Clustering is Snowflake’s automatic micro-partitioning optimization feature. When a table is clustered on a key column like `SALE_DATE`, Snowflake organizes the data in micro-partitions based on that column’s values. This allows for efficient pruning, meaning queries with predicates on `SALE_DATE` can skip reading entire micro-partitions that don’t contain relevant data, significantly improving performance without the manual maintenance of partitions or separate tables.

For a predictable, recurring, and complex analytical query (like a monthly report), a materialized view is the optimal solution. It pre-computes and stores the result of the underlying join and aggregation queries. When the finance team runs their report, they query the materialized view, which returns results in seconds. While it uses storage and compute for refresh, the trade-off is justified by the massive reduction in recurring compute time and improved user experience for a critical business process.

The `ON_ERROR = CONTINUE` option instructs the COPY command to load all valid rows and log any errors encountered (like malformed JSON) into a designated error table. `SKIP_FILE` would skip the entire file if any error is found, which is too coarse. `RETURN_ERRORS` is for validation only and doesn’t load data. `PURGE` relates to cleaning up staged files after ingestion.

Sharing a secure view is the core Snowflake pattern for providing filtered, column-level access to live data. The view definition (`SELECT PRODUCT_ID, NAME, PRICE FROM inventory WHERE REGION=’West’`) acts as a persistent filter. When shared via Snowflake’s direct, reader-account based secure data sharing, the partner sees the results of this view, which always reflects the current state of the underlying table. It requires no data duplication, ensures real-time access, and centralizes logic management on the provider’s side.

Time Travel allows you to query or restore a table to a specific point in the past within its retention period (typically 1 day by default for Standard Edition). You can either `CREATE TABLE restored_table CLONE original_table AT (OFFSET => -60*30);` to clone it from 30 minutes ago, or use `SELECT * FROM original_table AT (OFFSET => -60*30)` to query the old data directly. This is the primary, built-in method for point-in-time recovery from user errors.

Streams in Snowflake provide change data capture (CDC) by pointing to a transactional offset. If a stream is not consumed, it will show new data. High latency (data not appearing) is most often caused by a stream being blocked by a long-running, uncommitted transaction (e.g., a massive COPY or DELETE) on the source table. The stream cannot advance its offset until that transaction is committed or rolled back, holding up the change data visibility.

Privileges in Snowflake are hierarchical. To create an object (like a schema), you need the `CREATE

The `QUERY_HISTORY` view (in the `INFORMATION_SCHEMA` or `ACCOUNT_USAGE` schema) is the primary source for analyzing query performance. It contains critical execution details like `QUERY_TAG`, `EXECUTION_STATUS`, `WAREHOUSE_SIZE`, `TOTAL_ELAPSED_TIME`, `QUEUED_TIME`, and `COMPILATION_TIME`. High `QUEUED_TIME` specifically would indicate the query was waiting for warehouse resources (e.g., slots) to become available, pointing to warehouse contention as the likely cause.

Row Access Policies are a centralized, reusable policy-based access control feature. You can create a policy that uses a mapping table or logic (e.g., `CURRENT_ROLE() IN (‘HR’) OR DEPARTMENT = ‘Sales’`) and apply it to the target table. The policy is evaluated at query runtime, dynamically filtering rows based on the querying role’s characteristics. This is more scalable and maintainable than creating multiple secure views for different access patterns.

Clustering is Snowflake’s automatic micro-partitioning optimization feature. When a table is clustered on a key column like `SALE_DATE`, Snowflake organizes the data in micro-partitions based on that column’s values. This allows for efficient pruning, meaning queries with predicates on `SALE_DATE` can skip reading entire micro-partitions that don’t contain relevant data, significantly improving performance without the manual maintenance of partitions or separate tables.0