Study COF-C03 Query Performance, Caching, and Pruning: key concepts, common traps, and exam decision cues.
Performance questions are where many candidates waste points by jumping straight to “bigger warehouse.” Snowflake wants you to decide whether the real problem is compute, scan volume, repeated-result reuse, or idle-cost behavior.
Pruning: Snowflake behavior that avoids scanning irrelevant micro-partitions when the data layout and filters allow it.
| Symptom | Better first thought |
|---|---|
| repeated identical query with unchanged conditions | result cache or cache reuse |
| too much data scanned | pruning and data layout |
| more users or queries at once | warehouse concurrency or compute capacity |
| warehouse burning credits while unused | auto-suspend and auto-resume |
| If the clue is… | Better reading |
|---|---|
| “same query again” | check whether reuse or caching is relevant |
| “scans large volumes for filtered query” | think pruning first |
| “many concurrent users” | think compute or warehouse behavior |
| “high credit spend during idle periods” | think suspend and resume settings |
Performance questions usually punish reflexive warehouse scaling. Snowflake wants evidence-driven performance reasoning, not bigger-compute by default.
| Scenario | Better first move |
|---|---|
| repeated identical query under stable conditions | check cache reuse |
| filtered query scans large volumes | inspect pruning and layout |
| many sessions compete at once | inspect concurrency and compute |
| cost rises while activity is low | inspect auto-suspend and auto-resume |
| Trap | Better rule |
|---|---|
| more compute fixes every slow query | compute does not solve every scan problem |
| repeated-result question means larger warehouse | repeated-result clues often point to cache reuse |
| warehouse-spend clue means performance tuning | sometimes the problem is idle-cost behavior, not speed |