A workload runs
dbt, Fivetran, Snowpark, Dataform, a Dynamic Table, Cortex, Looker, Tableau. APEX reads how it ran. It does not stand in front of the query.
The workload runs first. APEX reads that run and recommends one action. Your team approves anything that changes production. Then the same workload is measured again.
dbt, Fivetran, Snowpark, Dataform, a Dynamic Table, Cortex, Looker, Tableau. APEX reads how it ran. It does not stand in front of the query.
Run, Optimize, Redirect, or Review. One choice, with the reason and the owner who can change it.
The change goes through the review you already use. APEX does not apply it.
Same workload. Did the cost fall, and does the performance requirement still hold?
One pass
The numbers are illustrative. They are not a customer result. The same shape applies to a Fivetran sync, an agent run, or a Looker query.
A dbt model on Snowflake, every hour, on an X-Large warehouse. About 1,540 credits a month. Eight minutes. No queue. No spill.
The same runs were checked on Large: about 11 minutes, still one query at a time, still no queue and no spill. The requirement is 45 minutes.
Use a Large warehouse. Cutting the warehouse in half did not double the runtime.
APEX recommends the smaller warehouse. Data engineering approves it. APEX does not change the warehouse.
The four answers
Leave the current path. It has earned its cost.
Same requirement, less compute. The team ships the change.
Name a better path, such as reserved capacity. The team makes the move.
Too new, or too unclear, to schedule. A person decides.