SnowPro Core overview

  • 200 practice questions
  • 32 flashcards
  • 5 study guide topics
  • 38 key terms

What you'll study

  • Snowflake AI Data Cloud Features and Architecture20%
  • Account Access and Security20%
  • Performance Optimization, Querying, and Data Transformation20%
  • Data Loading and Unloading20%
  • Data Protection and Data Sharing20%

The SnowPro Core Certification is Snowflake's foundational credential for practitioners who build, secure, and operate workloads on the Snowflake AI Data Cloud. It suits data engineers, analytics engineers, database administrators, data architects, and analysts who already work hands-on with Snowflake and want an objective benchmark of core platform knowledge, and it is commonly used by teams as a baseline before moving on to role-specific specializations. The exam is delivered in multiple-choice format and is organized around five equally listed knowledge domains: Snowflake AI Data Cloud features and architecture; account access and security; performance optimization, querying, and data transformation; data loading and unloading; and data protection and data sharing. Because the blueprint spans both conceptual architecture and day-to-day operational tasks, preparation works best as a blend of guided study and hands-on lab time - creating virtual warehouses, tuning queries, configuring roles and network policies, staging and copying data, and sharing objects between accounts. Note that public details such as item counts, time limits, passing thresholds, pricing, and retake rules were not verifiable from the official Snowflake pages at the time of writing, so this guide deliberately leaves those figures unpublished rather than risking stale numbers. Always confirm logistics with Snowflake directly before booking.

Common mistakes

Treats a clustering key as traditional static partitioning - expects defining one to physically split the table into partitions that must be maintained, and answers pruning questions as if partition creation is a DDL decision.

Most candidates arrive from Oracle/Teradata/Synapse where partitioning is an explicit physical layout chosen at CREATE TABLE time, so they map 'clustering' onto that model.

What to do instead

State the rule explicitly: every table is automatically stored in immutable, columnar micro-partitions of roughly 50–500 MB uncompressed. A clustering key only co-locates similar rows to reduce partition overlap and improve pruning; it creates no partitions and needs no maintenance. Then judge pruning from the Query Profile (partitions scanned vs. total), not from the DDL.

Restarts or resizes a warehouse to make an identical repeated query faster, believing query results are cached per warehouse.

Warehouse-centric mental model from other cloud data warehouses, plus the fact that Snowflake actually has three different caches that are rarely separated in study material.

What to do instead

Separate the three: the result cache lives in the Cloud Services layer at account level (up to 24 hours, reusable by any warehouse if the SQL text is identical and underlying data is unchanged), the local disk cache holds raw micro-partitions on warehouse compute nodes and is lost when the warehouse suspends, and the metadata cache serves statistics and clustering information. A result-cache hit needs no running warehouse at all.

Reads the role hierarchy backwards - expects a parent role such as SYSADMIN to gain privileges that were granted to a custom child role.

The word 'hierarchy' suggests the top role accumulates all power, but Snowflake privilege inheritance flows downward from the role that holds the grant.

What to do instead

Memorize the direction: privileges granted to a role are inherited by every role that has been granted that role. Granting SELECT to a custom role gives SYSADMIN nothing. To let SYSADMIN manage those objects, grant the custom role TO SYSADMIN - i.e., put the custom role underneath SYSADMIN in the tree.

Assumes SECURITYADMIN is the role that creates users and roles, and performs routine object grants as ACCOUNTADMIN because that always works.

The role names sound authoritative, and most tutorials run every statement as ACCOUNTADMIN, which hides the intended separation of duties.

What to do instead

USERADMIN creates and manages users and roles; SECURITYADMIN owns MANAGE GRANTS and administers grants (and inherits USERADMIN); SYSADMIN creates warehouses, databases and other objects; ACCOUNTADMIN is reserved for account-level settings and billing and should not own routine objects. Build a custom role tree and attach it under SYSADMIN.

Declares a query inefficient because it 'scanned millions of rows', without checking partitions scanned versus total or bytes spilled, and then recommends a bigger warehouse.

Row counts dominate EXPLAIN output in other engines; Snowflake's Query Profile reports partitions, bytes and cache percentages, and pruning - not row count - is the dominant cost driver.

What to do instead

Read the profile in a fixed order: partitions scanned vs. total (pruning quality), bytes scanned, percentage scanned from cache, then the operator-level time breakdown and any spilling to local or remote storage. Fix pruning first via clustering or better filter predicates; only then consider warehouse size for concurrency or spill-heavy operators.

Uses LATERAL FLATTEN and expects one output row per input row, or writes JSON paths in the wrong case and gets NULLs back.

Relational intuition says a lateral join preserves row counts, and standard SQL identifiers are case-insensitive - neither holds for a table function exploding VARIANT.

What to do instead

Remember FLATTEN is a table function producing one row per array element or object key; it belongs in the FROM clause with LATERAL and needs an input row to explode, so row counts multiply. Element and key names inside VARIANT are case-sensitive: reference them with the exact stored case (e.g. '$.userId') or the value resolves to NULL.

Writes PUT or GET inside a worksheet, stored procedure, or task and expects files to be staged.

PUT and GET look like ordinary SQL and the editor accepts the text, but they are client-side commands executed by the SnowSQL/connector process, not by the warehouse.

What to do instead

Run PUT/GET from SnowSQL, a driver, or the Snowsight upload control; they fail in worksheets, procedures and tasks. Also recall that a stage holds only a URL and file metadata - the bytes live in the external bucket or on the local filesystem, never inside the stage itself.

Assumes a COPY INTO that encounters a malformed row keeps loading the remaining rows, and skips a dry run before loading production files.

Many bulk loaders default to continue-on-error, so candidates assume Snowflake does too - its default is the opposite and the load stops.

What to do instead

ON_ERROR defaults to ABORT_STATEMENT; the other options are CONTINUE, SKIP_FILE, SKIP_FILE_, and SKIP_FILE_%. Set it deliberately, and validate files first with VALIDATION_MODE = RETURN_ERRORS or RETURN__ROWS so problems surface before any data lands.

Treats FAIL_SAFE as a queryable recovery window and assumes the default Time Travel retention is 90 days on every account.

Marketing-level familiarity with 'Time Travel and Fail-safe' blurs the line between the two, and most hands-on practice happens on trials or Enterprise accounts where longer retention is available.

What to do instead

Time Travel retention defaults to 0 or 1 day and is configurable up to 90 days only on Enterprise Edition and above, at table, schema or database level. FAIL_SAFE is a fixed 7-day period that begins after Time Travel ends, is not queryable, and can only be restored by Snowflake Support - never with SQL.

Believes a share copies data into the consumer account and that the consumer therefore pays for the storage, or that a secure view is always mandatory to share data.

Export/import and file-transfer mental models carry over from other platforms, and older documentation stressed secure views before direct table sharing was available.

What to do instead

A share is metadata plus grants: no data is copied, the provider pays storage and the consumer pays only for the warehouse used to query. Tables, secure views, secure UDFs and Iceberg tables can all be shared; a secure view is needed only when you must filter rows or columns. The consumer creates a database FROM SHARE and needs IMPORTED PRIVILEGES plus a warehouse.

Exam requirements can change. Confirm current details with the exam provider.

SnowPro Core practice

200 questions. Each one lists the choices, the correct answer, and an explanation. With JavaScript, this list opens one question at a time. The checker at the top of this page saves progress in this browser when it is available.

  1. Question 1Snowflake AI Data Cloud Features and Architecture

    Snowflake's architecture is organized into three layers. Which layer handles query parsing, optimization, access control, and metadata management?

    1. The storage layer, where micro-partition statistics are persisted
    2. The virtual warehouse layer, which also performs all metadata management
    3. The hybrid columnar layer, which manages object-level privileges
    4. The cloud services layer Correct answer.

    Explanation

    The cloud services layer is the brain of the account: it authenticates users, enforces access control, stores metadata, parses and optimizes queries, and coordinates transactions. Compute and storage are the other two layers and do not perform these coordination functions.

  2. Question 2Snowflake AI Data Cloud Features and Architecture

    Which factor determines the rate at which a running virtual warehouse consumes credits?

    1. The total number of rows returned by each query
    2. The number of concurrent users connected to the warehouse
    3. The warehouse size Correct answer.
    4. The retention period configured on the queried tables

    Explanation

    Credit consumption per hour is a function of warehouse size: each successive size doubles the credits consumed per hour, from X-Small upward. A suspended warehouse consumes no compute credits regardless of how much data it holds or how many users are assigned to it.

  3. Question 3Snowflake AI Data Cloud Features and Architecture

    Which description best characterizes a micro-partition in Snowflake?

    1. An immutable, compressed, columnar storage unit Correct answer.
    2. A mutable row block that is rewritten in place whenever a row is updated
    3. A fixed 16 MB block that holds exactly one column of a table
    4. A logical view definition cached by the cloud services layer

    Explanation

    Micro-partitions are immutable, compressed, columnar storage units of roughly 50 to 500 MB of uncompressed data that Snowflake creates automatically as rows are loaded. Because they are immutable, updates and deletes mark rows and write new micro-partitions rather than modifying existing ones.

  4. Question 4Snowflake AI Data Cloud Features and Architecture

    In an Enterprise Edition account, what is the maximum Time Travel retention period that can be set for a permanent table?

    1. 7 days
    2. 30 days
    3. 90 days Correct answer.
    4. 24 hours

    Explanation

    Standard Edition allows a Time Travel retention period of 0 to 1 day, while Enterprise Edition and above allow up to 90 days for permanent objects. Fail-safe adds a separate, non-configurable 7-day period after Time Travel expires.

  5. Question 5Snowflake AI Data Cloud Features and Architecture

    During the Fail-safe period, who can recover data that has been deleted or lost?

    1. Only Snowflake support Correct answer.
    2. A user holding the ACCOUNTADMIN role, using UNDROP
    3. Any role that has been granted the FAILSAFE privilege
    4. The table owner, by querying the FAILSAFE schema

    Explanation

    Fail-safe is a non-configurable 7-day disaster-recovery window that begins after Time Travel ends. Customers cannot query data during Fail-safe; only Snowflake support can restore it, so self-service recovery must happen while Time Travel is still active.

  6. Question 6Snowflake AI Data Cloud Features and Architecture

    A table T2 is created with CREATE TABLE T2 CLONE T1. Later, rows are deleted from T1 but T2 is untouched. What is the result?

    1. Snowflake copies every micro-partition of T1 into T2 at clone time
    2. T2 is also emptied, because clones share the same underlying storage
    3. T2 still returns its rows, because it references the original micro-partitions Correct answer.
    4. T2 becomes read-only until it is refreshed from the source table

    Explanation

    Zero-copy cloning copies metadata pointers, not data, so both tables initially reference the same micro-partitions. When either table changes, new micro-partitions are written for the affected rows and the other table continues to see the original data.

  7. Question 7Snowflake AI Data Cloud Features and Architecture

    Which sequence shows the correct containment hierarchy in Snowflake?

    1. Schema, then Database, then Table, then Column
    2. Database, then Schema, then Table, then Column Correct answer.
    3. Database, then Table, then Schema, then Column
    4. Account, then Database, then Table, then Schema

    Explanation

    Tables and views live inside schemas, schemas live inside databases, and databases live inside an account. Columns are contained within tables, making them the finest level of the hierarchy listed.

  8. Question 8Snowflake AI Data Cloud Features and Architecture

    Which prefix references a user's personal stage in commands such as PUT or COPY INTO?

    1. @~ Correct answer.
    2. @#
    3. @%
    4. @&

    Explanation

    Snowflake supports three stage types with distinct reference prefixes: @~ for the user stage, @% for a table stage, and @name for a named stage. The user stage exists automatically for every user and needs no CREATE statement.

  9. Question 9Snowflake AI Data Cloud Features and Architecture

    Queries on a warehouse are queuing behind one another at peak times, and the warehouse is already sized appropriately for the memory needs of each query. Which change most directly addresses the concurrency problem?

    1. Increase the statement timeout so that queued queries wait longer before failing
    2. Convert it to a multi-cluster warehouse with MAX_CLUSTER_COUNT above 1 Correct answer.
    3. Enable the query acceleration service on that warehouse
    4. Lower AUTO_SUSPEND so that idle compute is released back to the pool sooner

    Explanation

    Multi-cluster warehouses add or remove clusters of the same size to serve more concurrent queries, which is the intended remedy for queueing. Warehouse size controls resources per query rather than the number of queries that can run at once.

  10. Question 10Snowflake AI Data Cloud Features and Architecture

    How does the Economy scaling policy differ from the Standard policy for multi-cluster warehouses?

    1. It caps the warehouse at one cluster regardless of configuration
    2. It starts new clusters immediately whenever queries begin to queue
    3. It bills extra clusters at half the normal credit rate
    4. It keeps clusters running longer before shutting them down Correct answer.

    Explanation

    Economy favors keeping already-running clusters alive for several minutes of idle time, which avoids the cost of repeatedly provisioning clusters for fluctuating workloads. Standard starts clusters as soon as queries are queued and shuts them down more aggressively once they are idle.

  11. Question 11Snowflake AI Data Cloud Features and Architecture

    A user re-runs an identical SELECT 30 minutes after the first execution, the underlying data has not changed, and the warehouse has since been suspended. What is the outcome?

    1. Only the local disk cache rebuilds the result set for the user
    2. The result cache returns the output without a running warehouse Correct answer.
    3. The query re-reads every micro-partition from remote storage and rebuilds the result
    4. The query fails until the warehouse is resumed by an administrator

    Explanation

    The result cache holds query results for 24 hours and is served by the cloud services layer, so a suspended warehouse is not required. Results are reused only when the query text matches and the underlying data has not changed, and no compute credits are consumed.

  12. Question 12Snowflake AI Data Cloud Features and Architecture

    What does the local disk cache on a virtual warehouse hold, and when is it cleared?

    1. Micro-partition metadata used for pruning; cleared when the table is reclustered
    2. Query results; cleared every 24 hours regardless of activity
    3. Staged data files; cleared when the stage object is dropped
    4. Micro-partitions read by queries; cleared when the warehouse suspends Correct answer.

    Explanation

    The local disk cache (warehouse cache) holds micro-partitions that the warehouse has read from remote storage, so repeat queries scan locally. It exists only while the warehouse is running and is dropped on suspend, which is why resuming a warehouse can make the first queries slower.

  13. Question 13Snowflake AI Data Cloud Features and Architecture

    A 10 TB table is repeatedly filtered on ORDER_DATE, and pruning currently eliminates very little data from the scan. What is the recommended remedy?

    1. Create a materialized view containing every column of the table
    2. Define ORDER_DATE as the clustering key Correct answer.
    3. Sort the table's rows by ORDER_DATE before each query runs
    4. Increase the warehouse size to improve scan throughput

    Explanation

    A clustering key co-locates rows with similar values into the same micro-partitions, which lets pruning skip far more partitions on the filtered column. Sorting manually is not a Snowflake operation on standard tables, and a larger warehouse only scans the same volume faster.

  14. Question 14Snowflake AI Data Cloud Features and Architecture

    What does a stream object record for its source table?

    1. Only the DDL changes made to the table's schema
    2. A full copy of every row as it existed when the stream was created
    3. Aggregated counts of inserts per micro-partition
    4. Row-level change records with metadata columns Correct answer.

    Explanation

    Streams implement change data capture by exposing inserted, updated, and deleted rows along with metadata columns such as METADATA$ACTION and METADATA$ISUPDATE. Consuming a stream with DML advances its offset so the same changes are not processed twice.

  15. Question 15Snowflake AI Data Cloud Features and Architecture

    In a Snowflake task graph, what causes a child task to run?

    1. Successful completion of its predecessor task Correct answer.
    2. A manual EXECUTE TASK statement issued against the child task
    3. The account-level scheduler firing on its one-minute interval
    4. An INSERT into the stream that feeds the graph

    Explanation

    Child tasks are defined with AFTER to depend on a predecessor and run when that predecessor completes successfully. Only the root task of a graph carries a schedule, and inserting into a stream does not by itself trigger task execution.

  16. Question 16Snowflake AI Data Cloud Features and Architecture

    At which levels can a resource monitor be attached?

    1. Only at the database or schema level, scoped to objects inside them
    2. Only at the individual warehouse level, one monitor per warehouse
    3. Only at the account level, applying to every warehouse in the account
    4. At the account level or on specific warehouses Correct answer.

    Explanation

    A resource monitor can be set at the account level to track all compute credit usage, or attached to individual warehouses to track their consumption separately. Monitors can suspend warehouses or notify when credit thresholds are reached.

  17. Question 17Snowflake AI Data Cloud Features and Architecture

    A provider must share data with a partner organization that does not have a Snowflake account of its own. What should the provider create?

    1. A reader account for the consumer Correct answer.
    2. A share with PUBLIC access enabled so anyone can query it
    3. An external stage mounted by the partner over SFTP
    4. A new copy of the database deployed in the consumer's region

    Explanation

    Reader accounts give consumers without Snowflake accounts a way to query shared data; the provider creates the account and pays for the compute it consumes. Readers can query shared data but cannot load data or create shares themselves.

  18. Question 18Snowflake AI Data Cloud Features and Architecture

    Under what condition are cloud services credits billed to an account?

    1. They are billed at the same rate as compute credits in every case
    2. Only when they exceed 10% of daily compute credits Correct answer.
    3. Whenever a user queries the INFORMATION_SCHEMA views
    4. Only when a resource monitor threshold has been reached

    Explanation

    Snowflake does not charge for cloud services usage that stays below 10% of the total compute credits consumed that day. Any usage above that threshold is billed as cloud services credits.

  19. Question 19Snowflake AI Data Cloud Features and Architecture

    Which compute resources does Snowpipe use to load files into a table?

    1. The cloud services layer, which charges standard compute credits
    2. The virtual warehouse named in the COPY statement
    3. The warehouse assigned to the pipe owner's default role
    4. Snowflake-provided serverless compute Correct answer.

    Explanation

    Snowpipe loads files using serverless compute managed by Snowflake, so no user-managed warehouse needs to be running. Billing is based on the serverless compute time and the number of files loaded.

  20. Question 20Snowflake AI Data Cloud Features and Architecture

    A team needs a table that refreshes itself from a defined query as base data changes, with Snowflake scheduling and performing incremental refreshes automatically. Which object fits this requirement?

    1. A transient table refreshed by a scheduled task
    2. A materialized view over the base table
    3. A dynamic table with a target lag Correct answer.
    4. An external table over the base table's storage

    Explanation

    Dynamic tables declare a query plus a target lag, and Snowflake schedules and performs incremental refreshes to keep the result within that lag. They are suited to multi-step transformation pipelines that a single materialized view cannot express.

  21. Question 21Snowflake AI Data Cloud Features and Architecture

    Which workload is the best fit for the Query Acceleration Service?

