What you will be able to do
- Tell tag inheritance apart from automatic tag propagation, and predict which objects receive a tag under each
- Configure PROPAGATE and ON_CONFLICT on a tag, and say which object dependencies and data movements carry it
- Choose the right view or table function to audit tag assignments, including inherited ones
Key concept
Inheritance versus propagation — Snowflake moves a tag onto other objects in two separate ways. Inheritance always pushes a tag down the containment hierarchy (account, database, schema, table, column). Propagation is something you switch on per tag, and it copies the tag sideways onto views, dynamic tables and tables built from the tagged source.
1.How tags inherit down the object hierarchy
Tags are how a governance team marks data as sensitive, owned by a cost centre or under a regulation. They only help if they reach the objects people actually query. The first way a tag spreads is inheritance, which follows the securable object hierarchy. Put a tag on an account and every schema in the account carries it. Put it on a table and every column in that table carries it. You don't configure anything for this. It is how the hierarchy works.
Inheritance does not follow dependencies between objects. A view isn't a child of the table it reads from, so in the chain table_1 » view_1 » materialized_view_1 the materialized view inherits nothing from either source. If tags need to travel along that chain, you have to turn on propagation, which the next section covers.
An inherited value is only a default, and three things can replace it. You can set the tag directly on the object with a more specific value. A column that inherits cost_center = 'sales', for example, can be set to sales_na for the North America cost centre. Automatic propagation also overwrites an inherited value, and so does sensitive data classification when it tags a column. Whenever two values compete, an inherited value has the weakest claim.
Checkpoint 1 of 7· Check yourself
Table orders has the tag cost_center = 'sales', so its column region inherits that value. Later, automatic propagation applies cost_center = 'marketing' to region from another source object. What value does region carry?
An inherited value is only a default and has the weakest claim. Setting the tag directly, automatic propagation and sensitive data classification all replace it, so region ends up with 'marketing'.
“The value of an inherited tag is overwritten when the tag is applied to the object as a result of automatic propagation.”Source: docs.snowflake.com
Sources1
2.Automatic tag propagation on dependency and data movement
Automatic tag propagation copies a tag from a source object to the objects built from it. You turn it on with the PROPAGATE property in CREATE TAG or ALTER TAG. Only the tag owner can set it, and the owner must also hold the account-level APPLY TAG privilege. Running a pipeline as ACCOUNTADMIN does not make propagation happen. The property on the tag is what decides it.
The property has three settings: ON_DEPENDENCY, ON_DATA_MOVEMENT (spelled out in the combined value) and ON_DEPENDENCY_AND_DATA_MOVEMENT. They cover different events and behave differently once the tag has arrived on the target.
| Aspect | Object dependency | Data movement |
|---|---|---|
| Triggering statements | Creating a view, secure view, materialized view or dynamic table from the source | CREATE TABLE ... AS SELECT (CTAS), CREATE DYNAMIC TABLE, and the DML commands INSERT, MERGE, UPDATE, COPY INTO |
| After the tag lands | Kept in sync continuously: tags added, values changed and tags removed on the source are mirrored on the target | Not updated afterwards: later changes on the source don't reach the target |
| Dependence on the source | Propagates only while the source object exists | Applied when the data moves, for instance on each INSERT or MERGE, but never refreshed by later tag changes on the source |
CREATE DYNAMIC TABLE is the one statement listed under both triggers. Creating a dynamic table from a source counts as an object dependency, and executing the statement also counts as data movement.
Teams most often get this wrong with CTAS. Suppose a tag is set to PROPAGATE = ON_DEPENDENCY. A view on RAW.CUSTOMERS gets the column tags, but a table created with CREATE TABLE ... AS SELECT from the same source gets none, because CTAS counts as data movement. The fix is to change the tag to ON_DEPENDENCY_AND_DATA_MOVEMENT. Data retention settings, switching to INSERT, or running as a different role do not help.
CREATE TABLE ... CLONE and CREATE TABLE ... LIKE work differently. They ignore PROPAGATE and always copy the source's tags to the new table. A clone of a sensitive table therefore carries its tags even when the tag has no propagation set.
Propagation is what turns tagging into a control. Snowflake states the goal as ensuring that policies associated with the tags reach the target objects. Tag-based masking policies attach a masking policy to a tag instead of a column. Once that tag propagates, a view or CTAS copy built from a masked column is masked as well, and nobody has to remember to protect the derived object.
Checkpoint 2 of 7· Match them up
Match each statement run against a tagged source table to how the tag reaches the new object.
Tap a term, then the definition that fits it.
