Recover worker leases after empty event replay

This commit is contained in:
kami
2026-07-28 16:31:11 +04:00
parent 673a3ab244
commit bb43944572
2 changed files with 56 additions and 0 deletions
+33
View File
@@ -260,6 +260,16 @@ func (w *worker) once(ctx context.Context) error {
w.cursor = e.Seq
}
}
// A coordinator restart can restore its task snapshot without retaining
// the in-memory event tail. In that state a worker with a persisted cursor
// receives an empty page even though a lease is currently assigned to it.
// Reconcile the authoritative projection before treating an empty page as
// "nothing to do"; otherwise the lease remains invisible until expiry.
if len(es) == 0 {
if err := w.reconcileLeases(ctx); err != nil {
return err
}
}
// State is only a cache. If a lease survived but its TaskCreated event is
// older than the worker's cursor (or the event has been compacted), hydrate
// the authoritative task projection before deciding whether to start.
@@ -294,6 +304,29 @@ func (w *worker) once(ctx context.Context) error {
return w.api.Ack(ctx, w.cursor)
}
func (w *worker) reconcileLeases(ctx context.Context) error {
tasks, err := w.api.Tasks(ctx)
if err != nil {
return fmt.Errorf("reconcile leased tasks: %w", err)
}
active := make(map[string]lease)
for _, task := range tasks {
w.tasks[task.ID] = task
if task.State == domain.StateLeased && task.Lease != nil && task.Lease.HarnessID == w.harnessID {
active[task.ID] = lease{HandoffRef: task.HandoffRef}
}
}
for taskID := range w.leases {
if _, ok := active[taskID]; !ok {
delete(w.leases, taskID)
}
}
for taskID, l := range active {
w.leases[taskID] = l
}
return nil
}
// reRegisterAfterCoordinatorRestart restores the coordinator's in-memory
// worker registry. A worker must survive a server restart without operator
// intervention; its persisted cursor and sessions remain valid.