Add recovery for orphaned jobs

cmd/agent/cancel_test.go

@@ -0,0 +1,68 @@
+package main
+
+import (
+	"context"
+	"testing"
+
+	"gitea.dcglab.co.uk/steve/restic-manager/internal/api"
+)
+
+type recordingSender struct {
+	envelopes []api.Envelope
+}
+
+func (s *recordingSender) Send(env api.Envelope) error {
+	s.envelopes = append(s.envelopes, env)
+	return nil
+}
+
+func TestCancelCommandReportsUnknownJob(t *testing.T) {
+	d := &dispatcher{}
+	tx := &recordingSender{}
+	env, err := api.Marshal(api.MsgCommandCancel, "request-1", api.CommandCancelPayload{JobID: "missing-job"})
+	if err != nil {
+		t.Fatalf("marshal command: %v", err)
+	}
+
+	if err := d.handle(context.Background(), env, tx); err != nil {
+		t.Fatalf("handle cancel: %v", err)
+	}
+	if len(tx.envelopes) != 1 {
+		t.Fatalf("sent %d envelopes, want 1", len(tx.envelopes))
+	}
+	if tx.envelopes[0].Type != api.MsgCommandResult || tx.envelopes[0].ID != env.ID {
+		t.Fatalf("unexpected result envelope: %+v", tx.envelopes[0])
+	}
+	var result api.CommandResultPayload
+	if err := tx.envelopes[0].UnmarshalPayload(&result); err != nil {
+		t.Fatalf("unmarshal result: %v", err)
+	}
+	if result.JobID != "missing-job" || result.Accepted || result.Error != "job_not_found" {
+		t.Fatalf("unexpected result: %+v", result)
+	}
+}
+
+func TestCancelCommandCancelsTrackedJob(t *testing.T) {
+	d := &dispatcher{}
+	cancelled := false
+	d.trackJob("running-job", func() { cancelled = true })
+	tx := &recordingSender{}
+	env, err := api.Marshal(api.MsgCommandCancel, "request-2", api.CommandCancelPayload{JobID: "running-job"})
+	if err != nil {
+		t.Fatalf("marshal command: %v", err)
+	}
+
+	if err := d.handle(context.Background(), env, tx); err != nil {
+		t.Fatalf("handle cancel: %v", err)
+	}
+	if !cancelled {
+		t.Fatal("tracked job was not cancelled")
+	}
+	var result api.CommandResultPayload
+	if err := tx.envelopes[0].UnmarshalPayload(&result); err != nil {
+		t.Fatalf("unmarshal result: %v", err)
+	}
+	if !result.Accepted || result.Error != "" {
+		t.Fatalf("unexpected result: %+v", result)
+	}
+}

cmd/agent/main.go

@@ -276,7 +276,8 @@ func (d *dispatcher) handle(ctx context.Context, env api.Envelope, tx wsclient.S
 		if err := env.UnmarshalPayload(&p); err != nil {
 			return fmt.Errorf("command.cancel: %w", err)
 		}
-		if d.cancelJob(p.JobID) {
+		known := d.cancelJob(p.JobID)
+		if known {
 			slog.Info("ws agent: command.cancel applied", "job_id", p.JobID)
 		} else {
 			// Job already finished or was never seen on this agent.
@@ -284,6 +285,17 @@ func (d *dispatcher) handle(ctx context.Context, env api.Envelope, tx wsclient.S
 			// natural completion. Server-side state is authoritative.
 			slog.Info("ws agent: command.cancel for unknown job (already finished?)", "job_id", p.JobID)
 		}
+		result := api.CommandResultPayload{JobID: p.JobID, Accepted: known}
+		if !known {
+			result.Error = "job_not_found"
+		}
+		ack, err := api.Marshal(api.MsgCommandResult, env.ID, result)
+		if err != nil {
+			return fmt.Errorf("command.cancel result: %w", err)
+		}
+		if err := tx.Send(ack); err != nil {
+			return fmt.Errorf("command.cancel result send: %w", err)
+		}
 
 	case api.MsgTreeList:
 		// Synchronous RPC for the restore wizard's tree browser. The