Views, materialized views and dynamic tables count as object dependencies, so their tags stay in sync with the source. CTAS counts as data movement, so its tag is not refreshed after it lands. CLONE and LIKE ignore PROPAGATE and always copy the source's tags.
“tags applied to target objects when there is data movement are not continuously updated as tags change on the source object.”Source: docs.snowflake.com
Checkpoint 3 of 7· Exam question
A data engineering team set PROPAGATE = ON_DEPENDENCY on the tag pii_level. Source table RAW.CUSTOMERS has tagged columns. A pipeline now runs CREATE TABLE CURATED.CUSTOMERS_COPY AS SELECT * FROM RAW.CUSTOMERS, and the new table's columns carry no tags, while a view built on the same source does. What change makes the CTAS table receive the tags?
Correct answer: A — Switch the tag to PROPAGATE = ON_DEPENDENCY_AND_DATA_MOVEMENT, issued by the tag owner holding account-level APPLY TAG.
- A. Correct. ON_DEPENDENCY covers views and similar dependent objects only. Data movement such as CTAS needs ON_DATA_MOVEMENT or the combined mode, set by the tag owner with APPLY TAG.
- B. Incorrect. Time Travel retention has no effect on tag propagation and does not carry column tag metadata into a new table.
- C. Incorrect. DML only propagates tags when the tag's PROPAGATE setting includes data movement, so swapping statement types would not help while the setting stays ON_DEPENDENCY.
- D. Incorrect. The creating role is not the deciding factor; propagation on data movement is governed by the tag's PROPAGATE setting, not by who runs the CTAS.
Sources2
3.Resolving propagation conflicts with ON_CONFLICT
A conflict happens when one target receives the same tag from two sources that carry different values. A typical case is a view column computed from two tagged source columns. Two situations look like conflicts but aren't. If the target already has the tag set manually, the manual value is kept and the propagated value is skipped. If the target only inherited a value from higher in the hierarchy, the propagated value replaces it.
For real conflicts, the tag's ON_CONFLICT property decides the result. There are four options:
| Setting | Result on conflict | Constraint |
|---|---|---|
| Not set (default) | Tag value becomes the literal string CONFLICT | None |
| ON_CONFLICT = '<string>' | Tag value becomes your string, e.g. 'HIGHLY CONFIDENTIAL' | Used only when values actually conflict |
| ON_CONFLICT = ALLOWED_VALUES_SEQUENCE | The value that comes first in the tag's ALLOWED_VALUES list wins | Reordering ALLOWED_VALUES changes future results |
| ON_CONFLICT = MERGE | Conflicting values are combined into a set on the target | Requires MULTI_VALUE = TRUE |
CREATE TAG data_sensitivity ALLOWED_VALUES 'confidential', 'internal', 'public' PROPAGATE = ON_DEPENDENCY ON_CONFLICT = ALLOWED_VALUES_SEQUENCE;ALLOWED_VALUES_SEQUENCE is the only option that picks one of the competing values based on a ranking. Order the list deliberately, with the strictest value first, so the most restrictive value wins without anyone stepping in. The other options don't pick a winner. The default does not keep the first source's value; it writes CONFLICT. A string override writes one fixed value, and only on conflicting targets. MERGE keeps every value and needs a multi-value tag.
Conflicts and skipped propagations can be recorded in an event table, a table where Snowflake collects telemetry records such as these propagation events. Set ENABLE_TAG_PROPAGATION_EVENT_LOGGING at the account level. Events go to the event table set for the tag's database, or to the default event table if none is set. Records have scope snow.automatic_tag_propagation. The event types are CONFLICT, PROPAGATION_SKIPPED_MANUAL_TAG and TAG_PROPAGATION_LIMIT_EXCEEDED, which is logged when a tag would reach more than 10,000 targets and so is not propagated.
Each event type carries a severity, and the LOG_EVENT_LEVEL parameter governing the event table controls which severities are actually logged. If it is set above a severity, events of that severity never appear. A default-resolved conflict is WARN and the limit event is ERROR.
ALTER ACCOUNT SET ENABLE_TAG_PROPAGATION_EVENT_LOGGING = TRUE;Checkpoint 4 of 7· Fill the gap
A view column is computed from two source columns. One is tagged data_sensitivity = 'internal' and the other data_sensitivity = 'public'. The governance team wants the competing value that comes earliest in the tag's ranked list to win, with no manual step. Which value completes the tag definition?
