CODEWITHSUNDEEP
pythonapache-sparkpysparkapache-spark-sqlspark-structured-streaming
user3150037
user3150037

Reputation: 147

Spark structured streaming with python

I am trying to Spark structured streaming with Kafka and Python. Requirement: I need to process streaming data from Kafka (in JSON format) in Spark (perform transformations) and then store it in a database.

I have data in JSON format like, {"a": 120.56, "b": 143.6865998138807, "name": "niks", "time": "2012-12-01 00:00:09"}

I am planning to use spark.readStream for reading from Kafka like,

data = spark.readStream.format("kafka").option("kafka.bootstrap.servers", "localhost:9092").option("subscribe","test").load()

I referred this link for reference but didn't get how to parse JSON data. I tried this, 

data = data.selectExpr("CAST(a AS FLOAT)","CAST(b as FLOAT)", "CAST(name as STRING)", "CAST(time as STRING)").as[(Float, Float, String, String)]

but looks like it doesn't work.

Can anyone who has worked on Spark structured streaming with Python guide me to proceed with sample examples or links?

Using, 

schema = StructType([
    StructField("a", DoubleType()),
    StructField("b", DoubleType()),
    StructField("name", StringType()),
    StructField("time", TimestampType())])

inData = spark.readStream.format("kafka").option("kafka.bootstrap.servers", "localhost:9092").option("subscribe","test").load()
data = inData.select(from_json(col("value").cast("string"), schema))
query = data.writeStream.outputMode("Append").format("console").start()

Program runs but I am getting values on console as, 

+-----------------------------------+
|jsontostruct(CAST(value AS STRING))|
+-----------------------------------+
|               [null,null,null,2...|
|               [null,null,null,2...|
+-----------------------------------+

17/04/07 19:23:15 INFO StreamExecution: Streaming query made progress: {
  "id" : "8e2355cb-0fd3-4233-89d8-34a855256b1e",
  "runId" : "9fc462e0-385a-4b05-97ed-8093dc6ef37b",
  "name" : null,
  "timestamp" : "2017-04-07T19:23:15.013Z",
  "numInputRows" : 2,
  "inputRowsPerSecond" : 125.0,
  "processedRowsPerSecond" : 12.269938650306749,
  "durationMs" : {
    "addBatch" : 112,
    "getBatch" : 8,
    "getOffset" : 2,
    "queryPlanning" : 4,
    "triggerExecution" : 163,
    "walCommit" : 26
  },
  "eventTime" : {
    "watermark" : "1970-01-01T00:00:00.000Z"
  },
  "stateOperators" : [ ],
  "sources" : [ {
    "description" : "KafkaSource[Subscribe[test]]",
    "startOffset" : {
      "test" : {
        "0" : 366
      }
    },
    "endOffset" : {
      "test" : {
        "0" : 368
      }
    },
    "numInputRows" : 2,
    "inputRowsPerSecond" : 125.0,
    "processedRowsPerSecond" : 12.269938650306749
  } ],
  "sink" : {
    "description" : "org.apache.spark.sql.execution.streaming.ConsoleSink@6aa91aa2"
  }
}

Did I miss something here.

Upvotes: 1

Views: 3938

Answers (1)

zero323
zero323

Reputation: 330093

You can either use from_json with schema:

from pyspark.sql.functions import col, from_json
from pyspark.sql.types import *

schema = StructType([
    StructField("a", DoubleType()),
    StructField("b", DoubleType()), 
    StructField("name", StringType()), 
    StructField("time", TimestampType())])

data.select(from_json(col("value").cast("string"), schema))

or get individual fields as strings with get_json_object:

from pyspark.sql.functions import get_json_object

data.select([
    get_json_object(col("value").cast("string"), "$.{}".format(c)).alias(c)
    for c in ["a", "b", "name", "time"]])

and cast them later according to your needs.

Upvotes: 1

Related Questions