| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "bytes" |
| 5 | "context" |
| 6 | "crypto/sha256" |
| 7 | "encoding/hex" |
| 8 | "errors" |
| 9 | "fmt" |
| 10 | "io" |
| 11 | "os" |
| 12 | "path/filepath" |
| 13 | "strings" |
| 14 | "sync" |
| 15 | "testing" |
| 16 | "time" |
| 17 | |
| 18 | "github.com/carlosframework/platform/internal/activator" |
| 19 | "github.com/carlosframework/platform/internal/appconfig" |
| 20 | "github.com/carlosframework/platform/internal/hostbox" |
| 21 | "github.com/carlosframework/platform/internal/hostclaim" |
| 22 | "github.com/carlosframework/platform/internal/lease" |
| 23 | "github.com/carlosframework/platform/internal/moves" |
| 24 | "github.com/carlosframework/platform/internal/park" |
| 25 | "github.com/carlosframework/platform/internal/provision" |
| 26 | "github.com/carlosframework/platform/internal/registry" |
| 27 | "github.com/carlosframework/platform/internal/release" |
| 28 | "github.com/superfly/ltx" |
| 29 | ) |
| 30 | |
| 31 | // The Phase 1 harness: the legacy moveHarness, its fake mirror extended |
| 32 | // with the staging writer's guarded At methods (F17) and a call log, a |
| 33 | // lease.VersionedStore over the same buckets, a staging-lease fake, a |
| 34 | // stepping clock, a MoveStart'ed record, and the host claim the request |
| 35 | // door would have written. fakeMoveMirror deliberately does NOT gain |
| 36 | // CopyVersioned: MirrorPrefix would then run the legacy tests in its |
| 37 | // versioned mode. |
| 38 | |
| 39 | // stagingCall is one guarded staging request the fake received. |
| 40 | type stagingCall struct { |
| 41 | Op, Bucket, Key string |
| 42 | SignAt time.Time |
| 43 | } |
| 44 | |
| 45 | func (f *fakeMoveMirror) logAt(op string, signAt time.Time, bucket, key string) error { |
| 46 | f.mu.Lock() |
| 47 | defer f.mu.Unlock() |
| 48 | f.at = append(f.at, stagingCall{op, bucket, key, signAt}) |
| 49 | if f.failAt != nil { |
| 50 | return f.failAt(op, key) |
| 51 | } |
| 52 | return nil |
| 53 | } |
| 54 | |
| 55 | func (f *fakeMoveMirror) nextVersion() string { |
| 56 | f.mu.Lock() |
| 57 | defer f.mu.Unlock() |
| 58 | f.versions++ |
| 59 | return fmt.Sprintf("ver-%04d", f.versions) |
| 60 | } |
| 61 | |
| 62 | func (f *fakeMoveMirror) CopyVersionedAt(ctx context.Context, signAt time.Time, srcBucket, srcKey, srcETag, dstBucket, dstKey string) (string, string, error) { |
| 63 | if err := f.logAt("copy", signAt, dstBucket, dstKey); err != nil { |
| 64 | return "", "", err |
| 65 | } |
| 66 | etag, err := f.Copy(ctx, srcBucket, srcKey, srcETag, dstBucket, dstKey) |
| 67 | if err != nil { |
| 68 | return "", "", err |
| 69 | } |
| 70 | ver := f.nextVersion() |
| 71 | f.setVersion(dstBucket, dstKey, ver) |
| 72 | return etag, ver, nil |
| 73 | } |
| 74 | |
| 75 | func (f *fakeMoveMirror) putAt(op string, signAt time.Time, bucket, key string, body []byte, createOnly bool) (string, error) { |
| 76 | if err := f.logAt(op, signAt, bucket, key); err != nil { |
| 77 | return "", err |
| 78 | } |
| 79 | f.bmu.Lock() |
| 80 | _, exists := f.buckets[bucket][key] |
| 81 | f.bmu.Unlock() |
| 82 | if exists && createOnly { |
| 83 | return "", lease.ErrExists |
| 84 | } |
| 85 | ver := f.nextVersion() |
| 86 | f.store(bucket, key, fakeMoveObject{lease.MirrorObject{ETag: "etag-" + ver, Size: int64(len(body))}, string(body)}) |
| 87 | f.setVersion(bucket, key, ver) |
| 88 | return ver, nil |
