What causes "Collection {collection_name} does not exist, cannot insert." in microsoft/semantic-kernel?

Create the collection first: `await store.create_collection('docs', ndim=1536)`.

microsoft/semantic-kernel · error · ServiceResourceNotFoundError

Collection {collection_name} does not exist, cannot insert.

Error message

Collection {collection_name} does not exist, cannot insert.

What it means

Raised by USearchMemoryStore.upsert_batch (ServiceResourceNotFoundError) when the target collection is not in `self._collections`. The message says 'cannot insert' because upsert first removes existing labels then adds — so a missing collection is rejected before any index mutation. The name is matched case-insensitively (lowercased).

Source

Thrown at python/semantic_kernel/connectors/memory_stores/usearch/usearch_memory_store.py:332

        Args:
            collection_name (str): Name of the collection to search within.
            records (List[MemoryRecord]): Records to upsert.
            compact (bool, optional): Removes links to removed nodes (expensive). Defaults to False.
            copy (bool, optional): Should the index store a copy of vectors. Defaults to True.
            threads (int, optional): Optimal number of cores to use. Defaults to 0.
            log (Union[str, bool], optional): Whether to print the progress bar. Defaults to False.
            batch_size (int, optional): Number of vectors to process at once. Defaults to 0.

        Raises:
            KeyError: If collection not exist

        Returns:
            List[str]: List of IDs.
        """
        collection_name = collection_name.lower()
        if collection_name not in self._collections:
            raise ServiceResourceNotFoundError(f"Collection {collection_name} does not exist, cannot insert.")

        ucollection = self._collections[collection_name]
        all_records_id = [record._id for record in records]

        # Remove vectors from index
        remove_labels = [
            ucollection.embeddings_id_to_label[id] for id in all_records_id if id in ucollection.embeddings_id_to_label
        ]
        ucollection.embeddings_index.remove(remove_labels, compact=compact, threads=threads)

        # Determine label insertion points
        table_num_rows = ucollection.embeddings_data_table.num_rows
        insert_labels = np.arange(table_num_rows, table_num_rows + len(records))

        # Add embeddings to index
        ucollection.embeddings_index.add(
            keys=insert_labels,
            vectors=np.stack([record.embedding for record in records]),

View on GitHub (pinned to c028a0c7dc)

Solutions

Create the collection first: `await store.create_collection('docs', ndim=1536)`.
Guard with `if 'docs' not in ...` / does_collection_exist before upserting.
Match the name exactly (it is lowercased internally).

Example fix

// before
await store.upsert_batch('docs', records)
// after
if not await store.does_collection_exist('docs'):
    await store.create_collection('docs', ndim=1536)
await store.upsert_batch('docs', records)

Defensive patterns

Strategy: validation

Validate before calling

name = collection_name.lower()
if not await store.does_collection_exist(name):
    await store.create_collection(name, ndim=embedding_dim)
await store.upsert_batch(name, records)

Try / catch

from semantic_kernel.exceptions import ServiceResourceNotFoundError
try:
    await store.upsert_batch(name, records)
except ServiceResourceNotFoundError:
    await store.create_collection(name, ndim=dim)
    await store.upsert_batch(name, records)

Prevention

Create the collection with the correct ndim before the first upsert.
Keep collection names consistent (they are lowercased internally).
If persistence is off, recreate collections on each process start.

When it happens

Trigger: Calling `await store.upsert_batch('docs', records)` when 'docs' was never created in this process (and not loaded from disk). The existence check runs before removing/adding vectors.

Common situations: Forgot to call create_collection before upserting; collection name typo or case mismatch (names are lowercased); in-memory store where the collection was lost after a restart; persist_directory not set so nothing was loaded.

Related errors

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/3dfa2795ace2177f. Report an issue: GitHub.