Files
gitea-dynamic-runner/internal/assignmentqueue/jetstream.go
panxiao81 38e8d59541
test / go (pull_request) Successful in 3m10s
test / shell (pull_request) Failing after 10m4s
test / python (pull_request) Failing after 10m4s
feat: 增加 Runner 生命周期结构化日志
2026-09-21 09:31:46 +00:00

239 lines
7.1 KiB
Go

// Package assignmentqueue implements the durable assignment handoff with JetStream.
package assignmentqueue
import (
"context"
"errors"
"fmt"
"strings"
"sync"
"time"
"github.com/nats-io/nats.go"
"github.com/nats-io/nats.go/jetstream"
"git.ddupan.top/panxiao81/gitea-dynamic-runner/internal/taskassignment"
)
type publishAPI interface {
PublishMsg(context.Context, *nats.Msg, ...jetstream.PublishOpt) (*jetstream.PubAck, error)
}
// Publisher implements the scheduler dispatcher with one subject per backend.
type Publisher struct {
JetStream publishAPI
SubjectBase string
}
func (p Publisher) Dispatch(ctx context.Context, assignment taskassignment.Assignment) error {
if p.JetStream == nil {
return errors.New("JetStream publisher is required")
}
body, err := taskassignment.Marshal(assignment)
if err != nil {
return err
}
base := strings.TrimSuffix(p.SubjectBase, ".")
if base == "" {
return errors.New("assignment subject base is required")
}
message := &nats.Msg{
Subject: base + "." + string(assignment.Backend),
Header: nats.Header{jetstream.MsgIDHeader: []string{assignment.ID}},
Data: body,
}
if _, err := p.JetStream.PublishMsg(ctx, message); err != nil {
return fmt.Errorf("publish assignment %s: %w", assignment.ID, err)
}
return nil
}
type Accepter interface {
Accept(context.Context, taskassignment.Assignment) (bool, error)
}
type Claims interface {
Offer(taskassignment.Assignment) (<-chan struct{}, error)
WaitClaimed(context.Context, string) error
}
type Admission interface {
Acquire(string) bool
Release(string)
}
// Message is the subset of jetstream.Msg needed by one reconciliation.
type Message interface {
Data() []byte
DoubleAck(context.Context) error
NakWithDelay(time.Duration) error
TermWithReason(string) error
}
// Processor maps one delivery to one idempotent worker reconciliation.
type Processor struct {
TrustDomain string
Accepter Accepter
Claims Claims
Admission Admission
RetryDelay time.Duration
ClaimTimeout time.Duration
OnEvent func(Event)
}
// Event describes a non-sensitive assignment handoff transition. It never
// contains task payloads, credentials, capabilities, or workload identities.
type Event struct {
Name string
AssignmentID string
Backend taskassignment.Backend
RetryDelay time.Duration
}
func (p Processor) event(name string, assignment taskassignment.Assignment, retryDelay time.Duration) {
if p.OnEvent != nil {
p.OnEvent(Event{Name: name, AssignmentID: assignment.ID, Backend: assignment.Backend, RetryDelay: retryDelay})
}
}
func (p Processor) Process(ctx context.Context, message Message) error {
if p.Accepter == nil || p.Claims == nil || p.Admission == nil {
return errors.New("assignment accepter, claim registry, and backend admission pool are required")
}
assignment, err := taskassignment.Unmarshal(message.Data(), p.TrustDomain)
if err != nil {
return errors.Join(err, message.TermWithReason("invalid assignment"))
}
p.event("received", assignment, 0)
if _, err := p.Claims.Offer(assignment); err != nil {
return errors.Join(err, message.TermWithReason("conflicting assignment"))
}
if !p.Admission.Acquire(assignment.ID) {
delay := p.RetryDelay
if delay <= 0 {
delay = 2 * time.Second
}
p.event("capacity_wait", assignment, delay)
return message.NakWithDelay(delay)
}
p.event("capacity_acquired", assignment, 0)