CREATE TAG data_sensitivity ALLOWED_VALUES 'confidential', 'internal', 'public' PROPAGATE = ON_DEPENDENCY ON_CONFLICT = ? ;ALLOWED_VALUES_SEQUENCE is the only option that picks one of the competing values by its position in ALLOWED_VALUES, so this conflict resolves to 'internal'. MERGE would keep both values and also needs MULTI_VALUE = TRUE. A string override such as 'confidential' writes that fixed string, not either competing value.
Source: docs.snowflake.comCheckpoint 5 of 7· Exam question
A view joins HR.EMPLOYEES.SALARY_BAND (tagged sensitivity = 'internal') with FIN.PAYROLL.BONUS (tagged sensitivity = 'restricted') in a single output column through an expression. The tag sensitivity propagates on dependency, and the governance team wants the stricter value to win automatically instead of the default result. The tag is created with ALLOWED_VALUES 'public', 'internal', 'restricted'. Which configuration meets the requirement?
Correct answer: B — Set ON_CONFLICT = ALLOWED_VALUES_SEQUENCE on the propagation setting so that the value listed later in the allowed values wins when sources disagree.
- A. Incorrect. MERGE only applies to multi-value tags and concatenates values rather than choosing one, so it does not select the stricter classification here.
- B. Correct. ALLOWED_VALUES_SEQUENCE ranks conflicting values by their order in the tag's ALLOWED_VALUES, so listing values from least to most sensitive makes the strictest value win.
- C. Incorrect. A string override replaces the value only when sources conflict, but it hard-codes one label instead of choosing the stricter of the two competing values.
- D. Incorrect. By default a conflict yields the literal value CONFLICT, not the first source's value, which would leave the column unclassified for masking.
Sources2
4.Auditing tag assignments, with and without lineage
Auditing tags means picking a view or function that handles inheritance the way you need. The most common mistake is querying the Account Usage TAG_REFERENCES view and expecting inherited tags in the results. That view returns only direct assignments. A tag set on a schema shows up as one schema row, and none of the tables or columns that inherit it appear.
| Option | Scope | Includes inherited tags? |
|---|---|---|
| SNOWFLAKE.ACCOUNT_USAGE.TAGS (view) | Account: a catalog of current and deleted tags | Not applicable (lists tags, not assignments) |
| SNOWFLAKE.ACCOUNT_USAGE.TAG_REFERENCES (view) | Account | No |
| SNOWFLAKE.ACCOUNT_USAGE.TAG_REFERENCES_WITH_LINEAGE (table function) | Account, for one tag | Yes |
| INFORMATION_SCHEMA.TAG_REFERENCES (table function) | One database, for one object | Yes |
| INFORMATION_SCHEMA.TAG_REFERENCES_ALL_COLUMNS (table function) | Every column of one table or view | Yes |
| SYSTEM$GET_TAG (system function) | One tag on one object or column | Returns the assigned value |
For an account-wide list of everything that carries a tag, inherited assignments included, call TAG_REFERENCES_WITH_LINEAGE with the fully qualified tag name. Filtering TAG_REFERENCES by domain can't produce the missing rows, because the view never stored them. SHOW TAGS lists tag objects, not where they are assigned.
The word "lineage" in these names means tag inheritance, not the dependency chain between views and tables. "With lineage" and "with inherited tags" are the same thing here, so the function reports assignments that arrived down the containment hierarchy. It does not report tags that arrived through propagation along a view chain as a separate category.
SELECT * FROM TABLE( SNOWFLAKE.ACCOUNT_USAGE.TAG_REFERENCES_WITH_LINEAGE( 'my_db.my_schema.cost_center' ) );TAG_REFERENCES_ALL_COLUMNS has one quirk that trips people up: its domain argument is always 'table', even when you pass a view or a materialized view. Snowsight also has a no-SQL option, Governance & security » Tags & policies, with coverage and prevalence on a Dashboard and a filterable Tagged Objects list. It requires Enterprise Edition or higher and either ACCOUNTADMIN or a role granted SNOWFLAKE.GOVERNANCE_VIEWER and SNOWFLAKE.OBJECT_VIEWER directly. The Dashboard refreshes every 12 hours, and Tagged Objects can lag by up to two hours.
Checkpoint 6 of 7· Check yourself
An auditor set the tag pii_owner on the schema SALES.CRM. They now need one account-wide list of every table and column that carries pii_owner, including the ones that only inherit it from the schema. Which query gives them that list?
TAG_REFERENCES_WITH_LINEAGE works across the whole account for one tag and includes inherited assignments. The TAG_REFERENCES view stores only direct assignments, so filtering by domain cannot bring back the inherited rows. SHOW TAGS lists the tag objects, not where they are assigned, and TAG_REFERENCES_ALL_COLUMNS only covers one table or view.
