概要

この記事では、Azure Data Factory を使用して Oracle Database から Azure Databricks Delta Lake コネクターに書き込む際に発生する特定のエラーへの対応方法を説明します。具体的には、Oracle Database の NLS_CHARACTERSET が JA16SJISTILDE であり、データに改行（\n）が含まれている場合に次のようなエラーが発生します。対応策として、Azure Databricks Delta Lake コネクターのシンク時の設定に additionalFormatOptions を追加します。

ErrorCode=AzureDatabricksCommandError,Hit an error when running the command in Azure Databricks. Error details: org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in stage 15.0 failed 4 times, most recent failure: Lost task 0.3 in stage 15.0 (TID 18) (10.139.64.10 executor 0): org.apache.spark.SparkException: [TASK_WRITE_FAILED] Task failed while writing rows to dbfs:/user/hive/warehouse/multi_line_test.db/multiline_from_oracle.
Caused by: com.databricks.sql.io.FileReadException: Error while reading file wasbs:REDACTED_LOCAL_PART@{storage_name}.blob.core.windows.net/2bceab90-1216-4ea9-a6d2-4b846fea93a6/AzureDatabricksDeltaLakeImportCommand/SSMA.MULTILINE_TABLE.txt.
Caused by: org.apache.spark.SparkException: [MALFORMED_RECORD_IN_PARSING.WITHOUT_SUGGESTION] Malformed records are detected in record parsing: [F",null].
Caused by: org.apache.spark.sql.catalyst.util.BadRecordException: org.apache.spark.SparkRuntimeException: [MALFORMED_CSV_RECORD] Malformed CSV record: F"
Caused by: org.apache.spark.SparkRuntimeException: [MALFORMED_CSV_RECORD] Malformed CSV record: F"
Caused by: org.apache.spark.SparkException: [TASK_WRITE_FAILED] Task failed while writing rows to dbfs:/user/hive/warehouse/multi_line_test.db/multiline_from_oracle.
Caused by: com.databricks.sql.io.FileReadException: Error while reading file wasbs:REDACTED_LOCAL_PART@{storage_name}.blob.core.windows.net/2bceab90-1216-4ea9-a6d2-4b846fea93a6/AzureDatabricksDeltaLakeImportCommand/SSMA.MULTILINE_TABLE.txt.
Caused by: org.apache.spark.SparkException: [MALFORMED_RECORD_IN_PARSING.WITHOUT_SUGGESTION] Malformed records are detected in record parsing: [F",null].
Caused by: org.apache.spark.sql.catalyst.util.BadRecordException: org.apache.spark.SparkRuntimeException: [MALFORMED_CSV_RECORD] Malformed CSV record: F"
Caused by: org.apache.spark.SparkRuntimeException: [MALFORMED_CSV_RECORD] Malformed CSV record: F".

additionalFormatOptionsに関する詳細な説明については、以下の記事を参照してください。本記事では、手順のみ紹介します。

Azure Data Factory にて改行を含む区切りテキストファイルを Databricks Delta Lake コネクターにより書き込むパイプラインの実装方法 #Databricks - Qiita

エラーコードと対応策

事前準備

Oracle Database のNLS_CHARACTERSET が JA16SJISTILDE であることを確認します。

SELECT
    *
FROM
    nls_database_parameters
WHERE
    parameter = 'NLS_CHARACTERSET';

Oracle Database でテーブルを作成し、データを挿入します。

/*
DROP TABLE control_character_table;
*/
CREATE TABLE multiline_table (
     text VARCHAR2(255),
     tgt_string VARCHAR2(10)
 )
;
TRUNCATE TABLE multiline_table; 
--
INSERT INTO multiline_table (text, tgt_string)
SELECT 'CR',  'C' || CHR(13) || 'R' FROM dual
UNION ALL
SELECT 'LF', 'L' || CHR(10) || 'F'  FROM dual
UNION ALL
SELECT'CRLF', 'CR' || CHR(13) || CHR(10) || 'LF' FROM dual
; 
COMMIT;
SELECT * FROM multiline_table 
;

