SnowPro Core Recertification — Free Practice Questions
10 free sample questions from a bank of 52, with the correct answers and explanations. No signup required — start practising right now.
1By default, how long is the standard retention period for Time Travel across all Snowflake accounts?
0 days
1 day
7 days
14 days
Answer: B
B is correct - the standard Time Travel retention period is 1 day.
The short version
Every Snowflake account gets 1 day of Time Travel out of the box, for tables, views, schemas and databases. Longer retention is something you configure, not a default.
Key concepts
Time Travel keeps historical versions of data for CLONE, UNDROP and AS OF TIME TRAVEL queries.
Default retention is 1 day; raise it with DATA_RETENTION_TIME_IN_DAYS on the object (90 days max) or at account/schema level.
After Time Travel, deleted data enters the 7-day Fail-safe period, which Snowflake Support manages - that is a different 7 days.
Setting retention to 0 puts an object straight into Fail-safe when dropped.
Why B is correct
B is the documented default. Snowflake's feature matrix lists 'Snowflake Time Travel (1 day standard for all accounts)' as a universal capability, so an account that has changed nothing behaves as if DATA_RETENTION_TIME_IN_DAYS = 1. This matters because the answer is easy to confuse with the two other seven-day numbers in the platform: Fail-safe retention after a drop, and the Snowsight Query History window.
Why the others are wrong
A. 0 days That is the setting that disables Time Travel, not the default. With 0 days an object has no AS OF TIME TRAVEL history and goes to Fail-safe as soon as it is dropped - a deliberate choice, not a platform default.
C. 7 days 7 days is the Fail-safe period, which begins after Time Travel expires and applies to dropped data that Snowflake Support can recover. It is not the Time Travel retention default.
D. 14 days 14 days is how far back the Snowsight Query History page reaches. It is unrelated to Time Travel retention, which is capped at 90 days when you raise it.
SnowPro-Core-Recertification exam tip
Fix two numbers in your head: Time Travel defaults to 1 day (max 90), Fail-safe is 7 days and is Support-only. 'Standard retention' questions are asking about the Time Travel default.
2At what level is the MIN_DATA_RETENTION_TIME_IN_DAYS parameter set?
Account
Database
Schema
Table
Answer: A
A is correct. MIN_DATA_RETENTION_TIME_IN_DAYS is an account parameter set by ACCOUNTADMIN that puts a floor under Time Travel retention account-wide. DATA_RETENTION_TIME_IN_DAYS sets retention on an individual object, and it can never drop below this account minimum.
3What type of function returns one value for each invocation?
Aggregate
Scalar
Table
Window
Answer: B
B is correct. A scalar function returns exactly one value per invocation, such as LENGTH(col) for each row. Aggregates collapse many rows into one value, table functions return a rowset, and window functions return a value per row only after an OVER clause partitions the input.
4What persistent data structures are used by the search optimization service to improve the performance of point lookups?
Micro-partitions
Clustering keys
Equality searches
Search access paths
Answer: D
D is correct - the search optimization service maintains search access paths.
The short version
When you add a search optimization on a table, Snowflake builds and keeps a persistent auxiliary structure - the search access path - so frequent selective predicates stop requiring a full scan.
Key concepts
Search access paths are separate persistent data structures covering the indexed columns and their values.
They are maintained automatically as the table changes, at an additional storage cost charged per TB.
They target frequent point lookups with selective WHERE predicates, not general analytics.
Clustering keys and micro-partitions serve pruning and data organisation, not search indexing.
Why D is correct
D is the term Snowflake uses. The service builds a persistent data structure called a search access path for the indexed table, and uses it to turn repeated selective predicates into efficient point lookups instead of scanning every micro-partition. That is why an SOS carries a storage charge: you are paying to maintain a second copy of indexed values beside the data.
Why the others are wrong
A. Micro-partitions Micro-partitions are Snowflake's physical storage units, compressed columnar chunks of roughly 50-500 MB. They are where data lives, not an index the search optimization service maintains; SOS works over them, skipping whole partitions via pruning.
B. Clustering keys Clustering keys re-sort data so similar values share micro-partitions, which helps range predicates and pruning. You define them yourself with CLUSTER BY, and they trade query speed against ongoing background reclustering credits - a different feature from SOS.
C. Equality searches That describes a query pattern the service accelerates, not a persistent data structure it maintains. The question asks what is stored, and 'searches' is nothing to store.
SnowPro-Core-Recertification exam tip
Store the service-to-structure pairing: search optimization service = search access path (built, maintained automatically, charged per TB). Clustering = clustering keys (your choice, trades speed for reclustering cost).
5In addition to performing all the standard steps to share data, which privilege must be granted on each database referenced by a secure view in order to be shared?
READ
REFERENCES
REFERENCE_USAGE
USAGE
Answer: C
C is correct - REFERENCE_USAGE must be granted on each additional database a shared secure view references.
The short version
When a secure view pulls objects out of databases other than its own, each of those databases must carry REFERENCE_USAGE granted to the share before the view itself can be shared. USAGE applies to the single database that contains the view.
