[
https://issues.apache.org/jira/browse/KUDU-767?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel
]
Todd Lipcon reopened KUDU-767:
------------------------------
I don't think this is a duplicate. This bug is about the handling of scans
marked explicitly as "do not cache". In these cases, we should not be updating
the recency information in the LRU cache policy on a hit. This is separate than
the general "we should have a better policy" ticket.
> Un-cached reads should not bump blocks to front of LRU list
> -----------------------------------------------------------
>
> Key: KUDU-767
> URL: https://issues.apache.org/jira/browse/KUDU-767
> Project: Kudu
> Issue Type: Improvement
> Components: cfile, perf
> Affects Versions: Private Beta
> Reporter: Todd Lipcon
> Priority: Major
> Fix For: NA
>
>
> When we're doing an un-cached read like an MR job or compaction, we should
> probably not be moving the blocks being accessed to the front of the LRU
> list. Because we do, this can cause hotter blocks to get pushed down to be
> towards the end of the LRU after the big scan.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