    1. Many small point-lookup queries returning a handful of rows
    2. Queries dominated by complex joins across many small tables
    3. Scan-heavy queries with selective filters Correct answer.
    4. Write-heavy ingestion pipelines loading files continuously

    Explanation

    The Query Acceleration Service offloads portions of scan and aggregation work to shared compute, so it benefits queries that scan large volumes with selective filters. It does not help point lookups or workloads dominated by small-table joins and writes.

  22. Question 22Snowflake AI Data Cloud Features and Architecture

    A large table is queried with highly selective equality predicates such as WHERE CUSTOMER_ID = 12345, and normal pruning still scans many micro-partitions. Which feature is designed for this pattern?

    1. The search optimization service Correct answer.
    2. A multi-cluster warehouse configuration
    3. The query acceleration service
    4. A clustering key on an unrelated high-cardinality column

    Explanation

    The search optimization service builds a search access path that lets Snowflake locate matching rows for equality, IN, and substring predicates without scanning every micro-partition. It targets selective point-lookup style queries rather than broad scans.

  23. Question 23Snowflake AI Data Cloud Features and Architecture

    Which statement about hybrid tables is accurate?

    1. They are a specialized form of external table that reads Iceberg files
    2. They can be created in any database and used inside shared databases
    3. They store data in micro-partitions and support Time Travel and cloning
    4. They use row-based storage and enforce primary keys Correct answer.

    Explanation

    Hybrid tables provide row-based storage with primary key and foreign key enforcement and row-level locking, aimed at transactional workloads that need low-latency single-row operations. They are distinct from standard micro-partition tables and are limited by constraints such as not being shareable.

  24. Question 24Snowflake AI Data Cloud Features and Architecture

    Which Snowflake edition is required for a secondary account to perform failover in a replication setup?

    1. Enterprise Edition
    2. Virtual Private Snowflake only
    3. Standard Edition
    4. Business Critical Edition or higher Correct answer.

    Explanation

    Database replication is available from Standard Edition upward, but failover and failback, including redirecting clients to a secondary account, require Business Critical Edition or higher. Snowflake recommends that primary and secondary accounts use the same edition.

  25. Question 25Snowflake AI Data Cloud Features and Architecture

    An Iceberg table is created in Snowflake using an external Iceberg catalog that is not Snowflake Open Catalog. What access does Snowflake provide for that table?

    1. Write access, but only through Snowpipe streaming
    2. Write access, but only when the table has a clustering key
    3. Read-only access to the Iceberg table Correct answer.
    4. Full read-write access with Snowflake-managed metadata

    Explanation

    When the Iceberg catalog is external, Snowflake can read the table's metadata and data files but cannot write to it. Write support requires Snowflake to manage the table's metadata through Snowflake Open Catalog or a Snowflake-managed Iceberg table.

  26. Question 26Snowflake AI Data Cloud Features and Architecture

    Snowflake separates storage from compute. Which component handles authentication, access control, query parsing and optimization, and metadata management?

    1. The remote storage layer
    2. The virtual warehouse layer
    3. The cloud services layer Correct answer.
    4. The local disk cache on warehouse nodes

    Explanation

    The cloud services layer is the brain of the account: it authenticates users, enforces access control, parses and optimizes SQL, and maintains metadata such as micro-partition statistics. It is provisioned by Snowflake and consumes cloud services credits rather than warehouse credits.

  27. Question 27Snowflake AI Data Cloud Features and Architecture

    A permanent table is created without specifying DATA_RETENTION_TIME_IN_DAYS, and the account-level parameter has not been changed. How much Time Travel retention does the table have?

    1. 0 days
    2. 1 day Correct answer.
    3. 90 days
    4. 7 days

    Explanation

    The default DATA_RETENTION_TIME_IN_DAYS for permanent objects is 1 day. Accounts on Enterprise Edition or higher can raise it to as much as 90 days, but only if the parameter is explicitly set.

  28. Question 28Snowflake AI Data Cloud Features and Architecture

    How is credit consumption metered for a running virtual warehouse?

    1. Per minute, with no minimum charge for short queries
    2. Per query, based on the number of rows returned
    3. Per hour, always rounded up to the next full hour
    4. Per second, with a 60-second minimum on each start Correct answer.

    Explanation

    Warehouses are billed on a per-second basis once running, but each time a warehouse starts it is charged for at least 60 seconds. This means very short queries still incur a full minute of credit consumption.

  29. Question 29Snowflake AI Data Cloud Features and Architecture

    When rows in a table are updated or deleted, what happens to the micro-partitions that currently hold those rows?

    1. They are merged with adjacent micro-partitions to keep the table compact
    2. New micro-partitions are written; existing ones are left unchanged Correct answer.
    3. They are updated in place, and partition metadata is recalculated immediately
    4. They are split into smaller micro-partitions to preserve column statistics

    Explanation

    Micro-partitions are immutable. DML operations write new micro-partitions containing the changed rows, and the older partitions are simply no longer referenced by the current table version, though they remain available for Time Travel.

  30. Question 30Snowflake AI Data Cloud Features and Architecture

    A virtual warehouse is resized from X-Small to Medium and runs a full hour at each size. How does credit consumption compare?

    1. Eight times as many credits per hour
    2. Four times as many credits per hour Correct answer.
    3. The same credits per hour, with twice the throughput
    4. Twice as many credits per hour

    Explanation

    Each size increase doubles the number of compute nodes, and credits per hour double with each step. X-Small consumes 1 credit per hour and Medium consumes 4, a fourfold increase across two steps.

  31. Question 31Snowflake AI Data Cloud Features and Architecture

    A team must query files in an S3 bucket from Snowflake without storing AWS access keys in a stage definition. Which Snowflake object enables this?

    1. A secret object referenced by a file format
    2. A storage integration referenced by an external stage Correct answer.
    3. A named internal stage with directory tables enabled
    4. A network policy attached to the warehouse

    Explanation

    A storage integration stores the cloud provider identity and trust configuration so Snowflake can access the bucket without embedding long-lived credentials. An external stage then references that integration along with the URL and file format.

  32. Question 32Snowflake AI Data Cloud Features and Architecture

    Which statement about Fail-safe is accurate?

    1. Fail-safe keeps data for seven days and is reachable only via Support Correct answer.
    2. Users can query Fail-safe data with AT or BEFORE clauses
    3. Fail-safe applies to transient and temporary tables as well as permanent tables
    4. Fail-safe retains data for 14 days after Time Travel ends

    Explanation

    After the Time Travel retention period passes, permanent tables enter a fixed seven-day Fail-safe period. The data can be recovered only by Snowflake Support; customers cannot query it directly.

  33. Question 33Snowflake AI Data Cloud Features and Architecture

    A developer wants to call a large language model from SQL to summarize text stored in a table. Which statement about Snowflake Cortex LLM functions is true?

    1. They run as serverless functions and need no warehouse to execute Correct answer.
    2. They are available only through the Snowpark Container Services runtime
    3. They copy the prompt text to an external model provider account
    4. They require a dedicated warehouse to be provisioned per model

    Explanation

    Cortex LLM functions are managed by Snowflake and execute serverlessly, so no virtual warehouse is required. Prompts and responses stay inside Snowflake's security and governance perimeter.

  34. Question 34Snowflake AI Data Cloud Features and Architecture

    A view is created with the SECURE keyword. What does this change?

    1. The view owner's privileges are no longer used when queried
    2. The view result is cached permanently until the base table changes
    3. The view runs on serverless compute rather than a warehouse
    4. The view definition and internal query details are hidden Correct answer.

    Explanation

    Secure views hide the view definition, the underlying objects, and internal query details from users who do not have sufficient privileges. They also prevent certain optimizer-driven behaviors that could leak information about the source data.

  35. Question 35Snowflake AI Data Cloud Features and Architecture

    Which objects can be added to a share in Snowflake?

    1. Individual tables, views, and schemas chosen one at a time
    2. Databases, which give the consumer access to the objects inside Correct answer.
    3. Warehouses, so the consumer can run queries without its own compute
    4. Stages, so the consumer can read the underlying files directly

    Explanation

    A share is populated by granting privileges on a database, and the consumer then sees the schemas and objects contained in that database. Compute is never shared, so each consumer uses its own warehouse.

  36. Question 36Snowflake AI Data Cloud Features and Architecture

    Which set of languages can be used to build Snowpark client applications that push DataFrame operations into Snowflake?

    1. Python, R, and Go
    2. Python, Java, and Scala Correct answer.
    3. SQL, Python, and Java
    4. Python, JavaScript, and C#

    Explanation

    Snowpark provides client-side APIs for Python, Java, and Scala, letting developers build dataframes and push processing into Snowflake. SQL is a query language rather than a Snowpark client API.

  37. Question 37Snowflake AI Data Cloud Features and Architecture

    Which statement about a masking policy applied to a column is accurate?

    1. A second masking policy can be added to the same column by a role with MANAGE GRANTS on the table
    2. One policy can apply to many columns, but a column has at most one masking policy Correct answer.
    3. The policy is evaluated only when rows are inserted or updated in the column
    4. A masking policy can be applied only to columns whose type is VARCHAR or NUMBER

    Explanation

    A masking policy is a reusable schema-level object that can be attached to many columns, including columns of different tables. Each column, however, can have only one masking policy attached at a time, and the policy is evaluated at query time.

  38. Question 38Snowflake AI Data Cloud Features and Architecture

    Which statement describes Tri-Secret Secure in Snowflake?

    1. It encrypts data three times before it is written to storage
    2. It requires three separate identity providers for login
    3. It combines a Snowflake key with a customer-managed key Correct answer.
    4. It stores three copies of every key in separate regions

    Explanation

    Tri-Secret Secure uses a composite master key made up of a Snowflake-managed key and a key held in the customer's own cloud key management service. Removing the customer key renders the account's data unreadable, and the feature requires Business Critical Edition or higher.

  39. Question 39Snowflake AI Data Cloud Features and Architecture

    In a schema with managed access enabled, which roles can grant privileges on objects in that schema?

    1. The schema owner role, not the owners of the individual objects Correct answer.
    2. Only ACCOUNTADMIN, because object ownership is transferred to it
    3. Any role that owns an object in the schema, as in a normal schema
    4. Any role that has been granted USAGE on the schema by its owner

    Explanation

    Managed access schemas centralize privilege management in the schema owner, who must hold the MANAGE GRANTS privilege globally or on the schema. Object owners no longer control grants on their own objects, which prevents privilege sprawl.

  40. Question 40Snowflake AI Data Cloud Features and Architecture

    How is Snowflake storage usage billed?

    1. By the total uncompressed size of every table, measured once at the end of each billing month
    2. By the average amount of compressed data stored per day, including Time Travel and Fail-safe Correct answer.
    3. By the number of read and write operations the account issues against the storage layer
    4. By the peak volume of data stored on the account on any single day within the billing month

    Explanation

    Storage is metered on the average daily volume of compressed data held in the account, computed from hourly measurements across the month. The total includes table data, Time Travel versions, Fail-safe data, and internal stages.

  41. Question 41Account Access and Security

    Which system-defined role is intended to own the warehouses, databases, and other objects that make up the data warehouse?

    1. SECURITYADMIN
    2. USERADMIN
    3. PUBLIC
    4. SYSADMIN Correct answer.

    Explanation

    SYSADMIN is the system-defined role designed to create warehouses, databases, and other objects, and it owns those objects once created. Security-related duties belong to SECURITYADMIN and USERADMIN, which keeps administrative concerns separated.

  42. Question 42Account Access and Security

    Which role is automatically granted to every user in a Snowflake account and cannot be revoked?

    1. PUBLIC Correct answer.
    2. ACCOUNTADMIN
    3. USERADMIN
    4. SYSADMIN

    Explanation

    PUBLIC is granted to every user and every role in the account, and it cannot be revoked from them. It grants no privileges by default until an administrator explicitly grants privileges to it.

  43. Question 43Account Access and Security

    Snowflake multi-factor authentication is available for accounts that authenticate using which method?

    1. Password-based Correct answer.
    2. SAML federation
    3. OAuth token
    4. Key pair

    Explanation

    MFA is supported for users who sign in with a Snowflake username and password. Users authenticated through federated SAML or key pair authentication rely on the external identity provider or key possession instead.

  44. Question 44Account Access and Security

    Which Snowflake object restricts connections to an account or user based on IP address ranges?

    1. Session policy
    2. Network policy Correct answer.
    3. Resource monitor
    4. Secondary role

    Explanation

    A network policy lists allowed and blocked IP address ranges and can be attached at the account level or to individual users. It is the mechanism Snowflake provides for IP allow-listing and block-listing.

  45. Question 45Account Access and Security

    In key pair authentication, what does the client use to sign each authentication request?

    1. The private key Correct answer.
    2. A one-time passcode
    3. A shared secret string
    4. The public key only

    Explanation

    The user's public key is stored on the Snowflake user object, while the client keeps the private key and uses it to sign the JWT sent during authentication. Snowflake verifies the signature with the stored public key.

  46. Question 46Account Access and Security

    Which privilege on a warehouse allows a role to start a suspended warehouse?

    1. USAGE
    2. OWNERSHIP
    3. MONITOR
    4. OPERATE Correct answer.

    Explanation

    OPERATE allows a role to start, stop, and abort queries on a warehouse. USAGE only lets the role run queries on a running warehouse, and MONITOR only exposes usage statistics.

  47. Question 47Account Access and Security

    A role must be able to pass a privilege it already holds on to another role. Which privilege modifier makes this possible?

    1. TRANSFER OWNERSHIP
    2. WITH GRANT OPTION Correct answer.
    3. FUTURE GRANTS
    4. GRANT ALL

    Explanation

    When a privilege is granted WITH GRANT OPTION, the receiving role can grant that same privilege to other roles. Without it, the role can use the privilege but cannot extend it to anyone else.

  48. Question 48Account Access and Security

    What does a future grant accomplish in Snowflake?

    1. It defers privilege checks until the object is queried
    2. It grants a privilege on all objects that already exist in the database
    3. It grants a privilege on objects created later in the schema Correct answer.
    4. It transfers ownership of new objects to a system role

    Explanation

    A future grant, written with ON FUTURE <object type> IN SCHEMA or IN DATABASE, applies a privilege automatically to objects of that type created afterward. Existing objects are unaffected and must be granted separately.

  49. Question 49Account Access and Security

    In a schema that has managed access enabled, which roles can grant privileges on the objects inside that schema?

    1. Any role that has been granted SELECT on the object
    2. The PUBLIC role
    3. The schema owner and roles with MANAGE GRANTS Correct answer.
    4. Any role that has USAGE on the parent database

    Explanation

    With managed access, privilege granting is centralized: only the schema owner, or a role holding the global MANAGE GRANTS privilege, can grant on objects in the schema. Individual object owners lose the ability to grant on their own objects.

  50. Question 50Account Access and Security

    A user without an authorized role queries a column protected by a masking policy. What does the user see?

    1. An error that blocks the query
    2. Nothing, because the column is dropped from the result
    3. The real value, flagged as sensitive
    4. A masked value in place of the real one Correct answer.

    Explanation

    A masking policy returns an alternate value, such as a fixed string, a hash, or NULL, when the querying role does not meet the policy condition. The query still runs and the column still appears, but its content is transformed.

  51. Question 51Account Access and Security

    A row access policy is attached to a table. What does that policy determine?

    1. Whether the table can be cloned
    2. Which roles can grant privileges on the table
    3. Which columns a role can select
    4. Which rows are visible to a given role Correct answer.

    Explanation

    A row access policy evaluates at query time and filters the rows a role can see, typically based on a mapping table or session context. Column-level protection is handled by masking policies instead.

  52. Question 52Account Access and Security

    Which settings can be enforced through a Snowflake password policy?

    1. The list of allowed IP addresses
    2. Only the session idle timeout in minutes
    3. Password length, age, and lockout limits Correct answer.
    4. Only the number of enrolled MFA devices

    Explanation

    A password policy defines requirements such as minimum length, character-class minimums, minimum and maximum password age, and lockout behavior after failed attempts. It can be set on the account or attached to individual users.

  53. Question 53Account Access and Security

    What does a session policy control?

    1. Idle timeout and UI session persistence Correct answer.
    2. Which IP addresses may connect
    3. Password history requirements
    4. Which roles may sign in

    Explanation

    A session policy sets the idle timeout for sessions and the UI idle timeout, and can control whether the Snowflake UI session stays alive. It addresses session lifetime rather than credentials or network origin.

  54. Question 54Account Access and Security

    Which requirement is addressed by an authentication policy?

    1. Row filtering in secure views
    2. MFA enrollment and allowed client types Correct answer.
    3. Password complexity and reuse requirements
    4. Grants applied to newly created tables

    Explanation

    An authentication policy lets administrators require MFA enrollment and restrict which client types, such as the Snowflake UI, drivers, or connectors, may authenticate. It can be set at the account level or attached to individual users.

  55. Question 55Account Access and Security

    With federated authentication using SAML 2.0, where are user credentials verified?

    1. In the user's default role
    2. In a Snowflake-managed password table
    3. In a Snowflake password policy
    4. In an external identity provider Correct answer.

    Explanation

    With SAML federation, Snowflake redirects the user to an external identity provider, which authenticates the user and returns a signed assertion. Snowflake trusts that assertion through a configured security integration.

  56. Question 56Account Access and Security

    What does a SCIM security integration provide for a Snowflake account?

    1. Kerberos-based single sign-on
    2. Automated user provisioning from an IdP Correct answer.
    3. Encryption of data at rest in the account
    4. Automatic creation of network policies

    Explanation

    SCIM lets an identity provider such as Okta or Microsoft Entra ID create, update, and deactivate Snowflake users and roles automatically. It complements SAML sign-in by automating the lifecycle of accounts and role assignments.

  57. Question 57Account Access and Security

    In an OAuth flow with Snowflake, what is the function of the authorization server?

    1. It encrypts query results at rest
    2. It issues the access tokens Snowflake validates Correct answer.
    3. It replicates the account to a secondary region
    4. It stores each user's private key pair

    Explanation

    In OAuth, an external authorization server authenticates the user and issues an access token, which Snowflake validates against the configured security integration before starting a session. Snowflake never handles the user's credentials in this flow.

  58. Question 58Account Access and Security

    A role has been granted USAGE on a database and USAGE on a schema. What additional privilege is needed to read rows from a table in that schema?

    1. CREATE TABLE on the schema
    2. No further privilege is needed
    3. OWNERSHIP of the schema
    4. SELECT on the table Correct answer.

    Explanation

    Privileges are checked at each level of the containment hierarchy: USAGE on the database and schema lets a role reach the table, and SELECT on the table permits reading its rows. Access is cumulative and each level must be satisfied.

  59. Question 59Account Access and Security

    Role A has been granted to Role B. What is the resulting effect?

    1. Users of Role A must also be granted Role B
    2. Role A inherits all privileges of Role B
    3. Role B inherits Role A's privileges Correct answer.
    4. Role B loses its own existing privileges

    Explanation

    Grants flow downward through the role hierarchy: granting Role A to Role B means Role B receives everything Role A holds. Role A gains nothing from Role B.

