apache-spark - 分区表上的 Hive 增量

Question

我正在努力在 Hive 表 A 上实施增量过程；表 A - 已在配置单元中创建，并在 YearMonth （ YYYYMM 列）上进行了分区，并具有完整的卷。

在持续的基础上，我们计划从源导入更新/插入并在 hive Delta Table 中捕获；

如下图所示，Delta 表表明新的更新与分区（ 201804 / 201611 / 201705 ）有关。

对于增量过程，我计划

1. 从受影响的原始表中选择 3 个分区。

INSERT INTO delta2 select YYYYMM from Table where YYYYMM in ( select distinct YYYYMM from Delta );

2. 将 Delta 表中的这 3 个分区与原始表中的相应分区合并。（我可以按照 Horton 的 4 步策略来应用更新）

       Merge Delta2 + Delta : = new 3 partitions.
   

3. 从原始表中删除 3 个分区

   Alter Table Drop partitions 201804 / 201611 / 201705
   

4. 将新合并的分区添加回原始表（具有新更新）

我需要自动化这个脚本 - 你能建议如何将上述逻辑放在 hive QL 或 spark 中 - 特别识别分区并将它们从原始表中删除。

Answer 1

you can build a solution using pyspark. I am explaining this approach with some basic example. you can re-modify it as per your business requirements.

Suppose you have a partitioned table in hive below configuration.

CREATE TABLE IF NOT EXISTS udb.emp_partition_Load_tbl (
 emp_id                 smallint
,emp_name               VARCHAR(30)
,emp_city               VARCHAR(10)
,emp_dept               VARCHAR(30)
,emp_salary             BIGINT
)
PARTITIONED BY (Year String, Month String)
ROW FORMAT DELIMITED FIELDS TERMINATED BY '|'
STORED AS ORC;

and you got some csv file with some input records which you want to load into your partitioned table

1|vikrant singh rana|Gurgaon|Information Technology|20000

dataframe = spark.read.format("com.databricks.spark.csv") \
  .option("mode", "DROPMALFORMED") \
  .option("header", "false") \
  .option("inferschema", "true") \
  .schema(userschema) \
  .option("delimiter", "|").load("file:///filelocation/userinput")

newdf = dataframe.withColumn('year', lit('2018')).withColumn('month',lit('01'))

+------+------------------+--------+----------------------+----------+----+-----+
|emp-id|emp-name          |emp-city|emp-department        |emp-salary|year|month|
+------+------------------+--------+----------------------+----------+----+-----+
|1     |vikrant singh rana|Gurgaon |Information Technology|20000     |2018|01   |
+------+------------------+--------+----------------------+----------+----+-----+

setting below properties to overwrite specific partitions data only.

spark.conf.set("spark.sql.sources.partitionOverwriteMode","dynamic")
spark.sql("set spark.hadoop.hive.exec.dynamic.partition=true");
spark.sql("set spark.hadoop.hive.exec.dynamic.partition.mode=nonstrict");

newdf.write.format('orc').mode("overwrite").insertInto('udb.emp_partition_Load_tbl')

lets say you got another set of data and want to insert into some other partitions

+------+--------+--------+--------------+----------+----+-----+
|emp-id|emp-name|emp-city|emp-department|emp-salary|year|month|
+------+--------+--------+--------------+----------+----+-----+
|     2|     ABC| Gurgaon|HUMAN RESOURCE|     10000|2018|   02|
+------+--------+--------+--------------+----------+----+-----+
newdf.write.format('orc').mode("overwrite").insertInto('udb.emp_partition_Load_tbl')

> show partitions udb.emp_partition_Load_tbl;
+---------------------+--+
|      partition      |
+---------------------+--+
| year=2018/month=01  |
| year=2018/month=02  |
+---------------------+--+

assuming you have got another set of records pertaining to existing partition.

3|XYZ|Gurgaon|HUMAN RESOURCE|80000

newdf = dataframe.withColumn('year', lit('2018')).withColumn('month',lit('02'))
+------+--------+--------+--------------+----------+----+-----+
|emp-id|emp-name|emp-city|emp-department|emp-salary|year|month|
+------+--------+--------+--------------+----------+----+-----+
|     3|     XYZ| Gurgaon|HUMAN RESOURCE|     80000|2018|   02|
+------+--------+--------+--------------+----------+----+-----+

newdf.write.format('orc').mode("overwrite").insertInto('udb.emp_partition_Load_tbl')


 select * from udb.emp_partition_Load_tbl where year ='2018' and month ='02';
+---------+-----------+-----------+-----------------+-------------+-------+--------+--+
| emp_id  | emp_name  | emp_city  |    emp_dept     | emp_salary  | year  | month  |
+---------+-----------+-----------+-----------------+-------------+-------+--------+--+
| 3       | XYZ       | Gurgaon   | HUMAN RESOURCE  | 80000       | 2018  | 02     |
| 2       | ABC       | Gurgaon   | HUMAN RESOURCE  | 10000       | 2018  | 02     |
+---------+-----------+-----------+-----------------+-------------+-------+--------+--+

you can see below that other partiion data was untouched.

> select * from udb.emp_partition_Load_tbl where year ='2018' and month ='01';

+---------+---------------------+-----------+-------------------------+-------------+-------+--------+--+
| emp_id  |      emp_name       | emp_city  |        emp_dept         | emp_salary  | year  | month  |
+---------+---------------------+-----------+-------------------------+-------------+-------+--------+--+
| 1       | vikrant singh rana  | Gurgaon   | Information Technology  | 20000       | 2018  | 01     |
+---------+---------------------+-----------+-------------------------+-------------+-------+--------+--+