-
Notifications
You must be signed in to change notification settings - Fork 29
/
Paso06_GeneracionVentasNuevasPorDia.py
61 lines (43 loc) · 2.53 KB
/
Paso06_GeneracionVentasNuevasPorDia.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
from pyspark.sql import *
from pyspark.sql.types import *
from pyspark.sql.functions import *
# Completar con la fecha nueva a ingestar:
fecha_nvo = '2021-01-01'
from pyspark.sql import SparkSession
spark = SparkSession.\
builder.\
appName("prueba-pyspark").\
master("spark://spark-master:7077").\
config("spark.executor.memory", "512m").\
getOrCreate()
cliente = spark.read.csv(path="hdfs://namenode:9000/data_nvo/Cliente.csv", inferSchema=True, sep=";", header=True)
empleado = spark.read.csv(path="hdfs://namenode:9000/data_nvo/Empleado.csv", inferSchema=True, sep=",", header=True)
producto = spark.read.csv(path="hdfs://namenode:9000/data_nvo/Producto.csv", inferSchema=True, sep=",", header=True)
cliente.createOrReplaceTempView("cliente")
cliente_nvo = spark.sql("SELECT ID AS IdCliente FROM cliente WHERE RAND() < 0.025")
cliente_nvo.createOrReplaceTempView("cliente_nvo")
empleado.createOrReplaceTempView("empleado")
empleado_nvo = spark.sql("SELECT CodigoEmpleado AS IdEmpleado, IdSucursal FROM empleado WHERE RAND() < 0.25")
empleado_nvo.createOrReplaceTempView("empleado_nvo")
producto.createOrReplaceTempView("producto")
producto_nvo = spark.sql("SELECT ID_PRODUCTO as IdProducto, Precio FROM producto WHERE RAND() < 0.2")
producto_nvo.createOrReplaceTempView("producto_nvo")
consulta = "SELECT '" + str(fecha_nvo) + "' AS Fecha, "
consulta = consulta + "ROUND(RAND() * 4 + 1, 0) AS Dias_Entrega, "
consulta = consulta + "ROUND(RAND() * 2 + 1, 0) AS IdCanal, "
consulta = consulta + "c.IdCliente, "
consulta = consulta + "e.IdSucursal, "
consulta = consulta + "e.IdEmpleado, "
consulta = consulta + "p.IdProducto, "
consulta = consulta + "p.Precio, "
consulta = consulta + "ROUND(RAND() * 3 + 1, 0) AS Cantidad "
consulta = consulta + "FROM cliente_nvo c, producto_nvo p, empleado_nvo e "
consulta = consulta + "WHERE RAND() < 0.0005"
venta_nvo = spark.sql(consulta)
venta_nvo = venta_nvo.withColumn("IdCanal",col("IdCanal").cast("int"))
venta_nvo = venta_nvo.withColumn("Cantidad",col("Cantidad").cast("int"))
venta_nvo = venta_nvo.withColumn("Dias_Entrega",col("Dias_Entrega").cast("int"))
venta_nvo = venta_nvo.withColumn("Fecha",col("Fecha").cast("date"))
venta_nvo.createOrReplaceTempView("venta_nvo")
venta_nvo = spark.sql("SELECT 0 AS IdVenta, Fecha, DATE_ADD(Fecha, Dias_Entrega) AS Fecha_Entrega, IdCanal, IdCliente, IdSucursal, IdEmpleado, IdProducto, Precio, Cantidad FROM venta_nvo")
venta_nvo.repartition(1).write.csv(path="hdfs://namenode:9000/data_nvo/venta_nvo", sep=",", header=True, mode="overwrite")