  60. Question 60Account Access and Security

    A user holds several roles but wants a single session to use the privileges of all of them at once. Which command accomplishes this?

    1. GRANT ALL ROLES TO SESSION
    2. USE SECONDARY ROLES ALL Correct answer.
    3. USE ROLE ALL
    4. SET ROLE HIERARCHY ALL

    Explanation

    USE SECONDARY ROLES ALL activates all roles granted to the user as secondary roles for the session, in addition to the primary role. Secondary roles must be enabled this way because only the primary role is active by default.

  61. Question 61Account Access and Security

    Which statement about the ACCOUNTADMIN role is accurate?

    1. It is created by the ORGADMIN role
    2. It can be owned by any custom role
    3. It combines SYSADMIN and SECURITYADMIN privileges Correct answer.
    4. It is granted to the PUBLIC role at account creation

    Explanation

    ACCOUNTADMIN encapsulates the SYSADMIN and SECURITYADMIN roles along with account-level privileges such as billing and account parameter management. Because of that breadth, it should be granted only to a small number of administrators.

  62. Question 62Account Access and Security

    What is the purpose of a database role in Snowflake?

    1. To authenticate users through SAML SSO
    2. To replace the PUBLIC role across the whole account
    3. To manage account-level billing data
    4. To hold privileges scoped to one database Correct answer.

    Explanation

    Database roles are contained inside a single database and can hold privileges on objects in that database, which lets teams manage access without granting account-level roles. They are granted to account roles to make their privileges usable.

  63. Question 63Account Access and Security

    Which role is required to create a new account within a Snowflake organization?

    1. SYSADMIN
    2. ORGADMIN Correct answer.
    3. ACCOUNTADMIN
    4. SECURITYADMIN

    Explanation

    ORGADMIN is the organization-level role used to create, view, and manage accounts in an organization, and to view organization-level usage. Account-level roles such as ACCOUNTADMIN only govern what happens inside a single account.

  64. Question 64Account Access and Security

    What does Snowflake Tri-Secret Secure combine?

    1. Three Snowflake-managed keys rotated per account
    2. Three network policies applied in sequence
    3. Password, MFA, and key pair authentication
    4. Snowflake key plus a customer-managed key Correct answer.

    Explanation

    Tri-Secret Secure builds a composite master key from a Snowflake-managed key and a key held in the customer's own cloud key management service. Both keys are required to decrypt data, so revoking the customer key removes access.

  65. Question 65Account Access and Security

    Which practice follows the principle of least privilege when designing Snowflake access?

    1. Grant PUBLIC every privilege and revoke as needed
    2. Grant only the privileges each role needs Correct answer.
    3. Grant ACCOUNTADMIN to all analysts for convenience
    4. Assign every user the SYSADMIN role

    Explanation

    Least privilege means each role receives only the privileges required for its tasks, usually through purpose-built custom roles granted to users. Broad grants such as ACCOUNTADMIN or blanket PUBLIC grants expand the blast radius of any mistake or compromised credential.

  66. Question 66Account Access and Security

    Which system-defined role should a custom role be granted to so that objects the custom role creates stay manageable by the SYSADMIN role?

    1. ACCOUNTADMIN
    2. SECURITYADMIN
    3. USERADMIN
    4. SYSADMIN Correct answer.

    Explanation

    Granting a custom role to SYSADMIN places that role inside the SYSADMIN hierarchy, so a SYSADMIN user inherits the custom role's privileges over the objects it owns. The reverse grant, SYSADMIN to the custom role, does not place the objects under SYSADMIN control.

  67. Question 67Account Access and Security

    Which statement about the PUBLIC role in Snowflake is accurate?

    1. It must be granted explicitly to each new user by SECURITYADMIN.
    2. It holds all account privileges until custom roles are created.
    3. It is active only while the user has no other role selected.
    4. Every user is granted it automatically, and it cannot be dropped. Correct answer.

    Explanation

    PUBLIC is a system-defined role that is granted to every user by default and cannot be dropped. Privileges granted to PUBLIC are therefore available to every user in the account, which is why it is kept minimal.

  68. Question 68Account Access and Security

    A user has been granted several roles. Which roles can the user activate with USE ROLE during a session?

    1. Only the role configured as the user's default role
    2. Only roles that hold USAGE on an accessible warehouse
    3. Any role that exists in the account, if the user knows its name
    4. Any role granted to the user directly or through a role hierarchy Correct answer.

    Explanation

    A user may switch to any role that has been granted to that user, including roles inherited through a chain of role grants. Roles that were never granted, directly or indirectly, cannot be activated.

  69. Question 69Account Access and Security

    During key-pair authentication, what does Snowflake verify?

    1. A one-time code produced by an authenticator app
    2. A shared symmetric secret stored for the user
    3. A signature created with the user's private key Correct answer.
    4. A signed SAML assertion from the identity provider

    Explanation

    In key-pair authentication the client signs a token with its private key, and Snowflake validates the signature using the public key previously assigned to the user. The private key never leaves the client.

  70. Question 70Account Access and Security

    An account-level network policy and a user-level network policy both apply to a user. How does Snowflake apply them?

    1. The two policies are combined, and an address allowed by either is admitted
    2. The account-level policy takes precedence over the user-level policy
    3. Neither policy applies until the user completes multi-factor authentication
    4. The user-level policy takes precedence over the account-level policy Correct answer.

    Explanation

    When a user has a network policy assigned directly, that policy overrides the account-level policy for that user's connections. The account-level policy applies to users who have no user-level policy of their own.

  71. Question 71Account Access and Security

    Which role does SECURITYADMIN inherit, giving it the ability to create users and roles?

    1. ACCOUNTADMIN
    2. PUBLIC
    3. SYSADMIN
    4. USERADMIN Correct answer.

    Explanation

    SECURITYADMIN inherits the privileges of USERADMIN, so it can create and manage users and roles in addition to its account-wide authority over grants. SYSADMIN and ACCOUNTADMIN sit in a different part of the hierarchy.

  72. Question 72Account Access and Security

    Which account-level privilege lets a role grant or revoke privileges on objects that another role owns?

    1. CREATE SHARE
    2. APPLY MASKING POLICY
    3. CREATE ROLE
    4. MANAGE GRANTS Correct answer.

    Explanation

    MANAGE GRANTS is an account-level privilege that authorizes a role to grant and revoke privileges on any object in the account, including objects it does not own. It is the privilege that makes SECURITYADMIN useful for access administration.

  73. Question 73Account Access and Security

    What does marking a view as SECURE change?

    1. Querying roles must also hold SELECT on the referenced tables
    2. The view definition and referenced objects are hidden from non-owners Correct answer.
    3. Only roles granted the view before its creation can query it
    4. The result set is encrypted with a per-query key before it is returned

    Explanation

    A secure view prevents users other than the view owner from seeing the view definition, the SQL it contains, and the objects it references. This protects intellectual property and underlying schema details from users who only need the query results.

  74. Question 74Account Access and Security

    A secure view must read a table that the querying roles are not permitted to access directly. Which approach makes this possible?

    1. Remove the SECURE property so the view runs with owner rights
    2. Grant SELECT on the table to the PUBLIC role so all users can read it
    3. Grant the required privileges to the view itself Correct answer.
    4. Grant the view to the ACCOUNTADMIN role

    Explanation

    An authorized view is a secure view that has been granted the privileges it needs on the underlying objects, so it can access those objects on behalf of querying roles. Querying roles then need privileges only on the view itself.

  75. Question 75Account Access and Security

    Which Account Usage view reports, for each query, the objects and columns that were read?

    1. QUERY_HISTORY
    2. ACCESS_HISTORY Correct answer.
    3. LOGIN_HISTORY
    4. GRANTS_TO_ROLES

    Explanation

    ACCESS_HISTORY records, per query, the objects and columns that were accessed, distinguishing direct sources from base sources reached through views. It is the standard view for auditing column-level data access.

  76. Question 76Account Access and Security

    Which role owns a table created by a user in a session where SYSADMIN is the active role?

    1. ACCOUNTADMIN, since it owns every object in the account
    2. The user, as a personal object owner
    3. The role that owns the schema in which the table is created
    4. SYSADMIN, the active role at creation time Correct answer.

    Explanation

    Objects are owned by the role that is active when the object is created, not by the user and not by the owner of the parent schema. The owning role can later be changed only by transferring ownership.

  77. Question 77Account Access and Security

    Which Snowflake capability provides private network connectivity to an account without sending traffic over the public internet?

    1. Private connectivity (PrivateLink) Correct answer.
    2. External OAuth
    3. A network policy
    4. Tri-Secret Secure

    Explanation

    Private connectivity services such as AWS PrivateLink or Azure Private Link let clients reach Snowflake through a private endpoint inside their own cloud network. Traffic between the client network and Snowflake does not traverse the public internet.

  78. Question 78Account Access and Security

    Which statement about the OWNERSHIP privilege on a table is accurate?

    1. It cannot be revoked; it moves only through a transfer of ownership Correct answer.
    2. It can be revoked, leaving the table temporarily without an owner
    3. It can be granted to several roles at once, one of which is primary
    4. It is held jointly by every role granted ALL PRIVILEGES on the table

    Explanation

    OWNERSHIP is a special privilege that cannot be taken away with REVOKE. It changes hands only when ownership is explicitly transferred to another role, and every object has exactly one owning role at any time.

  79. Question 79Account Access and Security

    An administrator tries to drop a role that owns several tables. What happens?

    1. The tables are dropped together with the role
    2. The role is dropped, and the tables become owned by ACCOUNTADMIN
    3. The role is dropped, and the tables become owned by PUBLIC
    4. The drop fails until ownership of those tables is transferred Correct answer.

    Explanation

    Snowflake will not drop a role while it still owns objects, because every object must have a valid owning role. The administrator must first transfer ownership of those objects to another role and then drop the role.

  80. Question 80Account Access and Security

    A user should have all of the roles granted to them active automatically at the start of every session. Which user property accomplishes this?

    1. DEFAULT_ROLE = ALL
    2. DEFAULT_WAREHOUSE = ALL
    3. DEFAULT_SECONDARY_ROLES = ('ALL') Correct answer.
    4. DEFAULT_NAMESPACE = ALL

    Explanation

    Setting the user's DEFAULT_SECONDARY_ROLES property to ALL enables every role granted to the user as a secondary role when a session starts. The primary role continues to come from the user's DEFAULT_ROLE setting.

  81. Question 81Performance Optimization, Querying, and Data Transformation

    Two identical SELECT statements are run against the same table ten minutes apart, and no DML has occurred on that table in between. Which statement describes the second execution?

    1. It fails unless the original warehouse is still running.
    2. It can return persisted results from the result cache without starting a warehouse. Correct answer.
    3. It reuses the warehouse disk cache and always consumes credits.
    4. It must rescan every micro-partition but consumes no credits.

    Explanation

    Snowflake persists the results of every query for 24 hours, and each reuse resets that window. If the text matches exactly and the underlying data has not changed, the results are served by the Cloud Services layer with no warehouse required. This means no compute credits are consumed for the repeated execution.

  82. Question 82Performance Optimization, Querying, and Data Transformation

    Snowflake stores table data in immutable micro-partitions. What is the size range of a single micro-partition before compression?

    1. 1 MB to 10 MB
    2. 500 MB to 1 GB
    3. 10 MB to 100 MB
    4. 50 MB to 500 MB Correct answer.

    Explanation

    Snowflake targets micro-partitions of roughly 50 MB to 500 MB of uncompressed data. The exact size varies because partitions are sized by row count and data characteristics, and Snowflake rebalances them in the background. This granularity is what makes partition pruning efficient.

  83. Question 83Performance Optimization, Querying, and Data Transformation

    A developer must return only the most recent order per customer using a window function. Which clause filters on the result of a window function without wrapping the query in a subquery?

    1. QUALIFY Correct answer.
    2. LIMIT
    3. WHERE
    4. HAVING

    Explanation

    QUALIFY is evaluated after window functions are computed, so it can filter on values such as ROW_NUMBER() or RANK() directly. WHERE and HAVING are evaluated before the window function results exist. QUALIFY removes the need for an extra subquery layer.

  84. Question 84Performance Optimization, Querying, and Data Transformation

    A column named PAYLOAD of type VARIANT holds a JSON array. Which construct turns each array element into its own row in the query output?

    1. TO_ARRAY(PAYLOAD)
    2. LATERAL FLATTEN(input => PAYLOAD) Correct answer.
    3. PAYLOAD[0]
    4. ARRAY_AGG(PAYLOAD)

    Explanation

    FLATTEN is a table function that expands an array or object into one row per element, and LATERAL lets it reference columns from the preceding table expression. It also emits path and index columns that identify each element. The other constructs return or aggregate a single value rather than expanding rows.

  85. Question 85Performance Optimization, Querying, and Data Transformation

    Which query is typically answered in milliseconds by the Cloud Services layer using only metadata, without a running warehouse?

    1. SELECT * FROM sales ORDER BY amount DESC LIMIT 10;
    2. SELECT DISTINCT region FROM sales;
    3. SELECT AVG(amount) FROM sales GROUP BY region;
    4. SELECT COUNT(*) FROM sales; Correct answer.

    Explanation

    An unfiltered COUNT(*) can be answered from metadata that Snowflake maintains about table row counts, so no warehouse is needed. Aggregations that must scan values, sorts, and duplicate elimination all require compute. This metadata-backed optimization is one reason simple profiling queries are so cheap.

  86. Question 86Performance Optimization, Querying, and Data Transformation

    A warehouse running large scans is resized from Small to Large. What is the typical effect?

    1. Credits per hour double and runtime for large scans typically halves. Correct answer.
    2. Credits per hour halve and runtime typically halves.
    3. Credits per hour stay the same and runtime stays the same.
    4. Credits per hour double and runtime typically doubles.

    Explanation

    Doubling the warehouse size doubles the number of credits consumed per hour because twice as many compute resources are attached. For a single large scan those resources work in parallel, so elapsed time for the query is roughly cut in half. The total credits consumed for that query therefore stay about the same.

  87. Question 87Performance Optimization, Querying, and Data Transformation

    How does defining a clustering key improve query performance on a very large table?

    1. It builds a B-tree index that points to individual rows.
    2. It compresses each column separately to reduce storage cost.
    3. It caches frequently used query results for repeated executions.
    4. It co-locates similar column values within micro-partitions so pruning reads fewer partitions. Correct answer.

    Explanation

    Snowflake does not use user-created indexes; instead it tracks the value ranges of each micro-partition in metadata. A cluster key physically groups similar values so those ranges overlap less, letting the optimizer skip partitions that cannot match a predicate. This pruning is the main performance benefit.

  88. Question 88Performance Optimization, Querying, and Data Transformation

    Many concurrent users queue on one warehouse during a two-hour peak each day, while the warehouse sits idle the rest of the time. Which configuration addresses this most directly?

    1. Increasing the statement timeout parameter
    2. Adding a clustering key to the largest table
    3. Configuring a multi-cluster warehouse with a minimum of 1 and a higher maximum Correct answer.
    4. Enabling the search optimization service

    Explanation

    A multi-cluster warehouse adds compute clusters automatically when queries queue and removes them when concurrency drops. Setting the minimum to one keeps off-peak cost low while the higher maximum absorbs the daily peak. This handles concurrency, whereas the other options address data layout or query behavior.

  89. Question 89Performance Optimization, Querying, and Data Transformation

    A 5 TB event table is almost always filtered on EVENT_TS (a timestamp with millions of distinct values) and only occasionally on STATUS (five distinct values). Which clustering key should be chosen?

    1. ROW_NUMBER() computed over the whole table
    2. STATUS
    3. No key is needed because Snowflake indexes every column automatically
    4. EVENT_TS Correct answer.

    Explanation

    Clustering works best on columns that are used in selective predicates and have high cardinality, because the value ranges of adjacent micro-partitions then overlap less. A low-cardinality column like STATUS creates large overlapping ranges and prunes poorly. EVENT_TS matches the dominant filter pattern.

  90. Question 90Performance Optimization, Querying, and Data Transformation

    A clustered table receives continuous inserts and updates. How does Snowflake keep the clustering effective, and how is that work billed?

    1. A serverless background service reclusters micro-partitions and bills separately for the credits consumed. Correct answer.
    2. It rewrites the table in a nightly maintenance window at no charge.
    3. It suspends the table until an administrator runs a manual recluster command.
    4. It uses the virtual warehouse that ran the DML, charging no additional credits.

    Explanation

    When automatic clustering is enabled, Snowflake runs a serverless service that reorganizes overlapping micro-partitions in the background. That service consumes Snowflake-managed credits, which appear as a separate line item. Because it is serverless, no user warehouse needs to be running.

  91. Question 91Performance Optimization, Querying, and Data Transformation

    Which workload benefits most from enabling the search optimization service?

    1. Loading staged files into tables with COPY INTO
    2. A nightly job that aggregates most rows of a large fact table
    3. Highly selective point lookups and substring searches on a large table Correct answer.
    4. Joins between two small dimension tables

    Explanation

    The search optimization service maintains a persistent index structure that lets the optimizer locate the few micro-partitions containing matching rows. It pays off when predicates are selective, such as equality on a lookup key or a substring pattern. Scans that touch most of a table gain little because almost every partition must be read anyway.

  92. Question 92Performance Optimization, Querying, and Data Transformation

    What problem is the query acceleration service designed to solve?

    1. Queueing caused by many small concurrent queries
    2. Low throughput when loading files from external stages
    3. Outlier queries with large scans and selective filters that run far longer than the rest of the workload Correct answer.
    4. Slow metadata operations handled by the Cloud Services layer

    Explanation

    The query acceleration service offloads portions of a query's scan and filter work to shared compute resources. It targets the occasional long-running outlier query on a warehouse rather than steady-state parallelism. The warehouse still performs the rest of the processing.

  93. Question 93Performance Optimization, Querying, and Data Transformation

    A materialized view is defined over a large base table, and a user submits a query that references the base table directly but matches the view's definition and filters. What can the optimizer do?

    1. Use the view only when the warehouse is configured as multi-cluster.
    2. Nothing, because users must name the materialized view explicitly to benefit.
    3. Use the view only when the base table has no clustering key.
    4. Rewrite the query to read the materialized view instead of the base table. Correct answer.

    Explanation

    Materialized views are maintained automatically, and the optimizer can recognize when a query against the base table can be answered from the view. It then rewrites the plan to read the smaller, precomputed result. Users do not have to reference the view by name to get the benefit.

  94. Question 94Performance Optimization, Querying, and Data Transformation

    A table has columns (SALE_DATE, PRODUCT, AMOUNT). A report needs one row per SALE_DATE with a separate column for each PRODUCT. Which transformation produces that shape?