Databricks で書き込み先のテーブルを作成します。

%sql
CREATE SCHEMA IF NOT EXISTS multi_line_test;
CREATE OR REPLACE TABLE multi_line_test.multiline_from_oracle (
  text string,
  tgt_string string
);

エラーとなるプログラムを確認

Oracle Database から Databricks への書き込みパイプラインを作成し、実行します。その後、エラーが発生することを確認します。

ErrorCode=AzureDatabricksCommandError,Hit an error when running the command in Azure Databricks. Error details: org.apache.spark.SparkException: Job aborted due to stage failure: Task 0 in stage 15.0 failed 4 times, most recent failure: Lost task 0.3 in stage 15.0 (TID 18) (10.139.64.10 executor 0): org.apache.spark.SparkException: [TASK_WRITE_FAILED] Task failed while writing rows to dbfs:/user/hive/warehouse/multi_line_test.db/multiline_from_oracle.
Caused by: com.databricks.sql.io.FileReadException: Error while reading file wasbs:REDACTED_LOCAL_PART@{storage_name}.blob.core.windows.net/2bceab90-1216-4ea9-a6d2-4b846fea93a6/AzureDatabricksDeltaLakeImportCommand/SSMA.MULTILINE_TABLE.txt.
Caused by: org.apache.spark.SparkException: [MALFORMED_RECORD_IN_PARSING.WITHOUT_SUGGESTION] Malformed records are detected in record parsing: [F",null].
Caused by: org.apache.spark.sql.catalyst.util.BadRecordException: org.apache.spark.SparkRuntimeException: [MALFORMED_CSV_RECORD] Malformed CSV record: F"
Caused by: org.apache.spark.SparkRuntimeException: [MALFORMED_CSV_RECORD] Malformed CSV record: F"
Caused by: org.apache.spark.SparkException: [TASK_WRITE_FAILED] Task failed while writing rows to dbfs:/user/hive/warehouse/multi_line_test.db/multiline_from_oracle.
Caused by: com.databricks.sql.io.FileReadException: Error while reading file wasbs:REDACTED_LOCAL_PART@{storage_name}.blob.core.windows.net/2bceab90-1216-4ea9-a6d2-4b846fea93a6/AzureDatabricksDeltaLakeImportCommand/SSMA.MULTILINE_TABLE.txt.
Caused by: org.apache.spark.SparkException: [MALFORMED_RECORD_IN_PARSING.WITHOUT_SUGGESTION] Malformed records are detected in record parsing: [F",null].
Caused by: org.apache.spark.sql.catalyst.util.BadRecordException: org.apache.spark.SparkRuntimeException: [MALFORMED_CSV_RECORD] Malformed CSV record: F"
Caused by: org.apache.spark.SparkRuntimeException: [MALFORMED_CSV_RECORD] Malformed CSV record: F".

対応策を実施

Azure Data Factory でパイプラインの右上に表示されている このソースの json コード表現を表示する をクリックして、定義の json を表示します。

次に、sink の項目を探し、importSettings に additionalFormatOptions を追加します。

"additionalFormatOptions": {
    "multiline": true
}

設定前後を比較すると、次のようになります。

"sink": {
    "type": "AzureDatabricksDeltaLakeSink",
    "importSettings": {
        "type": "AzureDatabricksDeltaLakeImportCommand"
    }
},

"sink": {
    "type": "AzureDatabricksDeltaLakeSink",
    "importSettings": {
        "type": "AzureDatabricksDeltaLakeImportCommand",
        "additionalFormatOptions": {
            "multiline": true
        }
    }
},

パイプラインを実行し、正常終了することを確認します。

最後に、Databricks でテーブルに書き込まれていることを確認します。

df = spark.table("multi_line_test.multiline_from_oracle")

print(df.collect())
df.display()

[
    Row(text="CR", tgt_string="C\rR"),
    Row(text="LF", tgt_string="L\nF"),
    Row(text="CRLF", tgt_string="CR\r\nLF"),
]