[
https://issues.apache.org/jira/browse/BEAM-6831?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17050880#comment-17050880
]
Keiji Yoshida commented on BEAM-6831:
-
That is because the bigquery.tables.get API is called every time a bundle in a
PCollection is processed in Apache Beam 2.10.0
([code|https://github.com/apache/beam/blob/v2.10.0/sdks/python/apache_beam/io/gcp/bigquery.py#L1365-L1367]).
In the latest version of Apache Beam (2.19.0), the bigquery.tables.get API is
not called as long as `create_disposition` is set to `CREATE_NEVER`
([code|https://github.com/apache/beam/blob/v2.19.0/sdks/python/apache_beam/io/gcp/bigquery.py#L989-L1009]).
So, you can avoid the rate limit error by using Apache Beam 2.19.0 and setting
`create_disposition` to `CREATE_NEVER`.
> python sdk WriteToBigQuery excessive usage of metered API
> -
>
> Key: BEAM-6831
> URL: https://issues.apache.org/jira/browse/BEAM-6831
> Project: Beam
> Issue Type: Bug
> Components: sdk-py-core
>Affects Versions: 2.10.0
>Reporter: Pesach Weinstock
>Assignee: Pablo Estrada
>Priority: Major
> Labels: bigquery, dataflow, gcp, python
> Attachments: apache-beam-py-sdk-gcp-bq-api-issue.png
>
>
> Right now, there is a potential issue with the python sdk where
> {{beam.io.gcp.bigquery.WriteToBigQuery}} calls the following api more often
> than needed:
> [https://www.googleapis.com/bigquery/v2/projects//datasets//tables/?alt=json|https://www.googleapis.com/bigquery/v2/projects/%3Cproject-name%3E/datasets/%3Cdataset-name%3E/tables/%3Ctable-name%3E?alt=json]
> The above request falls under specific bigquery API quotas which are excluded
> from bigquery streaming inserts. When used in a streaming pipeline, we hit
> this quota pretty quickly, and cannot proceed to write any further data to
> bigquery.
> Dispositions being used are:
> * create_disposition: {{beam.io.BigQueryDisposition.CREATE_NEVER}}
> * write_disposition: {{beam.io.BigQueryDisposition.WRITE_APPEND}}
> This is currently blocking us from using bigqueryIO in a streaming pipeline
> to write to bigquery, and required us to formally request an API quota
> increase from Google to temporarily correct the situation.
> Our pipeline uses DataflowRunner. Error seen is below, and in attached
> screenshot of stackdriver trace.
> {code:java}
> "errors": [
> {
> "message": "Exceeded rate limits: too many api requests per user per
> method for this user_method. For more information, see
> https://cloud.google.com/bigquery/troubleshooting-errors";,
> "domain": "usageLimits",
> "reason": "rateLimitExceeded"
> }
> ],
> {code}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
