Open lilinghai opened 3 years ago
revenue1
is a view that expands to a big join + agg. Total dataset sizes about 750GB. It is expected to have a large memory footprint for such a query.
We plan to have memory control but that will wait for 1 or 2 sprints.
Lower the priority to major.
Changing type to enhancement.
CH-benCHmark 10k warehouses data and tiflash replica is 2 in master cluster Execute the ap query
select count(*) from revenue1;
with one thread, The Tiflash memory raise about 130GB. If execute the sql with 2 threads, The Tiflash crash for oom.