diff --git a/tests/queries/0_stateless/00002_log_and_exception_messages_formatting.reference b/tests/queries/0_stateless/00002_log_and_exception_messages_formatting.reference
new file mode 100644
index 00000000000..75dd0426f4a
--- /dev/null
+++ b/tests/queries/0_stateless/00002_log_and_exception_messages_formatting.reference
@@ -0,0 +1,13 @@
+10	1
+20	1
+30	1
+40	1
+50	1
+60	1
+70	1
+80	1
+90	1
+100	1
+110	1
+120	1
+130	1
diff --git a/tests/queries/0_stateless/00002_log_and_exception_messages_formatting.sql b/tests/queries/0_stateless/00002_log_and_exception_messages_formatting.sql
new file mode 100644
index 00000000000..45862bb1717
--- /dev/null
+++ b/tests/queries/0_stateless/00002_log_and_exception_messages_formatting.sql
@@ -0,0 +1,51 @@
+-- Tags: no-parallel
+-- no-parallel because we want to run this test when most of the other tests already passed
+
+-- If this test fails, see the "Top patterns of log messages" diagnostics in the end of run.log
+
+drop table if exists logs;
+create view logs as select * from system.text_log where now() - toIntervalMinute(120) < event_time;
+system flush logs;
+
+
+-- Check that we don't have too many messages formatted with fmt::runtime or strings concatenation.
+-- 0.001 threshold should be always enough, the value was about 0.00025
+select 10, (sum(length(message_format_string) = 0) / count()) < 0.001 from logs;
+
+-- Check the same for exceptions. The value was 0.03
+select 20, (sum(length(message_format_string) = 0) / count()) < 0.05 from logs where message like '%DB::Exception%';
+
+-- Check that we don't have too many short meaningless message patterns.
+select 30, countDistinct(message_format_string) < 10 from logs where length(message_format_string) < 10;
+
+-- Same as above. Feel free to update the threshold or remove this query if really necessary
+select 40, countDistinct(message_format_string) < 35 from logs where length(message_format_string) < 16;
+
+-- Same as above, but exceptions must be more informative. Feel free to update the threshold or remove this query if really necessary
+select 50, countDistinct(message_format_string) < 90 from logs where length(message_format_string) < 30 and message ilike '%DB::Exception%';
+
+
+-- Avoid too noisy messages: top 1 message frequency must be less than 30%. We should reduce the threshold
+select 60, (select count() from logs group by message_format_string order by count() desc limit 1) / (select count() from logs) < 0.30;
+
+-- Same as above, but excluding Test level (actually finds top 1 Trace message)
+select 70, (select count() from logs where level!='Test' group by message_format_string order by count() desc limit 1) / (select count() from logs) < 0.16;
+
+-- Same as above for Debug
+select 80, (select count() from logs where level <= 'Debug' group by message_format_string order by count() desc limit 1) / (select count() from logs) < 0.08;
+
+-- Same as above for Info
+select 90, (select count() from logs where level <= 'Information' group by message_format_string order by count() desc limit 1) / (select count() from logs) < 0.04;
+
+-- Same as above for Warning
+select 100, (select count() from logs where level = 'Warning' group by message_format_string order by count() desc limit 1) / (select count() from logs) < 0.0001;
+
+-- Same as above for Error (it's funny that we have 100 time less warnings than errors)
+select 110, (select count() from logs where level = 'Warning' group by message_format_string order by count() desc limit 1) / (select count() from logs) < 0.001;
+
+-- Avoid too noisy messages: limit the number of messages with high frequency
+select 120, count() < 3 from (select count() / (select count() from logs) as freq, message_format_string from logs group by message_format_string having freq > 0.10);
+select 130, count() < 10 from (select count() / (select count() from logs) as freq, message_format_string from logs group by message_format_string having freq > 0.05);
+
+
+drop table logs;