beam-commits mailing list archives

Site index · List index
Message view « Date » · « Thread »
Top « Date » · « Thread »
From "Amit Sela (JIRA)" <>
Subject [jira] [Updated] (BEAM-848) A better shuffle after reading from within mapWithState.
Date Fri, 24 Feb 2017 22:24:44 GMT


Amit Sela updated BEAM-848:
    Issue Type: Improvement  (was: Bug)

> A better shuffle after reading from within mapWithState.
> --------------------------------------------------------
>                 Key: BEAM-848
>                 URL:
>             Project: Beam
>          Issue Type: Improvement
>          Components: runner-spark
>            Reporter: Amit Sela
>            Assignee: Amit Sela
> The SparkRunner uses {{mapWithState}} to read and manage CheckpointMarks, and this stateful
operation will be followed by a shuffle: 
> Since the stateful read maps "splitSource" -> "partition of a list of read values",
the following shuffle won't benefit in any way (the list of read values has not been flatMapped
yet). In order to avoid shuffle we need to set the input RDD ({{SourceRDD.Unbounded}}) partitioner
to be a default {{HashPartitioner}} since {{mapWithState}} would use the same partitioner
and will skip shuffle if the partitioners match.
> It would be wise to shuffle the read values _after_ flatmap.
> I will break this into two tasks:
> # Set default-partitioner to the input RDD.
> # Shuffle (using Coders) the input.

This message was sent by Atlassian JIRA

View raw message