Increase the search conditions to achieve the purpose of narrowing the search scope, but the result is that more data is parsed, and the query becomes slower

[MrSsunlight]

system: centos7
version: 20.3.11.97

The default.test_log table is a MergeTree() table,
The default.rule table is the Dictionary() table.

Query sql statement：

WITH 
	toDateTime('2020-08-05 00:00:00') As start_time, 
	toDateTime('2020-08-05 23:59:59') As end_time 
SELECT 
	tnow, 
	if(ipType = 0, IPv4NumToString(ipv4), IPv6NumToString(ipv6)) AS ip 
FROM default.test_log cl 
WHERE 
	cl.tnow >= start_time AND 
	cl.tnow <= end_time AND 
	ip in (
		SELECT 
			ra.addr 
		FROM default.rule ra 
		WHERE 
			(ra.is_deleted == 0 or ra.is_deleted == 1) and (ra.name = 'test')
	)
LIMIT 0, 10

The difference between the two query sql is that one more condition (ra.name ='test') is added；

set send_logs_level='trace'

A: Filter condition does not contain (ra.name='test')：

<Debug> InterpreterSelectQuery: MergeTreeWhereOptimizer: condition "(if(ipType = 0, IPv4NumToString(ipv4), IPv6NumToString(ipv6)) AS ip) IN ((SELECT ra.addr FROM default.rule AS ra WHERE (ra.is_deleted = 0) OR (ra.is_deleted = 1)) AS _subquery9)" moved to PREWHERE
<Trace> AccessRightsContext (admin): Access granted: SELECT(addr, is_deleted) ON default.rule
<Trace> AccessRightsContext (admin): Access granted: SELECT(addr, is_deleted) ON default.rule
<Trace> AccessRightsContext (admin): Access granted: SELECT(tnow, ipType, ipv4, ipv6) ON default.test_log
<Debug> default.test_log (SelectExecutor): Key condition: unknown, (column 0 in [1596556800, +inf)), (column 0 in (-inf, 1596643199]), and, and, unknown, and
<Debug> default.test_log (SelectExecutor): MinMax index condition: unknown, (column 0 in [1596556800, +inf)), (column 0 in (-inf, 1596643199]), and, and, unknown, and
<Debug> default.test_log (SelectExecutor): Selected 4 parts by date, 4 parts by key, 87898 marks to read from 4 ranges
<Trace> default.test_log (SelectExecutor): Reading approx. 1437378474 rows with 12 streams
<Trace> InterpreterSelectQuery: FetchColumns -> Complete
<Trace> CreatingSetsBlockInputStream: Creating set. 
<Trace> InterpreterSelectQuery: FetchColumns -> Complete
<Debug> CreatingSetsBlockInputStream: Created. Set with 477382 entries from 477382 rows. In 0.497 sec.
<Information> executeQuery: Read 903366 rows, 20.78 MiB in 1.300 sec., 695092 rows/sec., 15.99 MiB/sec.
<Debug> MemoryTracker: Peak memory usage (for query): 146.56 MiB.

B: Filter conditions include (ra.name ='test'):

<Debug> InterpreterSelectQuery: MergeTreeWhereOptimizer: condition "(if(ipType = 0, IPv4NumToString(ipv4), IPv6NumToString(ipv6)) AS ip) IN ((SELECT ra.addr FROM default.rule AS ra WHERE ((ra.is_deleted = 0) OR (ra.is_deleted = 1)) AND (ra.name = 'test')) AS _subquery10)" moved to PREWHERE
<Trace> AccessRightsContext (admin): Access granted: SELECT(addr, name, is_deleted) ON default.rule
<Trace> AccessRightsContext (admin): Access granted: SELECT(addr, name, is_deleted) ON default.rule
<Trace> AccessRightsContext (admin): Access granted: SELECT(tnow, ipType, ipv4, ipv6) ON default.test_log
<Debug> default.test_log (SelectExecutor): Key condition: unknown, (column 0 in [1596556800, +inf)), (column 0 in (-inf, 1596643199]), and, and, unknown, and
<Debug> default.test_log (SelectExecutor): MinMax index condition: unknown, (column 0 in [1596556800, +inf)), (column 0 in (-inf, 1596643199]), and, and, unknown, and
<Debug> default.test_log (SelectExecutor): Selected 4 parts by date, 4 parts by key, 87898 marks to read from 4 ranges
<Trace> default.test_log (SelectExecutor): Reading approx. 1437378474 rows with 12 streams
<Trace> InterpreterSelectQuery: FetchColumns -> Complete
<Trace> CreatingSetsBlockInputStream: Creating set. 
<Trace> InterpreterSelectQuery: FetchColumns -> Complete
<Debug> CreatingSetsBlockInputStream: Created. Set with 1 entries from 1 rows. In 0.644 sec.
<Debug> MergeTreeReadPool: Slow read, event №1: read 1048576 bytes in 1.081 sec., 0.970 MB/s.
<Debug> MergeTreeReadPool: Slow read, event №2: read 1048576 bytes in 1.687 sec., 0.622 MB/s.
<Debug> MergeTreeReadPool: Will lower number of threads to 11
<Information> executeQuery: Read 1412403247 rows, 27.63 GiB in 34.523 sec., 40912481 rows/sec., 819.54 MiB/sec.
<Debug> MemoryTracker: Peak memory usage (for query): 126.37 MiB.

The difference between A and B is CreatingSetsBlockInputStream: Created. Set with * entries from * rows ；

The number of results of '(ra.is_deleted == 0 or ra.is_deleted == 1) and (ra.name = 'test')' is much smaller than that of '(ra.is_deleted == 0 or ra.is_deleted == 1)', But the 'Created. Set entries' of A needs more than the query of B.<Debug> CreatingSetsBlockInputStream: Created. Set with 477382 entries from 477382 rows. In 0.497 sec and <Debug> CreatingSetsBlockInputStream: Created. Set with 1 entries from 1 rows. In 0.644 sec；

What is the logic of CreatingSetsBlockInputStream: Created. Set with 1 entrie ? Why I added a conditional restriction but got more data?