SnowPro Advanced: Architect Practice Exam

Free sample questions

1. Sample · question 1 · Ingestion architecture

   An Architect needs to design a pipeline that ingests JSON events from an Amazon S3 landing bucket into Snowflake. Files land every ~30 seconds and downstream users need to query the events within a minute of arrival. Which combination best meets the latency requirement while keeping operational overhead low?

   • A.Snowpipe with auto-ingest triggered by S3 event notifications, plus a task-scheduled MERGE to a curated table.correct
   • B.A COPY INTO command scheduled every 5 minutes on a warehouse, followed by a task that transforms the data.
   • C.An external table pointed at the S3 prefix and a materialized view on top for latency.
   • D.A Snowpark job that polls S3 every 60 seconds and inserts rows manually.

   Why: Snowpipe auto-ingest is purpose-built for continuous file-based ingestion — S3 events trigger loads within roughly a minute. Scheduled COPY INTO inherits the schedule as its latency floor. External tables add per-query metadata refresh cost. Custom polling is unnecessary reinvention of Snowpipe.

   Open this question on its own page →
2. Sample · question 2 · RBAC design

   An Architect is designing role hierarchy for a multi-team analytics account. Which pair of built-in roles is the recommended starting point for (a) creating databases, warehouses, and schemas versus (b) creating and managing users and roles?

   • A.(a) SYSADMIN, (b) SECURITYADMINcorrect
   • B.(a) ACCOUNTADMIN, (b) SYSADMIN
   • C.(a) USERADMIN, (b) SYSADMIN
   • D.(a) PUBLIC, (b) ACCOUNTADMIN

   Why: SYSADMIN owns the object hierarchy (databases, warehouses, schemas) and is intended for day-to-day resource creation. SECURITYADMIN is purpose-built for user and role management, delegating from USERADMIN. ACCOUNTADMIN has broader powers and should be used sparingly; PUBLIC is granted to every role and is only for default grants.

   Open this question on its own page →
3. Sample · question 3 · Secure data sharing

   A data provider wants to share a curated view of its customer data with an external partner who has their own Snowflake account, without physically copying the data. Which approach is the Snowflake-native way?

   • A.Create a Reader Account for the partner and grant them credentials.
   • B.Create a secure view and grant it via a Snowflake Share; the partner mounts the share as a database in their own account.correct
   • C.Use replication to copy the view's underlying tables to the partner's account nightly.
   • D.Export the view's rows to S3 and grant the partner pre-signed URLs.

   Why: Snowflake Secure Data Sharing lets two accounts in compatible regions share objects with zero data copy — the consumer mounts the share and queries live. Reader Accounts are for consumers who don't have their own Snowflake account. Replication is for cross-region availability, not partner sharing. Exporting to S3 defeats the no-copy requirement.

   Open this question on its own page →
4. Sample · question 4 · Warehouse concurrency

   An analytics team reports that queries queue during business hours even though a single query completes quickly. The warehouse is a Small single-cluster. The goal is to eliminate queuing without changing per-query performance. Which configuration change directly addresses this?

   • A.Increase the warehouse size to Medium or Large.
   • B.Enable multi-cluster mode in Auto-scale with min=1, max=4.correct
   • C.Add a clustering key to the largest fact table involved.
   • D.Increase MAX_CONCURRENCY_LEVEL to allow more parallel queries per cluster.

   Why: Queuing during business hours is a concurrency problem, not a per-query performance problem. Multi-cluster Auto-scale spins up additional clusters of the same size when queue depth grows, and scales back down when it drains. Scaling the warehouse up helps individual queries, not concurrency. Clustering keys improve pruning, not queueing. Raising MAX_CONCURRENCY_LEVEL packs more queries onto one cluster and often makes contention worse.

   Open this question on its own page →
5. Sample · question 5 · Cost controls

   An Architect needs a hard cap so that a specific virtual warehouse cannot consume more than 500 credits per month, and any queries that would push it past that threshold are stopped. Which Snowflake feature implements this?

   • A.A resource monitor with credit_quota=500 and a SUSPEND_IMMEDIATE action at 100%.correct
   • B.A statement timeout setting on the warehouse.
   • C.A row access policy on the tables the warehouse queries.
   • D.Setting the warehouse's AUTO_SUSPEND to a lower value.

   Why: Resource monitors enforce credit quotas and can perform actions at percentage thresholds — SUSPEND_IMMEDIATE at 100% stops the warehouse mid-query when the cap is hit. Statement timeout limits per-query time, not credits. Row access policies control data visibility. Auto-suspend controls idle behavior.