    1. PIVOT with PRODUCT as the pivot column and SUM(AMOUNT) as the aggregate Correct answer.
    2. GROUP BY SALE_DATE, PRODUCT
    3. LATERAL FLATTEN applied to PRODUCT
    4. UNPIVOT on PRODUCT

    Explanation

    PIVOT rotates distinct values from a column into separate output columns while aggregating the values in another column. Grouping by SALE_DATE and pivoting on PRODUCT with SUM(AMOUNT) yields the one-row-per-date layout. A GROUP BY would return one row per date-product pair instead.

  95. Question 95Performance Optimization, Querying, and Data Transformation

    A table has one row per customer with columns JAN_SALES, FEB_SALES and MAR_SALES. A report needs one row per customer per month. Which approach fits best?

    1. UNPIVOT the three columns into (MONTH, SALES) pairs Correct answer.
    2. PIVOT on the three month columns
    3. ARRAY_AGG applied to the three columns
    4. A self-join on CUSTOMER_ID

    Explanation

    UNPIVOT converts wide columns into name-value row pairs, which is exactly the transition from one row per customer to one row per customer-month. The column names can be emitted as a label column alongside the values. PIVOT performs the opposite transformation.

  96. Question 96Performance Optimization, Querying, and Data Transformation

    A VARIANT column V contains the value {"customer":{"name":"Ada","id":7}}. Which expression returns the name as a VARCHAR?

    1. CAST(V AS VARCHAR)
    2. GET(V, 'name')
    3. V:customer.name::string Correct answer.
    4. V['customer']['name']

    Explanation

    The colon notation navigates into a VARIANT object, and the ::string cast converts the extracted value to VARCHAR. Paths can be chained through nested keys with dots. This returns just the name value rather than the whole document.

  97. Question 97Performance Optimization, Querying, and Data Transformation

    A query profile shows significant 'bytes spilled to remote storage' on an aggregation step. Which action most directly addresses the cause?

    1. Enable the result cache for the session.
    2. Run the query on a larger warehouse or reduce the rows reaching the aggregation. Correct answer.
    3. Increase the statement timeout parameter.
    4. Add a clustering key to the aggregation output.

    Explanation

    Spilling means the operation ran out of memory in its warehouse and had to write intermediate data to local or remote storage. Giving the warehouse more memory or shrinking the input through earlier filters and pre-aggregation reduces the spill. Remote spilling in particular signals that local disk was also exhausted.

  98. Question 98Performance Optimization, Querying, and Data Transformation

    What does an average clustering depth of 1.0 indicate for a clustered table?

    1. The table contains only one micro-partition.
    2. Overlapping micro-partitions are minimal, so pruning is near-optimal. Correct answer.
    3. The table requires immediate manual reclustering.
    4. The table has exactly one clustering key column.

    Explanation

    Clustering depth measures how many micro-partitions must be touched to reach all rows sharing a given key value. An average depth close to 1 means values cluster tightly into very few partitions, so the optimizer can prune aggressively. Higher depth values mean more overlap and weaker pruning.

  99. Question 99Performance Optimization, Querying, and Data Transformation

    A multi-cluster warehouse is configured with the ECONOMY scaling policy. How does it differ from STANDARD?

    1. It runs extra clusters on discounted compute capacity.
    2. It starts additional clusters faster than STANDARD.
    3. It waits roughly six minutes before shutting down an idle cluster, favoring fewer start-stop cycles. Correct answer.
    4. It permanently caps the warehouse at two clusters.

    Explanation

    With ECONOMY, Snowflake keeps provisioned clusters running longer before shutting them down, which avoids repeated start-up delays when load fluctuates. STANDARD shuts clusters down more promptly when they are no longer needed. ECONOMY therefore trades slightly higher idle cost for fewer provisioning cycles.

  100. Question 100Performance Optimization, Querying, and Data Transformation

    What happens to a virtual warehouse's local disk cache when the warehouse is suspended?

    1. It persists for 24 hours in the same way the result cache does.
    2. It is dropped, so the first queries after resume re-read data from storage and may be slower. Correct answer.
    3. It is converted into a materialized view for future queries.
    4. It is moved to remote storage and restored automatically on resume.

    Explanation

    The local disk cache holds micro-partitions that the warehouse has already read, and it lives only as long as the warehouse is running. Suspending the warehouse discards that cache, so subsequent queries must fetch data again until the cache warms up. This is why keeping a warehouse running can help latency for repeated workloads.

  101. Question 101Performance Optimization, Querying, and Data Transformation

    A clustered table receives a large daily bulk load. Which statement about reclustering cost is most accurate?

    1. Rows appended in clustering order need little or no reclustering, while out-of-order inserts increase reclustering work and cost. Correct answer.
    2. Reclustering is free because it runs in the Cloud Services layer.
    3. Reclustering occurs only when a virtual warehouse is resumed.
    4. Every insert triggers a full rewrite of the table regardless of row order.

    Explanation

    When new rows arrive already aligned with the cluster key ranges, they land in partitions that overlap little, so the background service has little to reorganize. Loading rows that scatter values across existing ranges increases overlap and forces more partition rewrites. Ordering loads by the cluster key is therefore a common cost-control practice.

  102. Question 102Performance Optimization, Querying, and Data Transformation

    Which predicate allows the search optimization service to skip the fewest micro-partitions?

    1. An equality filter on a highly selective lookup key
    2. A substring search anchored to a specific prefix
    3. A filter that matches most of the rows in the table Correct answer.
    4. A geospatial intersection test against a small region

    Explanation

    The service accelerates queries by using its index to identify the small set of micro-partitions that can contain matches. When a predicate matches most rows, nearly every partition must still be read, so the index provides almost no pruning. Selective predicates are where the service delivers its benefit.

  103. Question 103Performance Optimization, Querying, and Data Transformation

    A GROUP BY on a column with hundreds of millions of distinct values spills heavily and runs slowly. Which change best reduces memory pressure for this single query?

    1. Filter or pre-aggregate the input so fewer distinct groups reach the final aggregation. Correct answer.
    2. Append a LIMIT clause to the query.
    3. Run the query on a multi-cluster warehouse with five clusters.
    4. Route the query through the query acceleration service.

    Explanation

    Aggregation memory scales with the number of distinct groups that must be tracked simultaneously. Reducing rows or cardinality before the final GROUP BY, for example with a coarser pre-aggregation or a tighter filter, shrinks the hash table. The query acceleration service targets scan-heavy work rather than aggregation state.

  104. Question 104Performance Optimization, Querying, and Data Transformation

    Under which condition does Snowflake stop reusing persisted query results before the 24-hour window ends?

    1. The underlying table data has changed since the results were persisted. Correct answer.
    2. A different user with the same role runs the identical statement.
    3. The warehouse that produced the results was suspended in between.
    4. The original query ran on a smaller warehouse than the current one.

    Explanation

    Persisted results are tied to the data they were computed from. Any DML or DDL that changes the underlying tables invalidates the cached result, and the query is re-executed on a warehouse. Session-independent factors such as warehouse state do not invalidate results.

  105. Question 105Performance Optimization, Querying, and Data Transformation

    A query profile shows a TableScan with partitions scanned equal to 500 out of 500 total. What conclusion is best supported?

    1. Pruning removed most partitions before the scan.
    2. The table contains exactly 500 rows of data.
    3. No pruning occurred, because all partitions were read. Correct answer.
    4. The result cache answered the query.

    Explanation

    The ratio of partitions scanned to partitions total shows how effective pruning was. Reading every partition means the optimizer could not eliminate any based on the predicates, so filtering or clustering could reduce the work. It says nothing about row counts or caching.

  106. Question 106Performance Optimization, Querying, and Data Transformation

    Which Snowflake clause filters rows based on the output of a window function without requiring a subquery or a CTE?

    1. HAVING
    2. QUALIFY Correct answer.
    3. WINDOW
    4. WHERE

    Explanation

    QUALIFY is evaluated after window functions are computed, so a predicate on something like ROW_NUMBER() OVER (...) can be applied in the same SELECT. WHERE runs before window functions exist, and HAVING applies to grouped aggregates. WINDOW only names a reusable window definition and performs no filtering.

  107. Question 107Performance Optimization, Querying, and Data Transformation

    Which mechanism lets Snowflake prune irrelevant data during a scan on a large table without any user-managed structures?

    1. Tables are hash-distributed across warehouses so each warehouse scans only a subset of rows.
    2. Secondary indexes are created on filtered columns and refreshed automatically after each load.
    3. Each micro-partition stores per-column min/max values that the optimizer uses to prune scans. Correct answer.
    4. The optimizer samples a fixed percentage of rows at run time to decide which columns to read.

    Explanation

    Tables are stored as immutable micro-partitions, and Snowflake records metadata such as the minimum and maximum value of each column in each partition. The optimizer compares filter predicates against that metadata to skip partitions that cannot contain matching rows. No user-created indexes or manually gathered statistics are involved.

  108. Question 108Performance Optimization, Querying, and Data Transformation

    A VARIANT column TAGS holds a JSON array. A query must return one output row for each element of that array. Which construct accomplishes this?

    1. Call TO_VARIANT on the column and include it in the SELECT list.
    2. Aggregate the column with ARRAY_AGG inside a GROUP BY.
    3. Cast the column to VARCHAR and split the resulting string on commas.
    4. Use LATERAL FLATTEN with the array path in the FROM clause. Correct answer.

    Explanation

    FLATTEN is a table function that expands a VARIANT array into one row per element, and LATERAL lets it reference the column of the table on its left. Each produced row exposes fields such as VALUE and INDEX. String manipulation or projection alone never multiplies rows.

  109. Question 109Performance Optimization, Querying, and Data Transformation

    A dashboard submits the identical query text every ten minutes while the underlying tables remain unchanged. How does the persisted result cache affect these repeated runs?

    1. Only the compilation step is skipped; the query still scans every micro-partition.
    2. Results are reused only if the warehouse has not been suspended between the two runs.
    3. Results are reused for one hour, after which the query is always fully re-executed.
    4. The stored result is returned directly until the data changes or 24 hours pass. Correct answer.

    Explanation

    When the query text and the underlying data are unchanged, Snowflake returns the previously computed result from the persisted result cache without executing the query, so no warehouse compute is consumed. The entry survives up to 24 hours and is invalidated early by changes to the referenced data or by non-deterministic functions in the statement.

  110. Question 110Performance Optimization, Querying, and Data Transformation

    A 200-column table occupying 2 TB is queried by a statement that references only 4 columns. Why does the query read far less than 2 TB of data?

    1. Unreferenced columns are compressed to zero bytes before the scan begins.
    2. Data is stored columnar inside micro-partitions, so only referenced columns are read. Correct answer.
    3. Only the first micro-partition of each column is scanned for narrow projections.
    4. The optimizer materializes a temporary projection of those columns once per day.

    Explanation

    Micro-partitions use a columnar layout, so a scan touches only the column data required by the query plus the metadata used for pruning. Wide tables therefore do not force the read of every column. Selecting fewer columns and avoiding SELECT * reduces I/O directly.

  111. Question 111Performance Optimization, Querying, and Data Transformation

    A table is clustered on ORDER_TS. A query using WHERE DATE_TRUNC('day', ORDER_TS) = '2024-03-01' scans nearly all micro-partitions. Which rewrite restores pruning?

    1. Move the filter into a HAVING clause evaluated after aggregation.
    2. Add a LIMIT clause to cap the number of rows the query returns.
    3. Wrap ORDER_TS in TO_CHAR and compare the formatted string to '2024-03-01'.
    4. Filter with ORDER_TS >= '2024-03-01' AND ORDER_TS < '2024-03-02'. Correct answer.

    Explanation

    Pruning compares constant values against the per-partition minimum and maximum of the column, which requires the column to appear on its own side of the predicate. A function such as DATE_TRUNC or TO_CHAR applied to the column hides that comparison, so every partition must be read. Expressing the condition as a range on the bare column keeps the min/max comparison usable.

  112. Question 112Performance Optimization, Querying, and Data Transformation

    An analyst needs quick exploratory statistics from a 10 TB events table, approximate answers are acceptable, and cost should be kept low. Which approach fits best?

    1. Query a fixed-percentage sample of the table using the SAMPLE clause. Correct answer.
    2. Enable the query acceleration service on the analyst's warehouse.
    3. Run the full aggregation on an X-Large warehouse and rely on the result cache.
    4. Create a materialized view over the base table and query the view instead.

    Explanation

    Sampling reads only a fraction of the micro-partitions, so exploration returns quickly at a fraction of the cost when approximate results are acceptable. A larger warehouse or materialized view still processes the full data set. The query acceleration service targets scans that are a small part of a much larger, complex query.

  113. Question 113Performance Optimization, Querying, and Data Transformation

    A query needs a running total of AMOUNT ordered by SALE_DATE, where each row shows the sum of itself and all earlier rows. Which window specification produces this?

    1. SUM(amount) OVER (ORDER BY sale_date RANGE BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING)
    2. SUM(amount) OVER (ORDER BY sale_date ROWS BETWEEN 1 PRECEDING AND 1 FOLLOWING)
    3. SUM(amount) OVER (ORDER BY sale_date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) Correct answer.
    4. SUM(amount) OVER (PARTITION BY sale_date ORDER BY sale_date)

    Explanation

    A cumulative total needs an ordered window whose frame starts at the first row of the partition and ends at the current row. ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW gives exactly that accumulation. Partitioning by SALE_DATE would produce per-date totals instead of a running sum.

  114. Question 114Performance Optimization, Querying, and Data Transformation

    A report needs one row per REGION with its sales total, plus a final grand total row, all from a single statement. Which clause achieves this?

    1. GROUP BY ROLLUP (region) Correct answer.
    2. ORDER BY region WITH SUBTOTALS
    3. GROUP BY region HAVING GROUPING(region) = 1
    4. SUM(amount) OVER (PARTITION BY region)

    Explanation

    ROLLUP produces the detail aggregation rows and then adds super-aggregate rows that roll the grouped columns up to a grand total. The extra rows can be labelled with the GROUPING function. Window functions add columns to existing rows and cannot create the additional summary rows.

  115. Question 115Performance Optimization, Querying, and Data Transformation

    A VARCHAR column RAW_QTY holds mostly numeric strings but includes values like 'N/A'. A transformation must yield NULL for those rows instead of failing. Which expression does this?

    1. TO_NUMBER(raw_qty, 10, 2)
    2. COALESCE(raw_qty, 0)::NUMBER
    3. CAST(raw_qty AS NUMBER)
    4. TRY_TO_NUMBER(raw_qty) Correct answer.

    Explanation

    The TRY_ variants of conversion functions return NULL when conversion fails rather than raising an error. CAST and TO_NUMBER raise an error on invalid input unless ON_ERROR is specified. COALESCE only handles NULL inputs and does not make a non-numeric string convertible.

  116. Question 116Performance Optimization, Querying, and Data Transformation

    An administrator must review query execution times over the past six months to identify tuning candidates. Which source retains that history?

    1. The LOCAL_DISK_CACHE view in INFORMATION_SCHEMA
    2. INFORMATION_SCHEMA.QUERY_HISTORY in the database being queried
    3. The ACCOUNT_USAGE.QUERY_HISTORY view in the shared SNOWFLAKE database Correct answer.
    4. A virtual warehouse's QUERY_LOG parameter output

    Explanation

    The ACCOUNT_USAGE schema in the shared SNOWFLAKE database retains query history for up to one year, which makes it the right source for long-range tuning analysis. The INFORMATION_SCHEMA QUERY_HISTORY view is limited to a short recent window. Data in ACCOUNT_USAGE views also appears with some latency.

  117. Question 117Performance Optimization, Querying, and Data Transformation

    Which parameter automatically cancels a statement that has been running longer than a configured number of seconds?

    1. LOCK_TIMEOUT
    2. STATEMENT_QUEUED_TIMEOUT_IN_SECONDS
    3. STATEMENT_TIMEOUT_IN_SECONDS Correct answer.
    4. ABORT_DETACHED_QUERY

    Explanation

    STATEMENT_TIMEOUT_IN_SECONDS sets the maximum run time for a statement and cancels it when the limit is exceeded; it can be set at the account, user, session, or warehouse level. The queued timeout applies only while a statement waits in a queue. ABORT_DETACHED_QUERY stops work whose client connection has gone away, not long-running statements.

  118. Question 118Performance Optimization, Querying, and Data Transformation

    A join intended only to add descriptive columns from a dimension table returns far more rows than the fact table contains, and the query becomes slow. What is the most likely cause?

    1. The dimension table lacks a clustering key, so its partitions are scanned twice.
    2. The statement omitted a WHERE clause, so the optimizer chose a cross join.
    3. The join key is not unique on the dimension side, so each fact row matches multiple rows. Correct answer.
    4. The warehouse is too small to hold the hash table built from the dimension table.

    Explanation

    Row multiplication after a join is the signature of a non-unique key on the side that should be one row per key. Each matching fact row is duplicated once per duplicate dimension row, which inflates downstream processing. Verifying uniqueness, or aggregating or deduplicating the dimension first, restores the expected row count.

  119. Question 119Performance Optimization, Querying, and Data Transformation

    A table EMP(EMP_ID, MANAGER_ID) stores an adjacency list. The query must return each employee together with the full chain of managers above them. Which approach fits Snowflake?

    1. Writing a recursive CTE whose UNION ALL branch joins EMP back to the CTE. Correct answer.
    2. Using a window with PARTITION BY manager_id and ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW.
    3. Repeating self-joins once per level, because a CTE cannot reference itself.
    4. Pivoting on MANAGER_ID so that each level becomes its own column.

    Explanation

    Snowflake supports recursive CTEs, where an anchor query produces the starting rows and a recursive branch joins the table back to the CTE until no new rows are produced. This walks the hierarchy to an arbitrary depth. Window functions and pivots operate on a fixed set of rows and cannot traverse levels.

  120. Question 120Performance Optimization, Querying, and Data Transformation

    A WITH clause defines an expensive aggregation that the main query references three times, and the query profile shows that aggregation executing three times. What explains this behavior?

    1. The optimizer converts each CTE reference into a stored procedure that runs separately.
    2. The result cache is disabled for statements that reference a CTE more than once.
    3. Snowflake materializes each CTE into a temporary table but drops it between references.
    4. CTEs are inlined by default, so each reference re-executes the CTE definition. Correct answer.

    Explanation

    In Snowflake, a CTE is a named subquery that is inlined into the statement unless it is not beneficial to do so, so referencing it three times can mean the underlying work happens three times. Materializing the intermediate result in a temporary table, or restructuring the query to scan once, removes the repeated work. The profile is the reliable way to confirm the repetition.

  121. Question 121Data Loading and Unloading

    What does the following command do? PUT file:///data/orders.csv @my_stage;

    1. Copies rows from the staged file into a table.
    2. Registers the local directory as an external stage.
    3. Uploads the local file to the internal stage. Correct answer.
    4. Downloads the staged file to the local file system.

    Explanation

    PUT transfers files from a local file system to an internal stage (user, table, or named). Files are compressed automatically during upload, and the command runs on the client rather than on a virtual warehouse.

