Snowflake event tables capture logs, traces, and metrics generated by Snowflake features such as stored procedures, functions, and dynamic tables. When a database does not have an explicitly associated event table, Snowflake writes telemetry data to the default account-level event table, which is exposed through SNOWFLAKE.TELEMETRY.EVENTS.
Since the problematic stored procedure resides in marketing_db, and that database is not associated with its own event table, the relevant telemetry data will not appear in sales_db’s event table. Instead, it will be recorded in the account-level telemetry view. Querying SNOWFLAKE.TELEMETRY.EVENTS allows the Architect to analyze logs and traces across databases within the same account.
Creating additional event tables is unnecessary for investigation, and querying MARKETING_DB.TELEMETRY.EVENTS is invalid because no such association exists. From a SnowPro Architect perspective, this question validates understanding of event table scope, default telemetry behavior, and effective troubleshooting of Snowflake-native observability features.
=========
QUESTION NO: 27 [Snowflake Data Engineering]
A data ingestion pipeline loads data randomly throughout the day. A requirement states that data must be processed within 10 minutes of landing in Snowflake.
What is the MOST cost-effective solution?
A. Use a third-party orchestration tool to run the pipeline every 5 minutes.
B. Create views containing transformation logic.
C. Create a task that runs every 5 minutes and processes new rows.
D. Create a stream on the table and a task that runs every 5 minutes using SYSTEM$STREAM_HAS_DATA.
Answer: D
Streams and tasks are Snowflake’s native mechanism for incremental data processing. A stream tracks row-level changes (inserts, updates, deletes) on a table, while tasks enable scheduled or conditional execution of SQL statements.
Using SYSTEM$STREAM_HAS_DATA allows the task to run only when new data is available, avoiding unnecessary compute usage (Answer D). This makes the solution both timely—meeting the 10-minute SLA—and cost-efficient, as compute resources are consumed only when there is work to do.
Views alone do not materialize transformations and still require compute at query time. Third-party orchestration introduces unnecessary operational overhead and cost. Running a task on a fixed schedule without checking stream data would waste compute cycles.
This pattern—streams + tasks with conditional execution—is a core SnowPro Architect design pattern for efficient, low-latency data pipelines.
=========
QUESTION NO: 28 [Performance Optimization and Monitoring]
A query shows a high number of bytes spilled to remote storage. The query includes a WHERE clause that filters out a large percentage of rows.
What actions can reduce spillage and improve performance? (Select TWO).
A. Increase the size of the virtual warehouse.
B. Increase the maximum number of clusters.
C. Define a clustering key on columns used for selective filtering.
D. Define primary and foreign keys.
E. Decrease the data retention period.
Answer: A, C
Bytes spilled to remote storage indicate memory pressure during query execution. Increasing the warehouse size provides more memory and compute resources, reducing the likelihood of spilling (Answer A).
Additionally, defining clustering keys on highly selective filter columns improves micro-partition pruning, which reduces the volume of data scanned and processed in memory (Answer C). Together, these actions address both the root cause (too much data scanned) and the symptom (insufficient memory).
Increasing clusters improves concurrency, not per-query memory. Primary and foreign keys are informational only in Snowflake and do not affect execution plans. Retention settings have no impact on query execution.
This question reinforces SnowPro Architect performance diagnostics: understand both physical execution (memory, spill) and logical data access (pruning).
