Question

我使用Spark sql DataSet将数据写入hive。如果模式相同，它的工作完美但如果我更改了avro模式，在其间添加新列，则显示错误（模式由模式注册表提供）

Error running job streaming job 1519289340000 ms.0 org.apache.spark.sql.AnalysisException: The column number of the existing table default.sample(struct<collection_timestamp:bigint,managed_object_id:string,managed_object_type:string,if_admin_status:string,date:string,hour:int,quarter:bigint>) doesn't match the data schema(struct<collection_timestamp:bigint,managed_object_id:string,if_oper_status:string,managed_object_type:string,if_admin_status:string,date:string,hour:int,quarter:bigint>);

if_oper_status是必须添加的新列。请建议。

StructType struct = convertSchemaToStructType(SchemaRegstryClient.getLatestSchema("simple"));
        Dataset<Row> dataset = getSparkInstance().createDataFrame(newRDD, struct);


        dataset=dataset.withColumn("date",functions.date_format(functions.current_date(), "dd-MM-yyyy"));
        dataset=dataset.withColumn("hour",functions.hour(functions.current_timestamp()));
        dataset=dataset.withColumn("quarter",functions.floor(functions.minute(functions.current_timestamp()).divide(5)));


        dataset
        .coalesce(1)
        .write().mode(SaveMode.Append)
        .option("charset", "UTF8")
        .partitionBy("date","hour","quarter")
        .option("checkpointLocation", "/tmp/checkpoint")
        .saveAsTable("sample");

Answer 1

我能够通过将架构从注册表保存到文件中并提供如下的avro.schema.url =文件路径来解决此问题。

注意：必须在saveAsTable("sample")

之前完成

dataset.sqlContext().sql("CREATE EXTERNAL TABLE IF NOT EXISTS sample PARTITIONED BY (dt STRING, hour STRING, quarter STRING ) ROW FORMAT SERDE 'org.apache.hadoop.hive.serde2.avro.AvroSerDe' STORED as INPUTFORMAT 'org.apache.hadoop.hive.ql.io.avro.AvroContainerInputFormat' OUTPUTFORMAT 'org.apache.hadoop.hive.ql.io.avro.AvroContainerOutputFormat' LOCATION 'hdfs://localhost:9000/user/root/sample'  TBLPROPERTIES ('avro.schema.url'='file://"+file.getAbsolutePath()+"')");

Answer 2

请参阅链接：https://github.com/databricks/spark-avro/pull/155。每次提交历史记录，支持不断发展的Avro架构的PR已添加到3.1版。什么是你在代码中使用的spark-avro版本？

如果在avro架构中添加了新列，则Spark sql saveAsTable创建表追加模式

2 个答案: