Documentation
¶
Overview ¶
Package drift is the shared vocabulary for divergence between what Miabi recorded and what Docker runs. Node housekeeping reports it for one node; the control manager watches it across the platform.
Index ¶
- Constants
- func OwnerOf(labels map[string]string) (kind string, id uint, ok bool)
- func Replaced(recorded, engineCreatedAt string) bool
- func VolumeOwner(labels map[string]string) (kind string, id uint, ok bool)
- func VolumeState(ctx context.Context, dc docker.Client, name, recordedCreatedAt string) (string, error)
- type Item
Constants ¶
const ( ClassOrphan = "orphan" // managed label present, DB record gone — still running on the node ClassMissing = "missing" // DB record expects it live, nothing running for it ClassUntracked = "untracked" // no miabi.* label — a hand-run resource // ClassReplaced is a resource whose record is intact but whose contents are not: a volume deleted and // recreated by hand keeps its name and its row, and holds none of the data they describe. ClassReplaced = "replaced" )
Classes.
const ( ActionRemove = "remove" // orphan → delete the lingering resource ActionRedeploy = "redeploy" // missing → redeploy from the owning resource ActionImport = "import" // untracked → adopt via the existing import flow // ActionRestore is for lost data: recreating the volume would give an empty one, so the way back is a // backup, not a redeploy. ActionRestore = "restore" // ActionNone is for drift Miabi must report but may not touch, because something else owns the resource — // the platform's own gateway, installed and owned by the Miabi stack. ActionNone = "none" )
Recommended actions per item.
const ( OwnerApp = "app" OwnerDatabase = "database" OwnerVolume = "volume" OwnerStack = "stack" OwnerConfig = "config" // OwnerNode is for what a node itself owns rather than a workspace: its gateway. OwnerNode = "node" )
Owner kinds: the DB record class a managed resource belongs to.
Variables ¶
This section is empty.
Functions ¶
func OwnerOf ¶
OwnerOf returns the owning DB record (kind and numeric id) encoded in a managed resource's labels, so drift can check whether that record still exists. ok is false when the resource is not orphan-eligible. Precedence matters: an app's container carries both app and stack labels, and the app is the owning record.
func Replaced ¶
Replaced reports whether a volume that exists under its own name is no longer the one Miabi recorded: its engine creation timestamp moved, so it was deleted and created again, and it holds none of the data the record describes. A volume Miabi never recorded a timestamp for is not replaced — most predate the column, and calling them all replaced would cry wolf about every one of them.
func VolumeOwner ¶
VolumeOwner returns the record a managed Docker volume backs: a volume row, or the database instance whose data it holds. App and stack labels are not owners here: neither record owns a volume's data.
func VolumeState ¶
func VolumeState(ctx context.Context, dc docker.Client, name, recordedCreatedAt string) (string, error)
VolumeState reports whether a volume still holds the data its record describes: "" when it does, ClassMissing when it is gone, ClassReplaced when it was deleted and recreated. An engine that cannot answer returns an error and no class — not knowing is never evidence of loss.
Types ¶
type Item ¶
type Item struct {
Class string `json:"class"`
Kind string `json:"kind"` // container | volume | config | service
Ref string `json:"ref"` // container ID or volume name (the apply handle)
Name string `json:"name"`
Image string `json:"image,omitempty"`
State string `json:"state,omitempty"`
OwnerKind string `json:"owner_kind,omitempty"` // app | database | volume | stack
OwnerID uint `json:"owner_id,omitempty"`
Action string `json:"action"`
}
Item is one resource that diverges from intent.