What you will be able to do
- Build a least-privilege custom role that can create, refresh or promote failover groups without holding ACCOUNTADMIN
- Choose which role should hold OWNERSHIP, REPLICATE, FAILOVER, MODIFY and MONITOR on a group, and grant them in both accounts
- Audit which roles can create or refresh a replication or failover group
- Limit where security objects can be replicated with ALLOWED_ACCOUNTS, OBJECT_TYPES and ALLOWED_INTEGRATION_TYPES
- Predict who owns objects a refresh creates in the target account, and which local objects a refresh drops
Key concept
Replication privileges on group objects — Replication and failover groups are securable objects with their own privileges. Creating a group is an account-level privilege. Refreshing, promoting, changing and viewing a group are each a separate privilege on the group itself. Each privilege can go to a different custom role, so no single role needs ACCOUNTADMIN to run disaster recovery.
1.Privileges that control replication and failover groups
A failover group decides which account your users, roles, integrations and databases are served from, so the privileges on it carry real weight. A role that can refresh a secondary group can overwrite the target account's copy of those objects. A role that can promote a group can make the target the live account. Snowflake does not bundle these powers into one privilege. Each operation has its own privilege, and that is how you keep disaster-recovery work out of ACCOUNTADMIN.
| Privilege | Granted on | Allows | Granted by |
|---|---|---|---|
| CREATE FAILOVER GROUP | Account | Creating a failover group | ACCOUNTADMIN only |
| CREATE REPLICATION GROUP | Account | Creating a replication group | ACCOUNTADMIN only |
| OWNERSHIP | Replication or failover group | Deleting, altering, and granting or revoking access to the group | ACCOUNTADMIN, a role with MANAGE GRANTS, or the owning role |
| REPLICATE | Replication or failover group | Refreshing a secondary group | The role with OWNERSHIP on the group |
| FAILOVER | Failover group | Promoting a secondary failover group to primary | The role with OWNERSHIP on the group |
| MODIFY | Replication or failover group | Changing the group's settings or properties | The role with OWNERSHIP on the group |
| MONITOR | Replication or failover group | Viewing details within the group | The role with OWNERSHIP on the group |
Because of the two creation privileges, ACCOUNTADMIN has to act once, and only once. It creates a custom role, grants that role CREATE FAILOVER GROUP on the account, and that role is granted to the disaster-recovery engineers and no one else. The role hierarchy behind this pattern is covered in RBAC role hierarchy and least privilege (system roles, functional vs access roles). Here the point is narrower: the custom role holds one account privilege, not the whole administrative tree.
USE ROLE ACCOUNTADMIN; CREATE ROLE myrole; GRANT CREATE FAILOVER GROUP ON ACCOUNT TO ROLE myrole;Each tempting shortcut gives away too much or does the wrong job. Handing out SECURITYADMIN or MANAGE GRANTS is far too broad: MANAGE GRANTS lets its holder see every object in the account. It also cannot stand in for the creation privilege, because only ACCOUNTADMIN can grant CREATE FAILOVER GROUP. REPLICATE is the wrong tool too. It is a privilege on a group, it lets a role refresh a secondary group, and it does nothing for creating one. Neither FAILOVER nor MONITOR allows a refresh.
The privileges on the group itself can be split further. The role that owns the group grants REPLICATE to a replication role that runs ALTER FAILOVER GROUP ... REFRESH, and FAILOVER to a separate role that is allowed to promote. MONITOR goes to anyone who only needs to watch. One detail catches teams out during a real outage: neither REPLICATE nor FAILOVER is replicated. Each must be granted on the group in the source account and again in the target account.
GRANT REPLICATE ON FAILOVER GROUP myfg TO ROLE my_replication_role;Auditing these grants takes two tools. SHOW GRANTS on a specific group lists the current privileges on that object, so its OWNERSHIP and REPLICATE rows tell you who controls and who can refresh that group. To find every role that can create groups, query SNOWFLAKE.ACCOUNT_USAGE.GRANTS_TO_ROLES, filtering PRIVILEGE on CREATE FAILOVER GROUP and CREATE REPLICATION GROUP and GRANTED_ON on the group object kind. The view can lag by up to two hours, so a grant made a few minutes ago may not appear in it yet.
Checkpoint 1 of 7· Match them up
A disaster-recovery runbook splits replication work across several custom roles. Match each privilege to the operation it allows.
Tap a term, then the definition that fits it.
Each replication operation has its own privilege. Creating a group is an account-level privilege. Refreshing, promoting, changing and viewing a group are separate privileges on the group itself, so each one can go to a different least-privilege role.
“FAILOVER | Failover Group | Grants the ability to promote a secondary failover group to serve as primary failover group.”Source: docs.snowflake.com
Checkpoint 2 of 7· Check yourself
In the source account, the role that owns failover group myfg granted REPLICATE on myfg to my_replication_role. During an outage, my_replication_role runs ALTER FAILOVER GROUP myfg REFRESH in the target account, and the command fails for lack of privileges. What fixes this while keeping least privilege?
