285 lines
12 KiB
Go
285 lines
12 KiB
Go
// workflow-recovery 注入本地渠道提交后断线，使用明确证据恢复，保留UNKNOWN尝试和暂停门控。
|
||
package main
|
||
|
||
import (
|
||
"bytes"
|
||
"context"
|
||
"encoding/json"
|
||
"errors"
|
||
"fmt"
|
||
"io"
|
||
"net/http"
|
||
"net/http/httptest"
|
||
"os"
|
||
"sync"
|
||
"sync/atomic"
|
||
"time"
|
||
|
||
"git.yuntuo.hk/k2safe/orbit/examples/internal/flowdemo"
|
||
orbit "git.yuntuo.hk/k2safe/orbit/sdk/go"
|
||
)
|
||
|
||
const exactSequence uint64 = 9007199254740993
|
||
|
||
type request struct {
|
||
OperationID string `json:"operation_id"` // OperationID 原外部操作身份，核查时保持不变。
|
||
Sequence uint64 `json:"sequence"` // Sequence 精确业务序号，不使用float64解析。
|
||
}
|
||
type receipt struct {
|
||
OperationID string `json:"operation_id"` // OperationID 已完成操作的稳定身份。
|
||
Sequence uint64 `json:"sequence"` // Sequence 渠道已保存的精确序号。
|
||
State string `json:"state"` // State 本夹具唯一可核实的成功状态CONFIRMED。
|
||
}
|
||
|
||
func main() {
|
||
if err := run(); err != nil {
|
||
fmt.Fprintln(os.Stderr, err)
|
||
os.Exit(1)
|
||
}
|
||
}
|
||
func run() error {
|
||
api, queue, operation, err := flowdemo.Client()
|
||
if err != nil {
|
||
return err
|
||
}
|
||
ctx, cancel := context.WithTimeout(context.Background(), 45*time.Second)
|
||
defer cancel()
|
||
var mu sync.Mutex
|
||
var committed json.RawMessage
|
||
var submissions atomic.Int32
|
||
var dropped atomic.Bool
|
||
// 本地夹具代表唯一受控外部系统：先保存结果，再关闭HTTP连接而不发送回执。
|
||
// 真实人工核查需要持久业务证据及执行已结束的证明，不能只凭GET成功推断所有执行已停止。
|
||
remote := httptest.NewServer(http.HandlerFunc(func(w http.ResponseWriter, r *http.Request) {
|
||
if r.URL.Path != "/objects/"+operation {
|
||
http.NotFound(w, r)
|
||
return
|
||
}
|
||
if r.Method == "GET" {
|
||
mu.Lock()
|
||
raw := append([]byte(nil), committed...)
|
||
mu.Unlock()
|
||
if len(raw) == 0 {
|
||
http.NotFound(w, r)
|
||
return
|
||
}
|
||
w.Header().Set("Content-Type", "application/json")
|
||
_, _ = w.Write(raw)
|
||
return
|
||
}
|
||
if r.Method != "POST" {
|
||
w.WriteHeader(http.StatusMethodNotAllowed)
|
||
return
|
||
}
|
||
raw, err := io.ReadAll(io.LimitReader(r.Body, 4097))
|
||
var in request
|
||
if err != nil || len(raw) > 4096 || json.Unmarshal(raw, &in) != nil || in.OperationID != operation || in.Sequence != exactSequence {
|
||
http.Error(w, "invalid command", 400)
|
||
return
|
||
}
|
||
if submissions.Add(1) != 1 {
|
||
http.Error(w, "unexpected resubmission", 409)
|
||
return
|
||
}
|
||
evidence, _ := json.Marshal(receipt{operation, exactSequence, "CONFIRMED"})
|
||
mu.Lock()
|
||
committed = evidence
|
||
mu.Unlock()
|
||
conn, _, err := w.(http.Hijacker).Hijack()
|
||
if err == nil {
|
||
dropped.Store(true)
|
||
_ = conn.Close()
|
||
}
|
||
}))
|
||
defer remote.Close()
|
||
httpClient := remote.Client()
|
||
httpClient.Timeout = 3 * time.Second
|
||
requestType := orbit.WorkflowType{Kind: "OBJECT", Fields: map[string]orbit.WorkflowType{"operation_id": {Kind: "STRING"}, "sequence": {Kind: "NUMBER"}}}
|
||
receiptType := orbit.WorkflowType{Kind: "OBJECT", Fields: map[string]orbit.WorkflowType{"operation_id": {Kind: "STRING"}, "sequence": {Kind: "NUMBER"}, "state": {Kind: "STRING", Enum: []string{"CONFIRMED"}}}}
|
||
publishHandler, recordHandler := "object.publish."+operation, "object.record."+operation
|
||
publish := orbit.WorkflowTask{QueueID: queue, Handler: publishHandler, HandlerVersion: "v1", PayloadVersion: "v1", InputType: requestType, Input: orbit.WorkflowRef("input"), ResultType: receiptType, TimeoutMS: 30000, Retry: orbit.RetryPolicy{MaxAttempts: 1, AttemptTimeoutMS: 5000}, RecoverySafety: orbit.RecoveryReconcileOnly}
|
||
record := publish
|
||
record.Handler = recordHandler
|
||
record.InputType = receiptType
|
||
record.Input = orbit.WorkflowRef("publish")
|
||
record.RecoverySafety = orbit.RecoveryIdempotent
|
||
spec := orbit.WorkflowSpecification{SchemaVersion: 1, InputType: requestType, TimeoutMS: 60000, FailurePolicy: "REVIEW", Body: orbit.WorkflowBlock{Steps: []orbit.WorkflowStep{{Key: "publish", Kind: "TASK", Task: &publish}, {Key: "record", Kind: "TASK", Task: &record}}, OutputType: receiptType, Output: orbit.WorkflowRef("record")}}
