RAG Interview Questions #11 - The Zombie Hub Trap

How soft-deleting vectors silently inflates latency and collapses recall, and why tracking your tombstone ratio as a first-class SLO is the only way to survive the inevitable memory spike.

You’re in a Senior AI Engineer interview at Meta and the interviewer asks:

“Your HNSW index takes a delete every few seconds in production. Do you remove the node and repair the graph?”

Don’t say: “Yes, delete the node, rewire its neighbors, keep the graph clean.”

That answer ends the interview. You just proposed a write lock on a live index, every few sec…