Skip to main content

Switch FAQs and Troubleshooting

Can Switch Support My Source System?​

A common question is whether Switch can handle sources beyond the built-in conversion types. The answer is: try it!

Switch already supports various source formats including SQL dialects (MySQL, Snowflake, Oracle, etc.), programming languages (Python scripts, Scala code), and workflows (Airflow DAGs).

For SQL-based sources: Creating a custom prompt YAML file should work well for most SQL dialects. Since LLMs understand SQL syntax patterns, you can typically achieve good results by:

  • Starting with a similar built-in dialect's YAML as a template
  • Adding specific syntax examples from your source system
  • Testing and iterating based on results

Tips for efficient prompt creation:

  • Quick baseline creation: Feed built-in prompts and your source dialect's representative features to an advanced LLM to quickly generate a baseline YAML configuration
  • Dialect-specific patterns: Reference open source projects like SQLGlot dialects for insights into dialect-specific transformation patterns

For other source formats: Switch's LLM-based architecture means it can potentially handle various conversions beyond the built-in types. Modern LLMs have strong comprehension capabilities across many languages and formats. You can experiment by:

  • Creating custom prompts that define your source format
  • Providing clear conversion examples in the few-shots section
  • Testing with representative source samples

Rather than waiting for additional built-in examples, we encourage experimentation with custom prompts. The flexibility of LLM-based conversion means many use cases are possible with the right prompt engineering.

Conversion Results and Troubleshooting​

Understanding Conversion Results​

After your Switch job completes, review the conversion results displayed at the end of the 00_main notebook execution. The results table shows the status of each input file:

  • Successfully converted files: Ready to use as Databricks notebooks
  • Files requiring attention: May need manual review or re-processing

If you encounter files that didn't convert successfully, here are the most common issues and their solutions:

Files Not Converting (Status: Not converted)​

These files were skipped during the conversion process, typically because they're too large for the model to process effectively.

Cause: Input files exceed the token count threshold

Solutions:

  • Split large input files into smaller, more manageable parts
  • Increase the token_count_threshold parameter if your LLM model can handle larger inputs

Conversion with Errors (Status: Converted with errors)​

These files were successfully processed by the LLM but the generated code contains syntax errors that need to be addressed.

Cause: Files were converted but contain syntax errors

Solutions:

  • Review syntax error messages in the result table's error_details column
  • Manually fix errors in the converted notebooks/files
  • Increase max_fix_attempts for more automatic error correction attempts

Export Failures (Status: Export failed or Not exported)​

These files were converted successfully but couldn't be exported to the output directory.

Causes:

  • Content exceeds 10MB size limit of Databricks notebooks
  • File system permissions issues
  • Invalid output paths

Solutions:

  • Check the export_error column in the result table for specific error details
  • For size issues: Manually split large converted content into smaller units
  • For permission issues: Verify workspace access to the output directory
  • For path issues: Ensure output directory paths are valid workspace locations