| 89 | } |
| 90 | |
| 91 | // setVersion records the current version of bucket/key. The legacy |
| 92 | // fakeMoveObject has no version field (its positional literals stay as |
| 93 | // they are), so versions live beside it. |
| 94 | func (f *fakeMoveMirror) setVersion(bucket, key, ver string) { |
| 95 | f.mu.Lock() |
| 96 | defer f.mu.Unlock() |
| 97 | if f.vers == nil { |
| 98 | f.vers = map[string]string{} |
| 99 | } |
| 100 | f.vers[bucket+"/"+key] = ver |
| 101 | if f.keepHistory { |
| 102 | if f.history == nil { |
| 103 | f.history = map[string]fakeMoveObject{} |
| 104 | } |
| 105 | f.bmu.Lock() |
| 106 | f.history[bucket+"/"+key+"@"+ver] = f.buckets[bucket][key] |
| 107 | f.bmu.Unlock() |
| 108 | } |
| 109 | } |
| 110 | |
| 111 | // older is a non-current version of bucket/key the fake kept (keepHistory), |
| 112 | // as a versioned bucket keeps every version a later write superseded. |
| 113 | func (f *fakeMoveMirror) older(bucket, key, ver string) (fakeMoveObject, bool) { |
| 114 | f.mu.Lock() |
| 115 | defer f.mu.Unlock() |
| 116 | o, ok := f.history[bucket+"/"+key+"@"+ver] |
| 117 | return o, ok && f.keepHistory |
| 118 | } |
| 119 | |
| 120 | func (f *fakeMoveMirror) versionOf(bucket, key string) string { |
| 121 | f.mu.Lock() |
| 122 | defer f.mu.Unlock() |
| 123 | return f.vers[bucket+"/"+key] |
| 124 | } |
| 125 | |
| 126 | func (f *fakeMoveMirror) PutVersionedAt(_ context.Context, signAt time.Time, bucket, key string, body []byte) (string, error) { |
| 127 | return f.putAt("put", signAt, bucket, key, body, false) |
| 128 | } |
| 129 | |
| 130 | func (f *fakeMoveMirror) PutReaderVersionedAt(_ context.Context, signAt time.Time, bucket, key string, r io.Reader, size int64, sha256Hex string) (string, error) { |
| 131 | body, err := io.ReadAll(io.LimitReader(r, size+1)) |
| 132 | if err != nil { |
| 133 | return "", err |
| 134 | } |
| 135 | if int64(len(body)) != size { |
| 136 | return "", fmt.Errorf("fake: body is %d bytes, declared %d", len(body), size) |
| 137 | } |
| 138 | if sum := sha256.Sum256(body); hex.EncodeToString(sum[:]) != sha256Hex { |
| 139 | return "", errors.New("fake: XAmzContentSHA256Mismatch") |
| 140 | } |
| 141 | return f.putAt("put", signAt, bucket, key, body, false) |
| 142 | } |
| 143 | |
| 144 | func (f *fakeMoveMirror) PutIfAbsentVersionedAt(_ context.Context, signAt time.Time, bucket, key string, body []byte) (string, error) { |
| 145 | return f.putAt("put-if-absent", signAt, bucket, key, body, true) |
| 146 | } |
| 147 | |
| 148 | func (f *fakeMoveMirror) DeleteAt(ctx context.Context, signAt time.Time, bucket, key string) error { |
| 149 | if err := f.logAt("delete", signAt, bucket, key); err != nil { |
| 150 | return err |
| 151 | } |
| 152 | return f.Delete(ctx, bucket, key) |
| 153 | } |
| 154 | |
| 155 | var _ lease.StagingWriter = (*fakeMoveMirror)(nil) |
| 156 | |
| 157 | // stagingCalls is the guarded call log, by op and key. |
| 158 | func (f *fakeMoveMirror) stagingCalls() []stagingCall { |
| 159 | f.mu.Lock() |
| 160 | defer f.mu.Unlock() |
| 161 | return append([]stagingCall(nil), f.at...) |
| 162 | } |
| 163 | |
| 164 | // fakeVersioned is a lease.VersionedStore over the fake mirror's buckets, |
| 165 | // reading by exact version. deny refuses a call the way an IAM policy does. |
| 166 | type fakeVersioned struct { |
| 167 | f *fakeMoveMirror |
| 168 | mu sync.Mutex |
| 169 | calls []string |
