wait for jobs to complete when stopping the runner

When receiving a signal (INT or TERM) wait for running jobs to
complete instead of terminating them right away.

The new shutdown_timeout configuration setting can be used to force
the termination after a grace delay. If not specified or zero it will
shutdown immediately, for backward compatibility. It will be the case
with existing configuration files or when a configuration file is not
specified.

The config.yml created with the generate-config subcommand will
however default shutdown_timeout to 3h (same as timeout) because it is
likely what a new admin would expect: shutting down waits for jobs to
complete and not abort them.

RELEASE-NOTES.md

@@ -1,7 +1,8 @@
 # Release Notes
 
-## 3.4.2
+## 3.5.0
 
+* [Allow graceful shutdowns](https://code.forgejo.org/forgejo/runner/pulls/202): when receiving a signal (INT or TERM) wait for running jobs to complete (up to shutdown_timeout).
 * [Fix label declaration](https://code.forgejo.org/forgejo/runner/pulls/176): Runner in daemon mode now takes labels found in config.yml into account when declaration was successful.
 * [Fix the docker compose example](https://code.forgejo.org/forgejo/runner/pulls/175) to workaround the race on labels.
 * [Fix the kubernetes dind example](https://code.forgejo.org/forgejo/runner/pulls/169).

internal/app/cmd/daemon.go

@@ -120,8 +120,18 @@ func runDaemon(ctx context.Context, configFile *string) func(cmd *cobra.Command,
 
 		poller := poll.New(cfg, cli, runner)
 
-		poller.Poll(ctx)
+		go poller.Poll()
 
+		<-ctx.Done()
+		log.Infof("runner: %s shutdown initiated, waiting [runner].shutdown_timeout=%s for running jobs to complete before shutting down", resp.Msg.Runner.Name, cfg.Runner.ShutdownTimeout)
+
+		ctx, cancel := context.WithTimeout(context.Background(), cfg.Runner.ShutdownTimeout)
+		defer cancel()
+
+		err = poller.Shutdown(ctx)
+		if err != nil {
+			log.Warnf("runner: %s cancelled in progress jobs during shutdown", resp.Msg.Runner.Name)
+		}
 		return nil
 	}
 }

internal/app/poll/poller.go

@@ -20,49 +20,100 @@ import (
 	"gitea.com/gitea/act_runner/internal/pkg/config"
 )
 
-type Poller struct {
+const PollerID = "PollerID"
+
+type Poller interface {
+	Poll()
+	Shutdown(ctx context.Context) error
+}
+
+type poller struct {
 	client       client.Client
-	runner       *run.Runner
+	runner       run.RunnerInterface
 	cfg          *config.Config
 	tasksVersion atomic.Int64 // tasksVersion used to store the version of the last task fetched from the Gitea.
+
+	pollingCtx      context.Context
+	shutdownPolling context.CancelFunc
+
+	jobsCtx      context.Context
+	shutdownJobs context.CancelFunc
+
+	done chan any
 }
 
-func New(cfg *config.Config, client client.Client, runner *run.Runner) *Poller {
+func New(cfg *config.Config, client client.Client, runner run.RunnerInterface) Poller {
-	return &Poller{
+	return (&poller{}).init(cfg, client, runner)
-		client: client,
-		runner: runner,
-		cfg:    cfg,
-	}
 }
 
-func (p *Poller) Poll(ctx context.Context) {
+func (p *poller) init(cfg *config.Config, client client.Client, runner run.RunnerInterface) Poller {
+	pollingCtx, shutdownPolling := context.WithCancel(context.Background())
+
+	jobsCtx, shutdownJobs := context.WithCancel(context.Background())
+
+	done := make(chan any)
+
+	p.client = client
+	p.runner = runner
+	p.cfg = cfg
+
+	p.pollingCtx = pollingCtx
+	p.shutdownPolling = shutdownPolling
+
+	p.jobsCtx = jobsCtx
+	p.shutdownJobs = shutdownJobs
+	p.done = done
+
+	return p
+}
+
+func (p *poller) Poll() {
 	limiter := rate.NewLimiter(rate.Every(p.cfg.Runner.FetchInterval), 1)
 	wg := &sync.WaitGroup{}
 	for i := 0; i < p.cfg.Runner.Capacity; i++ {
 		wg.Add(1)
-		go p.poll(ctx, wg, limiter)
+		go p.poll(i, wg, limiter)
 	}
 	wg.Wait()
+
+	// signal the poller is finished
+	close(p.done)
 }
 
