COF-C03 Snowflake SnowPro Core Certification 2026 Exam Free Practice Exam Questions (2026 Updated)

The clearly correct answer from the listed options is A. ARRAY_AGG .

Important correction:

The provided answer A, D is not fully correct if the question is asking about aggregate functions . ARRAY_CONSTRUCT creates an array from explicitly supplied values, but it does not aggregate values across multiple rows.

Why A is correct:

ARRAY_AGG is an aggregate function. It aggregates input values from multiple rows into a single array.

Example:

SELECT ARRAY_AGG(product_id)

FROM orders;

This returns an array containing product_id values from multiple rows.

Why D is not an aggregate function:

ARRAY_CONSTRUCT creates an array from the values passed into the function.

Example:

SELECT ARRAY_CONSTRUCT( ' A ' , ' B ' , ' C ' );

This creates an array from listed expressions, but it does not aggregate values across rows.

Why the other options are incorrect:

B. ARRAY_SLICE returns a subset of an existing array.

C. ARRAY_APPEND appends a value to an existing array.

E. ARRAY_TO_STRING converts an array to a string.

Official Snowflake documentation reference:

Snowflake documentation describes ARRAY_AGG as an aggregate function that returns the input values, pivoted into an array. ARRAY_CONSTRUCT is documented as a scalar function that constructs an array from input expressions.

Snowflake ' s Cloud Services layer is responsible for managing various aspects of the platform that are not directly related to computing or storage. Specifically, it handles authorisation, ensuring that users have appropriate access rights to perform actions or access data. Additionally, it takes care of query parsing and optimization, interpreting SQL queries and optimizing their execution plans for better performance. This layer abstracts much of the platform ' s complexity, allowing users to focus on their data and queries without managing the underlying infrastructure.

The correct answers are A. JSON null values and E. Timestamps within strings .

Important correction:

The provided answer A, B is not the best answer. Snowflake documentation specifically calls out that certain values in semi-structured data may not be extracted into separate columnar storage, which can result in slower queries and increased storage usage. Two commonly tested examples are JSON null values and date or timestamp values stored as strings.

Why A is correct:

When a semi-structured element contains JSON null values, Snowflake might not extract that element into a separate column. This can reduce pruning efficiency and require Snowflake to scan more of the VARIANT structure during query execution.

Why E is correct:

Date and timestamp values inside semi-structured data that are stored as strings may not be optimized in the same way as native typed values. Storing timestamps as strings can increase storage and slow query performance because Snowflake may need to parse or cast the values at query time.

Why the other options are incorrect:

B. Floating-point numbers are supported inside VARIANT; they are not the main documented issue in this context.

C. Arrays are supported in semi-structured data. While very complex nested structures can affect query patterns, arrays alone are not the best answer here.

D. Vectors are a Snowflake data type used for vector operations and are not the standard answer for this semi-structured data optimization question.

Official Snowflake documentation reference:

Snowflake documentation explains that for semi-structured data, elements containing JSON null values may not be extracted into separate columns. It also recommends storing date and timestamp values in native typed columns when possible rather than as strings inside VARIANT, to improve query performance and storage efficiency.

In Snowflake, the default access level for any securable object (such as a table, view, or schema) is " No access " until explicit access is granted. This means that when an object is created, only the owner of the object and roles with the necessary privileges can access it. Other users or roles will not have any form of access to the object until it is explicitly granted.

This design adheres to the principle of least privilege, ensuring that access to data is tightly controlled and that users and roles only have the access necessary for their functions. To grant access, the owner of the object or a role with the GRANT option can use the GRANT statement to provide specific privileges to other users or roles.

For example, to grant SELECT access on a table to a specific role, you would use a command similar to:

GRANTSELECTONTABLEmy_tableTOROLE my_role;

In Auto-scale mode, a multi-cluster virtual warehouse can automatically scale out by starting additional clusters as needed to handle concurrent queries . This is ideal for unpredictable or bursty workloads.

Clusters spin up based on queue length and system load .

They automatically shut down when not in use .

TheMIN_DATA_RETENTION_TIME_IN_DAYSparameter is set at theAccountlevel in Snowflake. This parameter specifies the minimum number of days Snowflake retains the historical data for time travel, which allows users to access and query data as it existed at previous points in time.

Here ' s how to understand and adjust this parameter:

Purpose of MIN_DATA_RETENTION_TIME_IN_DAYS:This parameter is crucial for managing data lifecycle and compliance requirements within Snowflake. It determines the minimum time frame for which you can perform operations like restoring deleted objects or accessing historical versions of data.

Setting the Parameter:Only account administrators can set or modify this parameter. It is done at the account level, impacting all databases and schemas within the account. The setting can be adjusted based on the organization ' s data retention policy.

Adjusting the Parameter:

To view the current setting, use:

SHOWPARAMETERSLIKE ' MIN_DATA_RETENTION_TIME_IN_DAYS ' ;

To change the setting, an account administrator can execute:

ALTERACCOUNTSETMIN_DATA_RETENTION_TIME_IN_DAYS = < number_of_days > ;

The default authentication mechanism for Snowflake clients (including JDBC) is the snowflake method, which uses username and password authentication.

The correct answer is C. Clustering .

Materialized views store precomputed results and can improve query performance. Snowflake also supports clustering with materialized views in applicable scenarios, which can further improve pruning and query performance.

Why C is correct:

A materialized view can be clustered to help optimize access patterns. Clustering can improve performance when queries filter or join on specific columns.

Why the other options are incorrect:

A. Time Travel is not the main supported feature associated with standard materialized views in this question.

B. Fail-safe applies to protected table data storage, not as a feature used directly with standard materialized views for query optimization.

D. Data sharing generally requires secure objects for sharing views. A standard materialized view is not the best answer for this question.

Official Snowflake documentation reference:

Snowflake documentation explains materialized view behavior and supports clustering considerations for materialized views.

The correct answer is D. File-level metadata of all data files in an internal or external stage .

A Snowflake directory table is associated with a stage and stores file-level metadata for files in that stage. It lets users query staged file information using SQL.

Why D is correct:

Directory tables hold metadata about staged files, such as file name, relative path, size, last modified time, and file URL information.

Example concept:

SELECT *

FROM DIRECTORY(@my_stage);

This returns metadata for files in the stage associated with the directory table.

Why the other options are incorrect:

A. Directory tables do not store metadata about views in a schema.

B. Directory tables do not store metadata about regular Snowflake tables in a schema.

C. Directory tables do not store clustering key metadata.

Official Snowflake documentation reference:

Snowflake documentation describes directory tables as objects that store file-level metadata about staged files in cloud storage.

The correct answer is B. Restrict access to Snowflake to known networks using a network policy .

A network policy improves account access security by allowing or blocking access based on IP address ranges. This helps ensure that users can connect to Snowflake only from approved networks.

Why B is correct:

Network policies can restrict inbound access to Snowflake by defining allowed and blocked IP address ranges. This is a direct security control for user access to a Snowflake account.

Example concept:

CREATE NETWORK POLICY corp_network_policy

ALLOWED_IP_LIST = ( ' 203.0.113.0/24 ' );

ALTER ACCOUNT SET NETWORK_POLICY = corp_network_policy;

Why the other options are incorrect:

A. A UDF is not the correct tool for managing account access security. Monitoring security parameters may provide visibility, but it does not directly restrict access.

C. A short session timeout can reduce risk from idle sessions, but making it “very short” can harm usability and is not as direct as network restrictions.

D. Every user already has access to the PUBLIC role by default. Setting restrictive permissions on PUBLIC is good practice, but it is not as direct as restricting access through a network policy.

Official Snowflake documentation reference:

Snowflake documentation describes network policies as security controls that restrict access to Snowflake based on allowed and blocked IP address lists.

The correct answer is A. The table data does not require Fail-safe .

Transient tables are designed for data that does not need the same level of data recovery protection as permanent tables. Unlike permanent tables, transient tables do not have a Fail-safe period.

Why A is correct:

Fail-safe storage can add cost because Snowflake retains historical data for recovery by Snowflake support after the Time Travel retention period ends. Since transient tables do not have Fail-safe, they can reduce storage costs for data that does not require this additional protection.

Common use cases include:

Temporary ETL data

Staging data

Intermediate transformation data

Reproducible data

Non-critical working data

Why the other options are incorrect:

B. Long-term critical storage is usually better suited for permanent tables because they support Fail-safe.

C. Frequent access and modification alone does not make a transient table the best cost choice. The key factor is whether Fail-safe is needed.

D. Sharing data with external accounts is not the primary cost-effective use case for transient tables.

Official Snowflake documentation reference:

Snowflake documentation explains that transient tables persist until explicitly dropped, support Time Travel with limited retention, and do not have a Fail-safe period.

For minimizing the cost of large tables that are used to store transitory data, which does not need to be protected against failures because it can be reconstructed outside of Snowflake, the best table type to use isTransient. Transient tables in Snowflake are designed for temporary or transitory data storage and offer reduced storage costs compared to permanent tables. However, unlike temporary tables, they persist across sessions until explicitly dropped.

Why Transient Tables:Transient tables provide a cost-effective solution for storing data that is temporary but needs to be available longer than a single session. They have lower data storage costs because Snowflake does not maintain historical data (Time Travel) for as long as it does for permanent tables.

Creating a Transient Table:

To create a transient table, use theTRANSIENTkeyword in the CREATE TABLE statement:

CREATETRANSIENTTABLEmy_transient_table (...);

Use Case Considerations:Transient tables are ideal for scenarios where the data is not critical, can be easily recreated, and where cost optimization is a priority. They are suitable for development, testing, or staging environments where data longevity is not a concern.

To temporarily bypass an active network policy by configuring the user object propertyMINS_TO_BYPASS_NETWORK_POLICY, the minimum set of privileges required is having the ACCOUNTADMIN role. This role has the necessary privileges to make such changes, including modifying user properties that affect network policies.

Snowflake Documentation: Network Policy Management

The correct answer is B. Use the CREATE TABLE AS SELECT, or CTAS, command .

A transient table cannot be directly converted into a permanent table by changing a table property. To create a permanent version of the data, create a new permanent table from the transient table data.

Why B is correct:

A CREATE TABLE AS SELECT statement can create a new permanent table from the data in the transient table. Since permanent is the default table type unless otherwise specified, omitting TRANSIENT creates a permanent table.

Example:

CREATE TABLE permanent_table AS

SELECT *

FROM transient_table;

Why the other options are incorrect:

A. ALTER TABLE ... RENAME only changes the table name. It does not change table type.

C. ALTER TABLE ... SWAP WITH swaps table names and metadata between two existing tables, but it is not the method to convert a transient table into a permanent table.

D. CREATE TABLE ... CLONE preserves the source table type behavior in this context and is not the recommended answer for converting transient data into a permanent table.

Official Snowflake documentation reference:

Snowflake documentation explains that the table type cannot be changed after table creation. To change from transient to permanent, create a new permanent table and copy the data, such as with CTAS.

In addition to performing all the standard steps to share data, theUSAGEprivilege must be granted on each database referenced by a secure view in order to be shared. When sharing a database or specific objects like secure views, the receiving account needs to have theUSAGEprivilege on the database and schema to access the shared data. Thisprivilege enables the receiving account to access the database and its schemas but does not allow for any DML operations. It ' s a prerequisite for accessing any objects within the database.

For a secure view to be part of a share, not only does the view itself need to be shared, but the underlying database (and schema, if applicable) must also be accessible to the recipients. GrantingUSAGEprivilege on the database ensures that the receiving account can access the database in a read-only mode to utilize the shared view.

In Snowflake, shares are used to share data across different Snowflake accounts securely. When you create a share, you can include various database objects that you want to share with consumers. According to Snowflake ' s documentation, the types of objects that can be shared include tables, secure views, secure materialized views, and streams. Secure functions and stored procedures are not shareable objects. Tasks also cannot be shared directly. Therefore, the correct answers are streams (C) and tables (D).

To share a stream or a table, you use theGRANTstatement to grant privileges on these objects to a share. The syntax for sharing a table or stream involves specifying the type of object, the object name, and the share to which you are granting access. For example:

GRANTSELECTONTABLEmy_tableTOSHARE my_share;GRANTSELECTONSTREAM my_streamTOSHARE my_share;

These commands grant the SELECT privilege on a table namedmy_tableand a stream namedmy_streamto a share namedmy_share. This enables the consumer of the share to access these objects according to the granted privileges.

Fail-safe is a non-configurable feature in Snowflake that provides an additional layer of data protection beyond Time Travel. Time Travel allows users to access historical data within a configurable period (up to 90 days), while Fail-safe provides an additional 7-day period during which Snowflake retains historical data to recover from significant data loss or corruption incidents. This period is not accessible by users but can be used by Snowflake support to assist in data recovery efforts.

TheJAROWINKLER_SIMILARITYfunction in Snowflake returns an integer between 0 and 100, indicating the similarity of two strings based on the Jaro-Winkler similarity algorithm. This function is useful for comparing strings and determining how closely they match each other.

Understanding JAROWINKLER_SIMILARITY:The Jaro-Winkler similarity metric is a measure of similarity between two strings. The score is a number between 0 and 100, where 100 indicates an exact match and lower scores indicate less similarity.

Usage Example:To compare two strings and get their similarity score, you can use:

SELECTJAROWINKLER_SIMILARITY( ' string1 ' , ' string2 ' )ASsimilarity_score;

Application Scenarios:This function is particularly useful in data cleaning, matching, and deduplication tasks where you need to identify similar but not identical strings, such as names, addresses, or product titles.

When unloading data using the COPY INTO < location > command, Snowflake offers options to compress files. The default behavior is governed by the COMPRESSION parameter. Setting COMPRESSION = AUTO allows Snowflake to automatically use its default compression algorithm, which is GZIP for most formats like CSV.

The correct answer is C. Auto-ingest Snowpipe .

