| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "fmt" |
| 8 | "os" |
| 9 | "path/filepath" |
| 10 | "testing" |
| 11 | "time" |
| 12 | |
| 13 | "github.com/carlosframework/platform/internal/hostbox" |
| 14 | "github.com/carlosframework/platform/internal/hostclaim" |
| 15 | execlease "github.com/carlosframework/platform/internal/hostexec" |
| 16 | "github.com/carlosframework/platform/internal/registry" |
| 17 | ) |
| 18 | |
| 19 | type staleClaimReader func(context.Context, string) (hostclaim.Observed, error) |
| 20 | |
| 21 | func (f staleClaimReader) Read(ctx context.Context, host string) (hostclaim.Observed, error) { |
| 22 | return f(ctx, host) |
| 23 | } |
| 24 | |
| 25 | func TestPlan9PreviousAgentRetiredParticipation(t *testing.T) { |
| 26 | t.Parallel() |
| 27 | for _, mode := range []string{"free", "free-lock-held", "free-without-history", "replacement", "replacement-artifact", "missing-retirement", "artifact", "launch", "fresh-retirement-missing", "fresh-absent", "root", "adopt"} { |
| 28 | t.Run(mode, func(t *testing.T) { |
| 29 | f := newRecoveryFixture(t) |
| 30 | ctx := t.Context() |
| 31 | old := f.read() |
| 32 | inc := old.Claim.Holder.Incarnation |
| 33 | if mode == "launch" { |
| 34 | f.controller("boot-old", "controller-old", 901, "101").admit(t, f) |
| 35 | } |
| 36 | o, err := hostclaim.Update(ctx, f.claims, f.read(), hostclaim.BeginDelete("delete", "console", "record-etag", f.now, nil, nil, "")) |
| 37 | if err != nil { |
| 38 | t.Fatal(err) |
| 39 | } |
| 40 | o, err = hostclaim.Update(ctx, f.claims, o, hostclaim.ToParking("delete", nil, nil, f.newID)) |
| 41 | if err != nil { |
| 42 | t.Fatal(err) |
| 43 | } |
| 44 | for len(o.Claim.Obligations) > 0 { |
| 45 | chore := o.Claim.Obligations[0] |
| 46 | o, err = hostclaim.Update(ctx, f.claims, o, hostclaim.CompleteObligation(chore.Op, chore.ID)) |
| 47 | if err != nil { |
| 48 | t.Fatal(err) |
| 49 | } |
| 50 | } |
| 51 | o, err = hostclaim.Update(ctx, f.claims, o, hostclaim.FreeAfterPark("delete")) |
| 52 | if err != nil || !o.Claim.Free() { |
| 53 | t.Fatalf("FreeAfterPark: %+v %v", o, err) |
| 54 | } |
| 55 | if mode == "free-without-history" { |
| 56 | o, err = hostclaim.Update(ctx, f.claims, o, func(c *hostclaim.Claim) error { c.RetiredIncarnations = nil; return nil }) |
| 57 | if err != nil { |
| 58 | t.Fatal(err) |
| 59 | } |
| 60 | } |
| 61 | if mode == "replacement" || mode == "replacement-artifact" || mode == "missing-retirement" || mode == "fresh-retirement-missing" { |
| 62 | created := f.now.Add(time.Hour) |
| 63 | in := hostclaim.NewHolder{Family: "instance", Key: old.Claim.Holder.Key, Account: "bab", App: "jam", Host: recoveryHost, Region: "ie", RecordRegion: "ie", CreatedAt: created} |
| 64 | body := []byte(fmt.Sprintf(`{"schema":1,"host":%q,"account":"bab","app":"jam","region":"ie","created_at":%q}`, recoveryHost, created.Format(time.RFC3339))) |
| 65 | o, err = hostclaim.Update(ctx, f.claims, o, hostclaim.BeginCreate(in, body, "create-next", "console", created)) |
| 66 | if err != nil { |
| 67 | t.Fatal(err) |
| 68 | } |
| 69 | o, err = hostclaim.Update(ctx, f.claims, o, hostclaim.CompleteCreate("create-next", "box-2")) |
| 70 | if err != nil { |
| 71 | t.Fatal(err) |
| 72 | } |
| 73 | if mode == "missing-retirement" { |
| 74 | o, err = hostclaim.Update(ctx, f.claims, o, func(c *hostclaim.Claim) error { c.RetiredIncarnations = nil; return nil }) |
| 75 | if err != nil { |
| 76 | t.Fatal(err) |
| 77 | } |
| 78 | } |
| 79 | } |
| 80 | inventory, err := openLegacyInventory(t, filepath.Join(t.TempDir(), "inventory.db")) |
| 81 | if err != nil { |
| 82 | t.Fatal(err) |
| 83 | } |
| 84 | t.Cleanup(func() { inventory.Close() }) |
| 85 | reg, err := registry.Open(filepath.Join(t.TempDir(), "registry.db")) |
| 86 | if err != nil { |
| 87 | t.Fatal(err) |
| 88 | } |
| 89 | t.Cleanup(func() { reg.Close() }) |
| 90 | census := &hostbox.EmptyInstallationCensus{Inventory: inventory, Registry: reg, Launches: f.journal, Box: "box-1", Enrollment: "enrollment-1", DirectoryRevision: "7", Roots: map[string]string{"data": f.guard.DataDir, "opt": t.TempDir(), "units": t.TempDir()}, SharedPaths: map[string][]string{"data": {".hostexec", "hostguard"}}} |
