A red checkmark in Snowflake Notebooks indicates that the cell encountered anerrorduring execution. This visual indicator helps users quickly identify issues so they can inspect the error message and adjust code accordingly. Successful execution shows a green checkmark, while running cells show a spinner. Snowflake notebooks do not include a “paused” state for notebook cells.

Snowflake uses theVARIANTdata type as the native mechanism for storing semi-structured data formats such as JSON, XML, Avro, ORC, and Parquet. VARIANT supports hierarchical, nested structures and allows querying via dot or bracket path notation without requiring schema-on-write.

JSON is not a Snowflake data type—JSON documents arestored insideVARIANT.

VARIANT allows:

Mixed structured and semi-structured content

Efficient storage through Snowflake’s columnar compression

Schema-on-read flexibility

Querying nested structures without flattening

VARCHARstores plain text only.

BLOBis reserved for binary data and not suitable for interactive querying.

This design eliminates the need for preprocessing or schema enforcement at ingestion time.

The PARSE_DOCUMENT function is part of Snowflake Cortex AI and is designed specifically to extract text, layout information, and structured elements from unstructured documents, especially PDFs. It supports OCR-based extraction for scanned files and layout-aware extraction to preserve tables, headings, and format structure.

Its purpose is not PII detection; Snowflake does not provide built-in automatic PII identification via PARSE_DOCUMENT. It does not identify candidate data for directory tables and is unrelated to JSON parsing—Snowflake uses PARSE_JSON for JSON data.

PARSE_DOCUMENT is primarily used for workflows such as contract analysis, invoice extraction, document classification, compliance automation, and downstream AI enrichment.

The Cloud Services Layer is responsible for generating and managing metadata, including object definitions, table schemas, micro-partition statistics, column-level profiles, access control information, and query optimization metadata. Metadata plays a central role in Snowflake’s performance and functionality because it informs pruning, query planning, and efficient execution.

Dashboards are created in Snowsight or external BI tools. Virtual warehouses belong to the Compute Layer, providing processing resources. Micro-partitions are created in the Storage Layer, where Snowflake automatically organizes compressed columnar data for efficient access.

Consequently, the Cloud Services Layer is where metadata—not data, not compute resources—is created and managed.

By default, Snowflake usesON_ERROR = ABORT_STATEMENTfor COPY INTO. This means that if a single row fails to load due to issues such as type mismatches or malformed data, Snowflake immediately stops the entire load operation. This behavior ensures data integrity by preventing partial or incomplete loads when errors occur.

Other error-handling modes must be explicitly specified: CONTINUE skips erroneous rows, SKIP_FILE skips whole files with errors, and VALIDATE is part of the validation mode feature, not a default COPY behavior.

Thus, unless explicitly changed, COPY INTO always uses ABORT_STATEMENT.