Prove expiry

Phases 1 through 3 on the previous page showed TTL configuration, a first chat turn, and a successful resume while the checkpoint was still alive. This page runs the full demo without --skip-wait to prove what happens after the window closes.

Run the full lifecycle

1. From the repository root, run the demo without --skip-wait:

   uv run python cookbooks/expiring-chat-sessions/demo.py

   Phases 1 through 3 match the --skip-wait run: configure TTL, start the session, resume once (four messages total). Phase 4 then waits approximately 70 seconds (one-minute TTL plus a small buffer) so Aerospike can reclaim the records. Phase 5 checks that the checkpoint is gone and invokes the same thread_id again.

2. Review Phases 4 and 5 in the output:

   Watch two lines: get_tuple() is the LangGraph API for reading a checkpoint by ID. Returning None confirms the record is gone from Aerospike. messages stored : 2 confirms LangGraph started a fresh session with no prior history.

   ================================================================
   Phase 4 - Wait 70s for the TTL to elapse
   ================================================================
     expiry window elapsed.
   
   ================================================================
   Phase 5 - Prove the state expired
   ================================================================
     get_tuple()      : None
     raw record       : gone
     user      > Hello again?
     assistant > Hello again! This looks like a brand-new session to me.
     messages stored  : 2  (fresh session -- no prior history)
   
   ================================================================
   Result
   ================================================================
     Aerospike expired the abandoned session automatically.
     No cron job, no sweeper, no DELETE query.

3. Confirm expiry reclaimed the prior session:

   By the end of Phase 3 the thread held four messages under session-demo-001. Phase 4 waited past the one-minute TTL plus a small buffer without any application code touching the records. Aerospike deleted the checkpoint natively.

   Phase 5 called get_tuple() on that thread_id to confirm the checkpoint was gone, then invoked the graph with the same ID. With nothing left in Aerospike, LangGraph started a fresh two-message session with no prior history.

   • get_tuple() returns None after the wait, confirming Aerospike removed the checkpoint.
   • messages stored drops from 4 back to 2 on the fresh invoke.
   • The final assistant line is the third scripted reply (“brand-new session”). It only appears when prior checkpoints are gone.

   Aerospike handled expiry automatically. No cron job, sweeper process, or explicit DELETE query was required. That means fewer scripts to maintain and test across environments.

Note

The demo waits 70 seconds (one-minute TTL plus a small buffer). Message count is the primary invariant if TTL timing differs slightly.

You configured AerospikeSaver with a native TTL, resumed a thread from checkpoints, and verified Aerospike reclaims expired sessions without a cleanup job.

Stop Aerospike

1. Stop and remove the Aerospike container:

   docker rm -f aerospike

Explore the cookbook source at cookbooks/expiring-chat-sessions/ in the langgraph-aerospike repository for the full step-by-step code walkthrough.

Checklist

• I’ve run the full demo including the TTL wait.
• I’ve confirmed the post-expiry session has only two messages.
• I’ve stopped the Aerospike Docker container.

Share

Congratulations on completing this tutorial! Share your achievement with the world and let everyone know about your newly acquired skills.

Next Steps

• Learn how to fork from a saved checkpoint in the Tutorial: Agent path correction with LangGraph.

• See the langgraph-checkpoint-aerospike package on GitHub for API reference.

• Read the LangGraph persistence documentation for checkpoint concepts.