Declarative node configuration
Apply reviewed host settings from Git with ownership, drift, and maintenance controls.
Declarative node configuration maps a versioned definition to selected node IDs. A controller compares desired configuration with observed host state and proposes or applies supported changes under organization policy.
Connect a repository
Under Fleet → Configuration sources, choose an approved repository, pinned revision or tracked branch, definition path, and read credential. Verify the resolved commit and file digest before enabling application.
A repository credential does not grant node administration by itself. The controller identity needs the required fleet and maintenance permissions. Secrets remain references to approved providers rather than plaintext committed values.
Own specific fields
A definition declares which host fields it manages, such as admission labels, agent channel, supported service settings, or approved package policy. Unsupported arbitrary host commands are rejected by schema validation.
An owned field changed manually appears as drift. The configured mode either reports the difference or schedules reconciliation. Another controller cannot silently take ownership of the same field set; its conflict must be resolved through a reviewed ownership transfer.
Disruptive changes
A new Git commit does not bypass change windows or required approvals. A proposed kernel, hypervisor, storage, or mesh change can require an ordinary maintenance plan and its preflight.
Canary targets and rollout waves retain resolved node IDs. Nodes enrolled after approval are not silently added to a disruptive wave merely because they match the selector.
Failure and rollback
Each application records source commit, rendered configuration, node revision, operation, and observed result. A partial rollout retains successful changes while failed targets remain identified.
Reverting Git creates another desired revision. It does not guarantee that a package downgrade or persistent state reversal is supported. Use the component's recovery procedure when the old configuration cannot safely be reapplied.
Disable reconciliation and release field ownership before moving nodes to another source. Deleting a local checkout does not stop the server controller. See blueprints for project resources and maintenance plans for disruptive execution.