| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "encoding/base64" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "os" |
| 8 | "path/filepath" |
| 9 | "testing" |
| 10 | |
| 11 | "github.com/carlosframework/platform/internal/hostfloor" |
| 12 | ) |
| 13 | |
| 14 | func sharedGuardFixture(t *testing.T) (*applyFixture, string) { |
| 15 | t.Helper() |
| 16 | f, id := guardRecoveryFixture(t) |
| 17 | dir := systemGuardDir(f.a.optDir) |
| 18 | if err := os.Rename(f.a.floor.Dir, dir); err != nil { |
| 19 | t.Fatal(err) |
| 20 | } |
| 21 | f.a.floor.Dir = dir |
| 22 | f.a.regionHA = true |
| 23 | return f, id |
| 24 | } |
| 25 | |
| 26 | func TestSystemGuardRegionHAPreservesReservationThroughRecovery(t *testing.T) { |
| 27 | f, id := sharedGuardFixture(t) |
| 28 | a := f.a |
| 29 | if err := a.floor.Begin(t.Context(), id, a.dataDir); err != nil { |
| 30 | t.Fatal(err) |
| 31 | } |
| 32 | before, err := os.ReadFile(updateBegunPath(a.optDir)) |
| 33 | if err != nil { |
| 34 | t.Fatal(err) |
| 35 | } |
| 36 | if err := a.beginExecutableChange(t.Context(), "new-build", f.now); err != nil { |
| 37 | t.Fatal(err) |
| 38 | } |
| 39 | a.settleUpdateBegun(t.Context()) |
| 40 | after, err := os.ReadFile(updateBegunPath(a.optDir)) |
| 41 | if err != nil || string(after) != string(before) { |
| 42 | t.Fatalf("region HA changed the floor reservation: %s, %v", after, err) |
| 43 | } |
| 44 | if err := a.journal(systemJournal{Phase: systemPhaseBackingUp, Target: "new-build", PrevBuild: "old-build", StartedAt: f.now}); err != nil { |
| 45 | t.Fatal(err) |
| 46 | } |
| 47 | if err := a.Run(t.Context(), true); err != nil { |
| 48 | t.Fatal(err) |
| 49 | } |
| 50 | completed, err := os.ReadFile(filepath.Join(a.floor.Dir, "completed.json")) |
| 51 | if err != nil || string(completed) != id { |
| 52 | t.Fatalf("completion = %q, %v", completed, err) |
| 53 | } |
| 54 | if _, err := os.Stat(updateBegunPath(a.optDir)); !os.IsNotExist(err) { |
| 55 | t.Fatalf("marker remains after recovery: %v", err) |
| 56 | } |
| 57 | } |
| 58 | |
| 59 | func TestSystemGuardRegionHAKeepsFailedCompletionEvidence(t *testing.T) { |
| 60 | for _, failure := range []string{"completion", "controller", "marker"} { |
| 61 | t.Run(failure, func(t *testing.T) { |
| 62 | f, id := sharedGuardFixture(t) |
| 63 | a := f.a |
| 64 | if err := a.floor.Begin(t.Context(), id, a.dataDir); err != nil { |
| 65 | t.Fatal(err) |
| 66 | } |
| 67 | if err := os.MkdirAll(a.stagingDir(), 0755); err != nil { |
| 68 | t.Fatal(err) |
| 69 | } |
| 70 | evidence := filepath.Join(a.stagingDir(), "evidence") |
| 71 | if err := os.WriteFile(evidence, []byte("retain"), 0600); err != nil { |
| 72 | t.Fatal(err) |
| 73 | } |
| 74 | switch failure { |
| 75 | case "completion": |
| 76 | if err := os.Mkdir(filepath.Join(a.floor.Dir, "completed.json"), 0755); err != nil { |
| 77 | t.Fatal(err) |
| 78 | } |
| 79 | case "controller": |
| 80 | if err := os.WriteFile(filepath.Join(a.dataDir, hostexecDir, "pending.json"), []byte(`{"state":"pending"}`), 0600); err != nil { |
| 81 | t.Fatal(err) |
| 82 | } |
| 83 | if err := a.beginExecutableChange(t.Context(), "new-build", f.now); !errors.Is(err, errSystemDeferred) { |
| 84 | t.Fatalf("late participation = %v", err) |
| 85 | } |
| 86 | case "marker": |
| 87 | if err := os.WriteFile(updateBegunPath(a.optDir), []byte(`{"target":"new-build"}`), 0600); err != nil { |
| 88 | t.Fatal(err) |
| 89 | } |
| 90 | if err := a.beginExecutableChange(t.Context(), "new-build", f.now); err == nil { |
