Spark Delta Lake

rm -r dp-203 -f

git clone https://github.com/MicrosoftLearning/dp-203-azure-data-engineer dp-203

cd dp-203/Allfiles/labs/07

./setup.ps1

python 复制代码
%%pyspark
df = spark.read.load('abfss://files@datalakexxxxxxx.dfs.core.windows.net/products/products.csv', format='csv'
## If header exists uncomment line below
##, header=True
)
display(df.limit(10))
python 复制代码
%%pyspark
df = spark.read.load('abfss://files@datalakexxxxxxx.dfs.core.windows.net/products/products.csv', format='csv'
## If header exists uncomment line below
, header=True
)
display(df.limit(10))
python 复制代码
 delta_table_path = "/delta/products-delta"
 df.write.format("delta").save(delta_table_path)
  1. On the files tab, use the icon in the toolbar to return to the root of the files container, and note that a new folder named delta has been created. Open this folder and the products-delta table it contains, where you should see the parquet format file(s) containing the data.
python 复制代码
from delta.tables import *
from pyspark.sql.functions import *

 # Create a deltaTable object
deltaTable = DeltaTable.forPath(spark, delta_table_path)

 # Update the table (reduce price of product 771 by 10%)
deltaTable.update(
     condition = "ProductID == 771",
     set = { "ListPrice": "ListPrice * 0.9" })

 # View the updated data as a dataframe
deltaTable.toDF().show(10)
python 复制代码
 new_df = spark.read.format("delta").load(delta_table_path)
 new_df.show(10)
python 复制代码
 new_df = spark.read.format("delta").option("versionAsOf", 0).load(delta_table_path)
 new_df.show(10)
python 复制代码
deltaTable.history(10).show(20, False, True)
python 复制代码
 spark.sql("CREATE DATABASE AdventureWorks")
 spark.sql("CREATE TABLE AdventureWorks.ProductsExternal USING DELTA LOCATION '{0}'".format(delta_table_path))
 spark.sql("DESCRIBE EXTENDED AdventureWorks.ProductsExternal").show(truncate=False)

This code creates a new database named AdventureWorks and then creates an external tabled named ProductsExternal in that database based on the path to the parquet files you defined previously. It then displays a description of the table's properties. Note that the Location property is the path you specified.

sql 复制代码
%%sql

 USE AdventureWorks;

 SELECT * FROM ProductsExternal;
python 复制代码
 df.write.format("delta").saveAsTable("AdventureWorks.ProductsManaged")
 spark.sql("DESCRIBE EXTENDED AdventureWorks.ProductsManaged").show(truncate=False)

This code creates a managed tabled named ProductsManaged based on the DataFrame you originally loaded from the products.csv file (before you updated the price of product 771). You do not specify a path for the parquet files used by the table - this is managed for you in the Hive metastore, and shown in the Location property in the table description (in the files/synapse/workspaces/synapsexxxxxxx/warehouse path).

sql 复制代码
%%sql

 USE AdventureWorks;

 SELECT * FROM ProductsManaged;
sql 复制代码
%%sql

 USE AdventureWorks;

 SHOW TABLES;
sql 复制代码
%%sql

 USE AdventureWorks;

 DROP TABLE IF EXISTS ProductsExternal;
 DROP TABLE IF EXISTS ProductsManaged;
  1. Return to the files tab and view the files/delta/products-delta folder. Note that the data files still exist in this location. Dropping the external table has removed the table from the metastore, but left the data files intact.
  2. View the files/synapse/workspaces/synapsexxxxxxx/warehouse folder, and note that there is no folder for the ProductsManaged table data. Dropping a managed table removes the table from the metastore and also deletes the table's data files.
sql 复制代码
%%sql

 USE AdventureWorks;

 CREATE TABLE Products
 USING DELTA
 LOCATION '/delta/products-delta';
sql 复制代码
%%sql

 USE AdventureWorks;

 SELECT * FROM Products;
python 复制代码
 from notebookutils import mssparkutils
 from pyspark.sql.types import *
 from pyspark.sql.functions import *

 # Create a folder
 inputPath = '/data/'
 mssparkutils.fs.mkdirs(inputPath)

 # Create a stream that reads data from the folder, using a JSON schema
 jsonSchema = StructType([
 StructField("device", StringType(), False),
 StructField("status", StringType(), False)
 ])
 iotstream = spark.readStream.schema(jsonSchema).option("maxFilesPerTrigger", 1).json(inputPath)

