SnowPro Core Practice Questions (2026): 10 Free Snowflake COF-C03 Questions with Answers

The cloud services layer is the brain of Snowflake: it handles authentication, infrastructure and metadata management, query parsing and optimization, access control, and transactions, all on Snowflake-managed compute. The storage layer (A) holds compressed, columnar micro-partitions. The compute layer (B) is the virtual warehouses that execute queries. There is no standalone result-cache layer (D) — the result cache lives inside cloud services.

Each T-shirt size up doubles the compute and therefore doubles the credits per hour: XS = 1, S = 2, M = 4, L = 8, XL = 16. So Small to Medium goes 2 to 4 credits/hour — a doubling. Concurrency (A) is handled by multi-cluster warehouses, not by a single size step. One step does not quadruple (C) or halve (D). Auto-suspend and auto-resume keep a warehouse from burning credits while idle.

Snowflake automatically divides table data into micro-partitions of about 50–500 MB of uncompressed data, stored columnar and compressed, and they are immutable — updates write new micro-partitions rather than editing existing ones. Snowflake keeps per-partition metadata (min/max values, distinct counts) that powers partition pruning. There is no manual PARTITION BY (A), no fixed 1 GB size (C), and no VACUUM step (D) — reclustering is automatic.

Snowflake's result cache, in the cloud services layer, holds query results for 24 hours and returns them with no virtual warehouse — and therefore no compute credits — when the query text is identical, the underlying data is unchanged, and the role has access. No warehouse rescan (A) is needed. Results persist for 24 hours, not 5 minutes (C). The result cache is free, not billed as storage (D). This is distinct from the warehouse's local SSD cache and the metadata cache.

For a very large table where the natural load order does not match the filter column, defining a clustering key (here EVENT_DATE) lets Snowflake's automatic clustering service reorganize micro-partitions so queries prune far more data. Snowflake has no user-managed secondary indexes (A). Temporary tables (C) change a table's lifespan, not its clustering. A permanently oversized warehouse (D) throws compute at a data-layout problem and wastes credits.

FLATTEN is a table function that explodes an array or nested object inside a VARIANT into one row per element; combined with a LATERAL join it produces a row per line item that you reference through the VALUE column and dot/colon notation. PARSE_JSON (C) only converts a string into a VARIANT — it does not un-nest arrays. PIVOT (B) rotates rows into columns, the opposite of what is needed. GROUP BY (D) aggregates rows; it cannot expand them.

Snowpipe provides continuous, serverless micro-batch loading: cloud event notifications tell Snowpipe when new files arrive, and it loads them using Snowflake-managed compute billed per second — no warehouse to size. Bulk COPY INTO (A) is batch-oriented and ties up a user warehouse. PUT (C) only uploads files to an internal stage; it does not load them into a table. An external table (D) queries files in place but never ingests them into Snowflake storage.

USERADMIN is dedicated to creating and managing users and roles (it holds CREATE USER and CREATE ROLE). SECURITYADMIN sits above USERADMIN and manages object grants globally. ACCOUNTADMIN (A) is the top-level superuser that encapsulates SECURITYADMIN and SYSADMIN and should be used sparingly. SYSADMIN (B) creates warehouses, databases, and other objects. PUBLIC (D) is the default role granted to every user and carries no administrative rights.

A network policy defines allowed (and blocked) IP ranges and can be applied at the account or user level, restricting where connections may originate. MFA (B) strengthens who signs in but does not limit the source network. A resource monitor (C) caps credit usage, not access. A masking policy (D) is column-level governance for hiding values — unrelated to network access.

Secure Data Sharing exposes selected objects through a share; the consumer queries live data with no copy and no extra storage cost, paying only for the compute they use, while the provider controls access through privileges. Exporting files (A) creates stale copies and storage overhead. Replication (C) duplicates data across accounts. A zero-copy clone (D) is a metadata-only, point-in-time copy inside your own account (like Time Travel and Fail-safe protection) — not a cross-account sharing mechanism.