feat: add durable Worker execution and WSS server

This commit is contained in:
2026-09-17 08:22:24 +08:00
parent 9701b7a071
commit cd0dc7011b
31 changed files with 3007 additions and 16 deletions

View File

@@ -0,0 +1,401 @@
package httpserver
import (
"bytes"
"context"
"encoding/json"
"errors"
"net/http"
"reflect"
"strings"
"sync"
"time"
"git.yuntuo.hk/k2safe/orbit/internal/fault"
"git.yuntuo.hk/k2safe/orbit/internal/jsonvalue"
core "git.yuntuo.hk/k2safe/orbit/model"
"git.yuntuo.hk/k2safe/orbit/platform/auth"
"git.yuntuo.hk/k2safe/orbit/pulse/worker"
"github.com/coder/websocket"
)
const WorkerSubprotocol = "pulse.worker.v1"
// WorkerHub 管理节点内连接生命周期;任务归属和容量仍由 MySQL 决定。
type WorkerHub struct {
service *worker.Service
ctx context.Context
cancel context.CancelFunc
slots chan struct{}
mu sync.Mutex
closing bool
wg sync.WaitGroup
poll time.Duration
}
func NewWorkerHub(parent context.Context, service *worker.Service) *WorkerHub {
ctx, cancel := context.WithCancel(parent)
return &WorkerHub{service: service, ctx: ctx, cancel: cancel, slots: make(chan struct{}, 128), poll: time.Second}
}
func RegisterWorkers(server *http.Server, hub *WorkerHub) {
server.Handler.(*http.ServeMux).Handle("GET /v1/worker/connect", hub)
}
// Shutdown 停止派发并关闭本节点连接;不清除持久化租约,由重連核对或到期恢复处理。
func (h *WorkerHub) Shutdown(ctx context.Context) error {
h.mu.Lock()
h.closing = true
h.cancel()
h.mu.Unlock()
done := make(chan struct{})
go func() { h.wg.Wait(); close(done) }()
select {
case <-done:
return nil
case <-ctx.Done():
return ctx.Err()
}
}
type workerFrame struct {
Version string `json:"protocol_version"`
Type string `json:"type"`
MessageID core.ID `json:"message_id"`
ReplyTo core.ID `json:"reply_to,omitempty"`
SessionID core.ID `json:"session_id,omitempty"`
Body json.RawMessage `json:"body,omitempty"`
Error *fault.Error `json:"error,omitempty"`
}
type outboundFrame struct {
data []byte
done chan error
}
func (h *WorkerHub) ServeHTTP(w http.ResponseWriter, r *http.Request) {
requestID, _ := core.NewID()
w.Header().Set("X-Request-ID", string(requestID))
w.Header().Set("Cache-Control", "no-store")
if r.URL.RawQuery != "" {
writeError(w, string(requestID), fault.Invalid("Worker credentials and options must not be placed in URL"))
return
}
values := r.Header.Values("Authorization")
if len(values) != 1 {
writeError(w, string(requestID), fault.Unauthorized)
return
}
scheme, value, ok := strings.Cut(values[0], " ")
if !ok || !strings.EqualFold(scheme, "Bearer") {
writeError(w, string(requestID), fault.Unauthorized)
return
}
token, err := auth.Parse(value)
if err != nil {
writeError(w, string(requestID), err)
return
}
supported := false
for _, header := range r.Header.Values("Sec-WebSocket-Protocol") {
for _, protocol := range strings.Split(header, ",") {
supported = supported || strings.TrimSpace(protocol) == WorkerSubprotocol
}
}
if !supported {
writeJSON(w, http.StatusUpgradeRequired, map[string]any{"request_id": requestID, "error": fault.New("PROTOCOL_REQUIRED", "pulse.worker.v1 subprotocol is required")})
return
}
h.mu.Lock()
if h.closing || h.ctx.Err() != nil {
h.mu.Unlock()
writeError(w, string(requestID), fault.Unavailable)
return
}
select {
case h.slots <- struct{}{}:
h.wg.Add(1)
default:
h.mu.Unlock()
writeJSON(w, 429, map[string]any{"request_id": requestID, "error": fault.New("CAPACITY_EXCEEDED", "node connection capacity exceeded")})
return
}
h.mu.Unlock()
defer func() { <-h.slots; h.wg.Done() }()
authorize, cancel := context.WithTimeout(r.Context(), 5*time.Second)
err = h.service.Authorize(authorize, token)
cancel()
if err != nil {
writeError(w, string(requestID), err)
return
}
