BigQuery connector for pyspark via Hadoop Input Format example

Question

I have a large dataset stored into a BigQuery table and I would like to load it into a pypark RDD for ETL data processing.

I realized that BigQuery supports the Hadoop Input / Output format

https://cloud.google.com/hadoop/writing-with-bigquery-connector

and pyspark should be able to use this interface in order to create an RDD by using the method "newAPIHadoopRDD".

http://spark.apache.org/docs/latest/api/python/pyspark.html

Unfortunately, the documentation on both ends seems scarce and goes beyond my knowledge of Hadoop/Spark/BigQuery. Is there anybody who has figured out how to do this?

Answer 1

Google now has an example on how to use the BigQuery connector with Spark.

There does seem to be a problem using the GsonBigQueryInputFormat, but I got a simple Shakespeare word counting example working

import json
import pyspark
sc = pyspark.SparkContext()

hadoopConf=sc._jsc.hadoopConfiguration()
hadoopConf.get("fs.gs.system.bucket")

conf = {"mapred.bq.project.id": "<project_id>", "mapred.bq.gcs.bucket": "<bucket>", "mapred.bq.input.project.id": "publicdata", "mapred.bq.input.dataset.id":"samples", "mapred.bq.input.table.id": "shakespeare"  }

tableData = sc.newAPIHadoopRDD("com.google.cloud.hadoop.io.bigquery.JsonTextBigQueryInputFormat", "org.apache.hadoop.io.LongWritable", "com.google.gson.JsonObject", conf=conf).map(lambda k: json.loads(k[1])).map(lambda x: (x["word"], int(x["word_count"]))).reduceByKey(lambda x,y: x+y)
print tableData.take(10)