Auto Loader fails when reading json element containing space

Magnus
Contributor

‎07-11-2023 02:42 AM

I'm using Auto Loader as part of a Delta Live Tables pipeline to ingest json files, and today it failed with this error message:

om.databricks.sql.transaction.tahoe.DeltaAnalysisException: Found invalid character(s) among ' ,;{}()\n\t=' in the column names of your schema. 
org.apache.spark.sql.AnalysisException:  Column name "NotificationSettings.element.Microsoft Teams" contains invalid character(s). Please use alias to rename it.

The failing json file contains an element named "Microsoft Teams" which causes the pipeline to fail. How can I handle such elements? The error message mentions the use of an alias, but I can't find any info on how to implement this.

		"NotificationSettings": [
			{
				"NotificationType": "MissedActivityReminder",
				"Microsoft Teams": true
			},
...

 

0 Kudos

Magnus
Contributor
In response to Lakshay

‎07-14-2023 12:36 AM

I could not get the DeltaTable solution to work in combination with Auto Loader/DLT/Unity Catalog, since it expects a table location and I want the framework to handle that.

I also tried withColumnRenamed, but I can't get it to work either. I still get the error message shown in my original question.

withColumnRenamed("NotificationSettings.Microsoft Teams", "MicrosoftTeams")

 

0 Kudos

Tharun-Kumar
Databricks Employee
Databricks Employee

‎07-14-2023 04:02 AM

@Magnus 

  • You can read the input file using Pandas or Koalas (https://koalas.readthedocs.io/en/latest/index.html)
  • then rename the columns
  • then convert the Pandas/Koalas dataframe to Spark dataframe. You can write it back with the correct column name, so the next time you use it, the error will not happen.