python sdk WriteToBigQuery excessive usage of metered API

[kennknowles]

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

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.

  "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"
    }
  ],

Imported from Jira BEAM-6831. Original Jira may contain additional context.
Reported by: pesach.