  122. Question 122Data Loading and Unloading

    A COPY INTO statement hits a malformed row in one data file and no ON_ERROR option is specified. What happens?

    1. Valid rows are loaded and the bad row is skipped.
    2. The statement continues but the error rows are recorded.
    3. Only the failing file is skipped and other files still load.
    4. The statement aborts and the load is rolled back. Correct answer.

    Explanation

    The default value of ON_ERROR is ABORT_STATEMENT, which means the whole COPY statement fails and no rows are committed when an error is encountered. A different ON_ERROR value must be set explicitly to tolerate bad rows or files.

  123. Question 123Data Loading and Unloading

    Which statement unloads data from the table SALES into the named stage @out_stage?

    1. GET @out_stage FROM SALES;
    2. COPY INTO SALES FROM @out_stage;
    3. COPY INTO @out_stage FROM SALES; Correct answer.
    4. PUT @out_stage FROM SALES;

    Explanation

    In an unload, the stage location follows COPY INTO and the source table or query follows FROM. COPY INTO <table> FROM <stage> performs the opposite operation: it loads staged files into a table.

  124. Question 124Data Loading and Unloading

    Which file format type is used for delimited text such as comma-separated values?

    1. ORC
    2. PARQUET
    3. AVRO
    4. CSV Correct answer.

    Explanation

    TYPE = CSV describes delimited text and exposes options such as FIELD_DELIMITER and SKIP_HEADER. PARQUET, AVRO, and ORC are binary formats whose structure is self-describing rather than delimiter-based.

  125. Question 125Data Loading and Unloading

    Which metadata column returns the name of the staged file from which a row was loaded?

    1. METADATA$FILENAME Correct answer.
    2. METADATA$FILE_ROW_NUMBER
    3. METADATA$START_SCAN_TIME
    4. METADATA$FILE_LAST_MODIFIED

    Explanation

    METADATA$FILENAME returns the path and name of the staged file that produced each row. It is selected inside the FROM (SELECT ...) subquery of a COPY INTO statement when the file name must be captured in a table column.

  126. Question 126Data Loading and Unloading

    Which command lists the files currently stored in the named internal stage @mystage?

    1. DESCRIBE STAGE @mystage;
    2. SELECT * FROM @mystage;
    3. SHOW FILES @mystage;
    4. LIST @mystage; Correct answer.

    Explanation

    LIST (or its alias LS) returns one row per file in a stage, including name, size, MD5 checksum, and last modified time. Querying DIRECTORY(@mystage) is the alternative when a directory table is enabled on the stage.

  127. Question 127Data Loading and Unloading

    A file was already loaded successfully by an earlier COPY INTO statement. Which option forces Snowflake to load it again?

    1. VALIDATION_MODE = 'RETURN_ALL_ERRORS'
    2. OVERWRITE = TRUE
    3. FORCE = TRUE Correct answer.
    4. PURGE = TRUE

    Explanation

    Snowflake records loaded files in load metadata for 64 days and skips files that were already processed. FORCE = TRUE instructs COPY INTO to ignore that metadata and reload the staged files.

  128. Question 128Data Loading and Unloading

    Which type of stage is created automatically for every table and is dropped when the table itself is dropped?

    1. A named internal stage
    2. The user stage
    3. A named external stage
    4. The table stage Correct answer.

    Explanation

    Every table owns a table stage, referenced with @%table_name, which is created and dropped along with the table. It cannot be altered or dropped directly and supports no file format defaults, so the format must be specified in COPY statements.

  129. Question 129Data Loading and Unloading

    A COPY INTO statement uses ON_ERROR = CONTINUE. How does Snowflake handle the load?

    1. It loads all rows and inserts NULL for any column that fails to parse.
    2. It aborts the statement as soon as the first error row is found.
    3. It skips every file that contains at least one error row.
    4. It loads every row that has no error and skips the rows that fail. Correct answer.

    Explanation

    ON_ERROR = CONTINUE keeps loading the rows that parse correctly and omits only the rows that produce errors. Files processed this way are marked as partially loaded, and the failed rows must be corrected and reloaded separately.

  130. Question 130Data Loading and Unloading

    Which approach validates staged files and returns parsing errors without loading any rows into the target table?

    1. Query the COPY_HISTORY table function for the target table.
    2. Run COPY INTO with VALIDATION_MODE = 'RETURN_ERRORS'. Correct answer.
    3. Run COPY INTO with FORCE = TRUE and review the returned result.
    4. Run COPY INTO with ON_ERROR = CONTINUE and review the load history.

    Explanation

    Setting VALIDATION_MODE to RETURN_ERRORS makes COPY INTO check the files and report errors without inserting any data. The VALIDATE table function serves a related purpose by validating files from a previous load job.

  131. Question 131Data Loading and Unloading

    Which COPY INTO option restricts a load to staged files whose names match a regular expression?

    1. INCLUDE
    2. FILES
    3. PATTERN Correct answer.
    4. FILE_FORMAT

    Explanation

    PATTERN accepts a regular expression that is matched against the full staged file path, allowing wildcard-style selection of files. FILES is used instead when an explicit list of file names must be loaded.

  132. Question 132Data Loading and Unloading

    What does PURGE = TRUE in a COPY INTO <table> statement do?

    1. Clears the load metadata so the same files can be loaded again.
    2. Deletes all existing rows in the target table before loading.
    3. Removes the staged files after a successful load. Correct answer.
    4. Removes the staged files after validation but before loading.

    Explanation

    PURGE = TRUE deletes the staged data files once the corresponding rows have been loaded successfully. Files that fail to load are retained so the errors can be investigated.

  133. Question 133Data Loading and Unloading

    What is the role of a storage integration when a stage is created for an Amazon S3 bucket?

    1. It defines the default file format the stage will use.
    2. It removes the need for a virtual warehouse during loading.
    3. It grants Snowflake access to the bucket through a generated IAM entity. Correct answer.
    4. It encrypts files on the client before they are written to the bucket.

    Explanation

    A storage integration generates an IAM user and trust policy for your account so Snowflake can access the bucket without credentials embedded in each stage. Creating the integration object requires the ACCOUNTADMIN role.

  134. Question 134Data Loading and Unloading

    Which configuration lets Snowpipe load new files automatically as they arrive in an S3 bucket?

    1. Setting the pipe's file format type to AUTO.
    2. A scheduled task that runs COPY INTO every five minutes.
    3. Enabling PURGE = TRUE in the pipe definition.
    4. CREATE PIPE with AUTO_INGEST = TRUE plus an S3 event notification on the bucket. Correct answer.

    Explanation

    AUTO_INGEST = TRUE makes Snowflake create a notification channel (an SQS queue) that the bucket publishes event messages to. Each event triggers a serverless load of the new file, so no polling task is needed.

  135. Question 135Data Loading and Unloading

    How long does Snowflake retain the load metadata that prevents the same file from being loaded twice by COPY INTO?

    1. 30 days
    2. 64 days Correct answer.
    3. 14 days
    4. 7 days

    Explanation

    Load metadata is kept for 64 days, after which a file with the same path can be loaded again even without FORCE. This is why re-staging a previously loaded file shortly afterwards appears to be silently ignored.

  136. Question 136Data Loading and Unloading

    Which file size does Snowflake recommend when staging many files for a parallel bulk load?

    1. 10 to 20 MB compressed
    2. 100 to 250 MB compressed Correct answer.
    3. 1 to 2 GB compressed
    4. 5 GB or more uncompressed

    Explanation

    Splitting data into files of roughly 100 to 250 MB compressed lets COPY INTO distribute the work across many threads and nodes. Very large single files cannot be parallelized well, and very small files add per-file overhead.

  137. Question 137Data Loading and Unloading

    Which option is required when loading Parquet files so that table columns are matched by name rather than by position?

    1. STRIP_OUTER_ARRAY = TRUE
    2. PARSE_JSON = TRUE
    3. FIELD_OPTIONALLY_ENCLOSED_BY = '"'
    4. MATCH_BY_COLUMN_NAME = CASE_INSENSITIVE Correct answer.

    Explanation

    MATCH_BY_COLUMN_NAME instructs COPY INTO to map the columns of Parquet, Avro, or ORC files to table columns by name. The value controls case sensitivity, and CASE_INSENSITIVE ignores differences in case between file and table column names.

  138. Question 138Data Loading and Unloading

    An unload statement is producing one output file per processing thread. Which option writes the result to a single file instead?

    1. MAX_FILE_SIZE = 0
    2. OVERWRITE = FALSE
    3. DETAILED_OUTPUT = TRUE
    4. SINGLE = TRUE Correct answer.

    Explanation

    SINGLE = TRUE makes COPY INTO <location> write all unloaded rows to one file instead of splitting them across parallel output files. The trade-off is slower unload performance for large result sets.

  139. Question 139Data Loading and Unloading

    Which reference points to a user's personal stage in a PUT or COPY statement?

    1. @%orders
    2. @~ Correct answer.
    3. @~orders
    4. @/users

    Explanation

    The tilde character alone, as in @~, refers to the current user's stage. A percent sign followed by a table name refers to that table's stage, and a plain name refers to a named stage.

  140. Question 140Data Loading and Unloading

    Which statement loads only two columns, in swapped order, from a staged CSV file into the table t?

    1. COPY INTO t FROM @s FILE_FORMAT = (TYPE = CSV) COLUMNS (a, b);
    2. COPY INTO t (a, b) FROM (SELECT $2, $1 FROM @s); Correct answer.
    3. INSERT INTO t SELECT * FROM @s;
    4. COPY INTO t FROM @s WITH TRANSFORM = (SELECT $2, $1);

    Explanation

    A COPY INTO statement can read from a subquery over the stage, in which $1, $2, and so on reference the staged columns by position. The target column list maps the subquery output to specific table columns.

  141. Question 141Data Loading and Unloading

    Which ON_ERROR setting skips an entire file when at least ten percent of its rows contain errors, while other files still load?

    1. ON_ERROR = ABORT_STATEMENT
    2. ON_ERROR = SKIP_FILE
    3. ON_ERROR = SKIP_FILE_10% Correct answer.
    4. ON_ERROR = CONTINUE

    Explanation

    The SKIP_FILE_n% variants skip a file only when the percentage of rejected rows reaches the stated threshold. Files below the threshold still load, with their bad rows omitted.

  142. Question 142Data Loading and Unloading

    Snowpipe already loaded a file, and an identical file with the same path and content is staged again. What does Snowpipe do?

    1. It fails and suspends the pipe until the file is removed.
    2. It loads the file into a temporary table for manual review.
    3. It loads the file again and duplicates every row.
    4. It skips the file because it is already recorded in load metadata. Correct answer.

    Explanation

    Snowpipe uses file load metadata, based on the file path and its checksum or ETag, to avoid loading the same file twice. If the file content changes, the modified file is treated as new and is loaded.

  143. Question 143Data Loading and Unloading

    Files were added directly to the S3 bucket behind an external stage that has no automatic refresh, and they do not appear when the directory table is queried. Which action makes them visible?

    1. Run ALTER STAGE mystage REFRESH. Correct answer.
    2. Recreate the stage with CREATE OR REPLACE STAGE.
    3. Set DIRECTORY = TRUE in the stage's file format.
    4. Run LIST @mystage to populate the directory table.

    Explanation

    A directory table must be refreshed to pick up files added outside Snowflake when event-based refresh is not configured. ALTER STAGE ... REFRESH performs that metadata scan on demand.

  144. Question 144Data Loading and Unloading

    A COPY INTO statement with ON_ERROR = SKIP_FILE processes 100 files, and 5 of them contain at least one malformed row. What is the outcome?

    1. All rows from all 100 files load and the bad rows are discarded.
    2. The statement aborts as soon as it reaches the first failing file.
    3. The 95 clean files load; the 5 files with errors are skipped. Correct answer.
    4. No rows are loaded from any of the files in the statement.

    Explanation

    With SKIP_FILE, a file that contains any error row is not loaded at all, but the remaining files continue to be processed. This differs from CONTINUE, which loads the good rows inside a failing file.

  145. Question 145Data Loading and Unloading

    How does compute usage differ between Snowpipe and a bulk COPY INTO load?

    1. Snowpipe executes on the virtual warehouse named in the pipe definition.
    2. Snowpipe is billed per file, while COPY INTO usage is included at no cost.
    3. Both Snowpipe and COPY INTO require a running user-managed virtual warehouse.
    4. Snowpipe uses Snowflake-managed compute; COPY INTO uses a warehouse. Correct answer.

    Explanation

    Snowpipe runs on Snowflake-provided serverless compute and is billed per file loaded, so no warehouse needs to be running. COPY INTO consumes credits from the virtual warehouse that executes the statement.

  146. Question 146Data Loading and Unloading

    A user runs PUT file:///data/sales.csv @my_stage. By default, what does Snowflake do to the file during the upload?

    1. Encrypts it with a user-supplied PGP key
    2. Compresses it with gzip Correct answer.
    3. Splits it into 16 MB chunks
    4. Converts it to the Parquet format

    Explanation

    PUT compresses files automatically with gzip before staging them, unless the file is already compressed or AUTO_COMPRESS is set to FALSE. The compression reduces storage footprint and transfer volume without changing the logical content of the data.

  147. Question 147Data Loading and Unloading

    Which command copies files from an internal stage down to the local file system of the client machine?

    1. LIST
    2. PUT
    3. GET Correct answer.
    4. RM

    Explanation

    GET is the counterpart to PUT: it downloads files from an internal stage to a local directory on the client. It is used mainly for retrieving unloaded query results or staged files for inspection.

  148. Question 148Data Loading and Unloading

    Which SQL statement creates a reusable, named file format that later COPY INTO statements can reference?

    1. CREATE FILE FORMAT Correct answer.
    2. ALTER FILE FORMAT
    3. CREATE STAGE
    4. DEFINE FILE FORMAT

    Explanation

    A named file format is a schema-level object created with CREATE FILE FORMAT and then referenced as FILE_FORMAT = (FORMAT_NAME = ...) in load and unload statements. Reusing one object keeps parsing rules consistent across many COPY operations.

  149. Question 149Data Loading and Unloading

    Which statement copies the rows of an existing table out to a stage as data files?

    1. PUT @my_stage FROM my_table
    2. INSERT INTO @my_stage SELECT * FROM my_table
    3. GET @my_stage FROM my_table
    4. COPY INTO @my_stage FROM my_table Correct answer.

    Explanation

    Unloading uses COPY INTO <location> FROM <table>, which writes the query result as files into the stage. The same COPY INTO keyword is used for loading, but with the source and target reversed.

  150. Question 150Data Loading and Unloading

    An analyst runs COPY INTO t FROM @s1 and omits the ON_ERROR option. One row in the staged file holds a value that cannot be converted to the target column type. What happens?

    1. The bad value is stored as NULL and the load completes
    2. The bad row is skipped and the remaining rows load
    3. The statement fails with an error Correct answer.
    4. The file is skipped and the statement succeeds with a warning

    Explanation

    The default value of ON_ERROR is ABORT_STATEMENT, so the COPY statement stops and reports the error instead of continuing past the bad row. Explicit ON_ERROR settings such as CONTINUE or SKIP_FILE are required to make a load tolerate malformed data.

  151. Question 151Data Loading and Unloading

    When querying a stage directly, which metadata column returns the path of the staged file that each row came from?

    1. METADATA$ROW_ID
    2. METADATA$FILE_ROW_NUMBER
    3. METADATA$START_SCAN_TIME
    4. METADATA$FILENAME Correct answer.

    Explanation

    Staged files expose metadata columns that can be selected like table columns, and METADATA$FILENAME returns the full staged path of the source file for each row. This is commonly used to capture the source file name into a target column during a load.

  152. Question 152Data Loading and Unloading

    A CSV file stores missing values as the literal text NULL. Which file format option causes that text to be loaded as a SQL NULL?

    1. EMPTY_FIELD_AS_NULL = TRUE
    2. TRIM_SPACE = TRUE
    3. SKIP_BLANK_LINES = TRUE
    4. NULL_IF = ('NULL') Correct answer.

    Explanation

    NULL_IF lists the string values that should be interpreted as SQL NULL when a file is parsed. Listing the placeholder text used by the source system makes those fields load as true NULLs rather than as the string 'NULL'.

  153. Question 153Data Loading and Unloading

    Which table function takes a table name as an argument and reports the status, row counts, and error counts of that table's COPY INTO loads?

    1. COPY_HISTORY('t1') Correct answer.
    2. LOAD_HISTORY('t1')
    3. SYSTEM$PIPE_STATUS('t1')
    4. VALIDATE('t1')

    Explanation

    INFORMATION_SCHEMA.COPY_HISTORY accepts a table name and returns one row per load with status, row counts, error counts, and the file list. It is the standard way to audit which files loaded successfully into a specific table.

  154. Question 154Data Loading and Unloading

    A stage holds hundreds of files, and an analyst wants to load exactly three of them by name in a single COPY INTO statement. Which option accomplishes this?

    1. FILES = ('a.csv','b.csv','c.csv') Correct answer.
    2. FILE_LIST = 'a.csv,b.csv,c.csv'
    3. WHERE METADATA$FILENAME IN ('a.csv','b.csv','c.csv')
    4. INCLUDE = ('a.csv','b.csv','c.csv')

    Explanation

    The FILES option restricts a COPY INTO load to an explicit list of staged file names, and each entry must match the file path relative to the stage. It is the most direct way to load a known set of files.

  155. Question 155Data Loading and Unloading

    An unload to a stage produces many files because each processing thread writes its own output. Which option sets an upper size limit for each file written by the unload?

    1. MAX_FILE_SIZE = <bytes> Correct answer.
    2. MAX_FILE_COUNT = <number>
    3. BLOCK_SIZE = <bytes>
    4. SINGLE = TRUE

    Explanation

    MAX_FILE_SIZE caps the size of each file produced by an unload, and Snowflake splits the output into additional files once that limit would be exceeded. It is used when downstream consumers require files below a certain size.

  156. Question 156Data Loading and Unloading

    A pipe has been created and paused, and the stage it reads already contains a backlog of files that arrived before the pipe existed. Which statement loads that backlog?

    1. CALL SYSTEM$PIPE_STATUS('my_pipe')
    2. SYSTEM$PIPE_FORCE_RESUME('my_pipe')
    3. ALTER PIPE my_pipe RESUME
    4. ALTER PIPE my_pipe REFRESH Correct answer.

    Explanation

    ALTER PIPE ... REFRESH makes the pipe scan the stage for files still present in its load history window and ingest any that were never loaded. It is the standard way to backfill files that existed before the pipe was defined.

  157. Question 157Data Loading and Unloading

    A staged file contains a single JSON array whose elements are objects, and each object should become one table row. Which file format option should be set?

    1. STRIP_OUTER_ARRAY = TRUE Correct answer.
    2. COMPRESSION = AUTO
    3. SKIP_HEADER = 1
    4. TRIM_SPACE = TRUE

    Explanation

    By default Snowflake expects newline-delimited JSON, where each line is a separate document. STRIP_OUTER_ARRAY = TRUE removes the enclosing array so each element of the array is treated as its own document and row.