REPLICATE is not replicated with the group. The owning role has to grant it on the group again in the target account. MANAGE GRANTS is far too broad, and CREATE FAILOVER GROUP lets a role create groups, not refresh them.
“The REPLICATE privilege is currently not replicated and must be granted on a failover (or replication) group in both the source and target accounts.”Source: docs.snowflake.com
Checkpoint 3 of 7· Exam question
A platform team must let a disaster-recovery engineering group create failover groups in the production account, but the security standard forbids handing out ACCOUNTADMIN. Which approach follows least privilege?
Correct answer: C — Create a custom role, grant it CREATE FAILOVER GROUP on the account, and assign it only to the DR engineers.
- A. Incorrect: SECURITYADMIN does not hold CREATE FAILOVER GROUP by default and brings extensive user and role management rights.
- B. Incorrect: REPLICATE is a privilege on an existing group that permits refreshing it; it does not allow creating groups.
- C. Correct: CREATE FAILOVER GROUP is an account-level privilege held only by ACCOUNTADMIN by default, so it can be delegated narrowly to a dedicated custom role.
- D. Incorrect: MANAGE GRANTS lets a role grant any privilege, which is far broader than needed and defeats least privilege.
Checkpoint 4 of 7· Exam question
Which privilege must a role hold on a replication group before it can run a manual refresh of that group in a target account?
Correct answer: D — The REPLICATE privilege on the group, allowing the role to refresh the secondary group.
- A. Incorrect: FAILOVER authorizes promoting a secondary failover group to primary, not refreshing it.
- B. Incorrect: MONITOR only permits viewing group details and does not allow a refresh to be started.
- C. Incorrect: USAGE is not a privilege that governs refreshing replication or failover groups.
- D. Correct: REPLICATE on the group is the privilege that authorizes a manual refresh of the group's objects in the target.
Checkpoint 5 of 7· Exam question
An auditor asks which roles in the primary account can create replication or failover groups and which roles can refresh a particular group. Which TWO actions give accurate answers?(Select 2)
Correct answers: A, D — Run SHOW GRANTS ON REPLICATION GROUP for the group in question and review the REPLICATE and OWNERSHIP rows returned.; Query SNOWFLAKE.ACCOUNT_USAGE.GRANTS_TO_ROLES, filtering on the group creation privileges and the group object type.
- A. Correct: SHOW GRANTS ON the group object lists which roles hold REPLICATE and OWNERSHIP on it.
- B. Incorrect: LOGIN_HISTORY records authentication events and says nothing about privilege grants on groups.
- C. Incorrect: this view tracks usage and cost per group, not privilege grants.
- D. Correct: GRANTS_TO_ROLES records privilege grants, so filtering on the privilege and object type lists every role holding them.
- E. Incorrect: that view reports credits and bytes consumed by replication, not who holds privileges.
2.Limiting where replicated security objects can go
A primary group can only replicate to the accounts named in its ALLOWED_ACCOUNTS list, written as <org_name>.<target_account_name>. That list is the first thing to review if security objects might be sent to an account nobody expected. Two more gates come before it. An organization administrator must enable replication for both the source and the target account, and SHOW REPLICATION ACCOUNTS lists the accounts where replication is already enabled.
OBJECT_TYPES decides which kinds of objects travel at all. Users, roles, integrations, network policies and account parameters each require Business Critical Edition or higher. Adding INTEGRATIONS is not enough by itself. You must also set ALLOWED_INTEGRATION_TYPES, and only security, API, storage, external access and some notification integrations are supported. A group that lists only security integrations cannot carry a storage integration's cloud credentials to the target account by accident.
CREATE FAILOVER GROUP FG OBJECT_TYPES = users, roles, warehouses, resource monitors, integrations ALLOWED_INTEGRATION_TYPES = security integrations ALLOWED_ACCOUNTS = example.northamericaeast REPLICATION_SCHEDULE = '10 MINUTE';Network policies only arrive intact if the objects they depend on arrive with them. How allowed and blocked lists behave is taught in Network policies: allowed/blocked lists and where they apply. For replication, the rule is to include whatever each policy references. If a policy uses a network rule, include the database that holds the rule's schema. If the policy is set on the account, add ACCOUNT PARAMETERS. If it is set on users, add USERS. Leave one of these out and the refresh fails with a dangling-references error.
Group layout has hard limits too. An account can have only one replication or failover group that contains objects other than databases or shares. All your users, roles, integrations and policies therefore share one group. Some protections also do not travel with replication: Tri-Secret Secure and private connectivity such as AWS PrivateLink are not on by default in a target account. You must configure them there yourself.
Checkpoint 6 of 7· Check yourself
A failover group lists OBJECT_TYPES = USERS, ROLES, NETWORK POLICIES. The network policy set on the account uses a network rule stored in database sec_db. The next refresh fails with a dangling-references error. Which change to the primary group fixes the refresh?
A replicated network policy needs everything it references to travel with it. Because this policy is set on the account, the group needs ACCOUNT PARAMETERS, and because it uses a network rule, the group needs the database that holds the rule's schema. REPLICATE controls who may refresh, not what gets replicated. A policy created locally in the target account has no global identifier, so it would not supply the missing references.