|
||
input, _ := json.Marshal(request{operation, exactSequence})
|
||
v, err := flowdemo.PublishStart(ctx, api, operation, "recovery", spec, input)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
runner, err := flowdemo.StartWorker(ctx, api, queue,
|
||
orbit.WorkerHandler{Capability: orbit.WorkerCapability{Handler: publishHandler, HandlerVersion: "v1", PayloadVersion: "v1"}, Handle: func(ctx context.Context, e orbit.WorkerExecution) (json.RawMessage, error) {
|
||
req, err := http.NewRequestWithContext(ctx, "POST", remote.URL+"/objects/"+operation, bytes.NewReader(e.Task.Payload))
|
||
if err != nil {
|
||
return nil, err
|
||
}
|
||
resp, err := httpClient.Do(req)
|
||
if err != nil {
|
||
return nil, err
|
||
} // 回执丢失只能报UNKNOWN，不能声明明确失败以触发重发。
|
||
resp.Body.Close()
|
||
return nil, errors.New("fixture expected a lost response; inspect remote operation")
|
||
}},
|
||
orbit.WorkerHandler{Capability: orbit.WorkerCapability{Handler: recordHandler, HandlerVersion: "v1", PayloadVersion: "v1"}, Handle: func(ctx context.Context, e orbit.WorkerExecution) (json.RawMessage, error) {
|
||
if err := ctx.Err(); err != nil {
|
||
return nil, err
|
||
}
|
||
if err := verifyReceipt(e.Task.Payload, operation); err != nil {
|
||
return nil, &orbit.WorkerFailure{Code: "INVALID_RECEIPT"}
|
||
}
|
||
return e.Task.Payload, nil
|
||
}})
|
||
if err != nil {
|
||
return err
|
||
}
|
||
defer runner.Close()
|
||
step, err := runner.Step(ctx, api, v.ID, "publish", func(d orbit.WorkflowStepDetail) bool { return d.Step.State == "NEEDS_REVIEW" })
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if step.Step.TaskID == nil || step.Step.ActivationID == nil {
|
||
return errors.New("review source missing")
|
||
}
|
||
taskID, activation := *step.Step.TaskID, *step.Step.ActivationID
|
||
task, err := api.GetTask(ctx, taskID)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
attempts, err := api.ListTaskAttempts(ctx, taskID, orbit.HistoryPageRequest{Limit: 2})
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if task.State != orbit.TaskNeedsReview || len(attempts.Items) != 1 || attempts.Items[0].State != "UNKNOWN" || attempts.Items[0].EndedAt == nil || submissions.Load() != 1 || !dropped.Load() {
|
||
return errors.New("lost response did not retain one UNKNOWN execution")
|
||
}
|
||
if err := runner.Wait(ctx, func() (bool, error) {
|
||
s := runner.Worker.Stats()
|
||
return s.Running == 0 && s.ReportsConfirmed == 1, nil
|
||
}); err != nil {
|
||
return err
|
||
}
|
||
current, err := api.GetWorkflow(ctx, v.ID)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if _, err = api.ControlWorkflow(ctx, v.ID, operation+".pause", orbit.WorkflowControlRequest{Operation: "PAUSE", ExpectedControlRevision: current.ControlRevision, Reason: "hold successor while operator verifies evidence"}); err != nil {
|
||
return err
|
||
}
|
||
|
||
// 这里只在受控夹具证明“唯一提交已经结束、操作身份和结果一致”后模拟管理员决定。
|
||
req, err := http.NewRequestWithContext(ctx, "GET", remote.URL+"/objects/"+operation, nil)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
resp, err := httpClient.Do(req)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
evidence, readErr := io.ReadAll(io.LimitReader(resp.Body, 4097))
|
||
resp.Body.Close()
|
||
if readErr != nil {
|
||
return readErr
|
||
}
|
||
if resp.StatusCode != 200 || len(evidence) > 4096 {
|
||
return errors.New("verified external evidence unavailable")
|
||
}
|
||
if err := verifyReceipt(evidence, operation); err != nil {
|
||
return err
|
||
}
|
||
step, err = api.GetWorkflowStep(ctx, v.ID, "publish")
|
||
if err != nil {
|
||
return err
|
||
}
|
||
evidenceRef := "evidence:local-fixture/" + operation
|
||
decision := orbit.WorkflowRecoveryDecision{ExpectedRevision: step.Revision, Reason: "verified fixture commit, original operation and settled execution", EvidenceRef: evidenceRef}
|
||
command := orbit.ReviewWorkflowTaskRequest{WorkflowRecoveryDecision: decision, TaskID: taskID, ExpectedTaskRevision: task.Revision, ExecutionSettled: true, Outcome: orbit.TaskSucceeded, Result: evidence}
|
||
reviewed, err := api.ReviewWorkflowTask(ctx, v.ID, operation+".review", command)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