  158. Question 158Data Loading and Unloading

    Which query returns the row-level error records produced by the most recent COPY INTO load into table t1?

    1. SELECT * FROM TABLE(RESULT_SCAN(LAST_QUERY_ID()))
    2. SELECT * FROM t1 WHERE VALIDATION_MODE = 'RETURN_ERRORS'
    3. SELECT * FROM TABLE(VALIDATE(t1, JOB_ID => '_last')) Correct answer.
    4. SELECT * FROM TABLE(COPY_HISTORY(TABLE_NAME => 'T1'))

    Explanation

    The VALIDATE table function replays a completed load and returns the rejected rows along with the error message and file location for each. Passing JOB_ID => '_last' selects the most recent load of that table.

  159. Question 159Data Loading and Unloading

    Which function returns the current execution state of a pipe, including how many files it is currently processing?

    1. SYSTEM$COPY_STATUS('my_pipe')
    2. SYSTEM$LOAD_STATUS('my_pipe')
    3. SYSTEM$PIPE_REFRESH('my_pipe')
    4. SYSTEM$PIPE_STATUS('my_pipe') Correct answer.

    Explanation

    SYSTEM$PIPE_STATUS returns a JSON document describing the pipe's execution state, pending file count, and last ingested file details. It is the tool for confirming that a pipe is running and keeping up with the stage.

  160. Question 160Data Loading and Unloading

    An unload of a table to a stage must write each distinct value of the column region into its own subdirectory under the stage. Which COPY INTO <location> clause does this?

    1. CLUSTER BY region
    2. GROUP BY region
    3. SPLIT_BY = 'region'
    4. PARTITION BY region Correct answer.

    Explanation

    COPY INTO <location> accepts PARTITION BY, which writes the output into one subdirectory per partition value. The expression is often written as a concatenation such as 'region=' || region so the directory names are self-describing.

  161. Question 161Data Protection and Data Sharing

    Which Snowflake feature allows a user to query table data as it existed at an earlier point in time without restoring the table?

    1. Zero-copy cloning
    2. Fail-safe
    3. Time Travel Correct answer.
    4. Database replication

    Explanation

    Time Travel retains historical versions of data so queries can use AT or BEFORE clauses to read past states. It is user-accessible within the configured retention period. Fail-safe is a later recovery stage handled by Snowflake, not a query feature.

  162. Question 162Data Protection and Data Sharing

    What is the default Time Travel retention period for a permanent table?

    1. 0 days
    2. 1 day Correct answer.
    3. 90 days
    4. 7 days

    Explanation

    Permanent tables default to DATA_RETENTION_TIME_IN_DAYS = 1. The period can be changed per object or at the account level, within edition limits. Fail-safe adds a separate 7-day period after Time Travel ends.

  163. Question 163Data Protection and Data Sharing

    How long is the Fail-safe period for a permanent table?

    1. 7 days Correct answer.
    2. 90 days
    3. 14 days
    4. 1 day

    Explanation

    Fail-safe is a fixed 7-day period that begins after Time Travel retention ends. It is not configurable and is intended for recovery by Snowflake Support. Transient and temporary tables do not have Fail-safe.

  164. Question 164Data Protection and Data Sharing

    Which statement describes a zero-copy clone?

    1. It copies every row into new storage.
    2. It replicates rows to another region.
    3. It shares existing micro-partitions. Correct answer.
    4. It moves rows out of the source object.

    Explanation

    A zero-copy clone creates a new object whose metadata points to the source object's existing micro-partitions. No table data is copied at creation time. Storage is consumed only when the clone or source changes and new micro-partitions are written.

  165. Question 165Data Protection and Data Sharing

    Which command restores a dropped table while it is still within the Time Travel retention period?

    1. RECOVER TABLE
    2. UNDROP TABLE Correct answer.
    3. CLONE TABLE
    4. RESTORE TABLE

    Explanation

    UNDROP TABLE restores a dropped table using its Time Travel history. The table must still be within its retention period, and the name must be available. Similar UNDROP commands exist for schemas and databases.

  166. Question 166Data Protection and Data Sharing

    Which statement is true about a database that a consumer creates from a share?

    1. The consumer owns the provider's storage.
    2. The consumer can insert rows into shared tables.
    3. The consumer can re-share it to other accounts.
    4. The database is read-only for the consumer. Correct answer.

    Explanation

    A database created from a share is read-only in the consumer account. The provider retains ownership of the shared objects and storage. Consumers cannot re-share the database to other accounts.

  167. Question 167Data Protection and Data Sharing

    What is a reader account in Snowflake data sharing?

    1. A provider account that stores shared data.
    2. A consumer account that can create shares.
    3. A temporary role used during a share.
    4. A provider-managed consumer account. Correct answer.

    Explanation

    A reader account is a consumer account created and managed by the provider for users who do not have their own Snowflake account. It can consume shares but cannot create shares. Usage in a reader account is billed to the provider.

  168. Question 168Data Protection and Data Sharing

    Which type of view can be included in a share?

    1. Recursive view
    2. Secure view Correct answer.
    3. Materialized view
    4. Temporary view

    Explanation

    Only secure views can be shared because they hide the view definition and underlying data from consumers. A regular view would expose implementation details. Secure views are a standard way to share curated data.

  169. Question 169Data Protection and Data Sharing

    Which Time Travel clause returns table data from before a specified statement was executed?

    1. PRIOR (STATEMENT => 'query_id')
    2. OFFSET (STATEMENT => 'query_id')
    3. AFTER (STATEMENT => 'query_id')
    4. BEFORE (STATEMENT => 'query_id') Correct answer.

    Explanation

    BEFORE (STATEMENT => 'query_id') returns the table state immediately before the statement ran. AT (STATEMENT => ...) returns the state after the statement. This is useful for inspecting changes made by a known query.

  170. Question 170Data Protection and Data Sharing

    What is the maximum Time Travel retention period for a permanent table in Enterprise Edition?

    1. 7 days
    2. 1 day
    3. 90 days Correct answer.
    4. 365 days

    Explanation

    Enterprise Edition and above allow permanent tables to retain Time Travel data for up to 90 days. The default remains 1 day unless changed. Standard Edition is limited to 1 day.

  171. Question 171Data Protection and Data Sharing

    Who can recover data during the Fail-safe period?

    1. The table owner using UNDROP
    2. Any role with ACCOUNTADMIN
    3. Snowflake Support Correct answer.
    4. The account administrator using SQL

    Explanation

    Fail-safe data is not accessible through SQL and can be recovered only by Snowflake Support. It covers a fixed 7-day period after Time Travel ends. UNDROP works only within Time Travel retention.

  172. Question 172Data Protection and Data Sharing

    Which statement creates a clone of table SALES as it existed one hour ago?

    1. CREATE TABLE sales_clone CLONE sales BEFORE (OFFSET => -3600)
    2. CLONE TABLE sales TO sales_clone AT (OFFSET => -3600)
    3. CREATE TABLE sales_clone LIKE sales AT (OFFSET => -3600)
    4. CREATE TABLE sales_clone CLONE sales AT (OFFSET => -3600) Correct answer.

    Explanation

    Cloning with AT (OFFSET => -3600) creates the clone from the table state one hour ago. The syntax is CREATE TABLE ... CLONE ... AT (...). BEFORE is used for statement-based Time Travel, not a negative offset in this form.

  173. Question 173Data Protection and Data Sharing

    Which privilege on a source table is required to clone that table?

    1. SELECT Correct answer.
    2. DELETE
    3. OWNERSHIP
    4. INSERT

    Explanation

    Cloning a table requires SELECT on the source table and CREATE TABLE on the target schema. OWNERSHIP is not required to make a clone. The clone is a new object with its own ownership.

  174. Question 174Data Protection and Data Sharing

    Which command grants a share access to a table?

    1. GRANT SELECT ON TABLE t TO SHARE s; Correct answer.
    2. GRANT USAGE ON TABLE t TO SHARE s;
    3. GRANT REFERENCES ON TABLE t TO SHARE s;
    4. GRANT OWNERSHIP ON TABLE t TO SHARE s;

    Explanation

    A share is granted USAGE on the database and schema, and SELECT on each table or secure view. SELECT is the privilege that allows the consumer to query the shared object. Ownership is never granted to a share.

  175. Question 175Data Protection and Data Sharing

    Which shareable object allows a provider to expose logic without revealing its definition?

    1. A secure UDF Correct answer.
    2. A standard UDF
    3. A stored procedure
    4. An external function

    Explanation

    Secure UDFs hide their definition from consumers and can be included in shares. Standard UDFs are not shareable in this way. This lets providers share computation while protecting proprietary logic.

  176. Question 176Data Protection and Data Sharing

    Who is billed for compute used inside a reader account?

    1. The consumer's cloud provider
    2. The reader account's users directly
    3. Snowflake, at no charge
    4. The provider account Correct answer.

    Explanation

    Reader account usage is billed to the provider account that created it. The reader account does not pay Snowflake directly. This is why providers monitor reader account consumption.

  177. Question 177Data Protection and Data Sharing

    Which command refreshes a secondary database from its primary database?

    1. ALTER DATABASE db ENABLE REPLICATION;
    2. REFRESH DATABASE db FROM primary_db;
    3. ALTER DATABASE db REFRESH; Correct answer.
    4. CREATE DATABASE db FROM REPLICA primary_db;

    Explanation

    After a secondary database is created as a replica, ALTER DATABASE ... REFRESH pulls the latest changes from the primary. ENABLE REPLICATION is used on the primary to allow replication. The secondary database is read-only until a failover.

  178. Question 178Data Protection and Data Sharing

    What is the purpose of a dynamic data masking policy?

    1. To hide rows the role cannot see
    2. To remove columns from a share
    3. To mask column values by role Correct answer.
    4. To encrypt data at rest

    Explanation

    A masking policy transforms column values at query time according to conditions such as the current role. The underlying data is not changed. Row access policies handle row-level filtering, and encryption is separate.

  179. Question 179Data Protection and Data Sharing

    Which feature restricts which rows a role can see in a table?

    1. Secure view
    2. Dynamic data masking
    3. Object tagging
    4. Row access policy Correct answer.

    Explanation

    A row access policy returns a Boolean expression that filters rows based on role or session context. Dynamic masking changes column values, not row visibility. Secure views can hide data but are not the row-filtering policy feature.

  180. Question 180Data Protection and Data Sharing

    What is the maximum Time Travel retention for a transient table?

    1. 90 days
    2. 0 days
    3. 7 days
    4. 1 day Correct answer.

    Explanation

    Transient tables support a Time Travel retention of 0 or 1 day. They do not have Fail-safe. This makes them suitable for data that does not need long recovery windows.

  181. Question 181Data Protection and Data Sharing

    Which statement about Time Travel for a consumer querying shared data is correct?

    1. Consumers can query historical versions within the provider's retention.
    2. Consumers cannot use Time Travel on shared data. Correct answer.
    3. Consumers can set their own retention on the share.
    4. Consumers can use Fail-safe to recover shared rows.

    Explanation

    Time Travel is controlled by the provider, and consumers cannot query historical versions of shared data. The share exposes the current state of the shared objects. Consumers also cannot change retention settings on shared objects.

  182. Question 182Data Protection and Data Sharing

    What is a failover group in Snowflake?

    1. Objects replicated together for failover Correct answer.
    2. A cluster of warehouses used during an outage
    3. A set of roles that can fail over between regions
    4. A share that is replicated to a reader account

    Explanation

    A failover group defines a set of objects, such as databases and shares, that are replicated together to a secondary account. It supports promotion of the secondary during an outage. This requires appropriate edition and account configuration.

  183. Question 183Data Protection and Data Sharing

    A table is cloned with AT (OFFSET => -3600). What data does the clone contain?

    1. Rows inserted after the offset
    2. All rows ever written to the table
    3. Rows as of one hour ago Correct answer.
    4. Current rows only

    Explanation

    The AT clause makes the clone reflect the table state at the specified historical time, which is one hour ago here. Rows written after that time are not included. The clone then diverges independently from the source.

  184. Question 184Data Protection and Data Sharing

    How can a masking policy be applied to many columns without attaching it to each column individually?

    1. By enabling Fail-safe on the database
    2. By cloning the table into a transient schema
    3. By using tag-based masking Correct answer.
    4. By creating a secure view over the table

    Explanation

    Tag-based masking associates a masking policy with a tag, and columns carrying that tag inherit the policy. This scales masking across many columns. It requires the appropriate Snowflake edition and tagging setup.

  185. Question 185Data Protection and Data Sharing

    A provider shares a secure view that selects from a table. What can the consumer see?

    1. The base table rows directly
    2. The view definition and the base table
    3. The provider's account metadata
    4. Results the view exposes Correct answer.

    Explanation

    A secure view hides its definition and the underlying objects from the consumer. The consumer can query the view and see only the results the view exposes. Direct access to the base table is not granted unless it is also shared.

  186. Question 186Data Protection and Data Sharing

    How long does Snowflake retain a permanent table's data in Fail-safe once its Time Travel retention period has ended?

    1. Ninety days beyond the Time Travel retention period
    2. Seven days beyond the Time Travel retention period
    3. Until an administrator manually recreates the table
    4. One day beyond the Time Travel retention period Correct answer.

    Explanation

    Fail-safe is a fixed, non-configurable period of seven days that begins after the Time Travel retention period expires for a permanent object. During Fail-safe, data can be recovered only by Snowflake Support, not by account users.

  187. Question 187Data Protection and Data Sharing

    A table named ORDERS was dropped 30 minutes ago and its Time Travel retention is 24 hours. Which command restores the table?

    1. RECOVER TABLE ORDERS
    2. UNDROP TABLE ORDERS Correct answer.
    3. ALTER TABLE ORDERS UNDROP
    4. RESTORE TABLE ORDERS

    Explanation

    UNDROP TABLE restores a dropped table while its data is still inside the Time Travel retention window, and the restored table keeps its original privileges.

  188. Question 188Data Protection and Data Sharing

    What storage is consumed at the moment a table is created with CREATE TABLE ... CLONE?

    1. Storage equal to the source table's size divided by the number of existing clones
    2. No additional storage, because the clone shares the source micro-partitions Correct answer.
    3. One extra day of Time Travel storage for the cloned object
    4. Twice the storage of the source table, since all micro-partitions are copied

    Explanation

    A clone is a zero-copy operation: Snowflake copies metadata pointers to the existing micro-partitions rather than duplicating data. New storage is consumed only when the clone or the source diverges through DML.

  189. Question 189Data Protection and Data Sharing

    In Standard Edition, what is the maximum Time Travel retention period for a permanent table?

    1. 1 day Correct answer.
    2. 90 days
    3. 7 days
    4. 0 days, because Time Travel requires Enterprise Edition

    Explanation

    Standard Edition supports a Time Travel retention of 0 or 1 day for permanent objects. Retention up to 90 days requires Enterprise Edition or higher.

  190. Question 190Data Protection and Data Sharing

    Which is the minimum Snowflake edition that supports row access policies?

    1. Standard Edition
    2. Business Critical Edition
    3. Virtual Private Snowflake Edition
    4. Enterprise Edition Correct answer.

    Explanation

    Row access policies are an Enterprise Edition feature, as are dynamic data masking, object tagging, and multi-cluster warehouses. Higher editions include them as well, but Enterprise is the minimum.

  191. Question 191Data Protection and Data Sharing

    A table has 10 days of Time Travel retention. Which approach returns the rows that were inserted, updated, or deleted in the table over the past three days?

    1. Run three separate SELECT statements using AT (TIMESTAMP => ...) at daily intervals
    2. Query QUERY_HISTORY in INFORMATION_SCHEMA and parse the statement text
    3. Query the table with BEFORE(STATEMENT => ...) using the statement ID of the last UPDATE
    4. Query the table with CHANGES(INFORMATION => DEFAULT) together with a Time Travel clause Correct answer.

    Explanation

    The CHANGES clause returns CDC-style rows with metadata columns indicating inserts, updates, and deletes when paired with a Time Travel clause such as AT or BEFORE. Plain AT queries return a snapshot, not a change feed.

  192. Question 192Data Protection and Data Sharing

    A consumer account has been granted a share by a provider. Which privilege must a role in the consumer account hold to create a database from that share?

    1. SELECT on every table contained in the share
    2. IMPORT SHARE on the consumer account Correct answer.
    3. USAGE on the provider's source database
    4. CREATE SHARE on the consumer account

    Explanation

    The IMPORT SHARE privilege allows a role to create a database from an inbound share, and it is granted to ACCOUNTADMIN by default. USAGE on the share itself is also needed, but the import capability comes from IMPORT SHARE.

  193. Question 193Data Protection and Data Sharing

    A consumer account creates a database from a share and runs a query against it. Which virtual warehouse processes the query?

    1. A warehouse in the consumer account selected by the consumer's session Correct answer.
    2. A Snowflake-managed warehouse created automatically for each share
    3. A warehouse inside the provider's reader account
    4. A warehouse in the provider account assigned to the share

    Explanation

    With a standard share between two Snowflake accounts, the consumer queries the data using its own virtual warehouse and is billed for that compute. The provider is billed only for storage of the shared data.

  194. Question 194Data Protection and Data Sharing

    What happens to the values stored in a column when a masking policy is attached to it?

    1. Values are overwritten with masked values on disk
    2. Values are re-encrypted by the policy on every write
    3. Values are copied into a hidden masked column
    4. Values are unchanged; masking applies at query time Correct answer.

    Explanation

    A masking policy is evaluated when the column is queried, so the underlying data at rest is never modified. Users whose role satisfies the policy condition see the real value, and all other roles see the masked result.

  195. Question 195Data Protection and Data Sharing

    A table EMP is cloned, and then a row is deleted from the clone. What is the effect on the original table EMP?

    1. The deletion fails because cloned tables are read-only
    2. The original table is unaffected and keeps the row Correct answer.
    3. The same row is deleted from the original table
    4. The original table becomes read-only until the clone is dropped

    Explanation

    A clone is an independent object from the moment it is created. DML on the clone causes copy-on-write of the affected micro-partitions and never alters the source table.

  196. Question 196Data Protection and Data Sharing

    A provider granted SELECT on table SALES to a share. The provider now wants the consumer to keep reading the other tables in the share but to lose access to SALES. Which command accomplishes this?

    1. DROP SHARE <share_name>
    2. DELETE FROM SALES WHERE ROLE = <account_name>
    3. REVOKE SELECT ON TABLE SALES FROM SHARE <share_name> Correct answer.
    4. ALTER SHARE <share_name> REMOVE ACCOUNT=<account_name>

    Explanation

    REVOKE removes the object-level privilege from the share while leaving the share and its other grants in place. Dropping the share or removing the account would cut off every object in the share.

