

def vector_kwargs_are_clean(kw) -> bool:
    return "sortable" not in kw and "no_index" not in kw

from redis import DataError
try:
    VectorField("vec", "FLAT", attrs, sortable=True)
except DataError as e:
    if "Cannot set 'sortable' or 'no_index'" in str(e):
        VectorField("vec", "FLAT", attrs)
    else:
        raise

When it happens

Trigger: Construct VectorField(name, algorithm='FLAT', attributes={...}, sortable=True) or VectorField(..., no_index=True) — passing either flag through **kwargs.

Common situations: Applying a uniform 'all fields sortable' policy when building a schema that includes vectors; copy-pasting a TextField definition into a vector context with sortable=True left in.

Related errors

• Realtime vector indexing supporting 3 Indexing Methods:'FLAT
• Non-Sortable non-Indexable fields are ignored
• CLIENT KILL type must be one of {client_types!r}
• CLIENT KILL skipme must be a bool
• CLIENT KILL <filter> <value> ... ... <filter> <value> must s

AI-assisted analysis of redis/redis-py@da03cdc7e8 (2026-08-04). Data as JSON: /data/errors/3e5efbf8f2f26ec2.json. Report an issue: GitHub.