connector-x VS polars

This was originally a blocker, however, we managed to set up a multi-stage Docker build to build from source. Here is the Github issue where we, along with community members, managed to solve it.

There's packages like connector-x and polars that do a lot of what you're mentioning out of the box. I used these two to massively speed up an SQLalchemy + Pandas based ETL in the past as well.

Thanks for sharing connector-x, I will also start to use it. I wonder if there are a list of tools like that. I know Ruff, Polars, pydantic-core.

I was trying https://github.com/sfu-db/connector-x and hacking around with this https://github.com/spitz-dan-l/postgres-binary-parser but it turned out that a COPY to csv using asyncpg and then converting to parquet was the fastest.

if you store the OHLC data in a relational database, use connector-x to load the data into pandas dataframe

For SQL reading I'd really recommend connector-x, they do a great job preventing unneeded serialization and don't have to go through python.

Indeed, currently we do not support persistent connections among different queries. We target more on the bulk loading scenario where the bottleneck is caused by the data size and the connection construction overhead is negligible. However, one possible solution to the problem is to expose our connection pool object that we use inside Rust to users, so the next call could reuse the same pool. We do not plan for this yet, but happy to see whether this is a common need! Feel free to open an issue in our github repo: https://github.com/sfu-db/connector-x

Feel free to ask any questions here or open an issue in our github repo: https://github.com/sfu-db/connector-x . You can also join our discord community: https://discord.com/invite/xwbkFNk and ask question under connector channel!

This is because 0.1 is in actuality the floating point value value 0.1000000000000000055511151231257827021181583404541015625, and thus 1 divided by it is ever so slightly smaller than 10. Nevertheless, fpround(1 / fpround(1 / 10)) = 10 exactly.

I found out about this recently because in Polars I defined a // b for floats to be (a / b).floor(), which does return 10 for this computation. Since Python's correctly-rounded division is rather expensive, I chose to stick to this (more context: https://github.com/pola-rs/polars/issues/14596#issuecomment-...).

https://github.com/pola-rs/polars/releases/tag/py-0.19.0

That turned out to be related to pola-rs/polars#11912, and this linked comment provided a deceptively simple solution - use PARSE_DECLTYPES when creating the connection: