Getting Started & Migration
If a rolled-back durable agent has left partially completed steps, what manual cleanup approaches are safe to use without corrupting future durable execution state?
Use Catalyst’s built-in graceful shutdown combined with a compensating action pattern to revert completed side effects before tearing down state. For orphaned state, delete the execution’s history from the state store using the platform’s management API, ensuring no in-flight references remain. Manual database alterations are risky; rely on the API to avoid dangling pointers. Verify that any compensation itself is idempotent to prevent over-correction during future re-runs.
Was this article helpful?
Your feedback helps improve Diagrid's FAQ experience.
Keep reading
More Diagrid FAQ articles
- Getting Started & Migration
What does a minimal but realistic durable agent look like when I first run Catalyst locally?
A realistic durable agent prototype that progresses from a call to an LLM through a conditional action to an API, illustrating automatic state persistence.
- Getting Started & Migration
How do I iterate quickly on durable agent logic without deploying to a shared cluster?
Use a local Dapr sidecar to execute and debug agent workflows, enabling a fast edit-run-debug cycle without remote dependencies.
- Getting Started & Migration
My agent is already built with LangChain/LlamaIndex. Can I add durable execution without a rewrite?
Wrap an existing agent’s entry point in a Catalyst workflow to gain durability without modifying framework internals, then progressively refactor.