| 170 | deny func(op, bucket, key string) error |
| 171 | // skipDelete, when it answers true for a key, makes DeleteVersion |
| 172 | // return success without removing the version: a delete the store |
| 173 | // accepted but whose listing still shows the version (Plan 2 Task 10, |
| 174 | // the landing's Parked gate). |
| 175 | skipDelete func(key string) bool |
| 176 | } |
| 177 | |
| 178 | func (v *fakeVersioned) call(op, bucket, key string) error { |
| 179 | v.mu.Lock() |
| 180 | v.calls = append(v.calls, op+" "+bucket+"/"+key) |
| 181 | v.mu.Unlock() |
| 182 | if v.deny != nil { |
| 183 | return v.deny(op, bucket, key) |
| 184 | } |
| 185 | return nil |
| 186 | } |
| 187 | |
| 188 | func (v *fakeVersioned) ListVersions(_ context.Context, bucket, prefix string) ([]lease.ObjectVersion, error) { |
| 189 | if err := v.call("list-versions", bucket, prefix); err != nil { |
| 190 | return nil, err |
| 191 | } |
| 192 | var out []lease.ObjectVersion |
| 193 | for k, o := range v.f.buckets[bucket] { |
| 194 | if strings.HasPrefix(k, prefix) { |
| 195 | out = append(out, lease.ObjectVersion{Key: k, VersionID: v.versionID(bucket, k), ETag: o.ETag, Size: o.Size, IsLatest: true}) |
| 196 | } |
| 197 | } |
| 198 | return out, nil |
| 199 | } |
| 200 | |
| 201 | // versionID is an object's version as S3 lists it: one written with no |
| 202 | // version (the source's replica, stored directly by a test) is "null", |
| 203 | // as S3 lists an object written while a bucket was not versioning. |
| 204 | func (v *fakeVersioned) versionID(bucket, key string) string { |
| 205 | if ver := v.f.versionOf(bucket, key); ver != "" { |
| 206 | return ver |
| 207 | } |
| 208 | return "null" |
| 209 | } |
| 210 | |
| 211 | func (v *fakeVersioned) GetVersion(_ context.Context, bucket, key, versionID string) ([]byte, error) { |
| 212 | if err := v.call("get-version", bucket, key); err != nil { |
| 213 | return nil, err |
| 214 | } |
| 215 | o, ok := v.f.buckets[bucket][key] |
| 216 | if !ok || versionID == "" || v.versionID(bucket, key) != versionID { |
| 217 | if old, kept := v.f.older(bucket, key, versionID); kept && versionID != "" { |
| 218 | return []byte(old.body), nil |
| 219 | } |
| 220 | return nil, lease.ErrMirrorNotFound |
| 221 | } |
| 222 | return []byte(o.body), nil |
| 223 | } |
| 224 | |
| 225 | func (v *fakeVersioned) OpenVersion(ctx context.Context, bucket, key, versionID string) (io.ReadCloser, error) { |
| 226 | b, err := v.GetVersion(ctx, bucket, key, versionID) |
| 227 | if err != nil { |
| 228 | return nil, err |
| 229 | } |
| 230 | return io.NopCloser(bytes.NewReader(b)), nil |
| 231 | } |
| 232 | |
| 233 | // CopyVersion copies one exact version within the fake's buckets, as the |
| 234 | // park engine archives a landing's residue (Plan 2 Task 10). createOnly |
| 235 | // refuses an existing destination with lease.ErrDestinationExists. |
| 236 | func (v *fakeVersioned) CopyVersion(_ context.Context, srcBucket, srcKey, srcVersionID, dstBucket, dstKey string, createOnly bool) (string, string, error) { |
| 237 | if err := v.call("copy-version", dstBucket, dstKey); err != nil { |
| 238 | return "", "", err |
| 239 | } |
| 240 | o, ok := v.f.buckets[srcBucket][srcKey] |
| 241 | if !ok || srcVersionID == "" || v.versionID(srcBucket, srcKey) != srcVersionID { |
| 242 | old, kept := v.f.older(srcBucket, srcKey, srcVersionID) |
| 243 | if !kept || srcVersionID == "" { |
