[DocDB] Utilize Bloom filters for multi-row operations

[mbautin]

Jira Link: DB-9397

Description

In the context of discussing a perf regression related to Bloom filters not being checked, @ttyusupov mentioned:

The result is that there were two separate breaking changes. First of them made index table bloom filters to be not checked: https://github.com/yugabyte/yugabyte-db/commit/beeebbe873bef97561427bc13f4906d607340ba9 And the second one made also main table bloom filters to be not checked: https://github.com/yugabyte/yugabyte-db/commit/7b582dd6edf1d764292ca69d8bcb323f5e8c5db4

Related issue: #20398

Test case from @ttyusupov and @kmuthukk:

\timing on
drop table if exists test_table;
create extension if not exists pgcrypto;
create table test_table(k text, v text, PRIMARY KEY(k ASC));
create index index_v_test_table on test_table(v);

-- Load 10M rows
do $$
begin
  for counter in 1..100 loop
    raise notice 'counter: %', (counter * 10000);
    insert into test_table (select gen_random_uuid(), gen_random_bytes(50)::text
                        from generate_series(1, 10000) i);
    commit;
  end loop;
end $$;

and then

curl -s http://localhost:9000/prometheus-metrics | grep -E "test_table" | grep -wE "rocksdb_bloom_filter_useful|rocksdb_bloom_filter_checked|rocksdb_block_cache_data_hit|rocksdb_block_cache_data_miss"  | sed 's/expo.*9000..//' | sed 's/[0-9]*$//' | tr " " "\n"

Issue Type

kind/bug

Warning: Please confirm that this issue does not contain any sensitive information

• [X] I confirm this issue does not contain any sensitive information.

[mbautin]

Here is a graph of performance of the above test as a function of the number of Bloom filter keys per group:

[rthallamko3]

@mbautin , Can we close this. Were you planning to do additional things for this?