QueryChat(engine, *_raw) takes too long to run

[jnhyeon]

Thank you for creating and sharing this wonderful library!

The *_raw table in my DB is about 10GB in size.
As soon as I run QueryChat(engine, *_raw), the read rate on the D drive, where the DB is stored, goes up to 100% for about ten minutes.

If I'm doing something wrong, please let me know.
If not, please help me resolve this.

[gadenbuie]

Thanks @jnhyeon. Can you first try using the latest version of QueryChat from GitHub to see if the latest version fixes this.

My suspicion is that the read is caused by querychat's logic for describing the dataset in the system prompt. If the latest version returns the qc object quickly, try calling print(qc.system_prompt) to trigger the evaluation of the system prompt logic. If that reproduces the issue, we'll at least start narrowing down potential problems.

Another thing you could try would be to turn on query logging via log_destination, possibly in combination with log_min_duration_statement to only log log-running queries. (See https://www.postgresql.org/docs/current/runtime-config-logging.html#GUC-LOG-MIN_DURATION_STATEMENT). That would likely point directly at a problematic statement.

[jnhyeon]

Thank you for your reply.

I installed the latest QueryChat from GitHub, but it still takes a long time to execute on a large DB table.

So, I'm trying to limit the date range using Shiny's input_date_range, write a query with those dates, save it as a data frame, and then pass that data frame to QueryChat. I'm not sure if this is a good approach.
(I raised this question in Issue #184.)

Successfully installed querychat-0.3.0.9000

[cpsievert]

Hi @jnhyeon, this seems like a great case for the polars support coming soon in #191. That is, you could add additional filter operations to the .df(), then .collect() when you need to realize the data in a render.DataGrid()

[jnhyeon]

Thank you for your reply.
I will implement this with polars in the upcoming release.