[jira] [Assigned] (IMPALA-10391) LIRS cache edge case where there is exactly one unprotected element

["Joe McDonnell (Jira)" <ji...@apache.org>]

     [ https://issues.apache.org/jira/browse/IMPALA-10391?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]

Joe McDonnell reassigned IMPALA-10391:
--------------------------------------

    Assignee: Joe McDonnell

> LIRS cache edge case where there is exactly one unprotected element
> -------------------------------------------------------------------
>
>                 Key: IMPALA-10391
>                 URL: https://issues.apache.org/jira/browse/IMPALA-10391
>             Project: IMPALA
>          Issue Type: Bug
>          Components: Backend
>    Affects Versions: Impala 4.0
>            Reporter: Joe McDonnell
>            Assignee: Joe McDonnell
>            Priority: Major
>
> The fix for IMPALA-10127 introduced an issue in this code:
> {noformat}
>           // If an UNPROTECTED entry is not on the recency list, then its new reuse
>           // distance is longer than any of the PROTECTED entries. So, it remains
>           // an UNPROTECTED entry, but it should be added back to the recency list and
>           // readded as the most recent entry on the unprotected list.
>           MoveToRecencyListBack(&tstate, e, false);
>           RemoveFromUnprotectedList(e);
>           AddToUnprotectedList(e);{noformat}
> RemoveFromUnprotectedList would detect that this is the last unprotected entry and clear unprotected_list_front_. This doesn't decrement num_unprotected_ to 0, so when AddToUnprotectedList is called, it doesn't set unprotected_list_front_ when it adds the element back.
> This only applies when there is exactly one entry on the unprotected list. This can be fixed by either decrementing/incrementing the counters to match up with other locations or by skipping this code when num_unprotected_ == 1.



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-all-unsubscribe@impala.apache.org
For additional commands, e-mail: issues-all-help@impala.apache.org