Questions about the checkpoint_blobs table in Postgres

[kgreen24]

Hey all, few questions about the checkpoint_blobs table in Postgres:

• In general, what is this table used for?
• Is the a particular reason why these rows don't have a checkpoint_id column on them?

I'm particularly interested in ways to purge data from this database periodically, I saw the question from wassim-mechergui-shift
and associated answer in the comments section of this page https://langchain-ai.github.io/langgraph/concepts/persistence/, particularly this part:

Deleting checkpoints will prevent accurate time-travel, but your graph will run. Each checkpoint blob, write corresponds to the state slices(channels) at some point. You can keep the latest checkpoint and you'll get the graph state at that point.
You can discard all the checkpoints(+blobs and writes, by thread_id) you no longer require except the latest* if all you care about is the last historical value of the state.

I'm not sure I understand that, if I'm deleting a bunch of checkpoints, how do I know what blob rows to delete and which to keep? What rows of that table correspond to the "latest" checkpoint? Do I just delete all of the blob rows for the associated thread, even though the thread might be active? Some clarification here would be very helpful. Thanks.