-func (p *Poller) poll(ctx context.Context, wg *sync.WaitGroup, limiter *rate.Limiter) {
+func (p *poller) Shutdown(ctx context.Context) error {
+	p.shutdownPolling()
+
+	select {
+	case <-p.done:
+		log.Trace("all jobs are complete")
+		return nil
+
+	case <-ctx.Done():
+		log.Trace("forcing the jobs to shutdown")
+		p.shutdownJobs()
+		<-p.done
+		log.Trace("all jobs have been shutdown")
+		return ctx.Err()
+	}
+}
+
+func (p *poller) poll(id int, wg *sync.WaitGroup, limiter *rate.Limiter) {
+	log.Infof("[poller %d] launched", id)
 	defer wg.Done()
 	for {
-		if err := limiter.Wait(ctx); err != nil {
+		if err := limiter.Wait(p.pollingCtx); err != nil {
-			if ctx.Err() != nil {
+			log.Infof("[poller %d] shutdown", id)
-				log.WithError(err).Debug("limiter wait failed")
-			}
 			return
 		}
-		task, ok := p.fetchTask(ctx)
+		task, ok := p.fetchTask(p.pollingCtx)
 		if !ok {
 			continue
 		}
-		p.runTaskWithRecover(ctx, task)
+		p.runTaskWithRecover(p.jobsCtx, task)
 	}
 }
 
-func (p *Poller) runTaskWithRecover(ctx context.Context, task *runnerv1.Task) {
+func (p *poller) runTaskWithRecover(ctx context.Context, task *runnerv1.Task) {
 	defer func() {
 		if r := recover(); r != nil {
 			err := fmt.Errorf("panic: %v", r)
@@ -75,7 +126,7 @@ func (p *Poller) runTaskWithRecover(ctx context.Context, task *runnerv1.Task) {
 	}
 }
 
-func (p *Poller) fetchTask(ctx context.Context) (*runnerv1.Task, bool) {
+func (p *poller) fetchTask(ctx context.Context) (*runnerv1.Task, bool) {
 	reqCtx, cancel := context.WithTimeout(ctx, p.cfg.Runner.FetchTimeout)
 	defer cancel()
 
@@ -85,10 +136,15 @@ func (p *Poller) fetchTask(ctx context.Context) (*runnerv1.Task, bool) {
 		TasksVersion: v,
 	}))
 	if errors.Is(err, context.DeadlineExceeded) {
+		log.Trace("deadline exceeded")
 		err = nil
 	}
 	if err != nil {
+		if errors.Is(err, context.Canceled) {
+			log.WithError(err).Debugf("shutdown, fetch task canceled")
+		} else {
 			log.WithError(err).Error("failed to fetch task")
+		}
 		return nil, false
 	}
 

