| 1 | //go:build linux |
| 2 | |
| 3 | package hostexec |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "fmt" |
| 9 | "github.com/carlosframework/platform/internal/hostfloor" |
| 10 | "os" |
| 11 | "path/filepath" |
| 12 | "testing" |
| 13 | "time" |
| 14 | ) |
| 15 | |
| 16 | func TestPlan9PendingRecovery(t *testing.T) { |
| 17 | for _, tc := range []struct { |
| 18 | name string |
| 19 | boot string |
| 20 | start string |
| 21 | probeErr error |
| 22 | state string |
| 23 | legacy bool |
| 24 | busy bool |
| 25 | ok bool |
| 26 | }{ |
| 27 | {name: "dead", boot: "boot", probeErr: os.ErrNotExist, state: "pending", ok: true}, |
| 28 | {name: "reused-pid", boot: "boot", start: "456", state: "pending", ok: true}, |
| 29 | {name: "new-boot", boot: "next", start: "123", state: "pending", ok: true}, |
| 30 | {name: "live", boot: "boot", start: "123", state: "pending"}, |
| 31 | {name: "unreadable", boot: "boot", probeErr: os.ErrPermission, state: "pending"}, |
| 32 | {name: "active", boot: "next", state: "active"}, |
| 33 | {name: "legacy", boot: "next", state: "pending", legacy: true}, |
| 34 | {name: "lock-held", boot: "next", state: "pending", busy: true}, |
| 35 | } { |
| 36 | t.Run(tc.name, func(t *testing.T) { |
| 37 | cfg, _, _, _, _ := fixture(t) |
| 38 | d, err := openLocal(cfg.Guard.DataDir) |
| 39 | if err != nil { |
| 40 | t.Fatal(err) |
| 41 | } |
| 42 | defer d.close() |
| 43 | s := State{V: 1, Binding: cfg.Binding, State: tc.state, Expires: time.Now().Add(TTL), SafeUntil: 99, LeaseETag: "tag", Process: ControllerProcess{KernelBootID: "boot", PID: 123, Start: "123"}} |
| 44 | s.ControllerID = "dead-controller" |
| 45 | if tc.legacy { |
| 46 | s.Process = ControllerProcess{} |
| 47 | } |
| 48 | if err = d.write(s); err != nil { |
| 49 | t.Fatal(err) |
| 50 | } |
| 51 | if tc.busy { |
| 52 | unlock, e := tryAcquire(cfg.LockDir, s.Host) |
| 53 | if e != nil { |
| 54 | t.Fatal(e) |
| 55 | } |
| 56 | defer unlock() |
| 57 | } |
| 58 | archive := filepath.Join(t.TempDir(), "settled") |
| 59 | err = recoverPending(context.Background(), cfg.Guard, cfg.LockDir, archive, func() (string, error) { return tc.boot, nil }, func(int) (string, error) { return tc.start, tc.probeErr }) |
| 60 | if tc.ok { |
| 61 | if err != nil { |
| 62 | t.Fatal(err) |
| 63 | } |
| 64 | if _, err = d.read(s.Host); !errors.Is(err, os.ErrNotExist) { |
| 65 | t.Fatalf("pending retained: %v", err) |
| 66 | } |
| 67 | entries, err := os.ReadDir(archive) |
| 68 | if err != nil || len(entries) != 1 { |
| 69 | t.Fatalf("missing durable evidence: %v %v", entries, err) |
| 70 | } |
| 71 | // Model the crash after archive fsync but before the pending unlink. |
| 72 | if err = d.write(s); err != nil { |
| 73 | t.Fatal(err) |
| 74 | } |
| 75 | if err = recoverPending(context.Background(), cfg.Guard, cfg.LockDir, archive, func() (string, error) { return tc.boot, nil }, func(int) (string, error) { return tc.start, tc.probeErr }); err != nil { |
| 76 | t.Fatal(err) |
| 77 | } |
| 78 | } else { |
| 79 | if err == nil { |
| 80 | t.Fatal("unproven pending controller settled") |
| 81 | } |
| 82 | got, e := d.read(s.Host) |