again, err := api.ReviewWorkflowTask(ctx, v.ID, operation+".review", command)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if !again.Deduplicated || again.AuditID != reviewed.AuditID || reviewed.Task == nil || reviewed.TaskAuditID == nil {
|
||
return errors.New("review identity or dual audit missing")
|
||
}
|
||
step, err = runner.Step(ctx, api, v.ID, "publish", func(d orbit.WorkflowStepDetail) bool { return d.Revision > reviewed.Instance.Revision })
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if step.Step.ActivationID == nil || *step.Step.ActivationID != activation {
|
||
return errors.New("review changed activation")
|
||
}
|
||
decision.ExpectedRevision = step.Revision
|
||
resolved, err := api.ResolveWorkflowStep(ctx, v.ID, operation+".resolve", orbit.ResolveWorkflowStepRequest{WorkflowStepDecision: orbit.WorkflowStepDecision{WorkflowRecoveryDecision: decision, StepKey: "publish", ActivationID: activation, ExpectedStepRevision: step.Step.Revision}, ExecutionSettled: true, Outcome: "SUCCEEDED", Result: evidence})
|
||
if err != nil {
|
||
return err
|
||
}
|
||
current, err = api.GetWorkflow(ctx, v.ID)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
decision.ExpectedRevision = current.Revision
|
||
continued, err := api.ContinueWorkflow(ctx, v.ID, operation+".continue", orbit.ContinueWorkflowRequest{WorkflowRecoveryDecision: decision, ReviewResolved: true})
|
||
if err != nil {
|
||
return err
|
||
}
|
||
next, err := api.GetWorkflowStep(ctx, v.ID, "record")
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if !continued.Instance.Paused || next.Step.ActivationID != nil || next.Step.TaskID != nil || next.Step.State != "BLOCKED" {
|
||
return errors.New("manual recovery bypassed pause gate")
|
||
}
|
||
for action, id := range map[string]string{"workflow.task.review": reviewed.AuditID, "workflow.step.resolve": resolved.AuditID, "workflow.review.continue": continued.AuditID} {
|
||
audit, err := api.GetWorkflowOperation(ctx, v.ID, id)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if audit.Action != action || audit.EvidenceRef != evidenceRef || audit.Reason != decision.Reason || audit.ActorID == "" {
|
||
return errors.New("management evidence or actor missing")
|
||
}
|
||
}
|
||
current, err = api.GetWorkflow(ctx, v.ID)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if _, err = api.ControlWorkflow(ctx, v.ID, operation+".resume", orbit.WorkflowControlRequest{Operation: "RESUME", ExpectedControlRevision: current.ControlRevision, Reason: "verified review complete; explicitly release pause"}); err != nil {
|
||
return err
|
||
}
|
||
final, err := runner.Workflow(ctx, api, v.ID, "SUCCEEDED")
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if err := verifyReceipt(final.Output, operation); err != nil {
|
||
return err
|
||
}
|
||
old, err := api.GetTaskAttempt(ctx, taskID, attempts.Items[0].ID)
|
||
if err != nil {
|
||
return err
|
||
}
|
||
if old.State != "UNKNOWN" || old.EndedAt == nil || !old.EndedAt.Equal(*attempts.Items[0].EndedAt) || submissions.Load() != 1 || !final.DeadlineAt.Equal(v.DeadlineAt) {
|
||
return errors.New("recovery rewrote history, repeated submission or extended deadline")
|
||
}
|
||
if err := runner.Close(); err != nil {
|
||
return err
|
||
}
|
||
stats := runner.Worker.Stats()
|
||
if stats.Started != 2 || stats.ReportsConfirmed != 2 || stats.Running != 0 {
|
||
return errors.New("unexpected executions or incomplete drain")
|
||
}
|
||
fmt.Printf("recovery_verified=true state=%s submissions=%d unknown_history_preserved=true review_deduplicated=true pause_preserved=true audits=3 sequence=%d worker_drained=true\n", final.State, submissions.Load(), exactSequence)
|
||
return nil
|
||
}
|
||
|
||
func verifyReceipt(raw json.RawMessage, operation string) error {
|
||
var out receipt
|
||
if json.Unmarshal(raw, &out) != nil || out.OperationID != operation || out.Sequence != exactSequence || out.State != "CONFIRMED" {
|
||
return errors.New("external receipt identity or exact sequence mismatch")
|
||
}
|
||
return nil
|
||
}
|