feat: add cancellable queries

[ritchie46]

This gives you an InProcessQuery handler that allows you to cancel the query

import polars as pl
import time

def sleep(df: pl.DataFrame) -> pl.DataFrame:
    time.sleep(1)
    return df
    
ipq = pl.LazyFrame({"a": [1]}).map_batches(sleep).map_batches(sleep).collect(background=True)

# do some other work in the meantime

# check if query is finished
# Note that if you use python UDFs, you actually have to call this sometimes (or fetch_blocking)
# to give the running process a hold to the GIL. As long as you hold it we cannot run python UDFs.
# So minimize lambdas. :)
out = ipq.fetch()

if out is not None:
     print("query finished", df)

# we can cancel the query if we need to
# note that deleting the `InProcessQuery` object will also cancel it, so keep it alive
if condition():
   ipq.cancel()
    
# do more work

# nothing left to do, let's block the thread until the query is finished
ipq.fetch_blocking()

[ritchie46]

@maartenbreddels I think this should work.

[maartenbreddels]

Nice! I’ll take a look after the hollidays, certainly makes it fit well with solara !

[stinodego]

Perhaps an even clearer parameter name for this would be run_in_background, rather than background. A few characters longer, but immediately obvious.

Maybe the returned object could be named QueryInProcess rather than InProcessQuery - I like having Query be the first/main thing in the name.

[deanm0000]

Suppose you did sleep 60. What's ipq return before a minute is up?

I'm imagining something like

while True:
    # do something else
    if ipq.done()
        break

[ritchie46]

Suppose you did sleep 60. What's ipq return before a minute is up?

fetch will return None if it isn't ready. Note that it doesn't work super nicely with python because of the GIL, but normal polars queries will actually run in the background.

[ritchie46]

Perhaps an even clearer parameter name for this would be run_in_background, rather than background

I feel that that's almost a sentence that belongs in the docstring.