microsoft/semantic-kernel · error · ServiceResourceNotFoundError

Key not found

Error message

Key not found

What it means

Raised by PostgresMemoryStore.get() after the SELECT returns no row for the supplied key (cur.fetchone() is None). It is a ServiceResourceNotFoundError distinct from the collection-missing case: the collection is fine, but the specific row is absent. The message is the literal string 'Key not found' with no interpolation.

Source

Thrown at python/semantic_kernel/connectors/memory_stores/postgres/postgres_memory_store.py:278

        with self._connection_pool.connection() as conn, conn.cursor() as cur:
            if not await self.__does_collection_exist(cur, collection_name):
                raise ServiceResourceNotFoundError(f"Collection '{collection_name}' does not exist")
            cur.execute(
                SQL(
                    """
                        SELECT key, embedding, metadata, timestamp
                        FROM {scm}.{tbl}
                        WHERE key = %s
                        """
                ).format(
                    scm=Identifier(self._schema),
                    tbl=Identifier(collection_name),
                ),
                (key,),
            )
            result = cur.fetchone()
            if result is None:
                raise ServiceResourceNotFoundError("Key not found")
            return MemoryRecord.local_record(
                id=result[0],
                embedding=(
                    np.fromstring(result[1].strip("[]"), dtype=float, sep=",") if with_embedding else np.array([])
                ),
                text=result[2]["text"],
                description=result[2]["description"],
                additional_metadata=result[2]["additional_metadata"],
                timestamp=result[3],
            )

    async def get_batch(
        self, collection_name: str, keys: list[str], with_embeddings: bool = False
    ) -> list[MemoryRecord]:
        """Gets a batch of records.

        Args:
            collection_name: The name of the collection to get the records from.

View on GitHub (pinned to c028a0c7dc)

Solutions

1. Catch ServiceResourceNotFoundError around get() and treat as a cache miss.
2. Prefer get_batch() if partial misses are acceptable; it returns only found rows rather than throwing.
3. Normalize the key before get (strip, lowercase) to match what was stored.
4. Verify the row exists with a direct check or by re-upserting before read.

Example fix

// before
rec = await store.get('mycol', key)
// after
from semantic_kernel.exceptions import ServiceResourceNotFoundError
try:
    rec = await store.get('mycol', key)
except ServiceResourceNotFoundError:
    rec = None

from semantic_kernel.exceptions import ServiceResourceNotFoundError
try:
    record = await store.get(collection_name, key)
except ServiceResourceNotFoundError as e:
    if str(e) == 'Key not found':
        record = None
    else:
        raise

When it happens

Trigger: Awaiting `store.get(collection_name, key)` against an existing collection where no row has `key = <key>`. Common when the record was never upserted, was removed, or the key string differs (whitespace, case, type).

Common situations: Reading a key right after upsert that silently failed; key generated client-side differs from stored id; record deleted by remove()/remove_batch(); concurrent writer deleted the row between operations.

Related errors

• No match found
• Failed to create Postgres settings.
• Collection '{collection_name}' does not exist
• Upsert failed
• Dimensionality of {dimension_num} exceeds the maximum allowe

AI-assisted analysis of microsoft/semantic-kernel@c028a0c7dc (2026-08-13). Data as JSON: /api/errors/7fdf993ec7626085. Report an issue: GitHub.