7.2 Last-Week Review Map

7.2 Last-Week Review Map

The final week is for consolidation, not new material. Stop opening new courses. Instead, walk the five exam domains in weight order, and for each one lock in the single fact the exam tests most. SnowPro Core weights the domains roughly as follows (use these as study-effort guides, not exact counts):

Lead with whichever high-weight domain your error log flags as weak. A 5% weakness in the 31%-weighted Architecture domain costs more than mastery of the 10% Collaboration domain.

Architecture and storage recap

Three-layer architecture (most-tested). Snowflake separates: (1) the cloud services layer (the brain — authentication, metadata, query optimization, infrastructure management, access control); (2) the query processing / compute layer, made of virtual warehouses; and (3) the centralized (database) storage layer, where data sits compressed in micro-partitions on cloud object storage. The layers scale independently — this decoupling of storage from compute is Snowflake's defining design.

Virtual warehouse scaling — know the two axes:

• Scale UP / DOWN (resize): change warehouse size (XS → S → M → L …). Each size up roughly doubles credits per hour and compute power. Use it for complex, slow individual queries.
• Scale OUT / IN (multi-cluster): add clusters (min/max) to handle concurrency — many users at once. Available on Enterprise edition and above. Modes: Maximized (all clusters run) vs Auto-scale (clusters spin up/down on demand).

Micro-partitions and pruning. Tables auto-divide into micro-partitions of ~50–500 MB of uncompressed data, stored columnar and compressed. Snowflake keeps metadata (min/max, distinct counts) per micro-partition and uses query pruning to skip partitions that cannot match a filter. There are no user-managed indexes. Clustering keys can co-locate related data to improve pruning on very large tables.

Editions ladder (recap): Standard → Enterprise (multi-cluster warehouses, up to 90-day Time Travel, materialized views) → Business Critical (HIPAA/PCI, customer-managed keys/Tri-Secret Secure) → Virtual Private Snowflake (VPS) (isolated environment).

Governance, performance, loading, and sharing recap

RBAC role hierarchy (security domain). Privileges are granted to roles, and roles are granted to users (and to other roles, forming a hierarchy). System roles, top to bottom: ORGADMIN (organization-level) — ACCOUNTADMIN (top account role; encapsulates SYSADMIN + SECURITYADMIN) — SECURITYADMIN (manages grants) — USERADMIN (creates users/roles) — SYSADMIN (creates warehouses/databases) — PUBLIC (default, every user). Best practice: do daily work in a custom role, not ACCOUNTADMIN.

Time Travel and Fail-safe (memorize the numbers):

• Time Travel: query/restore historical data. Standard = up to 1 day; Enterprise+ = 0 to 90 days (set via DATA_RETENTION_TIME_IN_DAYS). Powers AT/BEFORE, UNDROP, and zero-copy CLONE.
• Fail-safe: a fixed, non-configurable 7-day window after Time Travel ends, for Snowflake-assisted recovery only — you cannot query it. Transient and temporary tables have NO Fail-safe (and max 1-day Time Travel).

Loading — COPY INTO vs Snowpipe. COPY INTO does bulk batch loads using a user-managed warehouse. Snowpipe does continuous, near-real-time micro-batch loads (serverless, auto-triggered by cloud events or REST API). Result cache returns identical query results for 24 hours at zero compute cost if the underlying data and query are unchanged.

Secure data sharing. Sharing exposes selected objects to other accounts with no data copy and no storage cost to the consumer — the consumer queries the provider's live data using their own compute. The Marketplace and Data Exchange are built on this. Cross-region/cloud sharing requires replication.

How to spend the last week

• Days 1–3: hit your two weakest high-weight domains with focused review + short sets.
• Days 4–5: short mixed sets (20–30 items across all domains) to keep domain-switching sharp.
• Day 6: one full timed exam; review every miss.
• Day 7: light recap of this map, logistics, sleep. Add no new resources.

High-yield facts most often missed

Beyond the per-domain lock-in facts, a handful of details recur on the exam and trip up candidates. Drill these flashcard-style in the final week:

• Caching layers. Snowflake has (1) the result cache (cloud services layer, 24-hour reuse of identical query results, no compute), (2) the metadata cache, and (3) the local disk / warehouse cache (SSD on the warehouse, lost when it suspends). Know which survives a suspend: the result cache persists; the warehouse cache does not.
• Auto-suspend / auto-resume. Warehouses bill per second with a 60-second minimum each time they resume. Auto-suspend (default often 600s) controls idle cost; auto-resume restarts on the next query.
• Stages. User stage (@~), table stage (@%table), and named internal stage (@stage) hold files for loading; external stages point at S3/Azure/GCS.
• File formats & semi-structured data. VARIANT stores JSON/Avro/Parquet; query with : path and FLATTEN. COPY INTO uses a FILE FORMAT object.
• Cloning is zero-copy and metadata-only at creation; it relies on Time Travel and shares micro-partitions until data diverges.
• Zero-copy + Time Travel together power UNDROP TABLE, point-in-time restore, and dev/test clones at no storage duplication cost.

What to deliberately skip

Resist the urge to memorize obscure function syntax or every parameter default. The exam rewards conceptual command of architecture, security, performance, loading, and sharing far more than trivia. If a topic is not in the five domains or this review map, it is low-yield in the final week. Calm, targeted consolidation beats frantic breadth — protect your sleep the night before, because a rested brain reads stems more accurately than a crammed one.