Which description matches Snowflake's architecture?
- A hybrid that keeps one central copy of the data and processes queries on independent MPP clusters
- A federated engine that queries data in place inside customer-managed databases
- A shared-nothing design in which each compute node permanently stores and owns its own slice of the table data
- A shared-disk design in which all compute nodes contend for one storage device
A query ran at 09:00 and its result was persisted in the result cache. Which two later events mean the identical query will run on a warehouse again instead of reusing that result? (Choose two.)
- The stored result goes unused for more than 24 hours
- A pipeline inserts rows into one of the tables the query reads
- The warehouse that ran the query is resized from Small to Large
- A user with the same role runs the query on a different warehouse
- The warehouse that ran the 09:00 query is suspended
Which two statements need a running virtual warehouse? (Choose two.)
ALTER TABLE sales_db.core.orders RENAME TO orders_old;- A
CALL to a stored procedure whose body runs INSERT … SELECT SELECT * FROM sales_db.core.orders WHERE region = 'EU';LIST @sales_db.raw.landing_stage PATTERN = '.*2026.*[.]csv';SHOW TABLES IN SCHEMA sales_db.core;
What is the size of a Snowflake micro-partition, measured on uncompressed data?
- Between 1 GB and 2 GB
- Between 100 MB and 250 MB
- Exactly 16 MB, fixed by Snowflake
- Between 50 MB and 500 MB
Which three Snowflake features depend directly on micro-partitions being immutable? (Choose three.)
- Per-column compression inside each micro-partition
- Fail-safe
- Time Travel
- Multi-cluster warehouses
- Zero-copy cloning
Which statement about Snowflake's storage layer is correct?
- Customers can browse and download micro-partition files from their cloud console
- Table data is stored row by row so that single-row transactions run faster
- Data is stored as compressed, columnar micro-partitions that Snowflake manages
- The storage layer applies query filters before any data is sent to a warehouse
Which two functions belong to the cloud services layer? (Choose two.)
- Authenticating users and enforcing access control
- Caching recently read table data on local SSDs
- Parsing and optimizing SQL statements
- Executing joins and aggregations on micro-partitions
- Storing table data for the long term
A table has 120 columns. A query references 3 of them, and its filter lets Snowflake prune 95% of the table's micro-partitions. What does the warehouse read?
- All 120 columns from the 5% of micro-partitions that remain
- The referenced columns from the 5% of micro-partitions that remain
- The referenced columns from every micro-partition in the table
- Every column of every micro-partition, filtering rows in memory afterwards
An account is billed for cloud services almost every day. Which workload is the most likely cause?
- A tool that runs SHOW commands and catalog queries every few seconds
- Several terabytes of changed data held in Time Travel and Fail-safe
- A multi-cluster warehouse that scales out to three clusters at peak
- A few long aggregations running on an X-Large warehouse each night
A security team wants a key that it holds in its cloud provider's key management service to be part of the key protecting all account data, so that revoking it makes the data unreadable even to Snowflake. Which feature, at which minimum edition, meets this?
- Tri-Secret Secure, Business Critical edition
- Periodic rekeying, Enterprise edition
- Tri-Secret Secure, Enterprise edition
- Periodic rekeying, Business Critical edition
Which three features are first available in Enterprise edition? (Choose three.)
- Secure Data Sharing
- Materialized views
- Tri-Secret Secure
- Row access policies
- Multi-cluster warehouses
A data engineer compares two warehouse sizes by running the same query twice. The second run returns in 40 ms, and its Query Profile shows a single QUERY RESULT REUSE node. What should the engineer do before benchmarking again?
- Run the second query on a different warehouse
- Set
USE_CACHED_RESULT to FALSE for the session - Suspend and resume the warehouse between runs
- Add a LIMIT clause to the second run of the query
Why does CREATE TABLE … CLONE complete in seconds, even for a 50 TB table?
- It writes new metadata that points at the source table's existing micro-partitions
- It returns at once and copies the micro-partitions in the background afterward
- It restores a copy that Snowflake keeps ready from the table's nightly Fail-safe snapshot
- It copies the micro-partitions, and the cloud provider deduplicates the identical files at no charge
Which query can Snowflake answer from metadata alone, without a running warehouse?
SELECT AVG(amount) FROM sales;SELECT DISTINCT region FROM sales;SELECT MAX(order_date) FROM sales;SELECT COUNT(*) FROM sales WHERE region = 'EMEA';
A 3 TB table is loaded every day in event-time order. Queries filtering on event_date scan under 2% of micro-partitions, while queries filtering on customer_id scan almost all of them. What explains the difference?
- Each partition holds a narrow event_date range but a wide customer_id range
- The result cache serves date-filtered queries but not customer_id lookups
- Snowflake only prunes on DATE and TIMESTAMP columns, never on numeric IDs
- Snowflake created an index on event_date automatically during the first daily load
A regulator requires that a company's Snowflake environment share no metadata store and no compute resources with any other Snowflake customer. Which option meets this requirement?
