[
https://issues.apache.org/jira/browse/SPARK-9588?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=14654044#comment-14654044
]
Cheng Lian commented on SPARK-9588:
-----------------------------------
What we did for improving partitioning in 1.5 are partitioning predicate
push-down and partition discovery acceleration, neither of which relates to
partition level cache. However, re-caching partitioned tables should be faster
because of faster partition discovery.
> spark sql cache: partition level cache eviction
> -----------------------------------------------
>
> Key: SPARK-9588
> URL: https://issues.apache.org/jira/browse/SPARK-9588
> Project: Spark
> Issue Type: Improvement
> Components: SQL
> Reporter: Shenghu Yang
>
> In spark 1.4, we can only do 'cache table <table_name>'. However, if we have
> table which will get a new partition periodically, say every 10 minutes, we
> have to do 'uncache' & then 'cache' the whole table, taking long time.
> Things would be much faster if we can do:
> (1) cache table <table_name> partition <newest_partition>
> (2) uncache table <table_name> partition <oldest_partition>
> This way we will alway have a sliding window type of cached data.
--
This message was sent by Atlassian JIRA
(v6.3.4#6332)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
