| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "context" |
| 6 | "errors" |
| 7 | "path/filepath" |
| 8 | "reflect" |
| 9 | "strings" |
| 10 | "testing" |
| 11 | |
| 12 | "github.com/carlosframework/platform/internal/hostactors" |
| 13 | "github.com/carlosframework/platform/internal/hostenv" |
| 14 | "github.com/carlosframework/platform/internal/registry" |
| 15 | "github.com/carlosframework/platform/internal/release" |
| 16 | "github.com/carlosframework/platform/internal/status" |
| 17 | ) |
| 18 | |
| 19 | // withGroupIdentity makes IMDS answer account and instance (or fail) for |
| 20 | // one test, with no reference kept from before. |
| 21 | func withGroupIdentity(t *testing.T, account, instance string, err error) { |
| 22 | t.Helper() |
| 23 | was := groupIdentity |
| 24 | groupIdentity = func() (string, string, error) { return account, instance, err } |
| 25 | groupRef.Lock() |
| 26 | groupRef.ref = "" |
| 27 | groupRef.Unlock() |
| 28 | t.Cleanup(func() { |
| 29 | groupIdentity = was |
| 30 | groupRef.Lock() |
| 31 | groupRef.ref = "" |
| 32 | groupRef.Unlock() |
| 33 | }) |
| 34 | } |
| 35 | |
| 36 | func groupEnv(vals map[string]string) func(string) string { |
| 37 | return func(k string) string { return vals[k] } |
| 38 | } |
| 39 | |
| 40 | // A box its region's group launched takes its group reference everywhere |
| 41 | // it names itself: the identity record, mode selection, root's join. An |
| 42 | // identity IMDS has not given is no reference, never a label in its place. |
| 43 | func TestAGroupBoxTakesItsGroupReference(t *testing.T) { |
| 44 | group := groupEnv(map[string]string{hostenv.BoxReference: hostenv.GroupReference}) |
| 45 | withGroupIdentity(t, "", "", errors.New("IMDS answered 500")) |
| 46 | if got := metricsBoxLabel(group, "machine"); got != "" { |
| 47 | t.Fatalf("no identity: %q", got) |
| 48 | } |
| 49 | withGroupIdentity(t, "123456789012", "i-0a61f580df1acfa65", nil) |
| 50 | want := hostactors.GroupBoxReference("123456789012", "i-0a61f580df1acfa65") |
| 51 | if got := metricsBoxLabel(group, "machine"); got != want { |
| 52 | t.Fatalf("group box: %q, want %q", got, want) |
| 53 | } |
| 54 | } |
| 55 | |
| 56 | // statusFixture is a registry with one route and a status store holding |
| 57 | // another box's report under the label this box's machine id hashes to. |
| 58 | func statusFixture(t *testing.T) (*registry.Registry, *release.DirStore, []byte) { |
| 59 | t.Helper() |
| 60 | reg, err := registry.Open(filepath.Join(t.TempDir(), "registry.db")) |
| 61 | if err != nil { |
| 62 | t.Fatal(err) |
| 63 | } |
| 64 | if err := reg.Upsert(registry.Route{Host: "a.example.com", Kind: "static"}); err != nil { |
| 65 | t.Fatal(err) |
| 66 | } |
| 67 | s := release.NewDirStore(t.TempDir()) |
| 68 | victim := []byte(`{"label":"victim","metrics":{"instance_id":"i-victim"}}`) |
| 69 | if _, err := s.Put(context.Background(), "status/boxes/"+status.Label("machine-1")+".json", victim, ""); err != nil { |
| 70 | t.Fatal(err) |
| 71 | } |
| 72 | return reg, s, victim |
| 73 | } |
| 74 | |
| 75 | // T1 (census ruling region-ha-2b): a group box whose reference has not |
| 76 | // resolved publishes nothing, whatever the claim and region HA switches |
| 77 | // say; above all never under status.Label, where another box's report may |
| 78 | // stand. The reference is asked again each tick, and once it resolves the |
| 79 | // box publishes under it. Through statusLoop's own tick and WriteBox. |
| 80 | func TestAGroupBoxPublishesOnlyUnderItsReference(t *testing.T) { |
| 81 | for name, env := range map[string]map[string]string{ |
| 82 | "switches on": {hostenv.BoxReference: hostenv.GroupReference, "CARLOS_HOST_CLAIM": "1", "CARLOS_REGION_HA": "1"}, |
| 83 | "switches off": {hostenv.BoxReference: hostenv.GroupReference}, |
| 84 | } { |
| 85 | t.Run(name, func(t *testing.T) { |
| 86 | ctx := context.Background() |
| 87 | reg, s, victim := statusFixture(t) |
| 88 | withGroupIdentity(t, "", "", errors.New("IMDS answered 500")) |
| 89 | asked := 0 |
| 90 | groupIdentity = func() (string, string, error) { |
| 91 | asked++ |
| 92 | if asked < 3 { |
| 93 | return "", "", errors.New("IMDS answered 500") |
| 94 | } |
| 95 | return "123456789012", "i-0a61f580df1acfa65", nil |
| 96 | } |
| 97 | probes := status.Probes{MachineID: "machine-1"} |
