Jacek Woźniczak - LangGraph HANA Checkpointer in a CAP Project

Thanks to David Diaz's LangGraph HANA Checkpointer, SAP HANA can be used as a checkpointer for thread persistence. This post covers one missing point I had when using it with CAP projects.

The schema

By default, when you provide a database connection, the checkpoint tables are created in the schema associated with your user:

conn = dbapi.connect(address="...", port=443, user="...", password="...")
checkpointer = HANASaver(conn=conn)
checkpointer.setup()

If you take the connection details from the binding - the user and password - the tables will be created outside the CAP project schema.

Taking the user from the HANA binding:

...
"hana": [
  {
    "label": "hana",
    "name": "my-db",
    "tags": [
      "hana",
      "database",
      "relational"
    ],
    ...
    "credentials": {
      "database_id": "...",
      "host": "....hanacloud.ondemand.com",
      "port": "443",
      "driver": "com.sap.db.jdbc.Driver",
      "schema": "A66534E96NNNNNNNNNA245F35C6079C",
      "user": "A66NNNNNNNNNNNNNNNNNNNNNNNN_NNNNNNNNNNNNNNNNNNNNSPN606_RT",
      "password": "}(NZ~...."
      ...

The checkpointer tables will live in the A66NNNNNNNNNNNNNNNNNNNNNNNN_NNNNNNNNNNNNNNNNNNNNSPN606_RT schema, which is outside the A66534E96NNNNNNNNNA245F35C6079C CAP project HDI container. So when you connect via DB Explorer to the container using the quick path: DB Explorer -> log in to Cloud Foundry -> add your container, you won't see the tables.

The goal

My requirement was to keep the tables inside the CAP project schema. However, the RT/DT users from the binding do not have the CREATE privilege required to run the DDL statements used by the checkpointer setup method.

Here's the approach I'm currently evaluating:

Get the connection details from the binding. All details are extracted from the VCAP_SERVICES hana entry.

checkpointer = HANASaver.from_binding_with_db_deployer()

Instead of calling the setup method to execute the DDL, a dedicated .cds file has to be added under db. It contains the required tables, which will be created by the HDI deployer:

This way, the tables are created under the CAP project schema.

Checkpointer code: https://github.com/wozjac/langgraph-checkpoint-hana/blob/main/langgraph_checkpoint_hana/saver.py