| 244 | return "", "", lease.ErrMirrorNotFound |
| 245 | } |
| 246 | o = old |
| 247 | } |
| 248 | if _, exists := v.f.buckets[dstBucket][dstKey]; exists && createOnly { |
| 249 | return "", "", lease.ErrDestinationExists |
| 250 | } |
| 251 | ver := v.f.nextVersion() |
| 252 | v.f.store(dstBucket, dstKey, fakeMoveObject{lease.MirrorObject{ETag: "etag-" + ver, Size: o.Size}, o.body}) |
| 253 | v.f.setVersion(dstBucket, dstKey, ver) |
| 254 | return ver, "etag-" + ver, nil |
| 255 | } |
| 256 | |
| 257 | func (v *fakeVersioned) DeleteVersion(_ context.Context, bucket, key, versionID string) error { |
| 258 | if err := v.call("delete-version", bucket, key); err != nil { |
| 259 | return err |
| 260 | } |
| 261 | if v.skipDelete != nil && v.skipDelete(key) { |
| 262 | return nil |
| 263 | } |
| 264 | // "null" is how S3 names an object written while the bucket was not |
| 265 | // versioning (versionID), and deleting that version removes it. |
| 266 | if _, ok := v.f.buckets[bucket][key]; ok && v.versionID(bucket, key) == versionID { |
| 267 | delete(v.f.buckets[bucket], key) |
| 268 | } |
| 269 | return nil |
| 270 | } |
| 271 | |
| 272 | func (v *fakeVersioned) PutIfAbsent(_ context.Context, bucket, key string, body []byte) (string, error) { |
| 273 | if err := v.call("put-if-absent", bucket, key); err != nil { |
| 274 | return "", err |
| 275 | } |
| 276 | if _, ok := v.f.buckets[bucket][key]; ok { |
| 277 | return "", lease.ErrExists |
| 278 | } |
| 279 | ver := v.f.nextVersion() |
| 280 | v.f.store(bucket, key, fakeMoveObject{lease.MirrorObject{ETag: "etag-" + ver, Size: int64(len(body))}, string(body)}) |
| 281 | v.f.setVersion(bucket, key, ver) |
| 282 | return ver, nil |
| 283 | } |
| 284 | |
| 285 | func (v *fakeVersioned) callsOn(prefix string) []string { |
| 286 | v.mu.Lock() |
| 287 | defer v.mu.Unlock() |
| 288 | var out []string |
| 289 | for _, c := range v.calls { |
| 290 | if strings.Contains(c, prefix) { |
| 291 | out = append(out, c) |
| 292 | } |
| 293 | } |
| 294 | return out |
| 295 | } |
| 296 | |
| 297 | // fakeStagingLease is the staging-work lease: safe until `until`. |
| 298 | type fakeStagingLease struct { |
| 299 | mu sync.Mutex |
| 300 | until time.Time |
| 301 | acquired int |
| 302 | released int |
| 303 | // unsafeAfter, when positive, makes every SafeUntil read after that |
| 304 | // many answer the lease's start: too close to its end to sign. |
| 305 | reads int |
| 306 | unsafeAfter int |
| 307 | } |
| 308 | |
| 309 | func (l *fakeStagingLease) SafeUntil() (time.Time, error) { |
| 310 | l.mu.Lock() |
| 311 | defer l.mu.Unlock() |
| 312 | l.reads++ |
| 313 | if l.unsafeAfter > 0 && l.reads > l.unsafeAfter { |
| 314 | return l.until.Add(-moves.StagingLeaseTTL), nil |
| 315 | } |
| 316 | return l.until, nil |
| 317 | } |
| 318 | |
| 319 | // stepClock advances a millisecond each read, so every timed span is |
| 320 | // positive and the order of events is the order of reads. |
| 321 | type stepClock struct { |
| 322 | mu sync.Mutex |
| 323 | now time.Time |
| 324 | } |
| 325 | |
| 326 | func (c *stepClock) Now() time.Time { |
| 327 | c.mu.Lock() |
| 328 | defer c.mu.Unlock() |
| 329 | c.now = c.now.Add(time.Millisecond) |
| 330 | return c.now |
| 331 | } |
| 332 | |
| 333 | func (c *stepClock) Advance(d time.Duration) { |
| 334 | c.mu.Lock() |
| 335 | c.now = c.now.Add(d) |
| 336 | c.mu.Unlock() |
| 337 | } |
| 338 | |