conn, err := websocket.Accept(w, r, &websocket.AcceptOptions{Subprotocols: []string{WorkerSubprotocol}, CompressionMode: websocket.CompressionDisabled})
if err != nil {
return
}
defer conn.CloseNow()
conn.SetReadLimit(512 << 10)
ctx, cancel := context.WithCancel(h.ctx)
defer cancel()
helloCtx, helloCancel := context.WithTimeout(ctx, 5*time.Second)
kind, raw, err := conn.Read(helloCtx)
helloCancel()
if err != nil {
return
}
var hello workerFrame
var input worker.ConnectInput
if kind != websocket.MessageText || decodeWorkerJSON(raw, &hello) != nil || hello.Version != "1" || hello.Type != "worker.hello" || hello.MessageID.Validate() != nil || hello.SessionID != "" || hello.ReplyTo != "" || hello.Error != nil || decodeWorkerJSON(hello.Body, &input) != nil {
_ = conn.Close(websocket.StatusPolicyViolation, "invalid worker hello")
return
}
call, cancelCall := context.WithTimeout(ctx, 8*time.Second)
binding, welcome, err := h.service.Connect(call, token, input)
cancelCall()
if err != nil {
_ = writeWorkerFrame(ctx, conn, workerFrame{Version: "1", Type: "error", MessageID: requestID, ReplyTo: hello.MessageID, Error: publicWorkerError(err)})
return
}
welcomeRaw, err := json.Marshal(welcome)
if err != nil {
return
}
if err := writeWorkerFrame(ctx, conn, workerFrame{Version: "1", Type: "worker.welcome", MessageID: requestID, ReplyTo: hello.MessageID, SessionID: binding.SessionID, Body: welcomeRaw}); err != nil {
return
}
// 有界读写队列与 SQL 控制循环分开,处理器绝不运行在这里。
incoming := make(chan workerFrame, 4)
outgoing := make(chan outboundFrame, 8)
var loops sync.WaitGroup
loops.Add(2)
go func() {
defer loops.Done()
defer cancel()
for {
readCtx, stop := context.WithTimeout(ctx, 60*time.Second)
kind, raw, err := conn.Read(readCtx)
stop()
if err != nil {
return
}
var frame workerFrame
if kind != websocket.MessageText || decodeWorkerJSON(raw, &frame) != nil || frame.Version != "1" || frame.MessageID.Validate() != nil || frame.SessionID != binding.SessionID || frame.ReplyTo != "" || frame.Error != nil {
_ = conn.Close(websocket.StatusPolicyViolation, "invalid worker frame")
return
}
select {
case incoming <- frame:
case <-ctx.Done():
return
}
}
}()
go func() {
defer loops.Done()
defer cancel()
for {
select {
case <-ctx.Done():
return
case frame := <-outgoing:
writeCtx, stop := context.WithTimeout(ctx, 5*time.Second)
err := conn.Write(writeCtx, websocket.MessageText, frame.data)
stop()
if frame.done != nil {
frame.done <- err
}
if err != nil {
return
}
}
}
}()
defer func() { cancel(); _ = conn.CloseNow(); loops.Wait() }()
send := func(frame workerFrame, wait bool) bool {
frame.Version = "1"
frame.SessionID = binding.SessionID
if frame.MessageID == "" {
var err error
frame.MessageID, err = core.NewID()
if err != nil {
return false
}
}
raw, err := json.Marshal(frame)
if err != nil {
return false
}
out := outboundFrame{data: raw}
if wait {
out.done = make(chan error, 1)
}
select {
case outgoing <- out:
default:
return false
}
if !wait {
return true
}
select {
case err := <-out.done:
return err == nil
case <-ctx.Done():
return false
}
}
capacity := input.Capacity
draining := welcome.Draining
eligible := false
leases := map[core.ID]time.Time{}
ticker := time.NewTicker(h.poll)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case frame := <-incoming:
call, stop := context.WithTimeout(ctx, 8*time.Second)
var body any
var commandErr error
switch frame.Type {
case "worker.capacity":
var in worker.CapacityInput
if commandErr = decodeWorkerJSON(frame.Body, &in); commandErr == nil {
var out worker.Welcome
out, commandErr = h.service.Capacity(call, token, binding, in)
body = out
if commandErr == nil {
capacity = in.Capacity
draining = out.Draining
