Add Valkey backend support with conditional SORTBY handling#2
Conversation
| _never_supports_sortby, | ||
| raising=True, | ||
| ) | ||
|
|
There was a problem hiding this comment.
Suggestion: Add assertion/ test for sorting: scores = [extract_score(item) for item in res]; assert is_nondecreasing(scores), f"KNN results must be sorted: {scores}"
|
|
||
| _sortby_supported: bool | None = None | ||
|
|
||
| def _check_sortby_support(self, index_name: str, sort_field: str) -> bool: |
There was a problem hiding this comment.
Update docstring: Add "NOTE: Valkey KNN results are ALWAYS sorted by distance. 'The first name/value pair is for the distance computed' - Ref: https://github.com/valkey-io/valkey-search/blob/main/COMMANDS.md"
| self._sortby_supported = False | ||
| else: | ||
| self._sortby_supported = True | ||
| return self._sortby_supported |
There was a problem hiding this comment.
Suggestion: verify it's Valkey before skipping SORTBY, as it is uncertain if there might be other cases (not Valkey) where SORT BY is not supported and no default sorting is guaranteed.
except ResponseError as e:
if "SORTBY" in str(e):
try:
info = self._client.info("server")
is_valkey = info.get("server_name") == "valkey" or "valkey_version" in info
self._sortby_supported = not is_valkey
except Exception:
self._sortby_supported = False
else:
self._sortby_supported = True
return self._sortby_supported| * [x] Support `Chroma <https://github.com/chroma-core/chroma>`_\ , the AI-native open-source embedding database. | ||
| * [x] Support `DocArray <https://github.com/docarray/docarray>`_\ , DocArray is a library for representing, sending and storing multi-modal data, perfect for Machine Learning applications. | ||
| * [x] Support `Redis <https://redis.io/>`_. | ||
| * [x] Support `ValkeySearch <https://github.com/valkey-io/valkey-search/>`_\ , provided as a Valkey module, is a high-performance Vector Similarity Search engine optimized for AI-driven workloads. |
There was a problem hiding this comment.
Instead of ValkeySearch - suggesting to add Valkey (with valkey-search)
3 participants
This PR introduces Valkey integration for the Redis-based vector store in GPTCache. Key changes include:
Valkey compatibility:
Conditional SORTBY logic:
No breaking changes: