diff --git a/CHANGELOG.md b/CHANGELOG.md index 8522e83..392293d 100644 --- a/CHANGELOG.md +++ b/CHANGELOG.md @@ -1,5 +1,18 @@ # Changelog +## 开发快照(真实依赖跑出来的接线缺陷) + +### Fixed + +- **Docker inspect 反序列化**:`containerState` 把 `Config` 映射成 `map[string]string`, + 真实返回是对象,于是每次拓扑校验都报 `invalid docker response`——所有维护窗口被误拒。 +- **后台静默检查没接线**:`openLibraryMaintenance` 漏传目标库/队列,实例级窗口永远报 + "缺少后台静默检查"。防线代码在仓库里、运行时却没生效。 +- **恢复作业队列 id 超列宽**:`"librecover-"+窗口作业 id` 超过 `varchar(36)`, + 恢复入口投递即失败(409 enqueue failed),等价于没有恢复入口。 +- **恢复不再信任上一次窗口的 StopRequested**:写入者在窗口后被人工拉起来时, + `stopAll` 要按"现在是否真的在跑"再停一次;否则校验会跑在半开的目标上。 + ## 开发快照(维护收尾必须按预备份验证 / 后台任务静默 / 实例级互斥) ### Fixed diff --git a/internal/libraryfreeze/window.go b/internal/libraryfreeze/window.go index faa3319..17773c9 100644 --- a/internal/libraryfreeze/window.go +++ b/internal/libraryfreeze/window.go @@ -331,14 +331,26 @@ func (c *Coordinator) awaitBackgroundIdle(ctx context.Context, j *journal) error // stopAll 只对"原本在跑"的服务发停止指令,每发一个都先落盘意图: // 进程在停止过程中死掉时,磁盘上仍能看出"哪些服务被要求停过"。 func (c *Coordinator) stopAll(ctx context.Context, j *journal) error { + // 以"现在是否真的在跑"为准,而不是上次窗口留下的 StopRequested: + // 恢复阶段被人工拉起来的写入者必须再停一次,否则校验跑在半开的目标上。 + items, err := inventory(ctx, c.control) + if err != nil { + return err + } + running := map[string]bool{} + for _, s := range items { + running[s.ID] = s.Running + } var failures error for i := range j.Services { - if !j.Services[i].Running || j.Services[i].StopRequested { + if !running[j.Services[i].ID] { continue } - j.Services[i].StopRequested = true - if err := c.save(*j); err != nil { - return err + if !j.Services[i].StopRequested { + j.Services[i].StopRequested = true + if err := c.save(*j); err != nil { + return err + } } if err := c.control.Stop(ctx, j.Services[i].ID); err != nil { failures = errors.Join(failures, err) diff --git a/internal/libraryfreeze/window_test.go b/internal/libraryfreeze/window_test.go index a05465e..f352fe2 100644 --- a/internal/libraryfreeze/window_test.go +++ b/internal/libraryfreeze/window_test.go @@ -325,3 +325,33 @@ func TestWindowRefusesWhenGateNotDrained(t *testing.T) { t.Fatalf("writers were stopped before the ingress was drained: %v", c.calls) } } + +// 窗口开过之后被人工拉起来的写入者,恢复时必须再停一次: +// 上一次窗口写下的 StopRequested 不能当成"它现在还停着"。 +func TestStopAllStopsWriterRestartedAfterWindowOpened(t *testing.T) { + _, c, engine := instanceFixture(t) + j := journal{Version: 1, Source: "target-instance", Job: "job-restart", + Result: Result{State: "held"}, Level: "instance", + Services: []serviceRecord{{Service: Service{ID: "app", Running: true, Healthy: true}, StopRequested: true}}} + if err := engine.stopAll(context.Background(), &j); err != nil { + t.Fatalf("stopAll: %v", err) + } + items, err := c.Inventory(context.Background()) + if err != nil { + t.Fatal(err) + } + for _, s := range items { + if s.ID == "app" && s.Running { + t.Fatal("写入者被人工拉起后没有被重新停住") + } + } + found := false + for _, call := range c.calls { + if call == "stop:app" { + found = true + } + } + if !found { + t.Fatalf("没有发出停止指令:%v", c.calls) + } +}