close connections

llama_stack/providers/inline/agents/meta_reference/agents.py

@@ -314,7 +314,7 @@ class MetaReferenceAgentsImpl(Agents):
         return paginate_records(session_dicts, start_index, limit)
 
     async def shutdown(self) -> None:
-        pass
+        await self.persistence_store.close()
 
     # OpenAI responses
     async def get_openai_response(

llama_stack/providers/inline/batches/reference/batches.py

@@ -129,6 +129,7 @@ class ReferenceBatchesImpl(Batches):
             # don't cancel tasks - just let them stop naturally on shutdown
             # cancelling would mark batches as "cancelled" in the database
             logger.info(f"Shutdown initiated with {len(self._processing_tasks)} active batch processing tasks")
+        await self.kvstore.close()
 
     # TODO (SECURITY): this currently works w/ configured api keys, not with x-llamastack-provider-data or with user policy restrictions
     async def create_batch(

llama_stack/providers/inline/eval/meta_reference/eval.py

@@ -64,7 +64,8 @@ class MetaReferenceEvalImpl(
             benchmark = Benchmark.model_validate_json(benchmark)
             self.benchmarks[benchmark.identifier] = benchmark
 
-    async def shutdown(self) -> None: ...
+    async def shutdown(self) -> None:
+        await self.kvstore.close()
 
     async def register_benchmark(self, task_def: Benchmark) -> None:
         # Store in kvstore