eligible = true
}
}
case "worker.resume":
var ref worker.AttemptRef
if commandErr = decodeWorkerJSON(frame.Body, &ref); commandErr == nil {
var out worker.LeaseStatus
out, commandErr = h.service.Resume(call, token, binding, ref)
body = out
if commandErr == nil {
if out.Active {
leases[ref.AttemptID] = out.LeaseUntil
} else {
delete(leases, ref.AttemptID)
}
}
}
case "task.start", "task.renew":
var ref worker.AttemptRef
if commandErr = decodeWorkerJSON(frame.Body, &ref); commandErr == nil {
var out worker.CommandResult
if frame.Type == "task.start" {
out, commandErr = h.service.Start(call, token, binding, frame.MessageID, ref)
} else {
out, commandErr = h.service.Renew(call, token, binding, frame.MessageID, ref)
}
body = out
if commandErr == nil {
if out.Lease.Active {
leases[ref.AttemptID] = out.Lease.LeaseUntil
} else {
delete(leases, ref.AttemptID)
}
}
}
case "task.finish":
var in worker.FinishInput
if commandErr = decodeWorkerJSON(frame.Body, &in); commandErr == nil {
var out worker.CommandResult
out, commandErr = h.service.Finish(call, token, binding, frame.MessageID, in)
body = out
if commandErr == nil && !out.Lease.Active {
delete(leases, in.Ref.AttemptID)
}
}
default:
commandErr = fault.Invalid("unsupported worker message type")
}
stop()
reply := workerFrame{Type: "ack", ReplyTo: frame.MessageID}
if commandErr != nil {
reply.Type = "error"
reply.Error = publicWorkerError(commandErr)
} else {
reply.Body, commandErr = json.Marshal(body)
if commandErr != nil {
return
}
}
fatal := commandErr != nil && (fault.Code(commandErr) == "STALE_SESSION" || fault.Code(commandErr) == "UNAUTHENTICATED" || fault.Code(commandErr) == "PERMISSION_DENIED")
if !send(reply, fatal) || fatal {
return
}
case <-ticker.C:
// 本地记录仅减少空轮询;SQL 仍原子验证总容量、队列额度和有效会话。
now := time.Now()
for attempt, until := range leases {
if !until.After(now) {
delete(leases, attempt)
}
}
if !eligible || draining || capacity == 0 || len(leases) >= int(capacity) || len(outgoing) >= 4 {
continue
}
for burst := 0; burst < 4 && len(leases) < int(capacity) && len(outgoing) < 4; burst++ {
call, stop := context.WithTimeout(ctx, 8*time.Second)
assignment, err := h.service.Claim(call, token, binding)
stop()
if err != nil {
if fault.Code(err) == "STALE_SESSION" || fault.Code(err) == "UNAUTHENTICATED" || fault.Code(err) == "PERMISSION_DENIED" {
send(workerFrame{Type: "error", Error: publicWorkerError(err)}, true)
return
}
break
}
if assignment == nil {
break
}
raw, err := json.Marshal(assignment)
if err != nil {
return
}
leases[assignment.AttemptID] = assignment.LeaseUntil
if !send(workerFrame{Type: "task.assign", Body: raw}, false) {
return
}
}
}
}
}
func publicWorkerError(err error) *fault.Error {
var known *fault.Error
if errors.As(err, &known) {
return known
}
return fault.Unavailable
}
func writeWorkerFrame(ctx context.Context, conn *websocket.Conn, frame workerFrame) error {
raw, err := json.Marshal(frame)
if err != nil {
return err
}
bounded, cancel := context.WithTimeout(ctx, 5*time.Second)
defer cancel()
return conn.Write(bounded, websocket.MessageText, raw)
}
func decodeWorkerJSON(raw []byte, out any) error {
if len(raw) == 0 || len(raw) > 512<<10 {
return fault.Invalid("invalid worker message size")
}
if _, err := jsonvalue.Canonical(raw); err != nil {
return fault.Invalid("invalid worker JSON")
}
if err := exactFields(raw, reflect.TypeOf(out).Elem()); err != nil {
return fault.Invalid("unknown or invalid worker field")
}
decoder := json.NewDecoder(bytes.NewReader(raw))
decoder.DisallowUnknownFields()
if err := decoder.Decode(out); err != nil {
return fault.Invalid("invalid worker field value")
}
return nil
}