| 339 | const ( |
| 340 | p1Host = "a.go.example.com" |
| 341 | p1ID = "mv_abcdefghijklmnopqrstuvwxyz" |
| 342 | p1ID2 = "mv_bcdefghijklmnopqrstuvwxyza" |
| 343 | ) |
| 344 | |
| 345 | var p1Created = time.Date(2026, 10, 1, 0, 0, 0, 0, time.UTC) |
| 346 | |
| 347 | type phase1Harness struct { |
| 348 | *moveHarness |
| 349 | vs *fakeVersioned |
| 350 | lease *fakeStagingLease |
| 351 | clock *stepClock |
| 352 | inc string |
| 353 | route registry.Route |
| 354 | env map[string]string // the source's launch environment |
| 355 | optDir string |
| 356 | probe moves.Check |
| 357 | probePaths []string |
| 358 | strict moves.Check |
| 359 | // worker1 is the source box's worker and worker2 the target's, each |
| 360 | // kept across passes as one agent keeps its own. |
| 361 | worker1, worker2 *movesWorker |
| 362 | // src is the source box's placement code: "ie", or the home's "" for a |
| 363 | // move out of home (fromHome). |
| 364 | src string |
| 365 | // The source box, its data bucket and the bucket's AWS region: ie-box, |
| 366 | // data-ie and eu-west-1 by default. |
| 367 | srcBox, srcBucket, srcAWS string |
| 368 | // dst is the target box's placement code: "uk", or the home's "" for a |
| 369 | // move into home (intoHome). dstBox, dstBucket and dstAWS name the |
| 370 | // target box, its data bucket and the bucket's AWS region. |
| 371 | dst, dstBox, dstBucket, dstAWS string |
| 372 | // dstEngine is the target box's claim engine region, its roster name: |
| 373 | // "uk", or the home box's roster name "ie" for a move into home (claim |
| 374 | // Plan 13 keeps the roster name as a home box's engine region and grants |
| 375 | // it the @home alias). |
| 376 | dstEngine string |
| 377 | // homeRegion is the fleet's signed home designation, "" when the |
| 378 | // fixture names none. With it the claim's op carries home_region and the |
| 379 | // target's ParkHomeRegion is set, as armMovesParks sets it. |
| 380 | homeRegion string |
| 381 | // admitted starts the move as the claim-mode door does |
| 382 | // (hostclaim.BeginAdmittedMove with the roster's enrollments), not the |
| 383 | // plain BeginMove. |
| 384 | admitted bool |
| 385 | // roster is the signed roster the fixture's boxes resolve through. |
| 386 | roster p1Roster |
| 387 | // retain is the claim state seeded at start as an earlier stay left it |
| 388 | // (retainTouched, retainQuiet, retainChore). |
| 389 | retain p1Retained |
| 390 | // realEnv composes both sides' environments through the production |
| 391 | // composer (realLaunchEnv): the source's launch report is its |
| 392 | // ExecBackend's LaunchEnvironment, the target's held env composeHeldEnv |
| 393 | // with the target's own region, each box's agent-owned variables its |
| 394 | // own. |
| 395 | realEnv bool |
| 396 | // srcEnvFile is the source's own <host>.env, the instance layer the |
| 397 | // archive carries: "APP_MODE=prod\n" by default (sourceEnvFile). |
| 398 | srcEnvFile string |
| 399 | } |
| 400 | |
| 401 | // sourceEnvFile sets the source's own <host>.env. |
| 402 | func sourceEnvFile(body string) p1Option { return func(h *phase1Harness) { h.srcEnvFile = body } } |
| 403 | |
| 404 | // realLaunchEnv composes both boxes' environments through the production |
| 405 | // composer instead of the h.env fixture (home part 2, Task 2b): the source |
| 406 | // pins what its ExecBackend would launch, and the target composes its own, |
