[ 
https://issues.apache.org/jira/browse/BEAM-9010?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17001196#comment-17001196
 ] 

Reuven Lax commented on BEAM-9010:
----------------------------------

This change could be a major performance regression for production users, as 
quotas I believe are based on the full JSON size. This could cause pipelines 
that used to work to suddenly stop working. We might have to hold off upgrading 
the library until this can be resolved.

> BigQuery TableRow's size is toString().length() ?
> -------------------------------------------------
>
>                 Key: BEAM-9010
>                 URL: https://issues.apache.org/jira/browse/BEAM-9010
>             Project: Beam
>          Issue Type: Improvement
>          Components: runner-dataflow
>            Reporter: Tomo Suzuki
>            Priority: Minor
>
> The following tests failed when I tried to upgrade google-http-client 1.34.0 
> from 1.28.0:
> {noformat}
> org.apache.beam.sdk.io.gcp.bigquery.BigQueryIOReadTest.testEstimatedSizeWithoutStreamingBuffer
> org.apache.beam.sdk.io.gcp.bigquery.BigQueryIOReadTest.testEstimatedSizeWithStreamingBuffer
> org.apache.beam.sdk.io.gcp.bigquery.BigQueryUtilTest.testInsertAll
> {noformat}
> [https://builds.apache.org/job/beam_PreCommit_Java_Commit/9288/#showFailuresLink]
> h3. Reason of the test failures
> [org.apache.beam.sdk.io.gcp.testing.TableContainer|https://github.com/apache/beam/blob/6fa94c9/sdks/java/io/google-cloud-platform/src/main/java/org/apache/beam/sdk/io/gcp/testing/TableContainer.java#L43]
>  and 
> [org.apache.beam.sdk.io.gcp.bigquery.BigQueryServicesImpl|https://github.com/apache/beam/blob/c2f0d28/sdks/java/io/google-cloud-platform/src/main/java/org/apache/beam/sdk/io/gcp/bigquery/BigQueryServicesImpl.java#L758]
>  rely on {{TableRow.toString().length()}} to calculate the size. Example:
> {code:java}
>           dataSize += row.toString().length();
>           if (dataSize >= maxRowBatchSize
>               || rows.size() >= maxRowsPerBatch
>               || i == rowsToPublish.size() - 1) {
> {code}
> However, with [google-http-client's 
> PR#589|https://github.com/googleapis/google-http-java-client/pull/589/files#diff-914cd7ff18143b3d2398149e1cfb4f45R218],
>  the GenericData.toString output has changed since v1.29.0.
> In old google-http-client 1.28.0, an example row's toString returned:
> {noformat}
> {f=[{v=foo}, {v=1234}]}
> {noformat}
> In new google-http-client 1.29.0 and higher, the same row's toString returns:
> {noformat}
> GenericData{classInfo=[f], {f=[GenericData{classInfo=[v], {v=foo}}, 
> GenericData{classInfo=[v], {v=1234}}]}}
> {noformat}
> h1. Question:
> Is this right thing to rely on {{toString().length()}} in the BigQuery 
> classes?



--
This message was sent by Atlassian Jira
(v8.3.4#803005)

Reply via email to