 # Write some event data to the folder
 device_data = '''{"device":"Dev1","status":"ok"}
 {"device":"Dev1","status":"ok"}
 {"device":"Dev1","status":"ok"}
 {"device":"Dev2","status":"error"}
 {"device":"Dev1","status":"ok"}
 {"device":"Dev1","status":"error"}
 {"device":"Dev2","status":"ok"}
 {"device":"Dev2","status":"error"}
 {"device":"Dev1","status":"ok"}'''
 mssparkutils.fs.put(inputPath + "data.txt", device_data, True)
 print("Source stream created...")

Ensure the message Source stream created... is printed. The code you just ran has created a streaming data source based on a folder to which some data has been saved, representing readings from hypothetical IoT devices.

python 复制代码
 # Write the stream to a delta table
 delta_stream_table_path = '/delta/iotdevicedata'
 checkpointpath = '/delta/checkpoint'
 deltastream = iotstream.writeStream.format("delta").option("checkpointLocation", checkpointpath).start(delta_stream_table_path)
 print("Streaming to delta sink...")
python 复制代码
 # Read the data in delta format into a dataframe
 df = spark.read.format("delta").load(delta_stream_table_path)
 display(df)
python 复制代码
 # create a catalog table based on the streaming sink
 spark.sql("CREATE TABLE IotDeviceData USING DELTA LOCATION '{0}'".format(delta_stream_table_path))
sql 复制代码
 %%sql

 SELECT * FROM IotDeviceData;
python 复制代码
 # Add more data to the source stream
 more_data = '''{"device":"Dev1","status":"ok"}
 {"device":"Dev1","status":"ok"}
 {"device":"Dev1","status":"ok"}
 {"device":"Dev1","status":"ok"}
 {"device":"Dev1","status":"error"}
 {"device":"Dev2","status":"error"}
 {"device":"Dev1","status":"ok"}'''

 mssparkutils.fs.put(inputPath + "more-data.txt", more_data, True)
sql 复制代码
%%sql

 SELECT * FROM IotDeviceData;
python 复制代码
 deltastream.stop()
sql 复制代码
 -- This is auto-generated code
 SELECT
     TOP 100 *
 FROM
     OPENROWSET(
         BULK 'https://datalakexxxxxxx.dfs.core.windows.net/files/delta/products-delta/',
         FORMAT = 'DELTA'
     ) AS [result]
sql 复制代码
 USE AdventureWorks;

 SELECT * FROM Products;

Run the code and observe that you can also use the serverless SQL pool to query Delta Lake data in catalog tables that are defined the Spark metastore.

相关推荐
武子康22 分钟前
大数据-128 - Flink 并行度详解:从概念到最佳实践,一文读懂任务并行执行机制 代码示例与性能优化
大数据·后端·flink
望获linux2 小时前
【实时Linux实战系列】FPGA 与实时 Linux 的协同设计
大数据·linux·服务器·网络·数据库·fpga开发·操作系统
励志成为糕手3 小时前
宽依赖的代价:Spark 与 MapReduce Shuffle 的数据重分布对比
大数据·spark·mapreduce·分布式计算·sortshuffle
Elastic 中国社区官方博客7 小时前
根据用户行为数据中的判断列表在 Elasticsearch 中训练 LTR 模型
大数据·数据库·人工智能·elasticsearch·搜索引擎·ai·全文检索
点控云8 小时前
点控云智能短信:重构企业与用户的连接,让品牌沟通更高效
大数据·人工智能·科技·重构·外呼系统·呼叫中心
风清再凯10 小时前
04_es原理&filebeat使用
大数据·elasticsearch·搜索引擎
小小王app小程序开发10 小时前
盲盒小程序开发新视角:从用户体验到运营落地的分析拆解
大数据·ux
weixin_5259363312 小时前
部分Spark SQL编程要点
大数据·python·sql·spark
wan5555cn13 小时前
当代社会情绪分类及其改善方向深度解析
大数据·人工智能·笔记·深度学习·算法·生活
SirLancelot113 小时前
MongoDB-基本介绍(一)基本概念、特点、适用场景、技术选型
java·数据库·分布式·后端·mongodb·软件工程·软件构建