Is there any difference between performance of Python and SQL ?

[abd]

I read somewhere that Python code is converted to SQL at the end. So is it true or there is any difference in performance while working with Scala, Python or SQL ?

[CodeMartian]

In short, no there's no difference. However, there does need to be a translation, like you read somewhere, so it could add a negligible amount of time to the workload. However, the performance doesn't degrade significantly enough to matter.

[adamq · Accepted solution]

If you're using the DataFrame API it all gets run in the JVM, just like sql queries. The exception is UDFs which have to transfer data to Python land to execute.

[abd]

Can you provide any resource where I would be able to look into it ?

Just wondering is python code converted to SQL at the end ?

Or as the other person mentioned it is converted to scala.

[Yuanjian]

Python API have an extra layer in runtime, which leverage local socket to transfer data. So it might have some performance gap due to the transformation, but should not large for most of scenarios.

[Adamatg]

Quite a bit performance depending upon where you are running

[Kaniz]

Hi @Abdullah Durrani​ , We haven’t heard from you in the last few responses, and I was checking back to see if these suggestions helped you. Or else, If you have any solution, please share it with the community as it can be helpful to others.

[abd]

Hi just did it. Thanks

[JBOCACHICA]

There should not be difference between One or other, at the end, every code should be translated to machine language in orden to run on a computer, it’s possible that the translation process be harder in some cases that others, however, that translation process could be harder for python (some cases) and for SQL (some other cases).

My recomendation is that you use every language for every use case.

SQL as a first option and when you have to process bunch of data on a structured format.

Python when you have certain complexity not supported by SQL.

Regards

[ABHI_CT]

Python is the choice for the ML/AI workloads while SQL would be for data based MDM modeling. Pretty much similar performance with certain assumptions. DB optimization is a must assumption for performance benchmarking.

[Bhew]

Relatively similar performance for simple use cases. Higher end tasks and pythons the better bet

[Katieyangcanada]

I think it all gets converted to scala in the end. Shouldn’t be different.

[abd]

Can you provide any source ?

[Rheiman]

To add on the consideration of UDFs, try to consider using HOFs (Higher Order Functions) whenever possible first as there is a signifcant performance benefit as seen here.