| 407 | // so the box-owned variables differ between the two exactly as they do in |
| 408 | // production. |
| 409 | func realLaunchEnv(h *phase1Harness) { h.realEnv = true } |
| 410 | |
| 411 | // heldEnv is the target's held environment for rec: the h.env fixture with |
| 412 | // the target's region rule, or under realLaunchEnv the production |
| 413 | // composition over the carried env with the target's region. |
| 414 | func (h *phase1Harness) heldEnv(ctx context.Context, s release.Store, rec provision.Record, carried []byte) (map[string]string, map[string]bool, error) { |
| 415 | if h.realEnv { |
| 416 | env, instance, _, _, err := composeHeldEnv(h.cfg.SecretsPriv, h.dst, false)(ctx, s, rec, carried) |
| 417 | return env, instance, err |
| 418 | } |
| 419 | env := map[string]string{} |
| 420 | for k, v := range h.env { |
| 421 | env[k] = v |
| 422 | } |
| 423 | setHeldRegion(env, h.dst) |
| 424 | return env, nil, nil |
| 425 | } |
| 426 | |
| 427 | // p1Roster is the fixture's signed roster: each box's actor, as |
| 428 | // hostbox.Roster answers it. |
| 429 | type p1Roster map[string]hostbox.Actor |
| 430 | |
| 431 | func (r p1Roster) Actor(_ context.Context, box string) (hostbox.Actor, error) { |
| 432 | a, ok := r[box] |
| 433 | if !ok { |
| 434 | return hostbox.Actor{}, hostbox.ErrChanged |
| 435 | } |
| 436 | return a, nil |
| 437 | } |
| 438 | |
| 439 | // p1Retained is claim state an earlier stay left, seeded at start in |
| 440 | // whichever spelling of home the option names: touched.regions, a |
| 441 | // quiet_after key, and an earlier residue chore owed in the home region. |
| 442 | type p1Retained struct { |
| 443 | touched, quiet, chore string |
| 444 | } |
| 445 | |
| 446 | // retainedChoreOp is the earlier op whose residue chore retainChore seeds, |
| 447 | // and retainedChoreGeneration the earlier stay's generation it parks. |
| 448 | const ( |
| 449 | retainedChoreOp = "mv_earlierstayearlierstayearlie" |
| 450 | retainedChoreGeneration = "old" |
| 451 | ) |
| 452 | |
| 453 | // The fixture's enrollments (the roster's 32-hex tokens). |
| 454 | const ( |
| 455 | p1SrcEnrollment = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa1" |
| 456 | p1DstEnrollment = "aaaaaaaaaaaaaaaaaaaaaaaaaaaaaaa2" |
| 457 | ) |
| 458 | |
| 459 | // p1Option adjusts a Phase 1 harness before it is built. |
| 460 | type p1Option func(*phase1Harness) |
| 461 | |
| 462 | // fromHome builds the harness for a move out of the home placement: the |
| 463 | // source box, its record and its move name the home's code, "" (home-region |
| 464 | // parks, part 1). The source box keeps its name and bucket. |
| 465 | func fromHome(h *phase1Harness) { h.src = "" } |
| 466 | |
| 467 | // intoHome builds the harness for a move into the home placement, the |
| 468 | // mirror of fromHome for the target (home part 2). The source is placed in a |
| 469 | // named code that is not the home's name (ca-box in ca, bucket data-ca), so |
| 470 | // no fixture name suggests home. The target is home-box: placement "", |
| 471 | // bucket data-home in eu-west-1, an edges file with a data bucket and no |
| 472 | // placement, and a held environment with no CARLOS_REGION. The roster names |
| 473 | // home-box in region ie and designates ie the fleet's home; the claim's op |
| 474 | // carries home_region ie (BeginAdmittedMove), and the target's |