  197. Question 197Data Protection and Data Sharing

    A table inherits a database-level DATA_RETENTION_TIME_IN_DAYS of 1. The table must keep 20 days of Time Travel history. Which change is required?

    1. Set DATA_RETENTION_TIME_IN_DAYS to 20 on the account
    2. Set DATA_RETENTION_TIME_IN_DAYS to 20 on the table Correct answer.
    3. Clone the table with AT (OFFSET => -1728000)
    4. Recreate the database with a longer retention period

    Explanation

    DATA_RETENTION_TIME_IN_DAYS can be set at the account, database, schema, or table level, and the most specific setting wins. Setting it on the table overrides the inherited database value without affecting other tables.

  198. Question 198Data Protection and Data Sharing

    A provider accidentally dropped a table and must restore it with all of its original access privileges intact. Time Travel retention is 7 days and Fail-safe has not begun. Which approach satisfies the requirement?

    1. Create a new table with CLONE at an offset within the retention window
    2. Ask Snowflake Support to restore the table from Fail-safe
    3. Create a new table with CREATE TABLE AS SELECT from the table at an offset
    4. Run UNDROP TABLE on the dropped table Correct answer.

    Explanation

    UNDROP restores the dropped object itself, so its existing grants and properties come back with it. Cloning or CTAS creates a new object whose privileges must be granted again.

  199. Question 199Data Protection and Data Sharing

    A provider must let consumers in another region query a shared database that is refreshed on a schedule, and no account failover is required. Which object should the provider create?

    1. A secure view created inside each consumer account
    2. A replication group containing the database and the share Correct answer.
    3. A reader account located in the consumer's region
    4. A failover group containing the database and the share

    Explanation

    A replication group replicates databases and shares to accounts in other regions and supports scheduled refreshes without promoting the replica. Failover groups are for disaster recovery scenarios where a secondary account becomes primary.

  200. Question 200Data Protection and Data Sharing

    Which statement describes Tri-Secret Secure?

    1. It stores three copies of each micro-partition across three cloud regions
    2. It combines a Snowflake-managed key with a customer-managed key held in the customer's KMS Correct answer.
    3. It rotates the account's encryption keys automatically every 30 days
    4. It requires the consumer to supply a key when creating a database from a share

    Explanation

    Tri-Secret Secure uses a composite master key made from a Snowflake-managed key and a customer-managed key in the customer's cloud key management service. It is available with Business Critical Edition and higher, and revoking the customer key makes the account's data unreadable.

SnowPro Core flashcards

32 cards. Read the term and the definition here. The deck at the top of this page is optional.

Three Snowflake layers
Storage holds data in cloud object storage, compute runs virtual warehouses, and cloud services manages metadata, security, optimization, and transactions across the account.
Micro-partition size range
Snowflake stores table data in immutable, compressed, columnar micro-partitions typically between 50 and 500 MB, with metadata used for pruning and clustering.
Virtual warehouse sizing
Warehouses range from X-Small to 6X-Large. Larger sizes add compute for bigger workloads and consume more credits per hour; they do not always speed small queries.
Result cache lifetime
Query results are cached for 24 hours by default and reused when the same query text runs and underlying data has not changed, without using warehouse compute.
Multi-cluster warehouse purpose
A multi-cluster warehouse starts additional compute clusters automatically when concurrency rises. It reduces queuing for many concurrent queries but does not speed a single slow query.
Snowflake account editions
Standard, Enterprise, Business Critical, and Virtual Private Snowflake. Higher editions add capabilities such as longer Time Travel, multi-cluster warehouses, Tri-Secret Secure, and dedicated infrastructure.
ACCOUNTADMIN role
The highest-level system role with all account privileges. It should be granted sparingly; SECURITYADMIN manages grants, SYSADMIN manages objects, and USERADMIN manages users and roles.
Future grants
Privileges that apply automatically to objects created later in a database or schema. They reduce the need to grant permissions on each new table or view.
Network policy
An account-level or user-level control that allows or blocks access by IP address. A user-level policy overrides the account-level policy for that user.
Tri-Secret Secure
A Business Critical feature that combines a Snowflake-managed key with a customer-managed key. Both keys are required to decrypt data at rest.
Masking policy
A schema-level object that changes column values at query time based on the user's role. The stored data remains unchanged and access is controlled by the policy.
Query Profile spilling
Local or remote spilling shown in Query Profile indicates the warehouse lacks memory for the query. Increasing warehouse size adds memory and can reduce or eliminate spilling.
Clustering key
Defines the physical ordering of micro-partitions for a table. It improves pruning for large tables filtered on high-cardinality or range columns.
Search optimization service
Maintains a search access path that speeds selective point lookups and substring queries. It is most useful for equality and IN filters on large tables.
Materialized view
A precomputed query result that Snowflake refreshes automatically. It can accelerate repeated aggregations but adds storage cost and maintenance overhead.
Micro-partition pruning
Snowflake uses micro-partition metadata to skip partitions that cannot match filter predicates. Wrapping filter columns in functions can prevent pruning and slow queries.
FLATTEN table function
Explodes semi-structured arrays or objects into rows. Use LATERAL FLATTEN to join each flattened element with its source row.
QUALIFY clause
Filters the results of window functions without requiring a subquery. It is evaluated after window functions and before ORDER BY.
Zero-copy clone
CREATE TABLE ... CLONE creates a new table that shares existing micro-partitions. No data is copied until either table is modified.
Time Travel AT clause
Queries historical table data using AT or BEFORE with TIMESTAMP, OFFSET, or STATEMENT. Available history depends on the table's data retention period.
Stream object
Captures change data for a table, view, or directory. It records inserts, updates, and deletes so downstream tasks can process changed rows.
Task object
Schedules SQL statements such as INSERT, MERGE, or stored procedure calls. Tasks can run on warehouse or serverless compute and be chained into DAGs.
COPY INTO command
Loads staged files into a table or unloads table data to a stage. Key options include ON_ERROR, PURGE, VALIDATION_MODE, and FILE_FORMAT.
Snowpipe
A continuous ingestion service that uses serverless compute. It loads new staged files automatically when event notifications or REST calls signal availability.
Recommended file size
For bulk loading, compressed files of roughly 100 to 250 MB are recommended. Many tiny files add overhead and degrade pruning performance.
Storage integration
An account-level object that stores cloud storage credentials for external stages. It avoids embedding access keys directly in stage definitions.
VALIDATION_MODE option
A COPY INTO option that checks files without loading data. It returns errors and can be set to RETURN_ERRORS or RETURN_N_ROWS.
Fail-safe period
Seven days of recovery after Time Travel ends. Fail-safe data is not queryable and can be restored only by Snowflake Support.
Time Travel retention
Standard accounts allow 0 or 1 day; Enterprise and higher allow up to 90 days. Retention is set per table or schema.
Secure data sharing
Provides read-only access to a database without copying data. Consumers query shared data using their own virtual warehouses and pay for their compute.
Reader account
A separate Snowflake account created by a provider for consumers without their own account. The provider pays for storage and compute used by the reader account.
Database replication
Copies databases to another account or region for disaster recovery or sharing. Secondary databases are read-only until a failover promotes them.

SnowPro Core study guide

Snowflake AI Data Cloud Features and Architecture

The three layers

Snowflake's architecture separates storage (cloud object storage holding micro-partitions), compute (virtual warehouses), and cloud services (metadata, optimizer, transactions, security, authentication). Micro-partitions are immutable, columnar, compressed units of roughly 50–500 MB of uncompressed data. Snowflake records min/max values and distinct counts for each column in each partition, which is exactly what makes partition pruning possible without indexes or user maintenance.

Warehouses run from X-Small to 6X-Large, each step roughly doubling compute and credit burn. Scale up (bigger warehouse) for a single slow, heavy query; scale out (multi-cluster warehouse) for many concurrent queries. Warehouses also cache table data on local SSD - that cache belongs to the warehouse and does not follow you to another one. The result cache is account-wide, lasts 24 hours, and is invalidated when the underlying data changes.

Traps

  • A multi-cluster warehouse does not make one query faster; it reduces queueing.
  • Metadata-only operations (SHOW, DESCRIBE, DDL, unfiltered COUNT(*)) do not require a running warehouse.
  • Resizing helps a query that spills; it does not fix a table that cannot prune.

Rule of thumb

If the Query Profile shows "Bytes spilled to remote storage", the warehouse is too small - double the size before rewriting SQL.

Account Access and Security

Privileges, roles, and hierarchy

Snowflake applies discretionary access control on top of role-based access control. A privilege (USAGE, SELECT, INSERT, CREATE TABLE, OWNERSHIP…) is granted on a securable object (database, schema, table, warehouse) to a role; roles are granted to users and to other roles. Effective rights are the union of privileges held by the active role and every role beneath it in the hierarchy.

The built-in chain is ACCOUNTADMIN → SECURITYADMIN → USERADMIN, with SYSADMIN owning warehouses and databases and PUBLIC granted to every user. Best practice: build custom functional roles (e.g. ANALYST_RO), grant object privileges to them, then grant those roles up to SYSADMIN; keep user and role administration in SECURITYADMIN/USERADMIN.

Traps

  • Ownership carries every privilege including the right to grant, so it is usually the answer to "who can grant access?".
  • In a managed access schema, object owners cannot grant; the schema owner or a role with MANAGE GRANTS must.
  • Without GRANT ... ON FUTURE TABLES IN SCHEMA, newly created tables are invisible to a role.
  • A role must be active (or the user's default role) before its privileges apply.

Other controls

Network policies (IP allow/block lists), MFA, federated SAML SSO, key-pair and OAuth for service users, and masking/row access policies for column- and row-level protection.

Rule of thumb

When access "should work" but doesn't, walk the chain: privilege on the object → granted to a role → that role granted into the user's active role hierarchy → role actually active.

Performance Optimization, Querying, and Data Transformation

Performance Optimization

What makes a query fast

Almost all Snowflake tuning reduces bytes scanned. Because every micro-partition stores min/max metadata per column, the optimizer prunes partitions that cannot match a filter. Clustering - natural, from load order, or an explicit clustering key - keeps related rows in the same partitions so pruning stays effective as the table grows; clustering depth/overlap measures how well sorted the table is. Good clustering keys are high-cardinality columns used in WHERE clauses, not low-cardinality flags.

Use the Query Profile to find real cost: the most expensive operator, partitions scanned versus total, and spilling. Spilling goes to local disk first, then remote storage; remote spilling means the warehouse is undersized.

Tooling: search optimization service for selective point lookups and substring searches, materialized views for repeated aggregations, result and local disk caches for repeats, and EXPLAIN plans to inspect pruning.

Traps

  • Resizing a warehouse does not fix poor pruning; clustering does.
  • Multi-cluster warehouses solve queueing, not slow queries.
  • Automatic clustering runs in the background and consumes credits.
  • Adding a clustering key to a table that is already naturally sorted wastes credits.
  • An undersized warehouse on a huge scan can spill and run far longer than a bigger one, even at similar total credits.

Rule of thumb

Read the profile top-down: is the scan pruning badly (partitions scanned ÷ total), or is the join spilling? Fix pruning with clustering and selective filters; fix spilling with a bigger warehouse.

Querying and Data Transformation

SQL surface you must know

Snowflake supports ANSI SQL with strong analytical extensions: window functions, QUALIFY (filters window results without a subquery), CTEs, PIVOT/UNPIVOT, GROUPING SETS/ROLLUP/CUBE, ASOF joins, and time-travel clauses (AT, BEFORE) in the FROM clause.

Semi-structured data lives in VARIANT columns. Navigate with the colon path (src:customer.id::string) or bracket notation (src['customer']). LATERAL FLATTEN(input => src:items) explodes arrays into rows and must be lateral-joined to the source row. TRY_CAST/TRY_TO_NUMBER return NULL instead of failing on bad input.

Transformation patterns

  • Streams + tasks for change data capture and scheduled pipelines: a stream records DML on a table, a task runs SQL on a schedule or when the stream has data.
  • Dynamic tables for declarative, refresh-on-schedule transformations.
  • UDFs/UDTFs (SQL, JavaScript, Python) and Snowpark for procedural logic.
  • Zero-copy CLONE for dev/test copies of production data.

Traps

  • QUALIFY is evaluated after window functions but before ORDER BY; it is not a substitute for HAVING on aggregates.
  • FLATTEN without LATERAL can behave like a cross join against every row in the table.
  • VARIANT keys are case-sensitive: src:City and src:city are different paths.
  • A stream becomes stale if it is not consumed within the table's data retention period.

Rule of thumb

Filtering on ROW_NUMBER() or SUM() OVER ()? Write QUALIFY. Filtering on a grouped total? Write HAVING.

Data Loading and Unloading

Stages and the load path

Snowflake loads through stages. A user stage (@~) is private to one user, a table stage (@%table_name) is tied to a table and usable by anyone with INSERT rights, and a named stage (internal @my_stage, or external pointing at S3/Azure/GCS) is a database object you can grant privileges on. PUT uploads local files to an internal stage and GET downloads them; both run only through a client such as SnowSQL or the connectors - not from a worksheet.

COPY INTO <table> FROM @stage loads data under a file format (type, field delimiter, SKIP_HEADER, compression, error handling). COPY is idempotent: a file already loaded is skipped for 64 days unless FORCE = TRUE. ON_ERROR controls behavior (ABORT_STATEMENT is the default; also CONTINUE and SKIP_FILE). VALIDATION_MODE checks files without loading them. PURGE = TRUE deletes staged files after a successful load. Metadata columns such as METADATA$FILENAME let you audit source files.

Continuous and bulk

Snowpipe loads files automatically as they arrive, using file event notifications and Snowflake-managed compute. For unloading, COPY INTO @stage FROM table writes files; control output with SINGLE, MAX_FILE_SIZE, HEADER, OVERWRITE, and partition the output by column for efficient downstream reads.

Traps

  • Trying PUT inside the web worksheet.
  • Forgetting a file format when the stage has none, or mismatched delimiters.
  • Expecting Snowpipe to re-ingest a file it already processed.
  • Loading one giant file; splitting into 100–250 MB compressed files parallelizes far better.

Rule of thumb

Stage → validate → load: PUT, then COPY ... VALIDATION_MODE = 'RETURN_ERRORS', then the real load with ON_ERROR = 'SKIP_FILE'.

Data Protection and Data Sharing

Time Travel, Fail-safe, cloning

Time Travel lets you query or restore historical data: SELECT ... AT(OFFSET => -3600) or BEFORE(STATEMENT => ...), plus UNDROP TABLE/SCHEMA/DATABASE. Retention is 0–1 day on Standard and up to 90 days on Enterprise, configurable per table with DATA_RETENTION_TIME_IN_DAYS. Fail-safe is a fixed, non-configurable 7 days of recovery data that only Snowflake Support can access - it is not queryable and is not a backup you can restore yourself.

Zero-copy cloning (CREATE TABLE t2 CLONE t1) copies metadata only; storage is shared until either table's micro-partitions change. Clones are instant and cheap, which is why they underpin dev/test and fast recovery.

Sharing

Secure data sharing exposes database objects to another account through a share without copying data; the consumer creates a database from the share and queries it with their own warehouse. Providers can share with reader accounts for consumers who are not Snowflake customers. Secure views, masking policies, and row access policies protect data inside shares; replication extends objects and shares to other regions for business continuity. Snowflake Data Clean Rooms enable privacy-preserving collaboration.

Traps

  • Fail-safe cannot be queried, shortened, or extended, and is not a user-accessible backup.
  • Cloning does not duplicate data at clone time; storage grows only as the copies diverge.
  • Dropping a table does not immediately free storage while Time Travel is active.
  • A share consumer cannot see the underlying table definition behind a secure view.

Rule of thumb

Accidental DELETE/DROP inside retention → Time Travel or UNDROP. Beyond retention → Snowflake Support and Fail-safe. Need a copy for testing → CLONE, never CREATE TABLE AS SELECT.

SnowPro Core cheat sheet

Quick reference

Snowflake AI Data Cloud Features and Architecture

  • Snowflake uses three layers: cloud services (metadata, security, optimization), compute (virtual warehouses), and storage (cloud object storage).

  • Object hierarchy: Organization → Account → Database → Schema → objects such as tables, views, stages, and functions.

  • Virtual warehouse sizes range from X-Small to 6X-Large; resizing changes compute capacity and credit consumption per hour.

  • Micro-partitions are immutable, compressed, columnar storage units typically 50–500 MB; metadata enables pruning.

  • Multi-cluster warehouses add clusters automatically for concurrency; auto-suspend and auto-resume reduce idle cost.

  • Editions: Standard, Enterprise, Business Critical, and Virtual Private Snowflake; higher editions add features such as longer Time Travel and Tri-Secret Secure.

  • Result cache persists up to 24 hours by default and is reused when identical query text runs and data has not changed.

  • Snowpark, Cortex AI, Marketplace, Streamlit, and Iceberg tables are AI Data Cloud capabilities built on the core platform.

Quick reference

Account Access and Security

  • RBAC grants privileges to roles; roles are granted to users and to other roles to form a hierarchy.

  • System roles include ACCOUNTADMIN, SECURITYADMIN, USERADMIN, SYSADMIN, and PUBLIC.

  • Authentication options include password with MFA, SAML federated SSO, OAuth, and key-pair authentication.

  • Network policies allow or block IP ranges at account or user level; a user-level policy overrides the account default.

  • Encryption uses AES-256 at rest and TLS in transit; Tri-Secret Secure combines Snowflake and customer-managed keys in Business Critical.

  • Masking policies transform column values at query time; row access policies filter rows based on role or session context.

  • Future grants apply to objects created later in a database or schema; MANAGE GRANTS controls grant operations.

  • Secure views hide the view definition and underlying data from unauthorized users.

Quick reference

Performance Optimization

  • Larger warehouses add compute linearly but cost proportionally more credits per hour; test before resizing.

  • Use Query Profile to find pruning inefficiency, exploding joins, and local or remote spilling.

  • Spilling to local or remote storage suggests insufficient warehouse memory; increasing size reduces spills.

  • Cluster large tables on high-cardinality columns used in filters; automatic clustering can maintain keys.

  • Search optimization service benefits selective point lookups and substring searches.

  • Materialized views accelerate repeated aggregations but consume storage and require maintenance.

  • Avoid functions on filter columns and use explicit casts to preserve micro-partition pruning.

  • Result cache returns identical query results without using warehouse compute when data is unchanged.

  • Multi-cluster warehouses solve concurrency queuing, not slow single-query execution.

Quick reference

Querying and Data Transformation

  • Semi-structured data uses VARIANT; access fields with colon notation and array elements with brackets.

  • LATERAL FLATTEN explodes arrays or objects into rows and joins them to the source row.

  • QUALIFY filters window function results without a subquery; it runs after window evaluation.

  • Time Travel queries use AT or BEFORE with TIMESTAMP, OFFSET, or STATEMENT.

  • CREATE ... CLONE creates zero-copy clones that share micro-partitions until modified.

