Using MinMax skip index to improve partition reading

MinMax.md

create table test(tenant_id UInt32,  ts1 DateTime, timestamp DateTime)
engine=MergeTree
partition by toDate(ts1)
PRIMARY KEY (tenant_id, toStartOfHour(timestamp))
ORDER BY (tenant_id, toStartOfHour(timestamp),  timestamp)
as select 1, now() - number/10 x, x from numbers(1e8);

insert into test select number%1000, now() - number/10 x, 0 from numbers(1e8);

optimize table test final;

query reads 989.14 thousand rows / all partitions: 117 /117 parts by primary key

select count() from test 
where tenant_id =1 
AND `timestamp` BETWEEN toDateTime (1686020502) AND toDateTime (1686020562);
┌─count()─┐
│     610 │
└─────────┘
Selected 117/117 parts by partition key, 117 parts by primary key, 121/24422 marks by primary key, 121 marks to read from 117 ranges

1 row in set. Elapsed: 0.026 sec. Processed 989.14 thousand rows, 7.91 MB (38.19 million rows/s., 305.55 MB/s.)

if we use ts1 -- partition_by column, it reads 49.15 thousand rows / 1 parts by primary key

select count() from test 
where tenant_id =1 
AND `timestamp` BETWEEN toDateTime (1686020502) AND toDateTime (1686020562) 
and ts1>=toStartOfDay(toDateTime(1686020502));
┌─count()─┐
│     610 │
└─────────┘
Selected 1/117 parts by partition key, 1 parts by primary key, 5/156 marks by primary key, 5 marks to read from 1 ranges

1 row in set. Elapsed: 0.017 sec. Processed 49.15 thousand rows, 589.82 KB (2.94 million rows/s., 35.31 MB/s.)

let's add skip index on timestamp column

alter table test add index x1 timestamp type minmax GRANULARITY 1000;
alter table test materialize index x1 settings mutations_sync=2;

The original query even without ts1 reads only 40.96 thousand rows / 1 parts by primary key

select count() from test 
where tenant_id =1 
AND `timestamp` BETWEEN toDateTime (1686020502) AND toDateTime (1686020562);
┌─count()─┐
│     610 │
└─────────┘
Index `x1` has dropped 116/121 granules.
Selected 1/117 parts by partition key, 1 parts by primary key, 5/156 marks by primary key, 5 marks to read from 1 ranges

1 row in set. Elapsed: 0.032 sec. Processed 40.96 thousand rows, 327.68 KB (1.27 million rows/s., 10.16 MB/s.)

Huge GRANULARITY 1000 -- allows to reduce impact on insertion, the index is very light, and very small, but it works anyway.