Add arrow_reader_clickbench benchmark

[alamb]

Which issue does this PR close?

• Closes arrow_reader_row_filter benchmark doesn't capture page cache improvements #7460
• Part of [EPIC] Faster performance for parquet predicate evaluation for non selective filters #7456

Rationale for this change

We are trying to improve the performance of row filter application in the Parquet arrow reader and part of that is a benchmark that we can use to guide optimization efforts.

Mostly I want to be able to approve improvements to the reader that will not regress other queries.

However, as discussed in #7428 the arrow_reader_row_filter microbenchmark doesn't currently reflect the actual performance we see in our end to end application (DataFusion).

cargo bench --all-features --bench arrow_reader_row_filter

Thus, we think we need to create a benchmark that uses the actual ClickBench dataset with appropriate filtering

• See arrow_reader_row_filter benchmark doesn't capture page cache improvements #7460 for more details

What changes are included in this PR?

1. Adds a new arrow_reader_clickbench benchmark

This benchmark tests applying the actual clickbench filters (and column materialization) to

1. hits_1.parquet (one of the data files in ClickBench)
2. async and sync readers
3. All ClickBench query and materialization patterns

If we find additional discrepancies in performance we can increase the benchmark further.

Are there any user-facing changes?

New benchmark, and hopefully thus improved filter / projection performance, no actual code hanges

TODO

• Change String types to use Utf8View
• Add sync/async reader
• Add hits_partitioned / hits
• Complete other predicate types

[alamb]

This benchmark is now looking pretty nice -- it tests just the parquet reading and has all the query predicate patterns. Tomorrow I need to finish adding all the other query patterns and give it a final polish.

[alamb]

I have all the query patterns now, but some of the queries fail because the predicates refer to the wrong columns. I will work on fixing that.

[alamb]

I ran this several times on @zhuqi-lucas 's PRs and all in all it is looking quite good and I think will be a useful addition

• Poc for adaptive parquet predicate pushdown(bitmap/range) with page cache(3 data pages) #7454 (comment)

[zhuqi-lucas]

LGTM, thank you @alamb and the result is also reasonable for
#7454 (comment)

Because, the result for me here is compared the Unified select PR with the main branch(And no parquet filter pushdown).

#7454 (comment)

So when we improve most of the regression for filter push down compared to no pushdown, it may also cause some regression to the original default push down, we can improve it further.

And the sync is no change because we still don't implement the sync version for the improvement PR. The improvement is for async.

[zhuqi-lucas]

Very good! Thank you @alamb

[alamb]

@zhuqi-lucas or @Dandandan or @etseidl , could I possibly trouble one of you for a review of this PR?

I think it is ready for a review

[Dandandan]

Suggested change

	let mut indicices = vec![];
	let mut indices = vec![];

[alamb]

Thank you -- fixed in b9782b6

[alamb]

Thank you @Dandandan 🙏

I'll plan to merge this tomorrow unless anyone else would like more time to review. I am also happy to make changes in subsequent PRs as well

[alamb]

let's gogogogogo

[etseidl]

Sorry @alamb, I intended to review, but between vacation and pestilence contracted while on vacation 😷 I haven't had the bandwidth 😢. Hopefully I'll be back on line this week 🤞

[alamb]

Sorry @alamb, I intended to review, but between vacation and pestilence contracted while on vacation 😷 I haven't had the bandwidth 😢. Hopefully I'll be back on line this week 🤞

No worries and hope you feel better soon!