Auto-ingest Snowpipe continuously loads data from staged files as soon as notifications indicate that new files are available in cloud storage. It is designed for near-real-time ingestion from external stages.

Corrected wording:

A clearer version of the question is:

What can load data from an external stage shortly after the data arrives?

Why C is correct:

Auto-ingest Snowpipe uses cloud messaging notifications to detect when new files arrive in an external stage. Snowpipe then automatically loads the new files into a target Snowflake table using the COPY statement defined in the pipe.

Why the other options are incorrect:

A. A scheduled COPY command with a task can load data periodically, but it does not automatically load files right after arrival unless the schedule happens to align.

B. The Snowflake Kafka Connector streams data from Kafka topics into Snowflake, but the question specifically asks about loading from an external stage.

D. An external table lets Snowflake query data in external storage, but it does not load the data into Snowflake tables.

Official Snowflake documentation reference:

Snowflake documentation describes Snowpipe auto-ingest as a feature that loads data from staged files automatically after receiving event notifications from cloud storage.

A join operator producing dramatically more rows than its input tables usually indicates a join explosion, commonly caused by a missing, incomplete, or incorrect join condition. This can result in a Cartesian product where rows from one table are matched with many unrelated rows from another table.

The primary factor that determines the cost of using a virtual warehouse in Snowflake isD. The length of time the compute resources in each cluster run. Snowflake ' s pricing model for compute usage is based on the concept of Snowflake credits, which are consumed based on the time virtual warehouses are running and the size of the warehouses. The more compute resources are utilized and the longer they run, the more credits are consumed, which directly impacts the cost.

Snowflake ' s Streams feature is specifically designed for change data capture (CDC). A stream records insert, update, and delete operations performed on a table, and allows users to query these changes. This enables actions to be taken on the changed data, facilitating processes like incremental data loads and real-time analytics. Streams provide a powerful mechanism for applications to respond to data changes in Snowflake tables efficiently.

The correct answer is A. VARCHAR .

A clustering key can be defined on supported table columns or expressions to help Snowflake organize micro-partitions and improve pruning for certain query patterns. Standard scalar data types such as VARCHAR are supported.

Why A is correct:

VARCHAR is a standard scalar data type and can be used in a clustering key.

Example:

CREATE TABLE customer_orders (

customer_id VARCHAR,

order_date DATE,

amount NUMBER

)

CLUSTER BY (customer_id);

Why the other options are incorrect:

B. VARIANT stores semi-structured data. Clustering directly on raw semi-structured columns is not the best answer here; clustering keys should use supported scalar columns or expressions.

C. OBJECT is a semi-structured data type and is not the correct choice for a clustering key column in this question.

D. GEOGRAPHY is not the correct supported clustering key data type in this context.

Official Snowflake documentation reference:

Snowflake documentation explains that clustering keys can be defined using table columns or expressions, and standard scalar columns such as string, number, and date/time columns are commonly used for clustering.

A VARIANT column in Snowflake is specifically designed to store semi-structured data, such as JSON, and allows for the creation of a data hierarchy. Unlike string columns, VARIANT columns can natively handle JSON data structures, enabling complex querying and manipulation of hierarchical data using functions designed for semi-structured data.

Snowflake Documentation: VARIANT Data

Stored procedures in Snowflake can be created with either ' owner ' s rights ' or ' caller ' s rights ' . A stored procedure created with caller ' s rights executes with the privileges of the role that calls the procedure, not the privileges of the role that owns the procedure. This is particularly useful in scenarios where the procedure needs to perform operations that depend on the caller ' s access permissions, ensuring that the procedure can only access objects that the caller is authorized to access.

The correct answer is D. ON_ERROR = ABORT_STATEMENT .

ON_ERROR = ABORT_STATEMENT stops the load operation if an error is encountered. This helps avoid loading only part of a dataset when a file or row error occurs.

Why D is correct:

If the goal is to avoid a partial dataset, the load should fail rather than continue after errors. ABORT_STATEMENT prevents Snowflake from loading only the valid rows while skipping invalid ones.

Why the other options are incorrect:

A. FORCE = FALSE prevents reloading files that were already loaded, but it does not prevent partial loads caused by data errors.

B. RETURN_FAILED_ONLY affects command output reporting. It does not control whether partial data is loaded.

C. LOAD_UNCERTAIN_FILES = FALSE controls loading files with uncertain load status, but it is not the primary parameter for avoiding partial datasets caused by load errors.

D. ON_ERROR = ABORT_STATEMENT is the correct parameter because it stops the load when an error occurs.

Official Snowflake documentation reference:

Snowflake documentation for COPY INTO < table > describes ON_ERROR = ABORT_STATEMENT as the option that aborts the load operation when an error is encountered.