| 91 | t.Fatal("region marker accepted for an armed transaction") |
| 92 | } |
| 93 | } |
| 94 | a.settleUpdateBegun(t.Context()) |
| 95 | if err := a.Run(t.Context(), true); err == nil { |
| 96 | t.Fatal("unsafe recovery succeeded") |
| 97 | } |
| 98 | for _, path := range []string{updateBegunPath(a.optDir), evidence} { |
| 99 | if _, err := os.Stat(path); err != nil { |
| 100 | t.Fatalf("recovery discarded %s: %v", path, err) |
| 101 | } |
| 102 | } |
| 103 | }) |
| 104 | } |
| 105 | } |
| 106 | |
| 107 | func TestSystemGuardUnpublishedKeepsRegionHAParticipation(t *testing.T) { |
| 108 | f, _ := applyHappy(t) |
| 109 | a := f.a |
| 110 | a.regionHA = true |
| 111 | a.floor = &hostfloor.Disk{Dir: systemGuardDir(a.optDir), Bucket: "bucket", PublicKey: f.pub, Participation: a.participation} |
| 112 | if err := os.MkdirAll(a.floor.Dir, 0755); err != nil { |
| 113 | t.Fatal(err) |
| 114 | } |
| 115 | raw, _ := json.Marshal(hostfloor.Policy{Schema: 1, DeploymentBucket: "bucket", Mode: "unpublished"}) |
| 116 | if err := hostfloor.Durable(filepath.Join(a.floor.Dir, "policy.json"), raw, 0644); err != nil { |
| 117 | t.Fatal(err) |
| 118 | } |
| 119 | if err := os.WriteFile(filepath.Join(hostexec(t, f), "pending.json"), []byte(`{}`), 0600); err != nil { |
| 120 | t.Fatal(err) |
| 121 | } |
| 122 | if err := a.beginExecutableChange(t.Context(), "v2", f.now); !errors.Is(err, errSystemDeferred) { |
| 123 | t.Fatalf("unpublished floor bypassed participation: %v", err) |
| 124 | } |
| 125 | if _, err := os.Stat(updateBegunPath(a.optDir)); !os.IsNotExist(err) { |
| 126 | t.Fatalf("refused unarmed update retained marker: %v", err) |
| 127 | } |
| 128 | if err := os.Remove(filepath.Join(a.dataDir, hostexecDir, "pending.json")); err != nil { |
| 129 | t.Fatal(err) |
| 130 | } |
| 131 | if err := a.beginExecutableChange(t.Context(), "v2", f.now); err != nil { |
| 132 | t.Fatal(err) |
| 133 | } |
| 134 | raw, err := os.ReadFile(updateBegunPath(a.optDir)) |
| 135 | var begun updateBegun |
| 136 | if err != nil || json.Unmarshal(raw, &begun) != nil || begun.Target != "v2" { |
| 137 | t.Fatalf("unpublished marker = %s, %v", raw, err) |
| 138 | } |
| 139 | a.settleUpdateBegun(t.Context()) |
| 140 | if _, err := os.Stat(updateBegunPath(a.optDir)); !os.IsNotExist(err) { |
| 141 | t.Fatalf("completed unarmed update retained marker: %v", err) |
| 142 | } |
| 143 | if err := os.MkdirAll(a.stagingDir(), 0755); err != nil { |
| 144 | t.Fatal(err) |
| 145 | } |
| 146 | if err := os.WriteFile(a.file("journal.json"), []byte("{"), 0600); err != nil { |
| 147 | t.Fatal(err) |
| 148 | } |
| 149 | if err := a.Run(t.Context(), true); err == nil { |
| 150 | t.Fatal("malformed journal recovered") |
| 151 | } |
| 152 | if _, err := os.Stat(a.stagingDir()); err != nil { |
| 153 | t.Fatalf("unresolved root recovery removed staging: %v", err) |
| 154 | } |
| 155 | if _, err := os.Stat(a.file("journal.json")); err != nil { |
| 156 | t.Fatalf("malformed journal discarded: %v", err) |
| 157 | } |
| 158 | } |
| 159 | |
| 160 | func TestSystemGuardHookRecognizesOnlyLegacyRegionState(t *testing.T) { |
| 161 | for _, state := range []string{"absent", "empty", "region-marker", "floor-marker", "malformed-marker", "tmp-only", "marker+tmp", "floor-marker+tmp", "checker", "identities", "activation", "completed", "policy", "symlink"} { |
| 162 | t.Run(state, func(t *testing.T) { |
| 163 | f := newApplyFixture(t, "stable", "v1") |
| 164 | t.Setenv("CARLOS_RELEASE_PUBKEY", base64.StdEncoding.EncodeToString(f.pub)) |
