Hi team, we recently upgraded our shared-nothing cluster from StarRocks 3.3.18 to StarRocks 3.4.8, and saw an increase in CPU usage of about 20%. Any thoughts on why we see this performance change? I have noticed that

skip_pk_preload

is enabled by default in 3.4.8, and have disabled it again, but saw no difference. EDIT: So there were two contributing factors to this: • Starrocks 3.4.8 started using a different query plan for some SQL with lots of

NEST_LOOP_JOIN

operations. I tweaked my sql to replace subqueries with user-defined-variable hints and that solved the problem • colocation groups were put into

isStable: false

state, which broke some colocated joins