Question

A data engineer notices that the first execution of a heavy analytical query each morning takes 40 seconds and scans nearly all micro-partitions from remote storage. Subsequent runs of the same query during the day complete in about 6 seconds. The warehouse has AUTO_SUSPEND set to 60 seconds and is idle overnight. The query itself is not eligible for the result cache because it references CURRENT_TIMESTAMP(). Which explanation BEST accounts for the slow first-of-the-day run compared to later runs?

Accepted Answer

The warehouse's local disk (SSD) cache is empty after the warehouse suspends and resumes, so the first run must read data from remote storage; later runs reuse cached micro-partitions.

Answer

The result cache expired after 24 hours, forcing the first run to recompute while later runs are served from the result cache.

Answer

Pruning does not work on the first run of the day because the metadata cache is rebuilt only after the warehouse resumes.

Answer

The first run spills to remote storage because the warehouse starts at a smaller size and auto-scales up for later runs.