Are there query patterns which improve / worsen performance?

[shimanoeu]

Question: Are certain query patterns that would help improve arquero's performance (or conversely make it worse in which case we should avoid)?

Example: Let's say I can to compute some basic stats about 2 metrics:

Would arquero perform better if I did each metric separately such as:

dt.rollup({
    min:  d => op.min(d.metric_1),
    max:  d => op.max(d.metric_1),
    sum:  d => op.sum(d.metric_1),
    avg:  d => op.average(d.metric_1),
    med:  d => op.median(d.metric_1)
})

dt.rollup({
    min:  d => op.min(d.metric_2),
    etc...

Or would arquero perform better if combined them both in the same rollup?

dt.rollup({
    metric_1_min:  d => op.min(d.metric_1),
    metric_2_min:  d => op.min(d.metric_2),
    etc...
})

Background research: I looked for performance and optimiz... in the forum/docs and found nothing about query optimization:

• https://github.com/uwdata/arquero/discussions?discussions_q=performance = 0 results
• https://github.com/uwdata/arquero/discussions?discussions_q=optimiz = 0 results
• https://github.com/uwdata/arquero / https://uwdata.github.io/arquero/: "to run performance benchmarks"
• https://uwdata.github.io/arquero/api/: "Arquero unpacks columns with null entries or containing multiple record batches to optimize query performance"

About performance benchmarks: to me these serve a different purpose. They are meant to show how fast arquero performs certain tasks, not how to optimize performance for particular queries.

Any tips would be much appreciated: not criticizing arquero nor implying it's slow, just trying to get the most out of it 😃 Many thanks!