Applied AI
A cancelled LangGraph dev run may still finish its work
Test late cancellation in LangGraph's in-memory dev runtime and bound side effects inside long-running nodes.
November 18, 2026
An "interrupted" status can be reassuring while a long-running node is still active. An open LangGraph issue reports that, in the langgraph dev in-memory runtime, cancellation after roughly 240 seconds can report interruption while execution later reaches success. This is a narrow report about that runtime. It does not establish the same behavior for hosted deployments.
Test the late-cancel path
Build a node that waits longer than four minutes and writes one harmless marker at the end. Start a run, cancel it before and after the reported threshold, and compare the run status with the marker. Test the exact version and runtime used by the team. A successful early-cancel test does not cover the late path.
Put a deadline around side effects
For expensive API calls, messages, or writes, pass a cancellation or deadline signal into the node where possible. Check it before committing a side effect. Make repeated execution safe through idempotency keys or a durable operation record. A cancellation acknowledgement from the orchestrator is not the same as proof that external work stopped.
Distinguish status from outcome
Log both the requested cancellation time and the last external effect. If they disagree, treat the run as a reconciliation case. A job that sent a customer message after cancellation needs a different response from one that only burned extra compute. Keep the development runtime finding scoped; validate the production runtime independently.
Adovo’s view
Cancellation is an operational promise only when the side effect obeys it. The status label is a signal; the external outcome is the fact to reconcile.
