Hi, I am having this issue, where one of the nodes on my yarn cluster is behaving badly and all the containers assigned to it are extremely slow.
Now, what MR AM does when it sees one of the task attempt to be much slower and launches a speculative attempt, but that is assigned to the same slow node, which ends up making it even more slower. I am waiting for this job to finish, while I have my entire cluster lying idle. So, I am wondering if there is any setting/configuration, which will disallow assigning speculative attempts to the same node? Or should I create jira/issue for the same ? Thanks, Dharmesh
