[
https://issues.apache.org/jira/browse/HIVE-10063?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14377146#comment-14377146
]
Sergey Shelukhin commented on HIVE-10063:
-----------------------------------------
this is mostly by design, because we expect to cache single exact RGs and look
up /starting with/ exact RG offset (not end though, unfortunately). If we
cached e.g. entire file in big block that will be a problem.
Added a large comment to describe API behavior for partial matches, and
optional ability to match partial start blocks by looking at floorKey (not used
by default except in unit test)
> LLAP: LowLevelCacheImpl does not match cached ranges properly
> -------------------------------------------------------------
>
> Key: HIVE-10063
> URL: https://issues.apache.org/jira/browse/HIVE-10063
> Project: Hive
> Issue Type: Sub-task
> Reporter: Prasanth Jayachandran
> Assignee: Sergey Shelukhin
> Fix For: llap
>
>
> LowLevelCacheImpl uses skip list to map start disk range offset to disk
> ranges. Find for match in the skip list uses cache.subMap() to identify
> correct disk range. This will not work if the requested disk range offset
> does not exactly match the cache disk range offset.
> For example: If the cache disk range is 0 - 1000 bytes and if the requested
> disk range is 500 - 1000 there will not be cache hit.
> We should probably use floorEntry() to get the greatest entry less than or
> equal to provided key.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
