理想地,以下代码片段可以正常工作:
import kudu
from kudu.client import Partitioning
df = … #some spark dataframe
# Connect to Kudu master server
client = kudu.connect(host=‘…‘, port=7051)
# infer schema from spark dataframe
schema = df.schema
# Define partitioning schema
partitioning = Partitioning().add_hash_partitions(column_names=['key'], num_buckets=3)
# Create new table
client.create_table('dev.some_example', schema, partitioning)
但是client.create_table需要一个kudu.schema.Schema而不是数据帧中的结构。但是,在Scala中,您可以执行此操作(来自https://kudu.apache.org/docs/developing.html):
kuduContext.createTable(
"dev.some_example", df.schema, Seq("key"),
new CreateTableOptions()
.setNumReplicas(1)
.addHashPartitions(List("key").asJava, 3))
现在我想知道是否可以在不使用kudu模式生成器手动定义每个列的情况下使用PySpark进行相同的操作?
答案 0 :(得分:0)
因此,我为自己编写了一个帮助程序函数,用于将PySpark Dataframe模式转换为kudu.schema.Schema,希望对您有所帮助。反馈表示赞赏!
旁注,您可能想添加或编辑数据类型映射。
import kudu
from kudu.client import Partitioning
def convert_to_kudu_schema(df_schema, primary_keys):
builder = kudu.schema.SchemaBuilder()
data_type_map = {
"StringType":kudu.string,
"LongType":kudu.int64,
"IntegerType":kudu.int32,
"FloatType":kudu.float,
"DoubleType":kudu.double,
"BooleanType":kudu.bool,
"TimestampType":kudu.unixtime_micros,
}
for sf in df_schema:
pk = False
nullable=sf.nullable
if (sf.name in primary_keys):
pk = True
nullable = False
builder.add_column(
name=sf.name,
nullable=nullable,
type_=data_type_map[str(sf.dataType)]
)
builder.set_primary_keys(primary_keys)
return builder.build()
您可以这样称呼它:
kudu_schema = convert_to_kudu_schema(df.schema,primary_keys=["key1","key2"])
我仍然愿意寻求更优雅的解决方案。 ;)