Question

我正在尝试用Kafka和Python激发结构化流媒体。要求：我需要在Spark中处理来自Kafka（采用JSON格式）的流数据（执行转换），然后将其存储在数据库中。

我有JSON格式的数据，比如 {"a": 120.56, "b": 143.6865998138807, "name": "niks", "time": "2012-12-01 00:00:09"}

我打算使用spark.readStream来读取Kafka，

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

我引用了this link作为参考，但没有得到如何解析JSON数据。我试过这个，

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

但看起来不起作用。

任何使用过Python的Spark结构化流程的人都可以指导我继续使用示例或链接吗？

使用，

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()

程序运行，但我在控制台上获取值，

+-----------------------------------+
|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"
  }
}

我在这里想念一些东西吗？

Answer 1

您可以将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))与架构一起使用：

get_json_object

或将单个字段作为字符串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"]])：

cast

和html以后根据您的需要。

使用python进行Spark结构化流式传输

1 个答案: