The part that crystallizes this post the best and shows just how much further along delta is than iceberg for developers is the 2-liner you had on daft reading a parquet file and then writing it to the delta format:
- no catalog nonsense
- no extra package to convert from one df format to arrow
- no needing to lasso a table object
- like you said, working with iceberg (when not on pyspark) is a “daisy chain of crap” which is 100% facts
What about actually feature support comparison between both? As far as I was able to research, it seems Iceberg has additional features that can improve performance over Delta. Is that not true?
For completeness, here is a simple cli tool that let's you read and write Iceberg tables with Datafusion even without a catalog:
https://github.com/JanKaul/frostbow
Disclaimer: I'm the author of the tool.
The part that crystallizes this post the best and shows just how much further along delta is than iceberg for developers is the 2-liner you had on daft reading a parquet file and then writing it to the delta format:
- no catalog nonsense
- no extra package to convert from one df format to arrow
- no needing to lasso a table object
- like you said, working with iceberg (when not on pyspark) is a “daisy chain of crap” which is 100% facts
I wonder how this has changed over a year of development
What about actually feature support comparison between both? As far as I was able to research, it seems Iceberg has additional features that can improve performance over Delta. Is that not true?
Thanks for your point of view, I have enjoyed your article.
I just finished one about Deltalake too and took the liberty to list yours as related article. Let me know if this is an issue. https://womenunbound.substack.com/p/quant-chronicles-deltalakea-powerful