Add clickbench_pushdown benchmark

[alamb]

Which issue does this PR close?

• Related to of Enable parquet filter pushdown (filter_pushdown) by default #3463
• Closes Add a datafusion benchmark for filter_pushdown #16729

Rationale for this change

In order to enable filter_pushdown by default, we need to ensure it doesn't regress existing performance

However, it has been very hard to make forward progress on improving filter pushdown because all our benchmarks compare filter pushdown to not filter pushdown, so the bar for change is quite high.
Here is the most recent example:

• POC: Test DataFusion with experimental Parquet Filter Pushdown (try 4) #16711

It seems obvious but the the right metric for improvements to the filter pushdown are comparing when filter pushdown is already on. However, we don't have any such benchmark (see #16729 and #16730 for why the existing benchmarks are not good enough)

What changes are included in this PR?

Add a benchmark (clickbench_pushdown) that turns on filter_pushdown and reorder_filters on

You can run it like this:

`./benchmarks/bench.sh run clickbench_pushdown

Which then invokes

+ cargo run --release --bin dfbench -- clickbench --pushdown --iterations 5 --path /Users/andrewlamb/Software/datafusion/benchmarks/data/hits_partitioned --queries-path /Users/andrewlamb/Software/datafusion/benchmarks/queries/clickbench/queries -o /Users/andrewlamb/Software/datafusion/benchmarks/results/alamb_new_filter_pushdown/clickbench_partitioned.json

Are these changes tested?

I tested it manually . You can see Q30 increase in time when --pushdown is enabled, as expected

with --pushdown:

...     Running `target/profiling/dfbench clickbench --pushdown --iterations 5 --path /Users/andrewlamb/Software/datafusion/benchmarks/data/hits_partitioned --queries-path /Users/andrewlamb/Software/datafusion/benchmarks/queries/clickbench/queries -o /Users/andrewlamb/Software/datafusion/benchmarks/results/alamb_new_filter_pushdown/clickbench_partitioned.json --query 30`
Running benchmarks with the following options: RunOpt { query: Some(30), pushdown: true, common: CommonOpt { iterations: 5, partitions: None, batch_size: None, mem_pool_type: "fair", memory_limit: None, sort_spill_reservation_bytes: None, debug: false }, path: "/Users/andrewlamb/Software/datafusion/benchmarks/data/hits_partitioned", queries_path: "/Users/andrewlamb/Software/datafusion/benchmarks/queries/clickbench/queries", output_path: Some("/Users/andrewlamb/Software/datafusion/benchmarks/results/alamb_new_filter_pushdown/clickbench_partitioned.json") }
Q30: -- Must set for ClickBench hits_partitioned dataset. See https://github.com/apache/datafusion/issues/16591
-- set datafusion.execution.parquet.binary_as_string = true

SELECT "SearchEngineID", "ClientIP", COUNT(*) AS c, SUM("IsRefresh"), AVG("ResolutionWidth") FROM hits WHERE "SearchPhrase" <> '' GROUP BY "SearchEngineID", "ClientIP" ORDER BY c DESC LIMIT 10;

Query 30 iteration 0 took 546.0 ms and returned 10 rows
Query 30 iteration 1 took 503.1 ms and returned 10 rows
Query 30 iteration 2 took 488.2 ms and returned 10 rows
Query 30 iteration 3 took 462.6 ms and returned 10 rows
Query 30 iteration 4 took 462.3 ms and returned 10 rows
Query 30 avg time: 492.42 ms

Without pushdown

...
     Running `target/profiling/dfbench clickbench --iterations 5 --path /Users/andrewlamb/Software/datafusion/benchmarks/data/hits_partitioned --queries-path /Users/andrewlamb/Software/datafusion/benchmarks/queries/clickbench/queries -o /Users/andrewlamb/Software/datafusion/benchmarks/results/alamb_new_filter_pushdown/clickbench_partitioned.json --query 30`
Running benchmarks with the following options: RunOpt { query: Some(30), pushdown: false, common: CommonOpt { iterations: 5, partitions: None, batch_size: None, mem_pool_type: "fair", memory_limit: None, sort_spill_reservation_bytes: None, debug: false }, path: "/Users/andrewlamb/Software/datafusion/benchmarks/data/hits_partitioned", queries_path: "/Users/andrewlamb/Software/datafusion/benchmarks/queries/clickbench/queries", output_path: Some("/Users/andrewlamb/Software/datafusion/benchmarks/results/alamb_new_filter_pushdown/clickbench_partitioned.json") }
Q30: -- Must set for ClickBench hits_partitioned dataset. See https://github.com/apache/datafusion/issues/16591
-- set datafusion.execution.parquet.binary_as_string = true

SELECT "SearchEngineID", "ClientIP", COUNT(*) AS c, SUM("IsRefresh"), AVG("ResolutionWidth") FROM hits WHERE "SearchPhrase" <> '' GROUP BY "SearchEngineID", "ClientIP" ORDER BY c DESC LIMIT 10;

Query 30 iteration 0 took 305.7 ms and returned 10 rows
Query 30 iteration 1 took 289.1 ms and returned 10 rows
Query 30 iteration 2 took 287.7 ms and returned 10 rows
Query 30 iteration 3 took 266.3 ms and returned 10 rows
Query 30 iteration 4 took 268.3 ms and returned 10 rows
Query 30 avg time: 283.43 ms

Are there any user-facing changes?

No this is a development process change only

[zhuqi-lucas]

LGTM, thank you @alamb !

This is very helpful for pushdown case performance monitor!

[alamb]

I tested this benchmark with our filter pushdown work here, and I think it is useful

#16711 (comment)

Thank you @zhuqi-lucas for the review