Describe what's wrong
In master, parallel_view_processing doesn't have any effect and the MV are executed sequentially.
Does it reproduce on recent release?
Only master.
Taken from parallel_mv.xml perf test:
create table main_table (number UInt64) engine = MergeTree order by tuple();
create materialized view mv_1 engine = MergeTree order by tuple() as
select number, toString(number) from main_table where number % 13 != 0;
create materialized view mv_2 engine = MergeTree order by tuple() as
select number, toString(number) from main_table where number % 13 != 1;
create materialized view mv_3 engine = MergeTree order by tuple() as
select number, toString(number) from main_table where number % 13 != 3;
create materialized view mv_4 engine = MergeTree order by tuple() as
select number, toString(number) from main_table where number % 13 != 4;21.9.4:
- No parallel: Elapsed: 1.333 sec
- With parallel view processing: 0.588 sec
21.10.2.8264:
- No parallel: 1.372 sec
- Parallel: 0.582
21.11.1.8263 from master:
- No parallel: 1.415 sec
- With parallel: 1.423 sec
Additional context
Checking the stats of CI it points to #28582 as the culptrit.
Describe what's wrong
In master,
parallel_view_processingdoesn't have any effect and the MV are executed sequentially.Does it reproduce on recent release?
Only master.
Taken from parallel_mv.xml perf test:
21.9.4:
21.10.2.8264:
21.11.1.8263 from master:
Additional context
Checking the stats of CI it points to #28582 as the culptrit.