| 83 | if e != nil || got.ControllerID != s.ControllerID { |
| 84 | t.Fatalf("refusal changed record: %+v %v", got, e) |
| 85 | } |
| 86 | } |
| 87 | }) |
| 88 | } |
| 89 | } |
| 90 | |
| 91 | func TestPlan9PendingCapturesKernelIdentity(t *testing.T) { |
| 92 | p, err := currentControllerProcess() |
| 93 | if err != nil { |
| 94 | t.Fatal(err) |
| 95 | } |
| 96 | boot, err := BootID() |
| 97 | if err != nil { |
| 98 | t.Fatal(err) |
| 99 | } |
| 100 | start, err := controllerProcessStart(os.Getpid()) |
| 101 | if err != nil { |
| 102 | t.Fatal(err) |
| 103 | } |
| 104 | if p.PID != os.Getpid() || p.Start != start || p.KernelBootID != boot { |
| 105 | t.Fatalf("wrong process: %+v", p) |
| 106 | } |
| 107 | if err := proveControllerDead(p, boot, controllerProcessStart); err == nil { |
| 108 | t.Fatal("current process considered dead") |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | func TestPlan9FailedOpenAbortsOnlyJoinedPending(t *testing.T) { |
| 113 | for _, failedCleanup := range []bool{false, true} { |
| 114 | t.Run(fmt.Sprint(failedCleanup), func(t *testing.T) { |
| 115 | cfg, _, _, _, _ := fixture(t) |
| 116 | cfg.Clock = SystemClock{} // Persisted JSON strips monotonic time metadata. |
| 117 | cfg.UpdateGuardDir = t.TempDir() |
| 118 | cfg.UpdateJournal = filepath.Join(t.TempDir(), "journal.json") |
| 119 | cfg.verifyRunning = func(context.Context) error { return hostfloor.ErrFloor } |
| 120 | if failedCleanup { |
| 121 | cfg.Cleanup = func(context.Context) error { return errors.New("cleanup uncertain") } |
| 122 | } |
| 123 | c, err := Open(t.Context(), cfg) |
| 124 | if c != nil || !errors.Is(err, hostfloor.ErrFloor) { |
| 125 | t.Fatal(c, err) |
| 126 | } |
| 127 | d, err := openLocal(cfg.Guard.DataDir) |
| 128 | if err != nil { |
| 129 | t.Fatal(err) |
| 130 | } |
| 131 | defer d.close() |
| 132 | state, err := d.read(cfg.Binding.Host) |
| 133 | if failedCleanup { |
| 134 | if err != nil || state.State != "pending" { |
| 135 | t.Fatal(state, err) |
| 136 | } |
| 137 | } else if !errors.Is(err, os.ErrNotExist) { |
| 138 | t.Fatalf("joined aborted Open retained live-process pending: %+v %v", state, err) |
| 139 | } |
| 140 | }) |
| 141 | } |
| 142 | } |
| 143 | func TestPlan9PendingRecoveryDoesNotCreateLocks(t *testing.T) { |
| 144 | cfg, _, _, _, _ := fixture(t) |
| 145 | d, err := openLocal(cfg.Guard.DataDir) |
| 146 | if err != nil { |
| 147 | t.Fatal(err) |
| 148 | } |
| 149 | defer d.close() |
| 150 | s := State{V: 1, Binding: cfg.Binding, State: "pending", Expires: time.Now().Add(TTL), SafeUntil: 99, Process: ControllerProcess{KernelBootID: "old", PID: 123, Start: "123"}} |
| 151 | s.ControllerID = "dead" |
| 152 | if err = d.write(s); err != nil { |
| 153 | t.Fatal(err) |
| 154 | } |
| 155 | lockDir := filepath.Join(t.TempDir(), "absent") |
| 156 | if err = recoverPending(t.Context(), cfg.Guard, lockDir, filepath.Join(t.TempDir(), "settled"), func() (string, error) { return "next", nil }, controllerProcessStart); err != nil { |
| 157 | t.Fatal(err) |
| 158 | } |
| 159 | if _, err = os.Stat(lockDir); !errors.Is(err, os.ErrNotExist) { |
| 160 | t.Fatal("root created runtime locks", err) |
| 161 | } |
| 162 | } |
| 163 | |