| 1 | //go:build linux |
| 2 | |
| 3 | package rootparticipation |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "os" |
| 9 | "os/exec" |
| 10 | "path/filepath" |
| 11 | "syscall" |
| 12 | "testing" |
| 13 | "time" |
| 14 | ) |
| 15 | |
| 16 | type recoveryDisk struct{ testDisk } |
| 17 | |
| 18 | func (d *recoveryDisk) Write(name string, raw []byte) error { |
| 19 | f, err := os.CreateTemp(d.dir, "write-") |
| 20 | if err != nil { |
| 21 | return err |
| 22 | } |
| 23 | tmp := f.Name() |
| 24 | defer os.Remove(tmp) |
| 25 | if _, err = f.Write(raw); err == nil { |
| 26 | err = f.Sync() |
| 27 | } |
| 28 | err = errors.Join(err, f.Close()) |
| 29 | if err != nil { |
| 30 | return err |
| 31 | } |
| 32 | if err = os.Rename(tmp, filepath.Join(d.dir, name)); err != nil { |
| 33 | return err |
| 34 | } |
| 35 | dir, err := os.Open(d.dir) |
| 36 | if err != nil { |
| 37 | return err |
| 38 | } |
| 39 | defer dir.Close() |
| 40 | return dir.Sync() |
| 41 | } |
| 42 | func (d *recoveryDisk) Lock(ctx context.Context, name string) (func(), error) { |
| 43 | f, err := os.OpenFile(filepath.Join(d.dir, name), os.O_CREATE|os.O_RDWR, 0600) |
| 44 | if err != nil { |
| 45 | return nil, err |
| 46 | } |
| 47 | for { |
| 48 | err = syscall.Flock(int(f.Fd()), syscall.LOCK_EX|syscall.LOCK_NB) |
| 49 | if err == nil { |
| 50 | return func() { _ = syscall.Flock(int(f.Fd()), syscall.LOCK_UN); _ = f.Close() }, nil |
| 51 | } |
| 52 | if err != syscall.EWOULDBLOCK { |
| 53 | f.Close() |
| 54 | return nil, err |
| 55 | } |
| 56 | select { |
| 57 | case <-ctx.Done(): |
| 58 | f.Close() |
| 59 | return nil, ctx.Err() |
| 60 | case <-time.After(time.Millisecond): |
| 61 | } |
| 62 | } |
| 63 | } |
| 64 | |
| 65 | func TestPlan14ExecutionRecoveryAfterProcessCrash(t *testing.T) { |
| 66 | if dir := os.Getenv("PLAN14_EXEC_CRASH_DIR"); dir != "" { |
| 67 | d := &recoveryDisk{testDisk: testDisk{dir: dir}} |
| 68 | l := Ledger{State: d, VerifyLaunches: func(context.Context, uint64) error { return nil }} |
| 69 | crash := func() error { os.Exit(31); return nil } |
| 70 | var err error |
| 71 | if os.Getenv("PLAN14_EXEC_CRASH_AT") == "acquire" { |
| 72 | err = l.Acquire(t.Context(), crash) |
| 73 | } else { |
| 74 | err = l.BeginUpdateWith(t.Context(), "update", crash) |
| 75 | } |
| 76 | t.Fatal("did not reach crash", err) |
| 77 | } |
| 78 | for _, at := range []string{"acquire", "update"} { |
| 79 | t.Run(at, func(t *testing.T) { |
| 80 | l := censusFixture(t) |
| 81 | dir := l.State.(*testDisk).dir |
| 82 | l.State = &recoveryDisk{testDisk: testDisk{dir: dir}} |
| 83 | child := exec.Command(os.Args[0], "-test.run=^TestPlan14ExecutionRecoveryAfterProcessCrash$") |
| 84 | child.Env = append(os.Environ(), "PLAN14_EXEC_CRASH_DIR="+dir, "PLAN14_EXEC_CRASH_AT="+at) |
| 85 | child.Dir = dir |
| 86 | out, err := child.CombinedOutput() |
| 87 | var exit *exec.ExitError |
| 88 | if !errors.As(err, &exit) || exit.ExitCode() != 31 { |
| 89 | t.Fatalf("crash: %v %s", err, out) |
| 90 | } |
| 91 | if err := l.Acquire(t.Context(), func() error { t.Error("acquired across crash"); return nil }); err == nil { |
| 92 | t.Fatal("crash grant") |
| 93 | } |
| 94 | if at == "acquire" { |
| 95 | if _, err := l.Census(); err == nil { |
| 96 | t.Fatal("pending acquisition accepted") |
| 97 | } |
| 98 | l.VerifyLaunches = nil |
| 99 | if err := l.RecoverAcquisition(t.Context()); err == nil { |
| 100 | t.Fatal("absence replaced kernel completeness") |
| 101 | } |
| 102 | l.VerifyLaunches = func(context.Context, uint64) error { return nil } |
| 103 | if err := l.RecoverAcquisition(t.Context()); err != nil { |
| 104 | t.Fatal(err) |
| 105 | } |
| 106 | } else { |
| 107 | failed := errors.New("terminal write failed") |
| 108 | if err := l.EndUpdateWith(t.Context(), "update", func() error { return failed }); !errors.Is(err, failed) { |
| 109 | t.Fatal(err) |
| 110 | } |
| 111 | if _, err := l.UpdateID(); err != nil { |
| 112 | t.Fatal("terminal failure lost exclusion", err) |
| 113 | } |
| 114 | if err := l.EndUpdate(t.Context(), "update"); err != nil { |
| 115 | t.Fatal(err) |
| 116 | } |
| 117 | } |
| 118 | if err := l.Acquire(t.Context(), func() error { return nil }); err != nil { |
| 119 | t.Fatal("positive recovery did not reopen", err) |
| 120 | } |
| 121 | }) |
| 122 | } |
| 123 | } |
| 124 | |