You are viewing a plain text version of this content. The canonical link for it is here.
Posted to issues@spark.apache.org by "Sean Owen (JIRA)" <ji...@apache.org> on 2017/03/03 10:25:45 UTC

[jira] [Commented] (SPARK-19808) About the default blocking arg in unpersist

    [ https://issues.apache.org/jira/browse/SPARK-19808?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=15894087#comment-15894087 ] 

Sean Owen commented on SPARK-19808:
-----------------------------------

(Maybe you can rewrite this as a proposed change rather than question?)

They should be consistent, but I don't think they're worth changing now because it's a behavior change for little gain. Consider also the destroy() and unpersist() operations for broadcasts.

However I have never been sure why an application would want to block waiting on an unpersist operation. For that reason, I think most calls in Spark are blocking=false and I'd personally support making this consistent. That is, unless someone highlights why this sometimes isn't a good idea?


> About the default blocking arg in unpersist
> -------------------------------------------
>
>                 Key: SPARK-19808
>                 URL: https://issues.apache.org/jira/browse/SPARK-19808
>             Project: Spark
>          Issue Type: Question
>          Components: ML, Spark Core
>    Affects Versions: 2.1.0
>            Reporter: zhengruifeng
>            Priority: Minor
>
> Now, {{unpersist}} are commonly used with default value in ML.
> Most algorithms like {{KMeans}} use {{RDD.unpersisit}} and the default {{blocking}} is {{true}}
> And for meta algorithms like {{OneVsRest}}, {{CrossValidator}} use {{Dataset.unpersist}} and the default {{blocking}} is {{false}}
> Should the default value for {{RDD.unpersisit}} and {{Dataset.unpersist}} be consistent?
> And all the {{blocking}} arg in ML should be set {{false}}?
> [~srowen] [~mlnick] [~yanboliang]



--
This message was sent by Atlassian JIRA
(v6.3.15#6346)

---------------------------------------------------------------------
To unsubscribe, e-mail: issues-unsubscribe@spark.apache.org
For additional commands, e-mail: issues-help@spark.apache.org