| 98 | for i := 0; i < 2; i++ { |
| 99 | err := groupStatusTick(ctx, s, reg, probes, false, "", groupEnv(env)) |
| 100 | if !errors.Is(err, errNoGroupReference) || !strings.Contains(err.Error(), "IMDS answered 500") { |
| 101 | t.Fatalf("tick %d: %v, want errNoGroupReference with its cause", i, err) |
| 102 | } |
| 103 | } |
| 104 | keys, _ := s.List(ctx, "status/boxes/") |
| 105 | got, _, _ := s.Get(ctx, "status/boxes/"+status.Label("machine-1")+".json") |
| 106 | if len(keys) != 1 || !bytes.Equal(got, victim) { |
| 107 | t.Fatalf("published with no reference: %v, victim now %s", keys, got) |
| 108 | } |
| 109 | if err := groupStatusTick(ctx, s, reg, probes, false, "", groupEnv(env)); err != nil { |
| 110 | t.Fatal(err) |
| 111 | } |
| 112 | ref := hostactors.GroupBoxReference("123456789012", "i-0a61f580df1acfa65") |
| 113 | if _, _, err := s.Get(ctx, "status/boxes/"+ref+".json"); err != nil || asked != 3 { |
| 114 | t.Fatalf("after recovery: %v, IMDS asked %d times", err, asked) |
| 115 | } |
| 116 | if got, _, _ := s.Get(ctx, "status/boxes/"+status.Label("machine-1")+".json"); !bytes.Equal(got, victim) { |
| 117 | t.Fatalf("the victim's report was replaced: %s", got) |
| 118 | } |
| 119 | }) |
| 120 | } |
| 121 | } |
| 122 | |
| 123 | // The ruling's third test: a box that is not a group box keeps its label |
| 124 | // and its status key byte for byte: statusLoop's tick is statusTick's, with |
| 125 | // statusLoop's own label (CollectBox's pseudonym, or a fleet box's label). |
| 126 | func TestABoxThatIsNotAGroupBoxKeepsItsLabelAndKey(t *testing.T) { |
| 127 | withGroupIdentity(t, "123456789012", "i-0a61f580df1acfa65", nil) |
| 128 | for name, env := range map[string]map[string]string{ |
| 129 | "no marker": {}, |
| 130 | "another value": {hostenv.BoxReference: "instance-id"}, |
| 131 | "region HA, no marker": {"CARLOS_REGION_HA": "1", "CARLOS_HOST_CLAIM": "1"}, |
| 132 | } { |
| 133 | getenv := groupEnv(env) |
| 134 | if got := metricsBoxLabel(getenv, "machine-1"); got != status.Label("machine-1") { |
| 135 | t.Fatalf("%s: label %q, want %q", name, got, status.Label("machine-1")) |
| 136 | } |
| 137 | for _, label := range []string{"", "kitchen"} { |
| 138 | ctx := context.Background() |
| 139 | reg, _, _ := statusFixture(t) |
| 140 | mine, theirs := release.NewDirStore(t.TempDir()), release.NewDirStore(t.TempDir()) |
| 141 | probes := status.Probes{MachineID: "machine-1"} |
| 142 | if err := groupStatusTick(ctx, mine, reg, probes, false, label, getenv); err != nil { |
| 143 | t.Fatal(err) |
| 144 | } |
| 145 | if err := statusTick(ctx, theirs, reg, probes, false, label); err != nil { |
| 146 | t.Fatal(err) |
| 147 | } |
| 148 | a, _ := mine.List(ctx, "status/boxes/") |
| 149 | b, _ := theirs.List(ctx, "status/boxes/") |
| 150 | if !reflect.DeepEqual(a, b) || len(a) != 1 { |
| 151 | t.Fatalf("%s, label %q: keys %v, statusTick's %v", name, label, a, b) |
| 152 | } |
| 153 | } |
| 154 | } |
| 155 | fleet := groupEnv(map[string]string{"CARLOS_FLEET": "pi", "CARLOS_FLEET_BOX": "kitchen"}) |
| 156 | if got := metricsBoxLabel(fleet, "machine-1"); got != "kitchen" { |
| 157 | t.Fatalf("fleet box: %q", got) |
| 158 | } |
| 159 | } |
| 160 | |
| 161 | // Task 8's gate relies on this: a marked box's label is "" only until its |
| 162 | // reference resolves, asked again on every call, and then the reference, |
| 163 | // kept, even if IMDS fails afterwards. |
| 164 | func TestAGroupBoxLabelIsEmptyOnlyUntilItsReferenceResolves(t *testing.T) { |
| 165 | group := groupEnv(map[string]string{hostenv.BoxReference: hostenv.GroupReference}) |
| 166 | withGroupIdentity(t, "", "", nil) |
| 167 | asked := 0 |
| 168 | groupIdentity = func() (string, string, error) { |
| 169 | asked++ |
| 170 | if asked != 3 { |
| 171 | return "", "", errors.New("IMDS answered 500") |
| 172 | } |
| 173 | return "123456789012", "i-0a61f580df1acfa65", nil |
| 174 | } |
| 175 | want := hostactors.GroupBoxReference("123456789012", "i-0a61f580df1acfa65") |
| 176 | for i, w := range []string{"", "", want, want, want} { |
| 177 | if got := metricsBoxLabel(group, "machine-1"); got != w { |
| 178 | t.Fatalf("call %d: %q, want %q", i, got, w) |
| 179 | } |
| 180 | } |
| 181 | if asked != 3 { |
| 182 | t.Fatalf("IMDS asked %d times, want 3 (a resolved reference is kept)", asked) |
| 183 | } |
| 184 | } |
| 185 | |
amadan