| 91 | p := &productionClaimLocal{legacyParticipation: true, Resources: ClaimRuntimeResources{Mode: "agent", DataDir: f.guard.DataDir}, lifetime: ctx, assembly: &ClaimAssembly{Wake: &ClaimWakeAssembly{}, Launches: f.journal, Reconcile: &hostbox.ReconcileAdapter{Census: census}}} |
| 92 | state := execlease.State{V: 1, Binding: execlease.Binding{Host: recoveryHost, Box: "box-1", Enrollment: "enrollment-1", Incarnation: inc, Op: "old-op", BootID: "old-boot", ControllerID: "old-controller"}, State: "lost", SafeUntil: 1, Expires: time.Now()} |
| 93 | dir := filepath.Join(f.guard.DataDir, ".hostexec") |
| 94 | if err = os.Mkdir(dir, 0700); err != nil { |
| 95 | t.Fatal(err) |
| 96 | } |
| 97 | body, _ := json.Marshal(state) |
| 98 | path := filepath.Join(dir, recoveryHost+".json") |
| 99 | if err = os.WriteFile(path, body, 0600); err != nil { |
| 100 | t.Fatal(err) |
| 101 | } |
| 102 | if mode == "artifact" || mode == "replacement-artifact" { |
| 103 | // Even complete previous-installation metadata cannot bypass the full scan. |
| 104 | if err = inventory.WriteStopInventory(ctx, hostbox.StopInventory{Complete: true, Host: recoveryHost, Incarnation: inc, InstallationStamp: old.Claim.Holder.Stamp, Box: "box-1", Enrollment: "enrollment-1", DirectoryRevision: "7"}); err != nil { |
| 105 | t.Fatal(err) |
| 106 | } |
| 107 | if err = os.WriteFile(filepath.Join(census.Roots["opt"], "leftover"), []byte("data"), 0600); err != nil { |
| 108 | t.Fatal(err) |
| 109 | } |
| 110 | } |
| 111 | cfg := execlease.Config{Binding: state.Binding, Guard: f.guard, LockDir: t.TempDir()} |
| 112 | if err = census.Initialize(ctx, o, hostbox.RecordSelection{Host: recoveryHost, Incarnation: inc}); !errors.Is(err, hostbox.ErrChanged) { |
| 113 | t.Fatalf("cleanup scope initialized replacement inventory: %v", err) |
| 114 | } |
| 115 | if mode == "free-lock-held" { |
| 116 | original := cfg.Guard.Claims |
| 117 | competing := cfg |
| 118 | cfg.Guard.Claims = staleClaimReader(func(ctx context.Context, host string) (hostclaim.Observed, error) { |
| 119 | deadline, cancel := context.WithTimeout(ctx, 25*time.Millisecond) |
| 120 | defer cancel() |
| 121 | if err := execlease.ClearStale(deadline, competing); !errors.Is(err, context.DeadlineExceeded) { |
| 122 | t.Fatalf("fresh proof did not retain lifetime lock: %v", err) |
| 123 | } |
| 124 | return original.Read(ctx, host) |
| 125 | }) |
| 126 | } |
| 127 | if mode == "fresh-retirement-missing" || mode == "fresh-absent" { |
| 128 | cfg.Guard.Claims = staleClaimReader(func(ctx context.Context, host string) (hostclaim.Observed, error) { |
| 129 | live, err := execlease.InspectLocal(f.guard.DataDir, host) |
| 130 | if err != nil || live.State != "lost" { |
| 131 | t.Fatalf("fresh proof without durable denial: %+v %v", live, err) |
| 132 | } |
| 133 | if mode == "fresh-absent" { |
| 134 | return hostclaim.Observed{Host: host, Absent: true}, nil |
| 135 | } |
| 136 | fresh := o |
| 137 | c := *o.Claim |
| 138 | c.RetiredIncarnations = nil |
| 139 | fresh.Claim = &c |
| 140 | return fresh, nil |
| 141 | }) |
| 142 | } |
| 143 | actor := hostbox.Actor{Box: "box-1", Enrollment: "enrollment-1"} |
| 144 | if mode == "root" || mode == "adopt" { |
| 145 | p.Resources.Mode = mode |
| 146 | err = p.prepareWith(ctx, actor, o) |
| 147 | } else { |
| 148 | err = p.clearUntracked(ctx, actor, o, state, cfg) |
| 149 | } |
| 150 | wantOK := mode == "free" || mode == "free-lock-held" || mode == "free-without-history" || mode == "replacement" |
| 151 | if wantOK { |
| 152 | if err != nil { |
| 153 | t.Fatalf("positive stale cleanup refused: %v", err) |
| 154 | } |
| 155 | if _, err = os.Stat(path); !errors.Is(err, os.ErrNotExist) { |
| 156 | t.Fatalf("stale state retained: %v", err) |
| 157 | } |
| 158 | } else { |
| 159 | if err == nil { |
| 160 | t.Fatal("unproven stale cleanup accepted") |
| 161 | } |
| 162 | if _, err = os.Stat(path); err != nil { |
| 163 | t.Fatalf("refusal removed local denial: %v", err) |
| 164 | } |
| 165 | } |
| 166 | }) |
| 167 | } |
| 168 | } |
| 169 | |
amadan