feat: 为执行后端增加独立容量池

internal/taskscheduler/poller.go

@@ -4,8 +4,11 @@ import (
 	"context"
 	"errors"
 	"fmt"
+	"sync/atomic"
 	"time"
 
+	"golang.org/x/sync/errgroup"
+
 	runnerv1 "gitea.dev/actionslib/runner/v1"
 )
 
@@ -19,10 +22,12 @@ type PollerConfig struct {
 	Labels       []string
 	EmptyBackoff time.Duration
 	ErrorBackoff time.Duration
+	Capacity     int
 }
 
-// Poller is the scheduler component. Once Gitea assigns a task, it never
-// fetches another one until the current assignment is durably dispatched.
+// Poller is the scheduler component. Each fetcher keeps its assigned task
+// until that assignment is durably dispatched; all fetchers share one runner
+// declaration and a monotonic tasks version.
 type Poller struct {
 	Client    PollClient
 	Scheduler *Scheduler
@@ -49,9 +54,21 @@ func (p Poller) Run(ctx context.Context) error {
 	if errorBackoff <= 0 {
 		errorBackoff = 5 * time.Second
 	}
-	var tasksVersion int64
+	capacity := p.Config.Capacity
+	if capacity < 1 {
+		capacity = 1
+	}
+	var tasksVersion atomic.Int64
+	group, groupContext := errgroup.WithContext(ctx)
+	for range capacity {
+		group.Go(func() error { return p.runFetcher(groupContext, &tasksVersion, emptyBackoff, errorBackoff) })
+	}
+	return group.Wait()
+}
+
+func (p Poller) runFetcher(ctx context.Context, tasksVersion *atomic.Int64, emptyBackoff, errorBackoff time.Duration) error {
 	for {
-		response, err := p.Client.FetchTask(ctx, tasksVersion)
+		response, err := p.Client.FetchTask(ctx, tasksVersion.Load())
 		if err != nil {
 			if ctx.Err() != nil {
 				return nil
@@ -69,7 +86,7 @@ func (p Poller) Run(ctx context.Context) error {
 			}
 			continue
 		}
-		tasksVersion = response.GetTasksVersion()
+		storeMaximum(tasksVersion, response.GetTasksVersion())
 		task := response.GetTask()
 		if task == nil {
 			if !wait(ctx, emptyBackoff) {
@@ -93,6 +110,14 @@ func (p Poller) Run(ctx context.Context) error {
 	}
 }
 
+func storeMaximum(value *atomic.Int64, candidate int64) {
+	for current := value.Load(); candidate > current; current = value.Load() {
+		if value.CompareAndSwap(current, candidate) {
+			return
+		}
+	}
+}
+
 func (p Poller) report(err error) {
 	if p.OnError != nil {
 		p.OnError(err)

internal/taskscheduler/poller_test.go

@@ -115,3 +115,52 @@ func TestPollerRetriesAssignedTaskBeforeFetchingAnother(t *testing.T) {
 		t.Fatalf("dispatches=%d fetches-before-dispatch=%d declares=%d", dispatcher.calls, dispatcher.fetchesAtSuccess, client.declared)
 	}
 }
+
+type blockingPollClient struct {
+	mu       sync.Mutex
+	declared int
+	started  chan struct{}
+}
+
+func (c *blockingPollClient) Declare(context.Context, string, []string) error {
+	c.mu.Lock()
+	c.declared++
+	c.mu.Unlock()
+	return nil
+}
+
+func (c *blockingPollClient) FetchTask(ctx context.Context, _ int64) (*runnerv1.FetchTaskResponse, error) {
+	c.started <- struct{}{}
+	<-ctx.Done()
+	return nil, ctx.Err()
+}
+
+func TestPollerStartsConfiguredNumberOfFetchersAfterOneDeclare(t *testing.T) {
+	client := &blockingPollClient{started: make(chan struct{}, 3)}
+	poller := Poller{
+		Client:    client,
+		Scheduler: &Scheduler{TrustDomain: "ddupan.top", Dispatcher: &retryDispatcher{done: make(chan struct{})}},
+		Config: PollerConfig{
+			Version: "dev", Labels: []string{"self-hosted:host", "pod:host", "vm:host"}, Capacity: 3,
+		},
+	}
+	ctx, cancel := context.WithCancel(context.Background())
+	finished := make(chan error, 1)
+	go func() { finished <- poller.Run(ctx) }()
+	for range 3 {
+		select {
+		case <-client.started:
+		case <-time.After(time.Second):
+			t.Fatal("configured fetchers did not start")
+		}
+	}
+	cancel()
+	if err := <-finished; err != nil {
+		t.Fatal(err)
+	}
+	client.mu.Lock()
+	defer client.mu.Unlock()
+	if client.declared != 1 {
+		t.Fatalf("declares = %d", client.declared)
+	}
+}