You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@hive.apache.org by "Sergey Shelukhin (JIRA)" <ji...@apache.org> on 2018/03/01 23:01:00 UTC

[jira] [Created] (HIVE-18837) add a flag and disable some object pools in LLAP until further testing

Sergey Shelukhin created HIVE-18837:
---------------------------------------

             Summary: add a flag and disable some object pools in LLAP until further testing
                 Key: HIVE-18837
                 URL: https://issues.apache.org/jira/browse/HIVE-18837
             Project: Hive
          Issue Type: Bug
            Reporter: Sergey Shelukhin
            Assignee: Sergey Shelukhin


There appears to be some subtle concurrency issue in FixedSizedObjectPool that happens with multiple consumers where some object may be retrieved twice. 
Unfortunately running a load test for hour(s) does not trigger it for me and overall it happens extremely rarely on non-specific tests; adding debug info at this level is a little bit difficult to determine how it could have happened and interlocked operations in the trace may actually eliminate the issue. I suspect it has something to do with aggressive assumptions made for locking and array elements and the memory model. Maybe that can be simplified without much perf loss.
Anyway, for now we will disable the pools where multiple consumers use them.
Need to test perf to see if these two pools even matter; if so, we can simplify the model as per above or debug the issue in some way.



--
This message was sent by Atlassian JIRA
(v7.6.3#76005)