fix: clear consumed ephemeral registration state (#1256)

Ephemeral registrations are single-use, but the daemon kept `.runner` after the task, so a restarted container failed `Declare` with `unregistered runner` forever.

An ephemeral runner now removes its registration file as soon as it fetches a task, so a container killed mid-job registers afresh on restart. It also removes the file when Gitea answers `unregistered runner` at startup or while polling. A proxy's 401 does not count. Persistent runners keep their file, because Gitea sends the same reply for a token mismatch.

Closes https://gitea.com/gitea/runner/issues/818

---------

Co-authored-by: silverwind <[email protected]>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1256
Reviewed-by: silverwind <[email protected]>

internal/app/poll/poller.go

@@ -52,6 +52,8 @@ type Poller struct {
 
 	done chan struct{}
 
+	registrationRejected atomic.Bool
+
 	// unregistered is set when the server rejects the runner with an
 	// Unauthenticated response, meaning the runner is no longer registered.
 	unregistered       atomic.Bool
@@ -151,7 +153,7 @@ func (p *Poller) Poll() {
 	}
 }
 
-func (p *Poller) PollOnce() {
+func (p *Poller) PollOnce(onTaskFetched func()) {
 	defer close(p.done)
 	s := &workerState{}
 	for {
@@ -173,6 +175,7 @@ func (p *Poller) PollOnce() {
 			continue
 		}
 		s.resetBackoff()
+		onTaskFetched()
 		p.runTaskWithRecover(p.jobsCtx, task)
 		return
 	}
@@ -191,6 +194,11 @@ func (p *Poller) Unregistered() bool {
 	return p.unregistered.Load()
 }
 
+// RegistrationRejected reports whether polling stopped on an error matching client.IsRegistrationRejected.
+func (p *Poller) RegistrationRejected() bool {
+	return p.registrationRejected.Load()
+}
+
 // Ready reports whether the daemon can currently communicate with Gitea and
 // accept work, reusing the availability the poll loop last observed rather than
 // re-running the check. Transient transport failures are tolerated for grace.
@@ -368,9 +376,11 @@ func (p *Poller) fetchTask(ctx context.Context, s *workerState) (*runnerv1.Task,
 		// An Unauthenticated response means the server no longer knows this
 		// runner (e.g. it was deleted). Retrying forever is pointless, so stop
 		// polling and let the daemon exit with an error instead of spinning.
-		if isUnregistered(err) {
+		rejected := client.IsRegistrationRejected(err)
+		if rejected || connect.CodeOf(err) == connect.CodeUnauthenticated {
 			log.WithError(err).Error("server rejected the runner as unregistered, stopping poller")
 			p.unregistered.Store(true)
+			p.registrationRejected.Store(rejected)
 			p.shutdownPolling()
 			return nil, false
 		}
@@ -415,12 +425,6 @@ func (p *Poller) fetchTask(ctx context.Context, s *workerState) (*runnerv1.Task,
 	return resp.Msg.Task, true
 }
 
-func isUnregistered(err error) bool {
-	var connectErr *connect.Error
-	return errors.As(err, &connectErr) && (connectErr.Code() == connect.CodeUnauthenticated ||
-		connectErr.Code() == connect.CodeUnknown && connect.IsWireError(err) && connectErr.Message() == "rpc error: code = Unauthenticated desc = unregistered runner")
-}
-
 func (p *Poller) markHealthyPoll() {
 	p.lastHealthyPoll.Store(time.Now().UnixNano())
 	p.lastPollFailed.Store(false)

internal/app/poll/poller_test.go

@@ -106,13 +106,16 @@ func TestPoller_FetchTimeoutIsNoSignal(t *testing.T) {
 // response marks the runner as unregistered and cancels the polling context so
 // the daemon can exit instead of retrying forever.
 func TestPoller_FetchUnauthenticatedStopsPolling(t *testing.T) {
-	for name, fetchErr := range map[string]error{
-		"connect": connect_go.NewError(connect_go.CodeUnauthenticated, errors.New("unregistered runner")),
-		"gitea":   connect_go.NewWireError(connect_go.CodeUnknown, errors.New("rpc error: code = Unauthenticated desc = unregistered runner")),
+	for name, tc := range map[string]struct {
+		fetchErr error
+		rejected bool
+	}{
+		"connect": {connect_go.NewError(connect_go.CodeUnauthenticated, errors.New("unregistered runner")), false},
+		"gitea":   {connect_go.NewWireError(connect_go.CodeUnknown, errors.New("rpc error: code = Unauthenticated desc = unregistered runner")), true},
 	} {
 		t.Run(name, func(t *testing.T) {
 			client := mocks.NewClient(t)
-			client.On("FetchTask", mock.Anything, mock.Anything).Return(nil, fetchErr)
+			client.On("FetchTask", mock.Anything, mock.Anything).Return(nil, tc.fetchErr)
 
 			cfg, err := config.LoadDefault("")
 			require.NoError(t, err)
@@ -122,6 +125,7 @@ func TestPoller_FetchUnauthenticatedStopsPolling(t *testing.T) {
 			require.False(t, ok)
 
 			assert.True(t, p.Unregistered(), "runner should be marked unregistered")
+			assert.Equal(t, tc.rejected, p.RegistrationRejected())
 			assert.Equal(t, int64(0), s.consecutiveErrors, "unauthenticated must not drive error backoff")
 			select {
 			case <-p.pollingCtx.Done():
@@ -330,7 +334,7 @@ func TestPollerPollOnceCallsOnIdle(t *testing.T) {
 	poller := New(cfg, cli, runner)
 
 	var wg sync.WaitGroup
-	wg.Go(poller.PollOnce)
+	wg.Go(func() { poller.PollOnce(func() {}) })
 
 	require.Eventually(t, func() bool {
 		return runner.idleCalls.Load() > 0
@@ -342,6 +346,18 @@ func TestPollerPollOnceCallsOnIdle(t *testing.T) {
 	wg.Wait()
 }
 
+func TestPollerPollOnceCallsOnTaskFetchedBeforeRunningIt(t *testing.T) {
+	cli := mocks.NewClient(t)
+	cli.On("FetchTask", mock.Anything, mock.Anything).Return(connect_go.NewResponse(&runnerv1.FetchTaskResponse{Task: &runnerv1.Task{Id: 1}}), nil)
+	cfg, err := config.LoadDefault("")
+	require.NoError(t, err)
+	runner := &mockRunner{}
+	runsSeenByCallback := int64(-1)
+	New(cfg, cli, runner).PollOnce(func() { runsSeenByCallback = runner.running.Load() + runner.totalCompleted.Load() })
+	assert.Zero(t, runsSeenByCallback)
+	assert.Equal(t, int64(1), runner.totalCompleted.Load())
+}
+
 // TestPoller_ConcurrencyLimitedByCapacity verifies that with capacity=3 and
 // 6 available tasks, at most 3 tasks run concurrently, and FetchTask is
 // never called concurrently (single poller).