[BUG] Raise Error when can't deserialize configuration json from server, lazily load ef on CollectionModel, warn on api_key (#4471)

## Description of changes

This PR fixes a bug where it warns users if the configuration is not
deserializable rather than raising an error, giving the users an empty
config. Now, it raises an error if the configuration is not
deserializable. This will only happen when the client is ahead of the
server, which we do not need to support.

EF's from the config are now lazily loaded. only when ._embed is invoked
from the model is it tried to be built, and will take precedent over the
provided ef.

This also fixes a bug with list_collections, where if the user tries to
load collections without the proper embedding function parameters set
up, like api key env var, etc it would throw an hnsw and spann not
configured error, which is not the root cause. instead, it will now load
only the configuration_json, and at embed time try to load the ef, so
errors will only propogate at that point, and with much clearer error
messages.

Now, using api_key directly in embedding_functions will give a
deprecation warning saying it is not persisted.

## Test plan

_How are these changes tested?_

- [ ] Tests pass locally with `pytest` for python, `yarn test` for js,
`cargo test` for rust

## Documentation Changes

_Are all docstrings for user-facing APIs updated if required? Do we need
to make documentation changes in the [docs
section](https://github.com/chroma-core/chroma/tree/main/docs/docs.trychroma.com)?_

Author: jairad26

chromadb/api/collection_configuration.py

@@ -82,12 +82,24 @@ def load_collection_configuration_from_json(
             ef = None
         else:
             try:
-                ef = known_embedding_functions[ef_config["name"]]
-                ef = ef.build_from_config(ef_config["config"])  # type: ignore
+                ef_name = ef_config["name"]
+            except KeyError:
+                raise ValueError(
+                    f"Embedding function name not found in config: {ef_config}"
+                )
+            try:
+                ef = known_embedding_functions[ef_name]
             except KeyError:
                 raise ValueError(
-                    f"Embedding function {ef_config['name']} not found. Add @register_embedding_function decorator to the class definition."
+                    f"Embedding function {ef_name} not found. Add @register_embedding_function decorator to the class definition."
                 )
+            try:
+                ef = ef.build_from_config(ef_config["config"])  # type: ignore
+            except Exception as e:
+                raise ValueError(
+                    f"Could not build embedding function {ef_config['name']} from config {ef_config['config']}: {e}"
+                )
+
     else:
         ef = None
 

chromadb/api/models/CollectionCommon.py

@@ -129,18 +129,7 @@ def __init__(
         if embedding_function is not None:
             validate_embedding_function(embedding_function)
 
-        config_ef = self.configuration.get("embedding_function")
-        if config_ef is not None:
-            if embedding_function is not None and not isinstance(
-                embedding_function, ef.DefaultEmbeddingFunction
-            ):
-                if embedding_function.name() is not config_ef.name():
-                    raise ValueError(
-                        f"Embedding function name mismatch: {embedding_function.name()} != {config_ef.name()}"
-                    )
-            self._embedding_function = config_ef
-        else:
-            self._embedding_function = embedding_function
+        self._embedding_function = embedding_function
         self._data_loader = data_loader
 
     # Expose the model properties as read-only properties on the Collection class
@@ -567,6 +556,13 @@ def _embed_record_set(
         )
 
     def _embed(self, input: Any) -> Embeddings:
+        if self._embedding_function is not None and not isinstance(
+            self._embedding_function, ef.DefaultEmbeddingFunction
+        ):
+            return self._embedding_function(input=input)
+        config_ef = self.configuration.get("embedding_function")
+        if config_ef is not None:
+            return config_ef(input=input)
         if self._embedding_function is None:
             raise ValueError(
                 "You must provide an embedding function to compute embeddings."

chromadb/api/rust.py

@@ -17,7 +17,6 @@
     UpdateCollectionConfiguration,
     create_collection_configuration_to_json_str,
     update_collection_configuration_to_json_str,
-    load_collection_configuration_from_json,
 )
 from chromadb.auth import UserIdentity
 from chromadb.config import DEFAULT_DATABASE, DEFAULT_TENANT, Settings, System
@@ -191,9 +190,7 @@ def list_collections(
             CollectionModel(
                 id=collection.id,
                 name=collection.name,
-                configuration=load_collection_configuration_from_json(
-                    collection.configuration
-                ),
+                configuration_json=collection.configuration,
                 metadata=collection.metadata,
                 dimension=collection.dimension,
                 tenant=collection.tenant,
@@ -233,9 +230,7 @@ def create_collection(
         collection_model = CollectionModel(
             id=collection.id,
             name=collection.name,
-            configuration=load_collection_configuration_from_json(
-                collection.configuration
-            ),
+            configuration_json=collection.configuration,
             metadata=collection.metadata,
             dimension=collection.dimension,
             tenant=collection.tenant,
@@ -254,9 +249,7 @@ def get_collection(
         return CollectionModel(
             id=collection.id,
             name=collection.name,
-            configuration=load_collection_configuration_from_json(
-                collection.configuration
-            ),
+            configuration_json=collection.configuration,
             metadata=collection.metadata,
             dimension=collection.dimension,
             tenant=collection.tenant,

