在 pyspark 的 StructStreaming 中； 如何將 DataFrame 中的每一行（json 格式的字符串）轉換為多列

Question

我的 DataFrame 結構看起來像這樣

+--------------------+
|              values|
+--------------------+
|{"user_id":"00000...|
+--------------------+

然后這里的字符串結構是這樣的

{
    "user_id":"00000000002",
    "client_args":{
        "order_by":"id",
        "page":"4",
        "keyword":"Blue flowers",
        "reverse":"false"
    },
    "keyword_tokenizer":[
        "Blue",
        "flowers"
    ],
    "items":[
        "00000065678",
        "00000065707",
        "00000065713",
        "00000065741",
        "00000065753",
        "00000065816",
        "00000065875",
        "00000066172"
    ]
}

我希望這個 DataFrame 看起來像這樣

+---------------+-------------------+------------------+----------------------------+
|    user_id    |     client_args   | keyword_tokenizer|            items          |
+---------------+-------------------+------------------+----------------------------+
|00000000000001 |{"order_by":"",...}|["Blue","flowers"]|["000006578","00002458",...]|
+---------------+-------------------+------------------+----------------------------+

我的代碼看起來像這樣

lines = spark_session\
    .readStream\
    .format("socket")\
    .option("host", "127.0.0.1")\
    .option("port", 9998)\
    .load()

@f.udf("struct<user_id:string,client_args:string,keyword_tokenizer:array>")
def str_to_json(s):
    return json.loads(s)

lines.select(str_to_json(lines.values))

但這只會將它們轉換為 JSON，不能進行列拆分。 我應該怎么辦？

另外：我后來找到了這個方法來解決這個問題。 那效率低嗎？

schema = StructType([StructField("user_id",StringType()),
                     StructField("client_args", StructType([
                         StructField("order_by", StringType()),
                         StructField("page", StringType()),
                         StructField("keyword", StringType()),
                         StructField("reverse", StringType()),
                     ])),
                     StructField("keyword_tokenizer", ArrayType(StringType())),
                     StructField("items", ArrayType(StringType()))])

new_df = lines.withColumn("tmp", f.from_json(lines.values, schema))\
    .withColumn("user_id", f.col("tmp").getItem("user_id"))\
    .withColumn("client_args", f.col("tmp").getItem("client_args"))\
    .withColumn("keyword_tokenizer", f.col("tmp").getItem("keyword_tokenizer"))\
    .withColumn("items", f.col("tmp").getItem("items"))\
    .drop("value", "tmp")

Answer 1

使用 pyspark 讀取為 json 文件

df = spark.read.json("test.json") 
df.show()

+--------------------+--------------------+-----------------+-----------+
|         client_args|               items|keyword_tokenizer|    user_id|
+--------------------+--------------------+-----------------+-----------+
|[Blue flowers, id...|[00000065678, 000...|  [Blue, flowers]|00000000002|
+--------------------+--------------------+-----------------+-----------+