Azure Databricks 集群合并小文件与删除过多的历史版本

lbl2512024-03-13 17:34

原因是：databricks 集群的数据底层是HDFS虽然是spark做为引擎读写如果没有及时合并也一样会因为小文件问题造成大量的资源消耗，也就会越来越慢。目前采用的主要方式，定时合并，与版本删除

1、python 脚本如下有用到的同学可以参考下

bash 复制代码

# 合并
for database_name in database_list:
    sqlQueryShowTables = "SHOW TABLES FROM {0}".format(database_name)
    tablesDF = spark.sql(sqlQueryShowTables).collect()

    for table in tablesDF:
        sqlOptimizeTable = "OPTIMIZE {0}.{1}".format(database_name, table['tableName'])
        try:
            spark.sql(sqlOptimizeTable)
            print("INFO: Optimize table {0}.{1} completed.".format(database_name, table['tableName']))
        except Exception as e:
            print("ERROR: Optimize table {0}.{1} failed.".format(database_name, table['tableName']))

bash 复制代码

# 删除多的版本
for database_name in database_list:
    sqlQueryShowTables = "SHOW TABLES FROM {0}".format(database_name)
    tablesDF = spark.sql(sqlQueryShowTables).collect()

    for table in tablesDF:
        sqlVACUUMTable = "VACUUM {0}.{1} RETAIN 168 HOURS".format(database_name, table['tableName'])
        try:
            spark.sql(sqlVACUUMTable)
            print("INFO: VACUUM table {0}.{1} completed.".format(database_name, table['tableName']))
        except Exception as e:
            print("ERROR: VACUUM table {0}.{1} failed.".format(database_name, table['tableName']))

2、在workflows 设置好定时器就行了，