Grammarly wrote this excellent blog post on building ETL pipelines in Clojure. I've primarily written ETL pipelines in Python and wanted to see how a similar pipeline would compare in Python.

As of writing this I'm Clojure curious, but a noob so largely lifted the Clojure code from the Grammarly post. There are many aspects of Clojure that appeal to me, and I was optimistic that I would finish this exercise convinced Clojure would have definite advantages to writing ETL jobs over Python. However, the results are more mixed than expected.

$ lein run

--- Time how we do for a single file ---

Lazy:
"Elapsed time: 946.714266 msecs"

Parallel:
"Elapsed time: 840.418467 msecs"

--- What if we run on 10 files ---

Lazy:
"Elapsed time: 8832.775676 msecs"

Parallel:
"Elapsed time: 2281.054373 msecs"

$ python3 python/pipeline.py

--- Time how we do for a single file ---

Lazy:
Elapsed time: 900.0811576843262 msecs

Parallel:
Elapsed time: 995.0668811798096 msecs

--- What if we run on 10 files ---

Lazy:
Elapsed time: 9346.51803970337 msecs

Parallel:
Elapsed time: 2756.990909576416 msecs

I expected Clojure to be order of magnitudes faster than Python and was surprised that is only marginally faster in most cases. Why is this? Is there something in the Clojure that is making it much less efficient than it could be?

The Clojure program was slightly smaller which is a plus and the general composbility in a more functional paradigm is nice. However, in my current context this is not compelling enough to consider switching writing pipelines in Clojure vs Python. Is there something I'm missing? Are there compelling reasons I'm missing?