Key concepts
REFERENCE_USAGE: per-database privilege that lets a shared secure view expose objects in that database without granting SELECT on them.
USAGE can be granted on only ONE database per share - that is the database containing the view.
REFERENCE_USAGE must be granted BEFORE granting SELECT on the secure view to the share.
It cannot be granted to a database role, so cross-database sharing cannot be done with database roles.
Why C is correct
C is the privilege the Snowflake documentation names for exactly this case, and the wording of the question is lifted from that documentation. USAGE is required too, but only for the one database that houses the secure view; the databases the view merely references need REFERENCE_USAGE, granted individually on each one: GRANT REFERENCE_USAGE ON DATABASE db2 TO SHARE share1; before GRANT SELECT ON VIEW db3.sch.v TO SHARE share1;
Why the others are wrong
A. READ READ is a privilege on stages and tables (and on schemas for views), not the database-level grant used when sharing a view across databases. It plays no part in this scenario.
B. REFERENCES Not a Snowflake privilege. 'REFERENCES' is the SQL-standard privilege for creating foreign-key constraints, which Snowflake does not use at all.
D. USAGE USAGE is required, but only on the one database that contains the secure view, and only one USAGE grant per share is allowed. It cannot be the privilege covering 'each database referenced', which rules it out as the answer to the question as asked.
SnowPro-Core-Recertification exam tip
Mnemonic: USAGE = the database you are exposing; REFERENCE_USAGE = the databases the view reaches into. If a stem says 'each database referenced by a secure view', the answer is REFERENCE_USAGE, and the order matters - REFERENCE_USAGE first, SELECT on the view second.
6Which of the following SQL statements will list the version of the drivers currently being used?
Execute SELECT CURRENT_ODBC_CLIENT(); from the Web UI
Execute SELECT CURRENT_JDBC_VERSION(); from SnowSQL
Execute SELECT CURRENT_CLIENT(); from an application
Execute SELECT CURRENT-VERSION(); from the Python Connector
Answer: C
C is correct - SELECT CURRENT_CLIENT() from an application returns the driver/client version.
The short version
CURRENT_CLIENT is a context function: called from a JDBC or ODBC driver it returns that driver's version, and it also identifies the client that submitted the statement. The other options name functions that do not exist, or that cannot run in the client named.
Key concepts
CURRENT_CLIENT() returns the client type and version; from JDBC/ODBC it returns the driver version.
CURRENT_CLIENT_TYPE() and CURRENT_CLIENT_VERSION() are the other context functions.
The same information appears in Snowsight under Query Details as the 'Name and version of the client' field.
These are metadata functions: they need no warehouse and cost no credits.
Why C is correct
C is the documented function for this purpose - it returns the version of the client from which the function was called, and specifically the JDBC or ODBC driver version when the call arrives through one of those drivers. That is exactly what an administrator running connectivity tests needs, which is why the option is written as an application-side call.
Why the others are wrong
A. SELECT CURRENT_ODBC_CLIENT(); from the Web UI There is no CURRENT_ODBC_CLIENT function. And a query typed into the web UI is submitted by the web interface, not by an ODBC driver, so it could not report an ODBC version even in principle.
B. SELECT CURRENT_JDBC_VERSION(); from SnowSQL No such function exists. Worse, SnowSQL does not use JDBC - it is a Python client - so a JDBC version query is meaningless there.
D. SELECT CURRENT-VERSION(); from the Python Connector Hyphens are not legal in a Snowflake SQL function name, and CURRENT_VERSION() is not a Snowflake context function. The hyphen alone invalidates the statement.
SnowPro-Core-Recertification exam tip
The three context functions to know are CURRENT_CLIENT(), CURRENT_CLIENT_TYPE() and CURRENT_CLIENT_VERSION() - all free, all warehouse-free. Anything with a hyphen, or an underscore form you cannot place, is a distractor.
7What action can be performed using the GET command in Snowflake?
Automatically rename downloaded files.
Automatically decrypt downloaded data on a client machine.
Download data files from Snowflake internal stages to a local directory/folder.
Download data files from Snowflake external stages to a local directory/folder.
Answer: C
C is correct - GET downloads files from an internal stage to a local directory.
The short version
GET is a client-side command that copies files out of a Snowflake internal stage - named, table or user stage - onto your machine. It does not rename files, does not decrypt anything, and cannot read external stages.
Key concepts
GET runs in SnowSQL / the Snowflake client, not as SQL on the server.
Sources: named internal stage, a table's internal stage, or the user's own stage.
It is the download step that follows COPY INTO @stage, which produces the files.
External stages are unsupported by GET; for those you copy directly from your own bucket.
Why C is correct
C matches the documented scope: GET downloads data files from a named internal stage, a table's internal stage, or the current user's stage to a local directory or folder on a client machine. Understand it as the second half of an export - COPY INTO @stage writes the files, GET retrieves them.
Why the others are wrong
A. Automatically rename downloaded files GET has no rename behaviour. Its options cover parallelism, compression and overwrite of local copies; file naming comes from the stage, not from GET.
B. Automatically decrypt downloaded data on the client machine There is no client-side decryption step. Snowflake's internal encryption is handled server-side; what you get back are the staged bytes, and if you use client-side encryption keys those stay on your side throughout.