chromadb/api/segment.py

@@ -3,7 +3,7 @@
 from chromadb.api.collection_configuration import (
     CreateCollectionConfiguration,
     UpdateCollectionConfiguration,
-    load_collection_configuration_from_create_collection_configuration,
+    create_collection_configuration_to_json,
 )
 from chromadb.auth import UserIdentity
 from chromadb.config import DEFAULT_DATABASE, DEFAULT_TENANT, Settings, System
@@ -235,7 +235,7 @@ def create_collection(
             id=id,
             name=name,
             metadata=metadata,
-            configuration=load_collection_configuration_from_create_collection_configuration(
+            configuration_json=create_collection_configuration_to_json(
                 configuration or CreateCollectionConfiguration()
             ),
             tenant=tenant,
@@ -413,7 +413,9 @@ def _fork(
         tenant: str = DEFAULT_TENANT,
         database: str = DEFAULT_DATABASE,
     ) -> CollectionModel:
-        raise NotImplementedError("Collection forking is not implemented for SegmentAPI")
+        raise NotImplementedError(
+            "Collection forking is not implemented for SegmentAPI"
+        )
 
     @trace_method("SegmentAPI.delete_collection", OpenTelemetryGranularity.OPERATION)
     @override

chromadb/db/impl/grpc/server.py

@@ -2,9 +2,8 @@
 from typing import Any, Dict, List, cast
 from uuid import UUID
 from overrides import overrides
-from chromadb.api.collection_configuration import (
-    load_collection_configuration_from_json_str,
-)
+import json
+
 from chromadb.config import DEFAULT_DATABASE, DEFAULT_TENANT, Component, System
 from chromadb.proto.convert import (
     from_proto_metadata,
@@ -286,15 +285,13 @@ def CreateCollection(
                 f"Collection {collection_name} already exists",
             )
 
-        configuration = load_collection_configuration_from_json_str(
-            request.configuration_json_str
-        )
+        configuration_json = json.loads(request.configuration_json_str)
 
         id = UUID(hex=request.id)
         new_collection = Collection(
             id=id,
             name=request.name,
-            configuration=configuration,
+            configuration_json=configuration_json,
             metadata=from_proto_metadata(request.metadata),
             dimension=request.dimension,
             database=database,

chromadb/db/mixins/sysdb.py

@@ -38,7 +38,8 @@
     create_collection_configuration_to_json_str,
     load_collection_configuration_from_json_str,
     CollectionConfiguration,
-    load_collection_configuration_from_create_collection_configuration,
+    create_collection_configuration_to_json,
+    collection_configuration_to_json,
     collection_configuration_to_json_str,
     overwrite_collection_configuration,
     update_collection_configuration_from_legacy_update_metadata,
@@ -310,9 +311,7 @@ def create_collection(
         collection = Collection(
             id=id,
             name=name,
-            configuration=load_collection_configuration_from_create_collection_configuration(
-                configuration
-            ),
+            configuration_json=create_collection_configuration_to_json(configuration),
             metadata=metadata,
             dimension=dimension,
             tenant=tenant,
@@ -541,7 +540,9 @@ def get_collections(
                     Collection(
                         id=cast(UUID, id),
                         name=name,
-                        configuration=configuration,
+                        configuration_json=collection_configuration_to_json(
+                            configuration
+                        ),
                         metadata=metadata,
                         dimension=dimension,
                         tenant=str(rows[0][5]),

chromadb/proto/convert.py

@@ -1,13 +1,13 @@
 from typing import Dict, Optional, Sequence, Tuple, TypedDict, Union, cast
 from uuid import UUID
+import json
 
 import numpy as np
 from numpy.typing import NDArray
 
 import chromadb.proto.chroma_pb2 as chroma_pb
 import chromadb.proto.query_executor_pb2 as query_pb
 from chromadb.api.collection_configuration import (
-    load_collection_configuration_from_json_str,
     collection_configuration_to_json_str,
 )
 from chromadb.api.types import Embedding, Where, WhereDocument
@@ -239,9 +239,7 @@ def from_proto_collection(collection: chroma_pb.Collection) -> Collection:
     return Collection(
         id=UUID(hex=collection.id),
         name=collection.name,
-        configuration=load_collection_configuration_from_json_str(
-            collection.configuration_json_str
-        ),
+        configuration_json=json.loads(collection.configuration_json_str),
         metadata=from_proto_metadata(collection.metadata)
         if collection.HasField("metadata")
         else None,