api/v2: add device update-status

updating from a started update until the node runs another version,
or for at most 10 minutes; else not-updating. Kept in memory.

Updates #3629
This commit is contained in:
Kristoffer Dalby committed 2026-11-05 15:51:04 +01:00
1 parent 8230c58a0c
commit 7af29e8c45
4 files changed
+107 -1

No files matched your search

+37
View File
@@ -108,6 +108,11 @@ type DevicePostureAttributes struct {
Expiries map[string]time.Time `json:"expiries,omitempty"`
}
// DeviceUpdateStatus is the GET /device/{id}/update-status response.
type DeviceUpdateStatus struct {
Status string `enum:"updating,not-updating" json:"status"`
}
// Request bodies match the Tailscale SDK wire shapes and, like Tailscale's,
// ignore properties they do not define.
type (
@@ -186,6 +191,8 @@ type (
}
}
emptyOutput struct{ Body struct{} }
deviceUpdateStatusOutput struct{ Body DeviceUpdateStatus }
)
func registerDevices(api huma.API, b Backend) {
@@ -541,8 +548,38 @@ func registerDevices(api huma.API, b Backend) {
return nil, huma.Error404NotFound("not found")
}
b.State.ClientUpdateStarted(node.ID(), hi.IPNVersion())
return &emptyOutput{}, nil
})
huma.Register(api, requireScope(huma.Operation{
OperationID: "getDeviceUpdateStatus",
Method: http.MethodGet,
Path: "/api/v2/device/{id}/update-status",
Summary: "Get a device's client update status",
Description: "updating from a started client update until the device runs a new version, else not-updating.",
Tags: deviceTags,
Security: security,
Errors: []int{http.StatusUnauthorized, http.StatusForbidden, http.StatusNotFound},
}, scope.DevicesCoreRead), func(_ context.Context, in *deviceIDInput) (*deviceUpdateStatusOutput, error) {
node, err := lookupNode(b, in.DeviceID)
if err != nil {
return nil, err
}
var version string
if hi := node.Hostinfo(); hi.Valid() {
version = hi.IPNVersion()
}
status := "not-updating"
if b.State.ClientUpdating(node.ID(), version) {
status = "updating"
}
return &deviceUpdateStatusOutput{Body: DeviceUpdateStatus{Status: status}}, nil
})
}
// lookupNode resolves a device id to its NodeView, mapping a malformed or
+1
View File
@@ -125,6 +125,7 @@ func TestAPIv2OAuthMatrix_Enforcement(t *testing.T) {
{"updateDeviceClient", http.MethodPost, "/api/v2/device/1/update-client", scope.DevicesCore, nil, false},
{"setDeviceRoutes", http.MethodPost, "/api/v2/device/1/routes", scope.DevicesRoutes, map[string]any{}, false},
{"getDeviceRoutes", http.MethodGet, "/api/v2/device/1/routes", scope.DevicesRoutesRead, nil, false},
{"getDeviceUpdateStatus", http.MethodGet, "/api/v2/device/1/update-status", scope.DevicesCoreRead, nil, false},
{"getACL", http.MethodGet, "/api/v2/tailnet/-/acl", scope.PolicyFileRead, nil, false},
{"setACL", http.MethodPost, "/api/v2/tailnet/-/acl", scope.PolicyFile, map[string]any{}, false},
{"getSettings", http.MethodGet, "/api/v2/tailnet/-/settings", scope.FeatureSettingsRead, nil, false},
+52 -1
View File
@@ -10,14 +10,20 @@ import (
const pingIDLength = 16
// clientUpdateWindow is how long a started client update reads as updating
// while the node keeps its old version, so a failed update ends.
const clientUpdateWindow = 10 * time.Minute
// pingTracker correlates outgoing [tailcfg.PingRequest]s with incoming HEAD
// callbacks. Entries have no server-side TTL: callers are responsible
// for cleaning up via [pingTracker.cancel] or by reading from the response
// channel within their own timeout.
// channel within their own timeout. It also remembers the client updates
// nodes started over c2n.
type pingTracker struct {
mu sync.Mutex
pending map[string]*pendingPing
c2n map[string]*pendingC2N
updates map[types.NodeID]startedUpdate
}
type pendingPing struct {
@@ -33,10 +39,17 @@ type pendingC2N struct {
responseCh chan []byte
}
// startedUpdate is a client update a node started while running from.
type startedUpdate struct {
from string
at time.Time
}
func newPingTracker() *pingTracker {
return &pingTracker{
pending: make(map[string]*pendingPing),
c2n: make(map[string]*pendingC2N),
updates: make(map[types.NodeID]startedUpdate),
}
}
@@ -165,6 +178,31 @@ func (pt *pingTracker) cancelC2N(c2nID string) {
pt.mu.Unlock()
}
// updateStarted records that nodeID started a client update at now while
// running version.
func (pt *pingTracker) updateStarted(nodeID types.NodeID, version string, now time.Time) {
pt.mu.Lock()
pt.updates[nodeID] = startedUpdate{from: version, at: now}
pt.mu.Unlock()
}
// updating reports whether nodeID's started update still runs: the node
// still runs version and the window has not passed. An ended update is
// forgotten.
func (pt *pingTracker) updating(nodeID types.NodeID, version string, now time.Time) bool {
pt.mu.Lock()
defer pt.mu.Unlock()
u, ok := pt.updates[nodeID]
if ok && u.from == version && now.Sub(u.at) < clientUpdateWindow {
return true
}
delete(pt.updates, nodeID)
return false
}
// RegisterPing tracks a pending ping and returns its ID and a channel
// for the latency. Callers must defer [State.CancelPing] or read the channel
// within their own timeout; there is no server-side TTL.
@@ -207,3 +245,16 @@ func (s *State) CompleteC2N(c2nID string, body []byte) bool {
func (s *State) CancelC2N(c2nID string) {
s.pings.cancelC2N(c2nID)
}
// ClientUpdateStarted records that nodeID started a client update while
// running version.
func (s *State) ClientUpdateStarted(nodeID types.NodeID, version string) {
s.pings.updateStarted(nodeID, version, time.Now())
}
// ClientUpdating reports whether nodeID is still updating its client: it
// started an update while running version, still runs it, and
// [clientUpdateWindow] has not passed.
func (s *State) ClientUpdating(nodeID types.NodeID, version string) bool {
return s.pings.updating(nodeID, version, time.Now())
}
+17
View File
@@ -243,3 +243,20 @@ func TestPingTracker_DrainC2N(t *testing.T) {
assert.Empty(t, pt.c2n, "c2n map should be empty after drain")
}
func TestPingTracker_ClientUpdate(t *testing.T) {
pt := newPingTracker()
start := time.Now()
assert.False(t, pt.updating(types.NodeID(1), "1.0.0", start), "no update started")
pt.updateStarted(types.NodeID(1), "1.0.0", start)
assert.True(t, pt.updating(types.NodeID(1), "1.0.0", start.Add(time.Minute)), "still on the old version")
assert.False(t, pt.updating(types.NodeID(2), "1.0.0", start), "another node")
assert.False(t, pt.updating(types.NodeID(1), "1.2.0", start.Add(time.Minute)), "on a new version")
assert.False(t, pt.updating(types.NodeID(1), "1.0.0", start.Add(time.Minute)), "an ended update is forgotten")
pt.updateStarted(types.NodeID(1), "1.0.0", start)
assert.False(t, pt.updating(types.NodeID(1), "1.0.0", start.Add(clientUpdateWindow)), "past the window")
}