“If a network policy is associated with the account, include NETWORK POLICIES and ACCOUNT PARAMETERS in the OBJECT_TYPES list.”Source: docs.snowflake.com
3.Who owns replicated objects in the target account
Who owns a new object after a refresh depends on whether ROLES is in the group. With roles replicated, the target account gives OWNERSHIP to the same role that owns the object in the source, as soon as roles are next replicated. If the objects and their roles are in the same group, this happens in the same refresh. Without roles, there is no matching role to give it to, and GLOBALORGADMIN becomes the owner. To keep owner roles the same across both accounts, replicate ROLES in the group that also carries the objects.
Everything replicated into the target account is read-only there. If USERS is replicated, you cannot create or modify users in the target account. If ROLES is replicated, you cannot create or modify roles, and you cannot grant privileges to a role on a secondary object. Local objects are still yours to manage, including the secondary group itself. That is why the target account's group owner can grant REPLICATE and FAILOVER to replicated roles.
Objects created locally in the target account carry a real risk. Users or roles made there by a script have no global identifier. A refresh drops any object of a replicated type that lacks one. The first refresh that replicates USERS or ROLES may fail with an error instead of dropping them, which protects data linked to those users and roles. Treat that error as a warning: the next refresh can remove the provisioning you built locally.
Checkpoint 7 of 7· Check yourself
Before anyone created the secondary failover group, a provisioning script added several users and roles directly in the target account. The primary group replicates USERS and ROLES. What risk do those script-created users and roles face once refreshes begin?
Users and roles that a script creates in the target account have no global identifier. A refresh drops every object of a replicated type that lacks one. The first refresh that replicates USERS or ROLES may fail with an error instead, which protects the data linked to them, but later refreshes can still remove them. GLOBALORGADMIN ownership applies to new replicated objects whose roles were not replicated, not to objects created locally.
“the refresh operation drops any account objects of the types in the OBJECT_TYPES list in the target account that have no global identifier”Source: docs.snowflake.com
Sources
(7)
Every claim above is drawn from one of these pages, quoted as it was written on the date shown.
- 1.
“Grants the ability to refresh a secondary group.”
↩︎ Privileges that control replication and failover groups“Grants the ability to promote a secondary failover group to serve as primary failover group.”
↩︎ Privileges that control replication and failover groups“If a network policy uses a network rule, include the database that contains the schema where the network rule was created.”
↩︎ Limiting where replicated security objects can go“An account can only have one replication or failover group that contains objects other than databases or shares.”
↩︎ Limiting where replicated security objects can go“All secondary objects in a target account, including secondary databases and shares, are read-only.”
↩︎ Who owns replicated objects in the target account“privileges can be granted to (or revoked from) a secondary role on local objects”
↩︎ Who owns replicated objects in the target account“operations users can perform on replication and failover group objects in the system”
↩︎ Key concept“Must be granted by the ACCOUNTADMIN role.”
↩︎ Exam trap 2“the refresh operation drops any account objects of the types in the OBJECT_TYPES list in the target account that have no global identifier.”
↩︎ Exam trap 3“FAILOVER | Failover Group | Grants the ability to promote a secondary failover group to serve as primary failover group.”
↩︎ Checkpoint“If a network policy is associated with the account, include NETWORK POLICIES and ACCOUNT PARAMETERS in the OBJECT_TYPES list.”
↩︎ Checkpoint“the refresh operation drops any account objects of the types in the OBJECT_TYPES list in the target account that have no global identifier”
↩︎ Checkpoint - 2.
“The REPLICATE privilege is currently not replicated”
↩︎ Privileges that control replication and failover groups“The organization administrator must enable replication for the source and target accounts.”
↩︎ Limiting where replicated security objects can go“Target accounts do not have Tri-Secret Secure or private connectivity to the Snowflake service, such as AWS PrivateLink, enabled by default.”
↩︎ Limiting where replicated security objects can go“The FAILOVER privilege is currently not replicated and must be granted in each source and target account.”
↩︎ Exam trap 1“The REPLICATE privilege is currently not replicated and must be granted on a failover (or replication) group in both the source and target accounts.”
↩︎ Checkpoint - 3.
“The MANAGE GRANTS access privilege implicitly allows its holder to see every object in the account.”
↩︎ Privileges that control replication and failover groups - 4.
“to view the current set of privileges granted on an object, you can run the SHOW GRANTS command.”
↩︎ Privileges that control replication and failover groups - 5.
“Latency for the view may be up to 120 minutes (2 hours).”
↩︎ Privileges that control replication and failover groups - 6.
“If integration objects are included in the list of specified object types, the ALLOWED_INTEGRATION_TYPES parameter must be set.”
↩︎ Limiting where replicated security objects can go - 7.
“the OWNERSHIP privilege is granted to the same role on the target account as the role with the OWNERSHIP privilege in the source account”
↩︎ Who owns replicated objects in the target account“the OWNERSHIP privilege for the new objects is granted to the GLOBALORGADMIN role.”
↩︎ Prediction