Geoff Genz

Ah, that's unfortunate. Apparently the main performance hit is in concatenating the underlying pandas Series objects, not the DataFrame wrappers. I think I'll need to construct a useful test case...

Okay, I think that's the expected behavior if the `use_extended_dtypes` argument to `query_df` is set to True (which is the default). And yes, that's probably been updated since 0.6.16.

@cwegener I'll take a look -- do you know if there is there a way to fix the index by somehow either recalculating the index of the series that are...

@cwegener The change is somewhat more subtle than that. Yes, the complexity is still the same, I was just hoping that Series concat was more efficient than DataFrame concat, but...

I did add a limited test to test small blocks, using the ClickHouse setting `max_block_size`, and my observations are: - The change in `0.7.2` actually does improve performance by about...

This is the first time I've heard that ClickHouse can even support Kerberos over HTTP, so we definitely haven't planned it. We also stopped using the requests library, so we'd...

The library doesn't look too bad so this should be doable directly with urllib3.

Unfortunately there's no way to do this currently using existing Python http libraries and ClickHouse's HTTP 1.1 interface. While intermediate progress headers are returned by ClickHouse, neither the `requests` or...

Thanks for the analysis and including a workaround! I think you're the first person to open an issue about SQLAlchemy that's not related to Superset. I'd honestly recommend against trying...

Thanks for the context -- I don't see sqlalchemy as a requirement for the latest streamlit install, do you happen to know if it's still required or what else I'm...