为什么在 AWS Glue pyspark 中使用 UDF 添加派生列后，将 DataFrame 写入 S3（或将动态帧写入 Redshift）会出错？

Question

I had an AWS Glue Job with ETL script in pyspark which wrote dynamic frame to redshift as a table and to s3 as json. One of the column in this df is status_date . I had no issue in writing this df.

I then had a requirement to add two more columns financial_year and financial_qarter based on the status_date . For this I created a udf added two new columns. Using printSchema() and show() I saw the columns were successfully created and values in the columns were collect.

The problem came when I tried to write this to aws s3 and aws redshift. It is giving a weird error which I am not able to troubleshoot.

Error if I write to Redshift - An error occurred while calling o177.pyWriteDynamicFrame. File already exists:s3://aws-glue-temporary-***********-ap-south-1/ShashwatS/491bb37a-404a-4ec5-a459-5534d94b0206/part-00002-af529a71-7315-4bd1-ace5-91ab5d9f7f46-c000.csv An error occurred while calling o177.pyWriteDynamicFrame. File already exists:s3://aws-glue-temporary-***********-ap-south-1/ShashwatS/491bb37a-404a-4ec5-a459-5534d94b0206/part-00002-af529a71-7315-4bd1-ace5-91ab5d9f7f46-c000.csv

Error if I write to s3 as json - An error occurred while calling o175.json. File already exists:s3://bucket_name/folder1/folder2/folder3/folder4/24022020124055/part-00026-fd481713-2ccc-4d23-98b0-e96908cb708c-c000.json An error occurred while calling o175.json. File already exists:s3://bucket_name/folder1/folder2/folder3/folder4/24022020124055/part-00026-fd481713-2ccc-4d23-98b0-e96908cb708c-c000.json

As you can see both the errors are similar kind. Below is the error trace. Some help required.

> org.apache.spark.SparkException: Job aborted due to stage failure:
> Task 3 in stage 503.0 failed 4 times, most recent failure: Lost task
> 3.3 in stage 503.0 (, executor 15):
> org.apache.hadoop.fs.FileAlreadyExistsException: File already
> exists:s3://bucket_name/folder1/folder2/folder3/folder4/24022020150358/part-00003-055f9327-1368-4ba7-9216-6a32afac1843-c000.json
>   at
> com.amazon.ws.emr.hadoop.fs.s3.upload.plan.RegularUploadPlanner.checkExistenceIfNotOverwriting(RegularUploadPlanner.java:36)
>   at
> com.amazon.ws.emr.hadoop.fs.s3.upload.plan.RegularUploadPlanner.plan(RegularUploadPlanner.java:30)
>   at
> com.amazon.ws.emr.hadoop.fs.s3.upload.plan.UploadPlannerChain.plan(UploadPlannerChain.java:37)
>   

Answer 1

Are u trying to write to the same location in your code twice ?? Eg : If u have already written to location s3://bucket_name/folder1/folder2/folder3/folder4/24022020124055 once in your code and are trying to do it again in the same code ,it will not work.

Answer 2

I had this error and took me a couple of days to find the cause. My Issue was caused by the file format/data types of the S3 Files. Create UTF8 files or change all your data files format into UTF-8.

Or find the "bad" record by ordering the records in source S3 files by ID or some uniqueidentifier and then by looking at the last file created example s3://bucket_name/folder1/folder2/folder3/folder4/24022020150358/part-xxxx

find last record in this file and correspond that to the source file. In my case the prblm record was 3 to 7 lines past that last record that was imported. If its a special character then you need to change your file format.

Or a quick check is to just rmove that special character and recheck the output file if its gone past the previous "bad" record.