D. Download from Snowflake external stages The decisive distinction. GET supports only internal stages, because external stages are just pointers at your own S3/GCS/Azure bucket - you copy files from there with cloud tooling instead. This is also what separates GET from PUT, which works on both.
SnowPro-Core-Recertification exam tip
Internal vs external is the whole story for GET and PUT: GET works on internal stages only, PUT works on both, and neither renames or decrypts anything.
8What is the PRIMARY factor that determines the cost of using a virtual warehouse in Snowflake?
The type of SQL statements executed
The number of tables or databases queried
The amount of data stored in the warehouse
The length of time the compute resources in each cluster run
Answer: D
D is correct - a warehouse costs you for how long its compute runs, nothing else.
The short version
Snowflake meters a virtual warehouse in credits, and credits burn for as long as the warehouse's clusters are provisioned. Storage is billed on a completely separate, capacity-based line (per TB per month) and never appears as warehouse cost.
Key concepts
Credit = one unit of compute, consumed per second while a warehouse cluster is running.
Warehouse size sets the credits-per-hour rate; cluster count multiplies it; elapsed running time multiplies the total.
Storage is charged for bytes retained, at a much lower rate, whether or not a warehouse is running.
A suspended warehouse consumes zero credits; auto-suspend is the single biggest lever on this bill.
Why D is correct
D names the actual metering unit. Cost of a virtual warehouse = (size x cluster count) x time running. That is why an idle warehouse left running over a weekend dominates the compute bill while a petabyte of dormant data costs comparatively little: the data is not what the warehouse is billed for. Auto-resume plus auto-suspend and a right-sized warehouse attack D; nothing you do to your table definitions does.
Why the others are wrong
A. The type of SQL statements executed Statement type matters only indirectly. DDL/metadata statements run in cloud services and are not billed to a warehouse, while DML and SELECT are - but they are billed for the seconds they occupy, not for being DML or SELECT.
B. The number of tables or databases queried Breadth of query, not consumption. A statement joining ten tables in five databases on one cluster costs exactly the same credits for the same elapsed time as a single-table scan of the same duration. Table count affects the work inside the cluster, not the meter itself.
C. The amount of data stored in the warehouse This is the trap: storage is billed separately at roughly $/TB/month, and a virtual warehouse stores nothing itself - it only queries Snowflake's storage layer. Volume of stored data does not change the credits a warehouse burns.
SnowPro-Core-Recertification exam tip
Translate every cost question into credits x seconds. Anything about volume of data - bytes stored, tables touched, databases queried - is either a storage charge or a workload-shape distractor. Only time x size x clusters lands on the warehouse.
9What are potential impacts of storing non-native values like dates and timestamps in a VARIANT column in Snowflake?
Faster query performance and increased storage consumption
Slower query performance and increased storage consumption
Faster query performance and decreased storage consumption
Slower query performance and decreased storage consumption
Answer: B
B is correct. A date or timestamp held as a string inside VARIANT is not natively typed, so Snowflake stores it as serialised text and must parse or convert it at query time - slower execution and larger storage. Native DATE/TIMESTAMP columns are stored typed and columnar-compressed.
10A Snowflake user is actively logged into Snowflake when a user-level network policy is assigned to that user.
What will Snowflake do if the user's IP address does not match the user-level network policy rules?
Log the user out.
Deactivate the network policy.
Prevent the user from executing additional queries.
Allow the user to continue until the session or login token expires.
Answer: C
C is correct - Snowflake prevents the user from executing additional queries.
The short version
The existing session is not killed. Snowflake enforces the newly attached policy at the next statement boundary, so the user can still see the session but cannot run anything further.
Key concepts
Network policies restrict which IP addresses may access Snowflake, evaluated at login and again on subsequent activity.
A user-level policy (ALTER USER ... SET NETWORK_POLICY) overrides an account-level policy for that user.
If the session's IP no longer satisfies the rules, further queries are blocked.
MINS_TO_BYPASS_NETWORK_POLICY gives a Support-set temporary bypass for blocked programmatic access.
Why C is correct
C is the documented behaviour for exactly this scenario. Snowflake states that when a user-level network policy is associated with a user who is already logged in, and the user's network location does not match the policy's rules, Snowflake prevents the user from executing further queries. The session object survives, which is why 'log the user out' overstates it and 'allow until the token expires' understates the enforcement.
Why the others are wrong
A. Log the user out Too strong. Snowflake blocks new statements rather than tearing down the session; the user may still hold an open worksheet and simply cannot run new queries until the location complies.
B. Deactivate the network policy Backwards, and dangerous as a mental model. Deactivating a policy would remove the restriction entirely and open the account up; Snowflake enforces the policy you assigned.
D. Allow the user to continue until the session or login token expires This would make policies advisory only. Enforcement is not deferred to session expiry - new statements are prevented as soon as the mismatch is detected.
SnowPro-Core-Recertification exam tip
Memorise the direction of the rule: policy assignment is enforced going forward even mid-session, and mis-located sessions lose the ability to execute, they are not logged out.