feat(poller): support scheduler to fetch task and dispatch to worker

This commit is contained in:
Bo-Yi Wu 2022-11-17 10:49:32 +08:00 committed by Jason Song
parent 5fc35faf17
commit a13ed0c52e
2 changed files with 104 additions and 71 deletions

View File

@ -59,6 +59,7 @@ func runDaemon(ctx context.Context, envFile string) func(cmd *cobra.Command, arg
poller := poller.New(
cli,
runner.Run,
cfg.Runner.Capacity,
)
g.Go(func() error {
@ -81,7 +82,7 @@ func runDaemon(ctx context.Context, envFile string) func(cmd *cobra.Command, arg
Errorln("failed to update runner")
}
return poller.Poll(ctx, cfg.Runner.Capacity)
return poller.Poll(ctx)
})
g.Go(func() error {

View File

@ -3,6 +3,7 @@ package poller
import (
"context"
"errors"
"sync"
"time"
"gitea.com/gitea/act_runner/client"
@ -14,12 +15,13 @@ import (
var ErrDataLock = errors.New("Data Lock Error")
func New(cli client.Client, dispatch func(context.Context, *runnerv1.Task) error) *Poller {
func New(cli client.Client, dispatch func(context.Context, *runnerv1.Task) error, workerNum int) *Poller {
return &Poller{
Client: cli,
Dispatch: dispatch,
routineGroup: newRoutineGroup(),
metric: &metric{},
workerNum: workerNum,
}
}
@ -28,83 +30,60 @@ type Poller struct {
Filter *client.Filter
Dispatch func(context.Context, *runnerv1.Task) error
sync.Mutex
routineGroup *routineGroup
metric *metric
ready chan struct{}
workerNum int
}
func (p *Poller) Wait() {
p.routineGroup.Wait()
}
func (p *Poller) Poll(ctx context.Context, n int) error {
for i := 0; i < n; i++ {
func(i int) {
p.routineGroup.Run(func() {
func (p *Poller) schedule() {
p.Lock()
defer p.Unlock()
if int(p.metric.BusyWorkers()) >= p.workerNum {
return
}
select {
case p.ready <- struct{}{}:
default:
}
}
func (p *Poller) Poll(ctx context.Context) error {
l := log.WithField("func", "Poll")
for {
// check worker number
p.schedule()
select {
// wait worker ready
case <-p.ready:
case <-ctx.Done():
return nil
}
LOOP:
for {
select {
case <-ctx.Done():
log.Infof("stopped the runner: %d", i+1)
return
break LOOP
default:
if ctx.Err() != nil {
log.Infof("stopping the runner: %d", i+1)
return
}
if err := p.poll(ctx, i+1); err != nil {
log.WithField("thread", i+1).
WithError(err).Error("poll error")
select {
case <-ctx.Done():
return
case <-time.After(5 * time.Second):
}
}
}
}
})
}(i)
}
p.routineGroup.Wait()
return nil
}
func (p *Poller) poll(ctx context.Context, thread int) error {
l := log.WithField("thread", thread)
l.Info("poller: request stage from remote server")
reqCtx, cancel := context.WithTimeout(ctx, 5*time.Second)
defer cancel()
// request a new build stage for execution from the central
// build server.
resp, err := p.Client.FetchTask(reqCtx, connect.NewRequest(&runnerv1.FetchTaskRequest{}))
if err == context.Canceled || err == context.DeadlineExceeded {
l.WithError(err).Trace("poller: no stage returned")
return nil
}
if err != nil && err == ErrDataLock {
l.WithError(err).Info("task accepted by another runner")
return nil
}
task, err := p.pollTask(ctx)
if task == nil || err != nil {
if err != nil {
l.WithError(err).Error("cannot accept task")
return err
l.Errorf("can't find the task: %v", err.Error())
}
// exit if a nil or empty stage is returned from the system
// and allow the runner to retry.
if resp.Msg.Task == nil || resp.Msg.Task.Id == 0 {
return nil
time.Sleep(5 * time.Second)
break
}
runCtx, cancel := context.WithTimeout(ctx, time.Hour)
defer cancel()
// update runner status
// running: idle -> active
// stopped: active -> idle
if val := p.metric.IncBusyWorker(); val == 1 {
if _, err := p.Client.UpdateRunner(
ctx,
@ -116,24 +95,77 @@ func (p *Poller) poll(ctx context.Context, thread int) error {
}
l.Info("update runner status to active")
}
defer func() {
if val := p.metric.DecBusyWorker(); val != 0 {
return
p.routineGroup.Run(func() {
if err := p.dispatchTask(ctx, task); err != nil {
l.Errorf("execute task: %v", err.Error())
}
})
break LOOP
}
}
}
}
func (p *Poller) pollTask(ctx context.Context) (*runnerv1.Task, error) {
l := log.WithField("func", "pollTask")
l.Info("poller: request stage from remote server")
reqCtx, cancel := context.WithTimeout(ctx, 5*time.Second)
defer cancel()
// request a new build stage for execution from the central
// build server.
resp, err := p.Client.FetchTask(reqCtx, connect.NewRequest(&runnerv1.FetchTaskRequest{}))
if err == context.Canceled || err == context.DeadlineExceeded {
l.WithError(err).Trace("poller: no stage returned")
return nil, nil
}
if err != nil && err == ErrDataLock {
l.WithError(err).Info("task accepted by another runner")
return nil, nil
}
if err != nil {
l.WithError(err).Error("cannot accept task")
return nil, err
}
// exit if a nil or empty stage is returned from the system
// and allow the runner to retry.
if resp.Msg.Task == nil || resp.Msg.Task.Id == 0 {
return nil, nil
}
return resp.Msg.Task, nil
}
func (p *Poller) dispatchTask(ctx context.Context, task *runnerv1.Task) error {
l := log.WithField("func", "dispatchTask")
defer func() {
val := p.metric.DecBusyWorker()
e := recover()
if e != nil {
l.Errorf("panic error: %v", e)
}
p.schedule()
if val != 0 {
return
}
if _, err := p.Client.UpdateRunner(
ctx,
connect.NewRequest(&runnerv1.UpdateRunnerRequest{
Status: runnerv1.RunnerStatus_RUNNER_STATUS_IDLE,
}),
); err != nil {
log.Errorln("update status error:", err.Error())
l.Errorln("update status error:", err.Error())
}
l.Info("update runner status to idle")
}()
}()
return p.Dispatch(runCtx, resp.Msg.Task)
runCtx, cancel := context.WithTimeout(ctx, time.Hour)
defer cancel()
return p.Dispatch(runCtx, task)
}