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 <me@silverwind.io>
Reviewed-on: https://gitea.com/gitea/runner/pulls/1256
Reviewed-by: silverwind <2021+silverwind@noreply.gitea.com>
This commit is contained in:
bircni
2026-10-01 09:23:45 +00:00
parent 0cedd98051
commit d93b7d278a
6 changed files with 148 additions and 15 deletions

View File

@@ -22,7 +22,7 @@ func testEphemeralRunner(t *testing.T) {
case <-time.After(5 * time.Second):
t.Fatal("ephemeral runner was not deleted")
}
if !poller.Unregistered() {
if !poller.RegistrationRejected() {
t.Fatal("ephemeral runner stopped without being deleted")
}
}

View File

@@ -191,6 +191,9 @@ func runDaemon(ctx context.Context, daemArgs *daemonArgs, configFile *string) fu
return err
case err != nil:
log.WithError(err).Error("fail to invoke Declare")
if client.IsRegistrationRejected(err) {
removeEphemeralRegistration(reg, cfg.Runner.File)
}
return err
default:
log.Infof("runner: %s, with version: %s, with labels: %v, declare successfully",
@@ -215,7 +218,7 @@ func runDaemon(ctx context.Context, daemArgs *daemonArgs, configFile *string) fu
done := make(chan struct{})
go func() {
defer close(done)
poller.PollOnce()
poller.PollOnce(func() { removeEphemeralRegistration(reg, cfg.Runner.File) })
}()
// shutdown when we complete a job or cancel is requested
@@ -244,6 +247,10 @@ func runDaemon(ctx context.Context, daemArgs *daemonArgs, configFile *string) fu
log.Warnf("runner: %s cancelled in progress jobs during shutdown", resp.Msg.Runner.Name)
}
if poller.RegistrationRejected() {
removeEphemeralRegistration(reg, cfg.Runner.File)
}
if poller.Unregistered() {
return errors.New("runner is no longer registered with the server; please register it again")
}
@@ -252,6 +259,17 @@ func runDaemon(ctx context.Context, daemArgs *daemonArgs, configFile *string) fu
}
}
func removeEphemeralRegistration(reg *config.Registration, file string) {
if !reg.Ephemeral {
return
}
if err := os.Remove(file); err == nil {
log.Infof("removed the spent ephemeral registration %q", file)
} else if !os.IsNotExist(err) {
log.WithError(err).Warn("failed to remove the spent ephemeral registration")
}
}
type daemonArgs struct {
Once bool
Labels string