internal/app/poll/poller_test.go

@@ -0,0 +1,263 @@
+// Copyright The Forgejo Authors.
+// SPDX-License-Identifier: MIT
+
+package poll
+
+import (
+	"context"
+	"fmt"
+	"testing"
+	"time"
+
+	"github.com/bufbuild/connect-go"
+
+	"code.gitea.io/actions-proto-go/ping/v1/pingv1connect"
+	runnerv1 "code.gitea.io/actions-proto-go/runner/v1"
+	"code.gitea.io/actions-proto-go/runner/v1/runnerv1connect"
+	"gitea.com/gitea/act_runner/internal/pkg/config"
+
+	log "github.com/sirupsen/logrus"
+	"github.com/stretchr/testify/assert"
+)
+
+type mockPoller struct {
+	poller
+}
+
+func (o *mockPoller) Poll() {
+	o.poller.Poll()
+}
+
+type mockClient struct {
+	pingv1connect.PingServiceClient
+	runnerv1connect.RunnerServiceClient
+
+	sleep  time.Duration
+	cancel bool
+	err    error
+	noTask bool
+}
+
+func (o mockClient) Address() string {
+	return ""
+}
+
+func (o mockClient) Insecure() bool {
+	return true
+}
+
+func (o *mockClient) FetchTask(ctx context.Context, req *connect.Request[runnerv1.FetchTaskRequest]) (*connect.Response[runnerv1.FetchTaskResponse], error) {
+	if o.sleep > 0 {
+		select {
+		case <-ctx.Done():
+			log.Trace("fetch task done")
+			return nil, context.DeadlineExceeded
+		case <-time.After(o.sleep):
+			log.Trace("slept")
+			return nil, fmt.Errorf("unexpected")
+		}
+	}
+	if o.cancel {
+		return nil, context.Canceled
+	}
+	if o.err != nil {
+		return nil, o.err
+	}
+	task := &runnerv1.Task{}
+	if o.noTask {
+		task = nil
+		o.noTask = false
+	}
+
+	return connect.NewResponse(&runnerv1.FetchTaskResponse{
+		Task:         task,
+		TasksVersion: int64(1),
+	}), nil
+}
+
+type mockRunner struct {
+	cfg    *config.Runner
+	log    chan string
+	panics bool
+	err    error
+}
+
+func (o *mockRunner) Run(ctx context.Context, task *runnerv1.Task) error {
+	o.log <- "runner starts"
+	if o.panics {
+		log.Trace("panics")
+		o.log <- "runner panics"
+		o.panics = false
+		panic("whatever")
+	}
+	if o.err != nil {
+		log.Trace("error")
+		o.log <- "runner error"
+		err := o.err
+		o.err = nil
+		return err
+	}
+	for {
+		select {
+		case <-ctx.Done():
+			log.Trace("shutdown")
+			o.log <- "runner shutdown"
+			return nil
+		case <-time.After(o.cfg.Timeout):
+			log.Trace("after")
+			o.log <- "runner timeout"
+			return nil
+		}
+	}
+}
+
+func setTrace(t *testing.T) {
+	t.Helper()
+	log.SetReportCaller(true)
+	log.SetLevel(log.TraceLevel)
+}
+
+func TestPoller_New(t *testing.T) {
+	p := New(&config.Config{}, &mockClient{}, &mockRunner{})
+	assert.NotNil(t, p)
+}
+
+func TestPoller_Runner(t *testing.T) {
+	setTrace(t)
+	for _, testCase := range []struct {
+		name           string
+		timeout        time.Duration
+		noTask         bool
+		panics         bool
+		err            error
+		expected       string
+		contextTimeout time.Duration
+	}{
+		{
+			name:     "Simple",
+			timeout:  10 * time.Second,
+			expected: "runner shutdown",
+		},
+		{
+			name:     "Panics",
+			timeout:  10 * time.Second,
+			panics:   true,
+			expected: "runner panics",
+		},
+		{
+			name:     "Error",
+			timeout:  10 * time.Second,
+			err:      fmt.Errorf("ERROR"),
+			expected: "runner error",
+		},
+		{
+			name:     "PollTaskError",
+			timeout:  10 * time.Second,
+			noTask:   true,
+			expected: "runner shutdown",
+		},
+		{
+			name:           "ShutdownTimeout",
+			timeout:        1 * time.Second,
+			contextTimeout: 1 * time.Minute,
+			expected:       "runner timeout",
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			runnerLog := make(chan string, 3)
+			configRunner := config.Runner{
+				FetchInterval: 1,
+				Capacity:      1,
+				Timeout:       testCase.timeout,
+			}
+			p := &mockPoller{}
+			p.init(
+				&config.Config{
+					Runner: configRunner,
+				},
+				&mockClient{
+					noTask: testCase.noTask,
+				},
+				&mockRunner{
+					cfg:    &configRunner,
+					log:    runnerLog,
+					panics: testCase.panics,
+					err:    testCase.err,
+				})
+			go p.Poll()
+			assert.Equal(t, "runner starts", <-runnerLog)
+			var ctx context.Context
+			var cancel context.CancelFunc
+			if testCase.contextTimeout > 0 {
+				ctx, cancel = context.WithTimeout(context.Background(), testCase.contextTimeout)
+				defer cancel()
+			} else {
+				ctx, cancel = context.WithCancel(context.Background())
+				cancel()
+			}
+			p.Shutdown(ctx)
+			<-p.done
+			assert.Equal(t, testCase.expected, <-runnerLog)
+		})
+	}
+}
+
+func TestPoller_Fetch(t *testing.T) {
+	setTrace(t)
+	for _, testCase := range []struct {
+		name    string
+		noTask  bool
+		sleep   time.Duration
+		err     error
+		cancel  bool
+		success bool
+	}{
+		{
+			name:    "Success",
+			success: true,
+		},
+		{
+			name:  "Timeout",
+			sleep: 100 * time.Millisecond,
+		},
+		{
+			name:   "Canceled",
+			cancel: true,
+		},
+		{
+			name:   "NoTask",
+			noTask: true,
+		},
+		{
+			name: "Error",
+			err:  fmt.Errorf("random error"),
+		},
+	} {
+		t.Run(testCase.name, func(t *testing.T) {
+			configRunner := config.Runner{
+				FetchTimeout: 1 * time.Millisecond,
+			}
+			p := &mockPoller{}
+			p.init(
+				&config.Config{
+					Runner: configRunner,
+				},
+				&mockClient{
+					sleep:  testCase.sleep,
+					cancel: testCase.cancel,
+					noTask: testCase.noTask,
+					err:    testCase.err,
+				},
+				&mockRunner{},
+			)
+			task, ok := p.fetchTask(context.Background())
+			if testCase.success {
+				assert.True(t, ok)
+				assert.NotNil(t, task)
+			} else {
+				assert.False(t, ok)
+				assert.Nil(t, task)
+			}
+		})
+	}
+}

