[jira] [Commented] (BEAM-2532) BigQueryIO source should avoid expensive JSON schema parsing for every record

["Daniel Halperin (JIRA)" <ji...@apache.org>]

    [ https://issues.apache.org/jira/browse/BEAM-2532?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=16066925#comment-16066925 ] 

Daniel Halperin commented on BEAM-2532:
---------------------------------------

Oh, that is an ugly bug:

- We use a String because we can serialize it, but we can't serialize TableSchema.
- But then we deserialize on every row (ugh, nice catch!!)

Need a Supplier there I guess, so that we deserialize once in memory. Should be a straightforward fix, but I doubt I'll get to it in the next day or so :)

> BigQueryIO source should avoid expensive JSON schema parsing for every record
> -----------------------------------------------------------------------------
>
>                 Key: BEAM-2532
>                 URL: https://issues.apache.org/jira/browse/BEAM-2532
>             Project: Beam
>          Issue Type: Improvement
>          Components: sdk-java-gcp
>    Affects Versions: 2.0.0
>            Reporter: Marian Dvorsky
>            Assignee: Stephen Sisk
>            Priority: Minor
>
> BigQueryIO source converts the schema from JSON for every input row, here:
> https://github.com/apache/beam/blob/master/sdks/java/io/google-cloud-platform/src/main/java/org/apache/beam/sdk/io/gcp/bigquery/BigQuerySourceBase.java#L159
> This is the performance bottleneck in a simple pipeline with BigQueryIO source.



--
This message was sent by Atlassian JIRA
(v6.4.14#64029)