   Open this question on its own page →
6. Sample · question 6 · Multi-region failover

   A production Snowflake account in us-east-1 must have a warm-standby replica in eu-west-1 so the team can fail traffic over within minutes of a regional outage. Which combination of features implements this?

   • A.Zero-copy cloning of the databases into the target account.
   • B.Time Travel with a 30-day retention window on the source account.
   • C.Database replication into the target region combined with a failover group and client redirect (Snowflake connection URL that redirects on failover).correct
   • D.External stages backed by cross-region S3 buckets, with COPY INTO commands on a schedule.

   Why: Database replication keeps the target account's data in sync; a failover group promotes the standby account to primary on failover; client redirect lets applications keep the same connection URL and be transparently pointed at the new primary. The other options don't provide cross-region warm-standby with fast failover.

   Open this question on its own page →
7. Sample · question 7 · Network security

   A financial-services company's security team requires that all Snowflake traffic from their VPC in AWS travel over private connectivity (no public internet) AND that Snowflake accept connections only from that VPC. Which two Snowflake features together implement this? (Choose two.)

   • A.AWS PrivateLink for private connectivity between the VPC and Snowflake.correct
   • B.A network policy restricting the account to the VPC's PrivateLink endpoint IDs / IP ranges.correct
   • C.A row access policy applied to every base table.
   • D.A masking policy on every sensitive column.
   • E.Enabling public endpoint SSL with a custom TLS certificate.

   Why: PrivateLink establishes a private route between the VPC and Snowflake; a network policy enforces that only the PrivateLink endpoint IDs (or the VPC's private IPs) are allowed to connect. Row access and masking policies control data visibility, not network access. Public endpoints with SSL still traverse the internet.

   Open this question on its own page →
8. Sample · question 8 · Kafka ingestion

   An Architect is choosing between Snowpipe and Snowpipe Streaming for a Kafka-based clickstream pipeline that generates roughly 20,000 events per second. End-to-end latency requirement is under 5 seconds. Which statement is correct?

   • A.Use Snowpipe (file-based) because it has lower latency than Snowpipe Streaming.
   • B.Use Snowpipe Streaming; it ingests rows directly with sub-second flush, meeting the 5-second SLA.correct
   • C.Use Snowpipe with tiny 100 KB files rotated every second to approximate streaming.
   • D.Use an external table over Kafka Connect's S3 sink and query directly.

   Why: Snowpipe Streaming is built for row-level ingestion with sub-second flush semantics and is the Snowflake-native path when end-to-end latency must stay under a minute. File-based Snowpipe has a ~1-minute floor and generates excessive small-file overhead if you try to force it faster. External tables add per-query metadata refresh cost.

   Open this question on its own page →
9. Sample · question 9 · Environment cloning

   An Architect needs to give each Data Engineer a writable copy of the production database for testing. Storage cost must stay minimal, and each engineer must be able to modify their copy without touching production. What is the correct approach?

   • A.Export production to S3 and import into a per-engineer database.
   • B.Create a zero-copy clone of the production database for each engineer with CREATE DATABASE ... CLONE.correct
   • C.Grant each engineer SELECT/INSERT/UPDATE on the production database.
   • D.Replicate the production database into a QA account and let engineers share it.

   Why: Zero-copy cloning creates a metadata pointer to the existing micro-partitions. Clones take seconds, use zero additional storage until the engineer writes, and are fully writable and isolated from the source. Export/import duplicates all data (wasteful). Granting write access on production is unsafe. Replication is for cross-region availability, not per-user sandboxes.

   Open this question on its own page →
10. Sample · question 10 · Data quality observability

    An Architect wants a Snowflake-native mechanism that runs automatically on a schedule and alerts the team whenever the NULL count in a specific column exceeds a threshold, without adding external monitoring infrastructure. Which feature best satisfies this?

    • A.Attach the SNOWFLAKE.CORE.NULL_COUNT data metric function to the column and define an EXPECTATION with the threshold.correct
    • B.Write a stored procedure that queries the column every hour and posts to a Slack webhook if the count is high.
    • C.Attach a masking policy that returns NULL when the count is high.
    • D.Enable a resource monitor on the loading warehouse.

    Why: Data Metric Functions (DMFs) like NULL_COUNT run on a schedule, and EXPECTATIONs let you declaratively define the threshold — the platform emits events when it's violated. This is the purpose-built Snowflake-native path for column-level data-quality alerting. Custom procedures work but add code and maintenance. Masking policies and resource monitors don't address data-content quality.

    Open this question on its own page →