Performance Optimization, Querying, and Transformation
21%
Data Collaboration
10%
Free COF-C03 sample questions
The first questions of the free diagnostic. Pick your answer, then check it. Why every option is right or wrong, and the What-If twist, open in the workspace.
Question 1 · Data Loading, Unloading, and Connectivity
One vendor file in last night's load had 3 malformed rows out of 10,000, and the business must have every valid row in the table by morning without hand-editing files or rerunning the load. This command loaded the folder:
SQL
COPY INTO raw.orders
FROM @orders_stage/2026/
FILE_FORMAT = (TYPE = 'CSV' SKIP_HEADER = 1)
ON_ERROR = 'CONTINUE';
What happens to that file's rows?
AThe whole file is skipped and none of its rows load
BThe malformed rows load with NULL in the bad columns
CThe valid rows load and the malformed rows are skipped
DThe COPY statement fails and no file from the stage loads
Check the answer
Answer: C. ON_ERROR controls COPY's behavior on bad data: ABORT_STATEMENT, SKIP_FILE variants, or CONTINUE, which skips only the failing rows.
A bad UPDATE ran on ORDERS about half an hour ago, and an analyst must see the earlier rows without restoring or cloning anything. The table keeps the default retention, and the check must finish before the next load at 11:00. They run:
SQL
-- the bad UPDATE ran at about 10:00; it is now 10:30
SELECT *
FROM orders
AT(OFFSET => -60*30);
What does the query return?
AThe rows as they existed half an hour before the query ran
BOnly the rows that changed since then
CThe table as it was thirty hours earlier
DAn error, because OFFSET accepts only a timestamp
Check the answer
Answer: A. AT and BEFORE query historical data by TIMESTAMP, OFFSET in seconds, or STATEMENT id within the retention period.
Question 3 · Snowflake AI Data Cloud Features & Architecture
Nightly staging data can always be rebuilt from source files, and several downstream sessions read it during the night. The team must stop paying for protection storage on it without losing cross-session access, yet the bill still shows Fail-safe bytes for this table:
Snowflake architecture: storage, compute, cloud services and the object hierarchy
Storage, compute and cloud services scale independently, so adding a warehouse for a new team never copies data or slows the warehouses that other teams already run.
Micro-partitions are immutable, so updates and deletes write new micro-partitions while the old ones stay available to Time Travel until the retention period ends.
The cloud services layer parses and optimizes queries, enforces access control and stores metadata, which also lets some queries such as COUNT(*) return without a running warehouse.
Transient and temporary tables have no Fail-safe, so they cost less for rebuildable staging data but cannot be recovered by Snowflake Support after Time Travel ends.
Edition choice decides features such as multi-cluster warehouses on Enterprise and Tri-Secret Secure or account failover on Business Critical, so read edition clues in every scenario.
Virtual warehouses: sizing, scaling out, auto-suspend and per-second billing
Scaling up to a larger warehouse helps one complex query, while scaling out with more clusters helps many concurrent queries that are waiting in the queue.
Per-second billing has a minimum charge each time a warehouse starts, so suspending after a few seconds of idle time can add cost for workloads that resume constantly.
The STANDARD scaling policy starts extra clusters as soon as queries queue, while ECONOMY waits until it expects enough work to keep a new cluster busy.
Maximized mode sets the same minimum and maximum cluster count, so every cluster runs the whole time the warehouse is running and is billed continuously.
Snowpark-optimized warehouses provide much more memory per node, which suits model training and memory-heavy Python code that a standard warehouse would spill to storage.
Access control: role hierarchy, privileges, ownership, future grants and secondary roles
Privileges are granted to roles, never directly to users, and users receive roles; access then follows the role hierarchy upward to every role that inherits it.
A custom role that is not granted to SYSADMIN leaves its objects outside the system administrators' reach, which is a common cause of privilege errors.
Querying a table needs USAGE on its database and schema and SELECT on the table, plus USAGE on a warehouse to run the query at all.
Future grants cover objects created later in a schema or database, while a plain GRANT ON ALL TABLES only covers the tables that exist at that moment.
Secondary roles add their privileges to queries in the session, but objects created in the session are always owned by the primary role.
Authentication and network security: SSO, SCIM, key pairs, OAuth, MFA and network policies
SAML single sign-on authenticates people through the company identity provider, while SCIM provisioning creates, updates and removes Snowflake users and roles from that provider.
Key-pair authentication uses an RSA public key stored on the user, and a second public key slot lets a service rotate keys without any interruption.
Authentication policies can require MFA enrollment and limit authentication methods or client types, which turns MFA from a recommendation into an enforced rule for people.
Network policies list allowed and blocked network rules, and blocked entries always win, so an address in both lists is refused even though it is allowed.
A network policy on a user overrides the account-level policy for that user, so the most specific policy is the one Snowflake enforces.
Data governance: masking and row access policies, tags, tokenization, encryption and access history
A masking policy decides per query whether a column shows its real value, so one table can serve both privileged and restricted roles without separate copies.
Tag-based masking attaches the policy to a tag, so any column tagged as sensitive is masked automatically, including columns created after the policy was set.
A row access policy filters rows with a Boolean expression, commonly joining a mapping table, so each role sees only the regions or tenants it may read.
Tri-Secret Secure combines a customer-managed key with Snowflake's key, so the customer can make data unreadable by revoking its own key in the cloud provider.
ACCESS_HISTORY records the columns that queries read and wrote, which supports audits of who touched sensitive data and lineage of data movement.
COPY INTO keeps load metadata per table, so rerunning the same command skips files that already loaded and avoids duplicates unless FORCE is set to TRUE.
Snowpipe uses serverless compute triggered by event notifications or REST calls, so files load within minutes of arrival without any warehouse that the team must manage.
Snowpipe Streaming writes rows directly through the SDK or the Kafka connector with no staged files, which suits low-latency streams of small records.
VALIDATION_MODE reports errors without loading any data, but it cannot be used when the COPY statement transforms columns during the load.
A storage integration delegates authentication to a cloud identity, so external stages never contain secret keys that could leak through stage definitions.
The passing score for SnowPro Core Certification is 750 / 1000.
Which domains does COF-C03 cover?
The official exam guide lists 5 domains: Snowflake AI Data Cloud Features & Architecture (31%); Account Management and Data Governance (20%); Data Loading, Unloading, and Connectivity (18%); Performance Optimization, Querying, and Transformation (21%); Data Collaboration (10%).
Is there a free COF-C03 practice test?
Yes. Gyan AI's free diagnostic is 10 COF-C03 questions with no sign-up, each with an explanation for every option and a link to the official documentation. A free account unlocks the full question set and study notes.
Are these the real COF-C03 exam questions?
No. Real exam content is confidential under vendor exam agreements, and using leaked "dumps" can get a certification revoked. Gyan AI questions are original and written to the published skills outline.