Exam Focus · Part 1
Three layers that scale apart: storage keeps one copy, warehouses compute, cloud services coordinates. Name the layer and the answer follows.
Know these cold
| Topic | Remember |
|---|---|
| Architecture | Multi-cluster shared data: shared-disk's one copy plus shared-nothing's parallel compute |
| Cloud services layer | Optimizer, security, metadata and the result cache |
| Micro-partitions | 50–500 MB uncompressed, columnar, immutable, created automatically |
| Pruning and metadata | Min/ |
| Result cache | 24 h, reset on reuse, max 31 days. Shared across warehouses; role needs access |
| Local disk cache | Per-warehouse SSD, lost on suspend; a downsize drops part of it |
| Billing | Storage: compressed TB/ |
| Enterprise unlocks | Multi-cluster, 90-day Time Travel, MVs, SOS, QAS, masking, row access |
| Business Critical unlocks | Tri-Secret Secure, private connectivity, failover/ |
Traps
- Myth Suspending or resizing the warehouse clears the result cache.Fact It lives in cloud services. Changed data ends it.
- Myth Micro-partitions are a fixed 16 MB.Fact Size varies and you can't set it. You never define partitions on a standard table.
- Myth VPS is the on-premises option.Fact There's no on-premises Snowflake. VPS is a dedicated, isolated cloud environment.
- Myth Enterprise can extend Fail-safe.Fact It's a fixed 7 days on every edition, and only Support recovers from it.
Before you turn the page
- Explain multi-cluster shared data in one sentence.
- Name two events that end the result cache and two that don't.
- Name the lowest edition for multi-cluster warehouses, and the lowest for Tri-Secret Secure.