accepted, err := p.Accepter.Accept(ctx, assignment)
if err != nil {
p.Admission.Release(assignment.ID)
delay := p.RetryDelay
if delay <= 0 {
delay = 15 * time.Second
}
p.event("backend_retry", assignment, delay)
return errors.Join(err, message.NakWithDelay(delay))
}
if accepted {
p.event("backend_ready", assignment, 0)
timeout := p.ClaimTimeout
if timeout <= 0 {
timeout = 4 * time.Minute
}
claimContext, cancel := context.WithTimeout(ctx, timeout)
err := p.Claims.WaitClaimed(claimContext, assignment.ID)
cancel()
if err != nil {
delay := p.RetryDelay
if delay <= 0 {
delay = 2 * time.Second
}
p.event("claim_timeout", assignment, delay)
return errors.Join(err, message.NakWithDelay(delay))
}
p.event("runner_claimed", assignment, 0)
if err := message.DoubleAck(ctx); err != nil {
return fmt.Errorf("ack assignment %s: %w", assignment.ID, err)
}
p.event("acked", assignment, 0)
return nil
}
delay := p.RetryDelay
if delay <= 0 {
delay = 2 * time.Second
}
p.event("backend_pending", assignment, delay)
return message.NakWithDelay(delay)
}
type consumeAPI interface {
Consume(jetstream.MessageHandler, ...jetstream.PullConsumeOpt) (jetstream.ConsumeContext, error)
}
// ConsumerComponent runs bounded reconciliation goroutines for one durable
// backend consumer. The goroutine set is operational state, not task storage.
type ConsumerComponent struct {
Consumer consumeAPI
Processor Processor
Capacity int
OnError func(error)
}
type consumerManager interface {
CreateOrUpdateConsumer(context.Context, string, jetstream.ConsumerConfig) (jetstream.Consumer, error)
}
// OpenConsumer creates the durable backend cursor. Capacity is enforced both
// server-side and by ConsumerComponent's local semaphore.
func OpenConsumer(ctx context.Context, manager consumerManager, stream, subjectBase string, backend taskassignment.Backend, capacity int) (jetstream.Consumer, error) {
if manager == nil || stream == "" || strings.TrimSuffix(subjectBase, ".") == "" || capacity < 1 {
return nil, errors.New("JetStream manager, stream, subject base, and positive capacity are required")
}
if backend != taskassignment.BackendPod && backend != taskassignment.BackendVM {
return nil, fmt.Errorf("unsupported assignment backend %q", backend)
}
consumer, err := manager.CreateOrUpdateConsumer(ctx, stream, jetstream.ConsumerConfig{
Name: string(backend),
Durable: string(backend),
FilterSubject: strings.TrimSuffix(subjectBase, ".") + "." + string(backend),
AckPolicy: jetstream.AckExplicitPolicy,
AckWait: 5 * time.Minute,
MaxAckPending: capacity,
MaxDeliver: 1000,
})
if err != nil {
return nil, fmt.Errorf("open %s assignment consumer: %w", backend, err)
}
return consumer, nil
}
func (c ConsumerComponent) Run(ctx context.Context) error {
if c.Consumer == nil || c.Capacity < 1 {
return errors.New("JetStream consumer and positive capacity are required")
}
semaphore := make(chan struct{}, c.Capacity)
var workers sync.WaitGroup
consumeContext, err := c.Consumer.Consume(func(message jetstream.Msg) {
select {
case semaphore <- struct{}{}:
case <-ctx.Done():
return
}
workers.Add(1)
go func() {
defer workers.Done()
defer func() { <-semaphore }()
if err := c.Processor.Process(ctx, message); err != nil && !errors.Is(err, context.Canceled) && c.OnError != nil {
c.OnError(err)
}
}()
}, jetstream.PullMaxMessages(c.Capacity))
if err != nil {
return fmt.Errorf("start JetStream consumer: %w", err)
}
select {
case <-ctx.Done():
consumeContext.Stop()
<-consumeContext.Closed()
workers.Wait()
return nil
case <-consumeContext.Closed():
workers.Wait()
return errors.New("JetStream consumer stopped unexpectedly")
}
}