  • Streams capture inserts, updates, and deletes; tasks schedule SQL and can form DAGs.

  • Stages hold data files: user stage (@~), table stage (@%table), named internal stage, or external stage.

  • MERGE performs insert, update, and delete in one statement based on a join condition.

Quick reference

Data Loading and Unloading

  • Load with COPY INTO

    FROM @stage; unload with COPY INTO @stage FROM
    .

  • PUT uploads local files to an internal stage; GET downloads staged files to a local directory.

  • ON_ERROR options include CONTINUE, SKIP_FILE, and ABORT_STATEMENT; VALIDATION_MODE checks files without loading.

  • Snowpipe provides continuous serverless ingestion triggered by event notifications or REST calls.

  • Recommended compressed file size is roughly 100–250 MB; avoid many small files.

  • External stages require a storage integration or credentials for S3, Azure Blob, or GCS.

  • File formats include CSV, JSON, Avro, ORC, Parquet, and XML; FILE_FORMAT defines parsing options.

  • Stage metadata columns include METADATA$FILENAME and METADATA$FILE_ROW_NUMBER.

Quick reference

Data Protection and Data Sharing

  • Time Travel retention: 0–1 day in Standard, up to 90 days in Enterprise and higher.

  • Fail-safe adds 7 days after Time Travel; only Snowflake Support can restore fail-safe data.

  • Zero-copy cloning duplicates metadata, not data; changes to either object create new micro-partitions.

  • Secure data sharing grants read-only access without copying data; consumers use their own warehouses.

  • Reader accounts let providers share data with consumers who lack a Snowflake account; the provider pays compute.

  • Database replication copies objects to another region or account; secondary databases are read-only until failover.

  • Replication and failover groups support business continuity across accounts and regions.

  • Masking policies, row access policies, and secure views protect shared or sensitive data.

SnowPro Core key terms

Three-layer architecture

Snowflake separates storage, compute, and cloud services into independent layers, so each can scale and be billed on its own without affecting the others.

Related: Virtual warehouse, Cloud services layer, Micro-partition
Virtual warehouse

A named cluster of compute resources that executes queries and DML. Warehouses are started and stopped on demand and are billed per second while running, based on their size.

Related: Multi-cluster warehouse, Warehouse cache, Resource monitor
Micro-partition

The unit in which table data is stored: a contiguous, columnar, compressed group of rows, roughly 50 to 500 MB uncompressed, with metadata used for pruning.

Related: Clustering key, Partition pruning, Time Travel
Cloud services layer

The layer that coordinates the account: authentication, access control, metadata, query parsing and optimization, and transaction management. It runs continuously and is not tied to a warehouse.

Related: Result cache, Three-layer architecture
Multi-cluster warehouse

A warehouse that can run several clusters of the same size at once, adding clusters when queries queue and removing them when idle. Requires Enterprise edition or higher.

Related: Virtual warehouse, Resource monitor
Clustering key

A column or expression defined on a table that Snowflake uses to co-locate related rows in the same micro-partitions, improving pruning for large tables and selective filters.

Related: Micro-partition, Partition pruning
Role

A named set of privileges to which access is granted. Users receive roles, and roles can be granted to other roles, so permissions are never assigned to users directly.

Related: Role hierarchy, Privilege
Privilege

A specific permission to perform an action on an object, such as SELECT on a table or USAGE on a warehouse. Privileges are granted to roles.

Related: Role, Managed access schema
Role hierarchy

The tree of roles in an account. A role inherits every privilege of the roles granted to it, so privileges flow downward from ACCOUNTADMIN toward PUBLIC.

Related: Role, Privilege
Managed access schema

A schema where object owners cannot grant access to their objects; only the schema owner or a role with MANAGE GRANTS can, which centralizes control.

Related: Privilege, Role
Network policy

An account-level rule set listing allowed or blocked IP addresses, optionally per user, that limits where connections to the account can originate.

Row access policy

A schema-level policy attached to a table or view that decides, at query time and per role, which rows are returned; stored data is unchanged.

Related: Data masking policy, Role
Data masking policy

A schema-level policy applied to a column that returns masked values to roles without the policy's exemption, while the stored data stays intact.

Related: Row access policy, Role
Partition pruning

Snowflake's use of micro-partition metadata to skip partitions that cannot contain rows matching a filter, cutting the amount of data scanned by a query.

Related: Micro-partition, Clustering key
Query profile

A Snowsight view of a single query's execution plan, showing each operator's time, rows, bytes, and spilling, used to find performance bottlenecks.

Related: Spilling, Partition pruning
Result cache

A cloud services cache holding the results of recent queries for 24 hours. An identical query reuses them without a running warehouse if the underlying data has not changed.

Related: Cloud services layer, Warehouse cache
Warehouse cache

The local SSD cache on a warehouse's compute nodes that holds table data read by earlier queries. It is lost when the warehouse is suspended.

Related: Virtual warehouse, Result cache
Spilling

When an operation such as a sort or join exceeds warehouse memory, intermediate data is written to local disk and then remote storage, which slows the query.

Related: Virtual warehouse, Query profile
Search optimization service

An Enterprise feature that builds and maintains an auxiliary index so highly selective point lookups and substring searches avoid scanning whole tables.

Related: Partition pruning
Resource monitor

An account or warehouse-level object that tracks credit consumption and can suspend warehouses or notify users when spending reaches set thresholds.

Related: Virtual warehouse, Multi-cluster warehouse
Window function

An analytic function that computes a value across a set of related rows, such as a running total or rank, without collapsing those rows into one.

Related: QUALIFY
QUALIFY

A clause that filters rows after window functions are evaluated, letting you keep, for example, only the top-ranked row per group without a subquery.

Related: Window function
VARIANT

A semi-structured data type that stores JSON, Avro, ORC, or Parquet values up to 16 MB; elements are read with path notation or cast to other types.

Related: FLATTEN
FLATTEN

A table function that converts arrays or nested objects inside a VARIANT column into separate rows, usually combined with LATERAL to keep the parent row's columns.

Related: VARIANT
Dynamic table

A table whose content is defined by a query and refreshed automatically by Snowflake to a target lag, removing the need to schedule manual transformation jobs.

Related: Stream
Stored procedure

Procedural code written in Snowflake Scripting, JavaScript, Python, Java, or Scala and run with CALL, used for multi-step logic, loops, and error handling.

Stage

A named location that points to internal Snowflake storage or an external cloud bucket and is used as the source or target of loads and unloads.

Related: File format, COPY INTO, Storage integration
File format

A named object storing parsing rules such as type, field delimiter, header, and compression, referenced by stages and COPY statements.

Related: Stage, COPY INTO
COPY INTO

The command that loads files from a stage into a table, or unloads query results from a table to a stage, with options for error handling and file management.

Related: Stage, Snowpipe
Snowpipe

A continuous loading service that runs a pipe's COPY statement shortly after new files land in a stage, using Snowflake-managed compute instead of a user warehouse.

Related: COPY INTO, Stage
Storage integration

An account-level object holding a trust relationship and credentials for an external cloud storage provider, so stages can read buckets without storing keys in SQL.

Related: Stage, External table
External table

A read-only table whose rows come from files in an external stage rather than Snowflake storage; partition columns and file metadata are queried like normal columns.

Related: Stage, Storage integration
Time Travel

The ability to query, clone, or restore data as it looked at an earlier point, for 0 to 90 days depending on edition and the retention set on the object.

Related: Fail-safe, Zero-copy clone
Fail-safe

A fixed seven-day period after Time Travel ends during which Snowflake may recover data after a disaster. Users cannot query or restore from Fail-safe.

Related: Time Travel
Zero-copy clone

Creating a copy of a table, schema, or database that initially shares the same micro-partitions, so no data is duplicated until either copy is modified.

Related: Time Travel, Micro-partition
Stream

An object that records inserts, updates, and deletes on a table so downstream tasks can process only changed rows, enabling change data capture.

Related: Time Travel, Dynamic table
Secure data sharing

Sharing live database objects with other Snowflake accounts through a share, without copying data; consumers query with their own warehouses and pay their own compute.

Related: Reader account, Zero-copy clone
Reader account

A Snowflake account created and paid for by a provider so a consumer that does not own a Snowflake account can query the shared data.

Related: Secure data sharing

SnowPro Core study plan

8 weeks · About 7 hours per week

Week 1

Snowflake AI Data Cloud Features and Architecture

  • Diagram the Snowflake architecture layers (storage, compute, cloud services) and explain in writing how each layer scales independently

  • Build a trial account and provision three virtual warehouse sizes, then compare credit burn and query latency on the same 50-row and 50-million-row query

  • Describe the database/schema/table object hierarchy and create a working example of each object type

  • Explain micro-partitioning and columnar storage well enough to predict why a point-lookup query scans less data than a full-table aggregate

Your checkpoint

A one-page architecture cheat sheet plus a screenshot log of the warehouse-size experiment with measured credit usage

Week 2

Snowflake AI Data Cloud Features and Architecture & Account Access and Security

  • Compare Snowflake editions and list which features (Time Travel length, masking, replication) unlock at each tier

  • Create a role hierarchy with three custom roles and one functional role, then verify inherited privileges with a query against the grants views

  • Summarize how the cloud services layer handles metadata, query parsing, optimization, and access control

  • Distinguish between virtual warehouse caching, results caching, and metadata caching, and state when each is invalidated

Your checkpoint

A role-hierarchy diagram plus a saved worksheet of GRANT statements that reproduces it from scratch

Week 3

Performance Optimization, Querying, and Data Transformation

  • Diagnose a slow query using the query profile, identifying pruning, spilling, and exploding-join symptoms

  • Apply the clustering key to a large table and measure the change in partitions scanned before and after

  • Tune warehouse size and multi-cluster settings against a concurrency workload and record the tradeoffs

  • Read a query plan and explain when the optimizer prunes, filters, or rewrites a join

Your checkpoint

A before/after performance report for three queries with partitions-scanned and execution-time numbers

Week 4

Performance Optimization, Querying, and Data Transformation

  • Write queries using window functions, QUALIFY, and common table expressions to solve ranking and de-duplication problems

  • Transform semi-structured JSON with VARIANT, FLATTEN, and dot/bracket path notation

  • Implement a Type 2 slowly changing dimension or an incremental aggregate using streams and tasks

  • Explain how Snowpark, UDFs, and stored procedures extend SQL transformation inside Snowflake

Your checkpoint

A library of 10 saved transformation queries, each annotated with the scenario it solves and the output it produces

Week 5

Data Loading and Unloading

  • Load CSV and JSON files from an internal stage using COPY INTO with a defined file format and error-handling options

  • Configure an external stage backed by cloud storage and validate connectivity with a LIST command

  • Debug a failed load by inspecting the COPY history and the rejected-records output

  • Load data with schema evolution or a transformation clause without a separate staging table

Your checkpoint

An end-to-end load script that ingests three file formats into a raw schema and reports row counts plus rejected rows

Week 6

Data Loading and Unloading

  • Unload query results to an external stage in Parquet and CSV with COPY INTO and verify the file layout

  • Set up Snowpipe for continuous ingestion and confirm near-real-time loading with a file-drop test

  • Compare bulk COPY, Snowpipe, and Snowpipe Streaming on latency, cost, and setup complexity

  • Apply file sizing, compression, and partitioning practices that minimize load time and cost

Your checkpoint

A runbook documenting ingestion and unload patterns with a Snowpipe latency measurement from a timed file drop

Week 7

Data Protection and Data Sharing & Account Access and Security

  • Implement Time Travel and Fail-safe policies and recover a dropped table and a modified table with the correct syntax for each case

  • Clone a database and a schema, then prove that the clone is zero-copy by comparing storage usage

  • Create a secure data share with a secure view, and contrast a direct share with a reader account

  • Configure a network policy, MFA enforcement, and a masking policy, then verify each blocks the intended access path

Your checkpoint

A recovery-and-sharing lab write-up showing the restore statements, the clone verification query, and a diagram of the share consumers

Week 8

Account Access and Security & Data Protection and Data Sharing & Performance Optimization, Querying, and Data Transformation

  • Review replication and failover concepts for cross-region business continuity

  • Re-audit weak domains using missed-question tags from prior weeks and rewrite notes for each gap

  • Complete a timed 100-question practice set under exam conditions

  • Re-teach your three weakest topics aloud from memory without notes

Your checkpoint

A scored practice exam (target 80% or higher) with a written error log naming the root cause for every missed question

Milestones

Week 2: Can explain storage/compute separation and reproduce a working role hierarchy from memory without referencing notes

Week 4: Can read a query profile and name the specific fix for spilling, poor pruning, and exploding joins in under five minutes each

Week 6: Can build a stage-to-table pipeline and a table-to-stage unload from a blank worksheet in one sitting

Week 8: Scored 80% or higher on a full-length timed practice exam and closed every gap in the error log

SnowPro Core exam day

Your checklist

Government-issued photo ID

Bring the original, physical ID you registered with. The name on the ID must match your registration exactly - if you have a preferred name or a recently changed surname, resolve it with the provider before exam day rather than at check-in. Confirm the accepted ID types and whether a secondary ID is required with the provider (needs_review).

Appointment confirmation and login details

Have your registration confirmation, candidate/order ID, and the exam launch link or test-center address handy. Double-check the start time in your local time zone - the confirmation may display the provider's time zone.

Reliable internet connection

For a remotely proctored sitting, a wired ethernet connection is more stable than Wi-Fi. Have a backup plan (mobile hotspot, a second location) in case your primary connection fails at launch.

Working webcam, microphone, and speakers

If you are testing remotely, the proctor needs to see and hear you during check-in and the session. Confirm whether a headset is permitted or prohibited with the provider before you rely on one (needs_review).

Power and a charged device

Plug in your laptop and have the charger within reach. A mid-exam shutdown costs time you will not get back.

A quiet, private, well-lit room

Clear the desk completely. Expect a 360-degree room scan and a desk-surface check at check-in for remote sittings. Tell household members not to enter the room for the full duration.

Close every other application

Shut down email, chat, messaging, screen-sharing, note-taking, and virtual-machine software before launch. Disable OS notifications and any scheduled tasks that could pop a window onto the screen.

Complete the provider's system/equipment test in advance

Run the official readiness check at least a day early, not five minutes before your slot. It catches browser, camera-permission, and bandwidth problems while you still have time to fix them.

Confirm the scratch-paper / notes policy

Rules on physical paper, whiteboards, and on-screen notepads vary by delivery mode and proctor. Confirm what is allowed for your specific appointment with the provider; do not assume (needs_review).

Confirm the break and leave-the-room policy

Many proctored certification exams do not pause the clock for breaks, and leaving camera view may end the session. Confirm the rules for your appointment with the provider (needs_review).

Know the format before you sit down

Review the current official exam guide for question count, time limit, question types (single-answer and multi-select), and passing score. Figures circulate widely online and change over time - verify against the provider's current guide rather than a forum post (needs_review).

Water and anything permitted, ready in advance

If a clear container of water is allowed, have it poured before check-in. Anything not explicitly permitted should be off the desk and out of frame.

Plan the day

Final 72 hours

Stop adding new topics. Re-read your own summary notes, the official exam guide's domain weightings, and the areas you scored weakest on in practice. Do one timed, full-length practice set under exam conditions and review every miss, including the ones you got right by guessing.

Final 72 hours

Re-confirm logistics: appointment time and time zone, delivery mode, test-center address or remote-proctor requirements, ID validity and name match, and any permitted-items list. Resolve discrepancies now, not at check-in.

Final 24 hours

Run the official system/equipment test. Update your browser, clear permissions issues, and charge everything. Prepare your ID and confirmation in one place near your desk.

Final 24 hours

Do a light review only - architecture and object-model fundamentals, key distinctions you keep confusing, and a skim of your error log. No new material, no marathon practice exam. Sleep is worth more than another hour of cramming.

Morning of

Eat, hydrate, and avoid anything that will make you need the restroom mid-exam. Do a five-minute skim of your one-page summary, not a deep dive.

60-30 minutes before

Set up your workspace, close all other applications, plug in power, and log in early. Remote check-in with ID verification and a room scan routinely takes longer than expected; being early is free, being late is not.

During the test - first pass

Answer every question you can resolve quickly and confidently. For long scenario questions, identify what is actually being asked before reading the four options, then eliminate.

During the test - multi-select items

Read the stem for how many selections are required and treat the choices as independent true/false statements rather than a single best-fit answer. Partial credit is uncommon on certification exams, so a missed selection can forfeit the whole item.

During the test - pacing

Budget roughly one minute per question against the posted time limit and check your remaining time at the one-third and two-thirds marks. Flag and move on rather than burning five minutes on one item.

During the test - endgame

Keep a reserve for flagged items and re-check that every multi-select has the required number of selections. Only change an answer when you can articulate a specific reason the first choice was wrong.

Rules to check

  • Bring the exact ID you registered with, and register under your legal name. A mismatch between the name on your registration and the name on your ID is one of the most common reasons candidates are turned away, and it is not resolved at the check-in desk.

  • Do not communicate with anyone during the exam. For remote sittings, speaking aloud, gesturing to someone off-camera, or having another person enter the room can void your session. If you need to reach the proctor, use the designated chat or raise-hand control.

  • Assume the clock does not stop. Unless your specific appointment's rules say otherwise, treat any break as consuming exam time and expect that leaving the camera's view may end the session. Confirm the break policy with the provider before exam day (needs_review).

  • Nothing on the desk that is not explicitly permitted. Phones, smart watches, second monitors, printed notes, and handwritten material should be out of the room or out of frame - a proctor who spots one may invalidate the attempt regardless of intent.

  • Scoring is on the exam as a whole, not per domain. There is no partial credit for getting close, and there is no penalty for guessing on items you cannot resolve - but a blank is always wrong, so answer everything.

  • Your result and any domain-level performance breakdown are reported after the session; confirm how and when scores are released and how long the credential remains valid with the provider, since both are subject to change (needs_review).

  • If you need to reschedule, cancel, or retake, follow the provider's stated notice window and waiting period rather than assuming a grace period. Confirm the current retake and cancellation terms in your registration agreement (needs_review).

  • If something goes wrong - a disconnection, a crashed browser, a proctor dispute - stop and follow the provider's incident process immediately rather than improvising a workaround, and keep any case or ticket reference you are given.

SnowPro Core FAQ

How many SnowPro Core practice questions are available?

The practice set has 200 SnowPro Core questions. Each one includes the choices, the correct answer, and an explanation, and the set is free to read with no signup.

How many SnowPro Core flashcards are available?

The flashcard set has 32 SnowPro Core cards. Each card shows the term and its definition.

Which topics does the SnowPro Core study material cover?

The SnowPro Core study material has 5 sections, including Snowflake AI Data Cloud Features and Architecture, Account Access and Security, and Performance Optimization, Querying, and Data Transformation.

What format facts does the catalog list for the SnowPro Core exam?

The catalog lists these item types: multiple choice for the SnowPro Core exam.