[
https://issues.apache.org/jira/browse/HIVE-17669?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16190622#comment-16190622
]
Mithun Radhakrishnan commented on HIVE-17669:
---------------------------------------------
bq. weight based eviction could be a better approach (weight can be length of
string).
Ah, that's an interesting suggestion. Shouldn't we also consider the cost of
deserializing the sarg-string? On the one hand, perhaps the longer sarg-strings
take longer to deserialize, and might benefit from caching. But on the other,
they might dominate the cache. :/ I'll have to think this through.
Any recommendation on the value for {{CacheBuilder.maximumWeight()}}? :]
> Cache to optimize SearchArgument deserialization
> ------------------------------------------------
>
> Key: HIVE-17669
> URL: https://issues.apache.org/jira/browse/HIVE-17669
> Project: Hive
> Issue Type: Improvement
> Components: ORC, Query Processor
> Affects Versions: 2.2.0, 3.0.0
> Reporter: Mithun Radhakrishnan
> Assignee: Mithun Radhakrishnan
> Attachments: HIVE-17699.1.patch, HIVE-17699.2.patch
>
>
> And another, from [~selinazh] and [~cdrome]. (YHIVE-927)
> When a mapper needs to process multiple ORC files, it might land up having
> use essentially the same {{SearchArgument}} over several files. It would be
> good not to have to deserialize from string, over and over again. Caching the
> object against the string-form should speed things up.
--
This message was sent by Atlassian JIRA
(v6.4.14#64029)
