[ https://issues.apache.org/jira/browse/SPARK-16630?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16386789#comment-16386789 ]
Attila Zsolt Piros commented on SPARK-16630: -------------------------------------------- I have checked the existing sources and I would like to open a discussion about the possible solution. As I have seen YarnAllocator#processCompletedContainers could be extended to track the number of failures by host. Also YarnAllocator is responsible to update the task-level backlisted nodes with YARN (calling AMRMClient#updateBlacklist). So a relatively easy solution would be to have a separate counter here (which is independent from task level failures) with its own configured limit and updating YARN with the union of task-level backlisted nodes and "allocator-level" backlisted nodes. What is your opinion? > Blacklist a node if executors won't launch on it. > ------------------------------------------------- > > Key: SPARK-16630 > URL: https://issues.apache.org/jira/browse/SPARK-16630 > Project: Spark > Issue Type: Improvement > Components: YARN > Affects Versions: 1.6.2 > Reporter: Thomas Graves > Priority: Major > > On YARN, its possible that a node is messed or misconfigured such that a > container won't launch on it. For instance if the Spark external shuffle > handler didn't get loaded on it , maybe its just some other hardware issue or > hadoop configuration issue. > It would be nice we could recognize this happening and stop trying to launch > executors on it since that could end up causing us to hit our max number of > executor failures and then kill the job. -- This message was sent by Atlassian JIRA (v7.6.3#76005) --------------------------------------------------------------------- To unsubscribe, e-mail: issues-unsubscr...@spark.apache.org For additional commands, e-mail: issues-h...@spark.apache.org