Domain 5: Data Collaboration

Overview

Verified against the official Snowflake COF-C03 Exam Study Guide — follows guide sections 5.1-5.3. Smallest domain by weight, but note it's where the official guide files Cloning, Time Travel, and Fail-safe (not under Architecture) — don't under-study this section just because it looks small.

Data collaboration and protection

• Cloning (zero-copy clone): CREATE TABLE/SCHEMA/DATABASE ... CLONE .... Metadata-only operation at clone time — instant regardless of source size, no extra storage consumed until either the clone or the original diverges (copy-on-write). Gotcha: cloning a schema/table that contains a stream with unconsumed change records creates a stream re-initialized at the point of cloning — pending records from the original do not carry over to the clone (a documented official sample-question scenario; see 09_Official_Sample_Questions_Analysis.md).
• Time Travel: query/restore historical data via AT/BEFORE clauses and UNDROP. Default 1 day on Standard edition, configurable up to 90 days on Enterprise+ (DATA_RETENTION_TIME_IN_DAYS). Applies to tables, schemas, databases.
• Fail-safe: additional fixed 7-day period after Time Travel ends, Snowflake-support- only recovery (you cannot query/restore it yourself). Permanent tables only — transient and temporary tables skip Fail-safe entirely. Sequence: live data → Time Travel (0/1 up to 90 days) → Fail-safe (7 days, permanent tables only) → purged.
• Data replication and failover: replicate databases/accounts across regions for disaster recovery; failover promotes a secondary to primary. See also Domain 2 for the account-management framing of the same feature.
  • Replication groups vs. failover groups are two distinct object types (verified against current docs) — easy to conflate: a replication group keeps objects in sync (read-only copy) but does not support promotion/failover, and database/share replication on its own is available to any edition. A failover group additionally supports promoting the secondary to primary — but a failover group can only be created on (and only replicated to) Business Critical edition or higher. An object can belong to one or the other, never both at once. Each group (of either kind) has its own independent replication schedule.
• Secure data sharing features: see 5.2 below — the umbrella term for everything that follows.

Snowflake's data sharing capabilities

• Secure Data Sharing — the core mechanism: a provider account shares live, read-only access to database objects with a consumer account, with zero data copying/movement. The consumer queries the provider's storage directly using their own compute — the consumer's warehouse is billed for their queries, not the provider's. Changes on the provider side are visible to the consumer instantly. Only works within the same cloud region/provider unless the share is replicated first.
• Accounts involved:
  • Provider: creates the Share object, grants USAGE/SELECT, adds consumer accounts.
  • Consumer: gets a read-only database created from the inbound share, queries it like any other database.
  • Reader accounts: a special account type a provider creates and pays the compute for, on behalf of a consumer who has no Snowflake account of their own.
• Sharing and resharing: a consumer with appropriate privileges can, in some configurations, reshare data they've received onward to a further consumer — know this is possible and governed by the same grant/privilege model, not an automatic default.
• Direct shares: the straightforward provider→specific-named-consumer-account sharing pattern (as opposed to publishing broadly via a Marketplace listing) — the baseline mechanism before you add Marketplace/listing distribution on top.
• Secure views / secure UDFs: prefer these over sharing raw tables when the share should expose a filtered/masked/derived subset — the view definition itself is hidden from the consumer, preventing reverse-engineering the underlying logic or leaking rows via crafted queries.
• Data Clean Rooms: privacy-preserving joint analysis pattern — two parties run combined analysis without either seeing the other's raw rows, built on secure views/UDFs plus strict governance. Awareness-level only, given the domain's overall 10% weight.

Snowflake Marketplace and listings

• Snowflake Marketplace: the public marketplace where providers publish listings, discoverable and directly queryable by any Snowflake consumer with zero data movement.
• Listings — two visibility types (this is the guide's own terminology; supersedes the older "Data Exchange" branding):
  • Public listings: visible to any Snowflake customer via the Marketplace.
  • Private listings: visible only to specifically named consumer account(s) — the invite-only equivalent, for sharing within a company or a defined consortium.
  • Paid listings integrate with Snowflake billing so consumers can be charged for access.
• Native Apps: applications a provider builds and distributes through the Marketplace, which run entirely inside the consumer's own account (their compute, their data never leaves) — distinct from a plain data listing because it ships logic/UI, not just data.

Self-check before moving on

• Can you state, precisely, whose compute is billed when a consumer queries shared data?
• Can you explain what happens to a stream's pending records across a clone, and why?
• Can you state Time Travel's default/max retention and how Fail-safe differs (duration, who can restore it, which table types skip it)?
• Can you distinguish a public listing, a private listing, and a Native App by what each one actually distributes (data access vs. running application)?
• Can you explain why a provider would use a secure view instead of sharing a table directly?