[
https://issues.apache.org/jira/browse/HIVE-22583?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16993009#comment-16993009
]
Slim Bouguerra commented on HIVE-22583:
---------------------------------------
The fix looks good to me.
I try to avoid using TEZ counters as a part of the test validation since they
can change quite drastically if the plan changes but i understand that it is
the only way for now to test the data in the cache.
One of the idea i want to implement is to have a query flag that tells LLAP to
run a query against the cache content only and that can enable such tests and
will avoid to use the the TEZ counters.
> LLAP cache always misses with non-vectorized serde readers such as OpenCSV
> --------------------------------------------------------------------------
>
> Key: HIVE-22583
> URL: https://issues.apache.org/jira/browse/HIVE-22583
> Project: Hive
> Issue Type: Bug
> Components: llap
> Reporter: Ádám Szita
> Assignee: Ádám Szita
> Priority: Major
> Attachments: HIVE-22583.0.patch, HIVE-22583.1.patch,
> HIVE-22583.2.patch
>
>
> Although after the first read LLAP cache stores data of tables that are not
> using the LazySimple serde, the stored data is then never used in the future
> subsequent queries, causing a full cache miss and re-read each time.
> Problem is rooted in SerdeEncodedDataReader#cacheFileData is not taking care
> of creating an entry for the root/struct column of the table. The only cases
> this is taken care of are when a vectorized reader is used _(e.g.
> LazySimpleSerde's LazySimpleDeserializeRead)_, where
> SerdeEncodedDataReader#processAsyncCacheData takes care of this.
> This can be reproduced by either using a custom serde, like OpenCSV or using
> LazySimpleSerde, but turning off _hive.llap.io.encode.vector.serde.enabled_.
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