“TAG_REFERENCES_WITH_LINEAGE to determine all of the objects that have a given tag key and tag value, including objects that inherited tags”Source: docs.snowflake.com
Checkpoint 7 of 7· Exam question
An auditor applied the tag governance.tags.pii_level at schema level on SALES.CUSTOMERS and asks for a list of every table and column that carries the tag, inherited ones included, across the whole account. A query against SNOWFLAKE.ACCOUNT_USAGE.TAG_REFERENCES returns only the schema row. Which approach produces the complete list?
Correct answer: C — Call the TAG_REFERENCES_WITH_LINEAGE table function with the fully qualified tag name, which also returns objects that inherited the tag.
- A. Incorrect. SHOW TAGS lists tag definitions and does not report the objects that carry or inherit a tag value.
- B. Incorrect. The ACCOUNT_USAGE view records direct assignments only, so filtering on the column domain still excludes columns that merely inherit the schema's tag.
- C. Correct. The TAG_REFERENCES_WITH_LINEAGE function takes a tag name and returns every object that has the tag, including objects that only inherit it from a parent.
- D. Incorrect. OBJECT_DEPENDENCIES describes view and reference dependencies, not the securable hierarchy, so it cannot reconstruct which columns inherit a schema-level tag.
Sources3
Sources
(3)
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“A descendant of an object in the hierarchy inherits tags from its ancestors.”
↩︎ How tags inherit down the object hierarchy“the tag can be updated with a more specific tag string value such as sales_na”
↩︎ How tags inherit down the object hierarchy“The value of an inherited tag is overwritten by sensitive data classification.”
↩︎ How tags inherit down the object hierarchy“A descendant of an object in the hierarchy inherits tags from its ancestors.”
↩︎ Key concept“Tag inheritance does not include propagation to nested objects.”
↩︎ Exam trap 1“Tag inheritance does not include propagation to nested objects.”
↩︎ Prediction“The value of an inherited tag is overwritten when the tag is applied to the object as a result of automatic propagation.”
↩︎ Checkpoint - 2.
“Only the tag owner with the account-level APPLY TAG privilege can implement automatic tag propagation.”
↩︎ Automatic tag propagation on dependency and data movement“Creating a view, secure view, materialized view, or dynamic table from a source object is considered an object dependency.”
↩︎ Automatic tag propagation on dependency and data movement“Executing a CREATE DYNAMIC TABLE statement.”
↩︎ Automatic tag propagation on dependency and data movement“When you execute these statements, tags from the source are always assigned to the target object.”
↩︎ Automatic tag propagation on dependency and data movement“Ensuring that any policies associated with the tags are automatically applied to the target objects.”
↩︎ Automatic tag propagation on dependency and data movement“the existing tag value takes precedence over a propagated value so there is no conflict.”
↩︎ Resolving propagation conflicts with ON_CONFLICT“the value of the data_sensitivity tag will be internal because it comes before public in the list of allowed values.”
↩︎ Resolving propagation conflicts with ON_CONFLICT“This option is valid only when the tag has MULTI_VALUE = TRUE.”
↩︎ Resolving propagation conflicts with ON_CONFLICT“because there were more than 10,000 target objects.”
↩︎ Resolving propagation conflicts with ON_CONFLICT“You can use an event table to collect telemetry data related to tag propagation.”
↩︎ Resolving propagation conflicts with ON_CONFLICT“Tag propagation events are logged only if the LOG_EVENT_LEVEL parameter governing the table”
↩︎ Resolving propagation conflicts with ON_CONFLICT“tags applied to target objects when there is data movement are not continuously updated as tags change on the source object.”
↩︎ Exam trap 2“tags applied to target objects when there is data movement are not continuously updated as tags change on the source object.”
↩︎ Checkpoint - 3.
“including objects that inherited tags”
↩︎ Auditing tag assignments, with and without lineage“whether you want to track tag inheritance”
↩︎ Auditing tag assignments, with and without lineage“the domain TABLE must be used for all objects that contain columns, even if the object name is a view”
↩︎ Auditing tag assignments, with and without lineage“Snowsight updates the Dashboard every 12 hours.”
↩︎ Auditing tag assignments, with and without lineage“but does not include objects that inherited the tag”
↩︎ Exam trap 3“TAG_REFERENCES_WITH_LINEAGE to determine all of the objects that have a given tag key and tag value, including objects that inherited tags”
↩︎ Checkpoint