internal/app/run/runner.go

@@ -41,6 +41,10 @@ type Runner struct {
 	runningTasks sync.Map
 }
 
+type RunnerInterface interface {
+	Run(ctx context.Context, task *runnerv1.Task) error
+}
+
 func NewRunner(cfg *config.Config, reg *config.Registration, cli client.Client) *Runner {
 	ls := labels.Labels{}
 	for _, v := range reg.Labels {

internal/pkg/config/config.example.yaml

@@ -23,7 +23,13 @@ runner:
   # Please note that the Forgejo instance also has a timeout (3h by default) for the job.
   # So the job could be stopped by the Forgejo instance if it's timeout is shorter than this.
   timeout: 3h
-  # Whether skip verifying the TLS certificate of the Forgejo instance.
+  # The timeout for the runner to wait for running jobs to finish when
+  # shutting down because a TERM or INT signal has been received.  Any
+  # running jobs that haven't finished after this timeout will be
+  # cancelled.
+  # If unset or zero the jobs will be cancelled immediately.
+  shutdown_timeout: 3h
+  # Whether skip verifying the TLS certificate of the instance.
   insecure: false
   # The timeout for fetching the job from the Forgejo instance.
   fetch_timeout: 5s

internal/pkg/config/config.go

@@ -26,10 +26,11 @@ type Runner struct {
 	Envs            map[string]string `yaml:"envs"`             // Envs stores environment variables for the runner.
 	EnvFile         string            `yaml:"env_file"`         // EnvFile specifies the path to the file containing environment variables for the runner.
 	Timeout         time.Duration     `yaml:"timeout"`          // Timeout specifies the duration for runner timeout.
+	ShutdownTimeout time.Duration     `yaml:"shutdown_timeout"` // ShutdownTimeout specifies the duration to wait for running jobs to complete during a shutdown of the runner.
 	Insecure        bool              `yaml:"insecure"`         // Insecure indicates whether the runner operates in an insecure mode.
 	FetchTimeout    time.Duration     `yaml:"fetch_timeout"`    // FetchTimeout specifies the timeout duration for fetching resources.
 	FetchInterval   time.Duration     `yaml:"fetch_interval"`   // FetchInterval specifies the interval duration for fetching resources.
-	Labels        []string          `yaml:"labels"`         // Labels specifies the labels of the runner. Labels are declared on each startup
+	Labels          []string          `yaml:"labels"`           // Labels specify the labels of the runner. Labels are declared on each startup
 }
 
 // Cache represents the configuration for caching.