How to read Avro file in PySpark

Question

I am writing a spark job using python. However, I need to read in a whole bunch of avro files.

This is the closest solution that I have found in Spark's example folder. However, you need to submit this python script using spark-submit. In the command line of spark-submit, you can specify the driver-class, in that case, all your avrokey, avrovalue class will be located.

avro_rdd = sc.newAPIHadoopFile(
        path,
        "org.apache.avro.mapreduce.AvroKeyInputFormat",
        "org.apache.avro.mapred.AvroKey",
        "org.apache.hadoop.io.NullWritable",
        keyConverter="org.apache.spark.examples.pythonconverters.AvroWrapperToJavaConverter",
        conf=conf)

In my case, I need to run everything within the Python script, I have tried to create an environment variable to include the jar file, finger cross Python will add the jar to the path but clearly it is not, it is giving me unexpected class error.

os.environ['SPARK_SUBMIT_CLASSPATH'] = "/opt/cloudera/parcels/CDH-5.1.0-1.cdh5.1.0.p0.53/lib/spark/examples/lib/spark-examples_2.10-1.0.0-cdh5.1.0.jar"

Can anyone help me how to read avro file in one python script?

Answer 1

Spark >= 2.4.0

You can use built-in Avro support. The API is backwards compatible with the spark-avro package, with a few additions (most notably from_avro / to_avro function).

Please note that module is not bundled with standard Spark binaries and has to be included using spark.jars.packages or equivalent mechanism.

See also Pyspark 2.4.0, read avro from kafka with read stream - Python

Spark < 2.4.0

You can use spark-avro library. First lets create an example dataset:

import avro.schema
from avro.datafile import DataFileReader, DataFileWriter

schema_string ='''{"namespace": "example.avro",
 "type": "record",
 "name": "KeyValue",
 "fields": [
     {"name": "key", "type": "string"},
     {"name": "value",  "type": ["int", "null"]}
 ]
}'''

schema = avro.schema.parse(schema_string)

with open("kv.avro", "w") as f, DataFileWriter(f, DatumWriter(), schema) as wrt:
    wrt.append({"key": "foo", "value": -1})
    wrt.append({"key": "bar", "value": 1})

Reading it using spark-csv is as simple as this:

df = sqlContext.read.format("com.databricks.spark.avro").load("kv.avro")
df.show()

## +---+-----+
## |key|value|
## +---+-----+
## |foo|   -1|
## |bar|    1|
## +---+-----+