[
https://issues.apache.org/jira/browse/HIVE-16778?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16027079#comment-16027079
]
Sergey Shelukhin commented on HIVE-16778:
-----------------------------------------
[~prasanth_j] can you take a look? https://reviews.apache.org/r/59615/
This improves error handling for refcounts, and also removes the per-column RG
arrays (that are the legacy of high-level cache).
> LLAP IO: better refcount management
> -----------------------------------
>
> Key: HIVE-16778
> URL: https://issues.apache.org/jira/browse/HIVE-16778
> Project: Hive
> Issue Type: Bug
> Reporter: Sergey Shelukhin
> Assignee: Sergey Shelukhin
> Attachments: HIVE-16778.patch
>
>
> Looks like task cancellation can close the UGI, causing the background thread
> to die with an exception, leaving a bunch of unreleased cache buffers.
> Overall, it's probably better to modify how refcounts are handled - if
> there's some bug in the code we don't want to leak them.
--
This message was sent by Atlassian JIRA
(v6.3.15#6346)
