Skip to main content
Community Hub

Snowflake connector

TL;DR

How Atlan extracts metadata from Snowflake: the two extraction strategies, where extraction runs, and how lineage and tags flow between the two

Your AI can read this via Docs MCPcurl -fsSL "https://docs.atlan.com/install-docs-mcp" | bashConnect

When you connect Snowflake to Atlan, two independent decisions shape every crawl: where Atlan reads metadata from (the extraction strategy) and where the extraction runs (the execution mode). This page explains the trade-offs behind each choice and how lineage and governance data flow between Snowflake and Atlan once you connect the two.

Extraction strategy: Account usage or Information schema​

Atlan can read Snowflake metadata from two sources:

  • Account Usage: Queries the views in the SNOWFLAKE.ACCOUNT_USAGE schema. A single connection covers your entire Snowflake account and the grant setup is simpler. Account Usage views populate with a delay of 45 minutes to 3 hours, and the method has functional gaps: it doesn't capture external table location data or primary and foreign keys, and for Iceberg tables it reads only the table type.
  • Information Schema: Queries the INFORMATION_SCHEMA views of each database you crawl. This is the more comprehensive method with no data latency, and the only method that reads the full set of Iceberg table properties, but it requires grants on every database you include, and stored procedures appear only when you grant USAGE on them.

The crawler's Extraction method setting selects between the two. The Permissions reference lists the exact grants each method needs.

Execution mode: Direct or Agent​

Independent of the extraction strategy, you choose where the extraction runs:

  • Direct: Atlan connects to your Snowflake account and crawls metadata directly.
  • Agent: Self-Deployed Runtime executes the extraction inside your organization's environment, for on-premises or network-restricted setups.

Any extraction strategy can run in either execution mode. The setup flow uses the phrase extraction method for both choices: at the credentials step it means Direct or Agent, and on the crawler configuration it means Account Usage or Information Schema.

Lineage from query history​

Atlan builds column-level lineage for Snowflake automatically from query history. Instead of relying on manual mapping or ETL tool integrations, the lineage miner reads QUERY_HISTORY and ACCESS_HISTORY from your account's usage views and derives lineage edges from the queries your users and pipelines actually ran. Subsequent mining runs keep lineage current as your transformations change. See Mine Snowflake to configure mining.

Horizon tags​

Beyond metadata extraction, Atlan reads Snowflake Horizon object tags and synchronizes them bidirectionally: tags imported from Snowflake appear on your Atlan assets, and tag values updated in Atlan can be pushed back to Snowflake. The crawl reference lists every object Atlan extracts and the tag fields involved.

See also​