- An on-premises Snowflake installation
- Virtual Private Snowflake (VPS) edition
- Enterprise edition with AWS PrivateLink
- Business Critical edition with Tri-Secret Secure
Which two kinds of information does Snowflake keep in metadata for each micro-partition? (Choose two.)
- The minimum and maximum value of each column
- The SQL text of the statement that wrote it
- The number of distinct values in each column
- A B-tree index on the table's primary key
- The table's clustering key definition
A dashboard fires a burst of queries every 4 minutes against the same large tables. The team sets AUTO_SUSPEND = 60 on its warehouse. Every burst now runs noticeably slower than before. What is the most likely cause?
- Each auto-suspend cycle lowers the warehouse size by one step until it resumes
- Each resume bills a new 60-second minimum, which delays the first queries
- Each suspension drops the warehouse cache, so every burst starts with cold reads
- Each resume makes cloud services rebuild partition statistics before queries run
A data engineer updates one row in ORDERS, a large standard table. What happens at the storage level?
- Snowflake edits the bytes of that row inside the existing micro-partition
- Snowflake flags the row as deleted and keeps the partition as the current version
- Snowflake rewrites every micro-partition of the table in a single transaction
- Snowflake writes a new micro-partition and keeps the old one for Time Travel
A role ran a query on the FINANCE table an hour ago, and the result is in the result cache. An administrator then revokes that role's SELECT privilege on FINANCE. No data has changed, and no other role available to the user has access to FINANCE. What happens when the user runs the identical query again?
- The cached result is returned, because it was produced while access was valid
- The query fails, because privileges are checked again before a result is reused
- The cached result is returned, because cached results bypass privilege checks
- The query runs on the warehouse, because revoking a privilege clears the cache
A team is moving from an on-premises data warehouse to Snowflake. Which two tasks remain the team's responsibility after the move? (Choose two.)
- Choosing warehouse sizes and auto-suspend settings
- Granting privileges on objects to roles
- Adding storage capacity before a large load
- Updating table statistics after large loads
- Applying software upgrades and security patches
What is the maximum Time Travel retention period for a permanent table in Standard edition?
- 1 day
- 0 days, since Time Travel needs Enterprise
- 7 days
- 90 days
On one day, an account's warehouses consume 120 credits and its cloud services consume 15 credits. How many cloud services credits are billed for that day?
- 3
- 12
- 15
- 0
A company plans to replicate its databases to another region and promote the replicas to primary if its main region has an outage. What is the minimum edition for the failover part of this plan?
- Enterprise
- Virtual Private Snowflake
- Standard
- Business Critical
How does Snowflake bill for the storage of table data?
- On the uncompressed size of the data as it was loaded, excluding Time Travel
- On the compressed bytes stored, excluding data held for Time Travel and Fail-safe
- As credits consumed by the warehouse that loaded the data, billed per second
- On the compressed bytes stored, including data kept for Time Travel and Fail-safe
An Enterprise edition account must make sure data protected by keys retired more than a year ago is re-encrypted with new keys. An administrator points out that Snowflake already rotates its keys every 30 days. What is true?
- ACCOUNTADMIN must turn on periodic rekeying, because rotation leaves old data under its retired key
- Rotation already re-encrypts old data, so nothing more is needed
- Periodic rekeying needs Business Critical, so the account must upgrade
- The account needs Tri-Secret Secure, so it must move to Business Critical
When a query filters on a column, where is the decision made about which micro-partitions to skip?
- In the cloud services layer, from metadata, when the query is compiled
- In the virtual warehouse, as each node opens and reads its micro-partitions
- In the storage layer, which filters files before returning them
- In the virtual warehouse, from min/max statistics cached on its local disk
Which statement about Fail-safe for permanent tables is correct?
- Users can query it with AT or BEFORE clauses, like Time Travel
- It lasts 7 days and only Snowflake Support can recover from it
- ACCOUNTADMIN can set it to 0 days to lower storage costs
- Enterprise edition can extend it to up to 90 days per table
A warehouse that loaded and queried the SALES table for two years is dropped. What happens to the SALES table?
- It must be recovered from Fail-safe before another warehouse can use it
- It stays, but roles lose access to it until it's linked to another warehouse
- It stays, but its Time Travel history is dropped with the warehouse
- Nothing happens to it; any other warehouse can query it right away
Which two statements about Snowflake's three layers are correct? (Choose two.)
- The storage layer holds one copy of each table that every warehouse can read
- The cloud services layer handles authentication, metadata and optimization
- The storage layer runs the joins that the optimizer pushes down to it
- Virtual warehouses share CPU and memory with each other to raise utilization
- Query results are stored on the warehouse that ran them and lost on suspend