community[minor]: add additional support for BigQueryVectorSearch (#15904)

BigQuery vector search lets you use GoogleSQL to do semantic search,
using vector indexes for fast but approximate results, or using brute
force for exact results.

This PR:
1. Add `metadata[_job_ib]` in Document returned by any similarity search
2. Add `explore_job_stats` to enable users to explore job statistics and
better the debuggability
3. Set the minimum row limit for running create vector index.

docs/docs/integrations/vectorstores/bigquery_vector_search.ipynb

@@ -324,6 +324,24 @@
     "docs = store.similarity_search_by_vector(query_vector, filter={\"len\": 6})\n",
     "print(docs)"
    ]
+  },
+  {
+   "cell_type": "markdown",
+   "metadata": {},
+   "source": [
+    "### Explore job satistics with BigQuery Job Id"
+   ]
+  },
+  {
+   "cell_type": "code",
+   "execution_count": null,
+   "metadata": {},
+   "outputs": [],
+   "source": [
+    "job_id = \"\"  # @param {type:\"string\"}\n",
+    "# Debug and explore the job statistics with a BigQuery Job id.\n",
+    "store.explore_job_stats(job_id)"
+   ]
   }
  ],
  "metadata": {

libs/community/langchain_community/vectorstores/bigquery_vector_search.py

@@ -28,6 +28,7 @@ DEFAULT_METADATA_COLUMN_NAME = "metadata"  # document metadata
 DEFAULT_CONTENT_COLUMN_NAME = "content"  # text content, do not rename
 DEFAULT_TOP_K = 4  # default number of documents returned from similarity search
 
+_MIN_INDEX_ROWS = 5000  # minimal number of rows for creating an index
 _INDEX_CHECK_PERIOD_SECONDS = 60  # Do not check for index more often that this.
 
 _vector_table_lock = Lock()  # process-wide BigQueryVectorSearch table lock
@@ -192,6 +193,11 @@ class BigQueryVectorSearch(VectorStore):
         if self._have_index or self._creating_index:
             # Already have an index or in the process of creating one.
             return
+        table = self.bq_client.get_table(self.vectors_table)
+        if (table.num_rows or 0) < _MIN_INDEX_ROWS:
+            # Not enough rows to create index.
+            self._logger.debug("Not enough rows to create a vector index.")
+            return
         if (
             datetime.utcnow() - self._last_index_check
         ).total_seconds() < _INDEX_CHECK_PERIOD_SECONDS:
@@ -228,6 +234,10 @@ class BigQueryVectorSearch(VectorStore):
     def _create_index(self):
         from google.api_core.exceptions import ClientError
 
+        table = self.bq_client.get_table(self.vectors_table)
+        if (table.num_rows or 0) < _MIN_INDEX_ROWS:
+            # Not enough rows to create index.
+            return
         if self.distance_strategy == DistanceStrategy.EUCLIDEAN_DISTANCE:
             distance_type = "EUCLIDEAN"
         elif self.distance_strategy == DistanceStrategy.COSINE:
@@ -534,6 +544,7 @@ class BigQueryVectorSearch(VectorStore):
             else:
                 metadata = {}
             metadata["__id"] = row[self.doc_id_field]
+            metadata["__job_id"] = job.job_id
             doc = Document(page_content=row[self.content_field], metadata=metadata)
             document_tuples.append(
                 (doc, row[self.text_embedding_field], row["_vector_search_distance"])
@@ -833,3 +844,14 @@ class BigQueryVectorSearch(VectorStore):
         vs_obj = BigQueryVectorSearch(embedding=embedding, **kwargs)
         vs_obj.add_texts(texts, metadatas)
         return vs_obj
+
+    def explore_job_stats(self, job_id: str) -> Dict:
+        """Return the statistics for a single job execution.
+
+        Args:
+            job_id: The BigQuery Job id.
+
+        Returns:
+            A dictionary of job statistics for a given job.
+        """
+        return self.bq_client.get_job(job_id)._properties["statistics"]