View File

@@ -4,13 +4,24 @@
package cmd
import (
"context"
"errors"
"net/http"
"net/http/httptest"
"os"
"path/filepath"
"testing"
"time"
"gitea.com/gitea/runner/internal/pkg/config"
"gitea.com/gitea/runner/internal/pkg/report"
"connectrpc.com/connect"
runnerv1 "gitea.dev/actionslib/runner/v1"
"gitea.dev/actionslib/runner/v1/runnerv1connect"
log "github.com/sirupsen/logrus"
"github.com/stretchr/testify/require"
"google.golang.org/protobuf/types/known/structpb"
)
func TestResolveLabels(t *testing.T) {
@@ -70,3 +81,79 @@ func TestInitLoggingSetsLevelAndCaller(t *testing.T) {
// act plans a job on this logger, so a live task's secrets have to be masked out of it
require.IsType(t, report.MaskingFormatter(nil), log.StandardLogger().Formatter)
}
var errUnregisteredRunner = errors.New("rpc error: code = Unauthenticated desc = unregistered runner")
type daemonTestService struct {
runnerv1connect.UnimplementedRunnerServiceHandler
declareErr, fetchErr error
beforeTask bool
cancel context.CancelFunc
}
func (s *daemonTestService) Declare(context.Context, *connect.Request[runnerv1.DeclareRequest]) (*connect.Response[runnerv1.DeclareResponse], error) {
if s.declareErr != nil {
return nil, s.declareErr
}
return connect.NewResponse(&runnerv1.DeclareResponse{Runner: &runnerv1.Runner{Name: "test"}}), nil
}
func (s *daemonTestService) FetchTask(context.Context, *connect.Request[runnerv1.FetchTaskRequest]) (*connect.Response[runnerv1.FetchTaskResponse], error) {
if s.fetchErr != nil {
return nil, s.fetchErr
}
if s.beforeTask {
s.cancel()
return nil, context.Canceled
}
return connect.NewResponse(&runnerv1.FetchTaskResponse{Task: &runnerv1.Task{Id: 1, Context: &structpb.Struct{}, WorkflowPayload: []byte("invalid: [")}}), nil
}
func (*daemonTestService) UpdateTask(_ context.Context, req *connect.Request[runnerv1.UpdateTaskRequest]) (*connect.Response[runnerv1.UpdateTaskResponse], error) {
return connect.NewResponse(&runnerv1.UpdateTaskResponse{State: req.Msg.State}), nil
}
func (*daemonTestService) UpdateLog(_ context.Context, req *connect.Request[runnerv1.UpdateLogRequest]) (*connect.Response[runnerv1.UpdateLogResponse], error) {
return connect.NewResponse(&runnerv1.UpdateLogResponse{AckIndex: req.Msg.Index + int64(len(req.Msg.Rows))}), nil
}
func TestDaemonRemovesOnlyConsumedEphemeralRegistration(t *testing.T) {
for _, tc := range []struct {
name string
ephemeral, beforeTask bool
declareErr, fetchErr error
removed bool
}{
{"ephemeral failed task", true, false, nil, nil, true},
{"ephemeral interrupted before task", true, true, nil, nil, false},
{"persistent once", false, false, nil, nil, false},
{"ephemeral rejected on fetch", true, false, nil, errUnregisteredRunner, true},
{"ephemeral proxy 401 on fetch", true, false, nil, connect.NewError(connect.CodeUnauthenticated, errors.New("HTTP status 401 Unauthorized")), false},
{"ephemeral rejected on declare", true, false, errUnregisteredRunner, nil, true},
{"persistent rejected on declare", false, false, errUnregisteredRunner, nil, false},
} {
t.Run(tc.name, func(t *testing.T) {
ctx, cancel := context.WithTimeout(t.Context(), 5*time.Second)
defer cancel()
_, handler := runnerv1connect.NewRunnerServiceHandler(&daemonTestService{declareErr: tc.declareErr, fetchErr: tc.fetchErr, beforeTask: tc.beforeTask, cancel: cancel})
server := httptest.NewServer(http.StripPrefix("/api/actions", handler))
defer server.Close()
dir := t.TempDir()
regFile := filepath.Join(dir, "registration.json")
configFile := filepath.Join(dir, "config.yaml")
require.NoError(t, os.WriteFile(configFile, []byte("runner:\n file: "+regFile+"\n idle_cleanup_interval: 0s\ncache:\n enabled: false\n"), 0o600))
require.NoError(t, config.SaveRegistration(regFile, &config.Registration{Address: server.URL, UUID: "test", Name: "test", Token: "test", Labels: []string{"host:host"}, Ephemeral: tc.ephemeral}))
err := runDaemon(ctx, &daemonArgs{Once: true}, &configFile)(nil, nil)
if tc.declareErr == nil && tc.fetchErr == nil {
require.NoError(t, err)
} else {
require.Error(t, err)
}
if tc.removed {
require.NoFileExists(t, regFile)
} else {
require.FileExists(t, regFile)
}
})
}
}

View File

@@ -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)

View File

@@ -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).

View File

@@ -5,6 +5,7 @@ package client
import (
"context"
"errors"
"connectrpc.com/connect"
"gitea.dev/actionslib/runner/v1"
@@ -18,3 +19,10 @@ type Client interface {
UpdateLog(context.Context, *connect.Request[runnerv1.UpdateLogRequest]) (*connect.Response[runnerv1.UpdateLogResponse], error)
UpdateTask(context.Context, *connect.Request[runnerv1.UpdateTaskRequest]) (*connect.Response[runnerv1.UpdateTaskResponse], error)
}
// IsRegistrationRejected reports whether Gitea itself answered "unregistered runner", unlike a proxy's bare 401.
func IsRegistrationRejected(err error) bool {
var connectErr *connect.Error
return errors.As(err, &connectErr) && connectErr.Code() == connect.CodeUnknown && connect.IsWireError(err) &&
connectErr.Message() == "rpc error: code = Unauthenticated desc = unregistered runner"
}