Postmortem of Spark Executor's death

A question I kept coming back to while comparing Spark with BigQuery was this: If Spark executors write shuffle data to disk, and that disk still exists, why can’t other executors read that data when one executor dies? At first glance, it feels lik...