We have noticed some lost data for a Spark application running on EMR using spot instances. Spark being resilient, any terminated spot instances should not result in any lost data per our understanding. I have 2 questions -
(1) Is this expected or we are missing something? (2) Does AWS reuse the EBS volume for interrupted instances and attach the same volume to the newly allocated instance for the cluster? In other words, is there any difference in AWS interrupted instances eventually terminated for the cluster and us manually terminating instances from the cluster? Thanks.