| 165 | t.Setenv("CARLOS_DEPLOYMENT_BUCKET", "bucket") |
| 166 | dir := systemGuardDir(f.a.optDir) |
| 167 | if state == "symlink" { |
| 168 | if err := os.Symlink(t.TempDir(), dir); err != nil { |
| 169 | t.Skip(err) |
| 170 | } |
| 171 | } else if state != "absent" { |
| 172 | if err := os.MkdirAll(dir, 0755); err != nil { |
| 173 | t.Fatal(err) |
| 174 | } |
| 175 | } |
| 176 | name, body := "", "" |
| 177 | switch state { |
| 178 | case "region-marker": |
| 179 | name, body = updateBegunFile, `{"target":"v2"}` |
| 180 | case "floor-marker": |
| 181 | name, body = updateBegunFile, `{"id":"activation"}` |
| 182 | case "malformed-marker": |
| 183 | name, body = updateBegunFile, `{` |
| 184 | case "tmp-only": |
| 185 | name, body = updateBegunFile+".tmp", `{"tar` |
| 186 | case "marker+tmp", "floor-marker+tmp": |
| 187 | name, body = updateBegunFile, `{"target":"v2"}` |
| 188 | if state == "floor-marker+tmp" { |
| 189 | body = `{"id":"activation"}` |
| 190 | } |
| 191 | if err := os.WriteFile(filepath.Join(dir, updateBegunFile+".tmp"), []byte(`{"tar`), 0644); err != nil { |
| 192 | t.Fatal(err) |
| 193 | } |
| 194 | case "checker": |
| 195 | name = "carlos" |
| 196 | case "identities": |
| 197 | if err := os.Mkdir(filepath.Join(dir, "identities"), 0755); err != nil { |
| 198 | t.Fatal(err) |
| 199 | } |
| 200 | case "activation", "completed", "policy": |
| 201 | name = state + ".json" |
| 202 | } |
| 203 | if name != "" { |
| 204 | if err := os.WriteFile(filepath.Join(dir, name), []byte(body), 0644); err != nil { |
| 205 | t.Fatal(err) |
| 206 | } |
| 207 | } |
| 208 | err := installFloorHook(f.a) |
| 209 | // A crash inside main's durable marker write leaves the .tmp or a |
| 210 | // torn marker; neither is floor state (Opus merge review M1-1). |
| 211 | legacy := state == "absent" || state == "empty" || state == "region-marker" || state == "malformed-marker" || state == "tmp-only" || state == "marker+tmp" |
| 212 | if legacy { |
| 213 | if err != nil || f.a.guarded() { |
| 214 | t.Fatalf("legacy state armed floor: %v", err) |
| 215 | } |
| 216 | if err := f.a.Run(t.Context(), true); err != nil { |
| 217 | t.Fatal(err) |
| 218 | } |
| 219 | if _, err := os.Stat(updateBegunPath(f.a.optDir)); !os.IsNotExist(err) { |
| 220 | t.Fatalf("legacy recovery kept marker: %v", err) |
| 221 | } |
| 222 | } else if err == nil { |
| 223 | t.Fatal("missing or invalid floor policy accepted") |
| 224 | } |
| 225 | }) |
| 226 | } |
| 227 | } |
| 228 | |
| 229 | // The floor decision is read under the apply lock: a locked-out run never |
| 230 | // looks at a guard directory another run may be writing (Opus merge M1-1). |
| 231 | func TestSystemFloorHookRunsOnlyUnderTheApplyLock(t *testing.T) { |
| 232 | f := newApplyFixture(t, "stable", "v1") |
| 233 | f.promote("v2", []byte("x"), false) |
| 234 | called := 0 |
| 235 | f.a.floorHook = func(*systemApplier) error { called++; return nil } |
| 236 | held, err := os.OpenFile(f.a.lockPath(), os.O_CREATE|os.O_RDWR, 0o600) |
| 237 | if err != nil { |
| 238 | t.Fatal(err) |
| 239 | } |
| 240 | if err := flockExclusiveNonblock(held); err != nil { |
| 241 | t.Fatal(err) |
| 242 | } |
| 243 | if err := f.a.Run(t.Context(), false); err != nil || called != 0 { |
| 244 | t.Fatalf("locked-out run: err=%v hook calls=%d", err, called) |
| 245 | } |
| 246 | held.Close() |
| 247 | if err := f.a.Run(t.Context(), false); err != nil || called != 1 { |
| 248 | t.Fatalf("lock-holding run: err=%v hook calls=%d", err, called) |
| 249 | } |
| 250 | } |
| 251 | |