| 475 | // ParkHomeRegion is ie, as armMovesParks sets it. |
| 476 | func intoHome(h *phase1Harness) { |
| 477 | h.src, h.srcBox, h.srcBucket, h.srcAWS = "ca", "ca-box", "data-ca", "ca-central-1" |
| 478 | h.dst, h.dstBox, h.dstBucket, h.dstAWS, h.dstEngine = "", "home-box", "data-home", "eu-west-1", "ie" |
| 479 | h.homeRegion, h.admitted = "ie", true |
| 480 | } |
| 481 | |
| 482 | // retainTouched seeds touched.regions with region, an earlier stay's |
| 483 | // spelling of home ("ie" or "@home"). |
| 484 | func retainTouched(region string) p1Option { |
| 485 | return func(h *phase1Harness) { h.retain.touched = region } |
| 486 | } |
| 487 | |
| 488 | // retainQuiet seeds a past quiet_after key under region, either spelling. |
| 489 | func retainQuiet(region string) p1Option { return func(h *phase1Harness) { h.retain.quiet = region } } |
| 490 | |
| 491 | // retainChore seeds an earlier op's residue chore owed under region (the |
| 492 | // roster name "ie" for home), after the start, as an earlier stay left it: |
| 493 | // a park-generation of the generation retainedChoreGeneration, scoped to |
| 494 | // the holder's location, the shape hostclaim.QueueResidueParks queues. |
| 495 | func retainChore(region string) p1Option { return func(h *phase1Harness) { h.retain.chore = region } } |
| 496 | |
| 497 | func newPhase1Harness(t *testing.T, opts ...p1Option) *phase1Harness { |
| 498 | t.Helper() |
| 499 | h := &phase1Harness{moveHarness: newMoveHarness(t), lease: &fakeStagingLease{}, clock: &stepClock{now: time.Date(2026, 10, 7, 9, 0, 0, 0, time.UTC)}, src: "ie", |
| 500 | srcEnvFile: "APP_MODE=prod\n", srcBox: "ie-box", srcBucket: "data-ie", srcAWS: "eu-west-1", dst: "uk", dstBox: "uk-box", dstBucket: "data-uk", dstAWS: "eu-west-2", dstEngine: "uk"} |
| 501 | for _, o := range opts { |
| 502 | o(h) |
| 503 | } |
| 504 | h.cfg.Region = h.src |
| 505 | h.cfg.OwnBucket, h.cfg.OwnBucketRegion = h.srcBucket, h.srcAWS |
| 506 | h.roster = p1Roster{ |
| 507 | h.srcBox: {Box: h.srcBox, Enrollment: p1SrcEnrollment, Region: park.PlacementRegion(h.src), FleetHomeRegion: h.homeRegion, Bucket: h.srcBucket, Eligible: true}, |
| 508 | h.dstBox: {Box: h.dstBox, Enrollment: p1DstEnrollment, Region: h.dstEngine, FleetHomeRegion: h.homeRegion, Bucket: h.dstBucket, Eligible: true}, |
| 509 | } |
| 510 | if h.dst == "" { |
| 511 | // The home box's roster entry equals the fleet's home designation: |
| 512 | // the claim grants it the @home alias (claim Plan 13). |
| 513 | a := h.roster[h.dstBox] |
| 514 | a.HomeRegion = h.homeRegion |
| 515 | h.roster[h.dstBox] = a |
| 516 | } |
| 517 | h.vs = &fakeVersioned{f: h.f} |
| 518 | h.lease.until = h.clock.now.Add(moves.StagingLeaseTTL) |
| 519 | inc, err := hostclaim.Incarnation("bab", "app", p1Host, p1Created) |
| 520 | if err != nil { |
| 521 | t.Fatal(err) |
| 522 | } |
| 523 | h.inc = inc |
| 524 | privB64, _, err := appconfig.GenerateKey() |
| 525 | if err != nil { |
| 526 | t.Fatal(err) |
| 527 | } |
| 528 | priv, err := appconfig.ParsePrivateKey(privB64, "test") |
| 529 | if err != nil { |
| 530 | t.Fatal(err) |
| 531 | } |
| 532 | h.optDir = t.TempDir() |
| 533 | h.route = registry.Route{Host: p1Host, Kind: "instance", App: "app", Account: "bab", Hibernate: true, Status: "ready", |
| 534 | Socket: provision.DefaultSocket(p1Host), DBPath: provision.DefaultDBPath(h.cfg.DataDir, p1Host), Provisioned: true} |
| 535 | h.env = map[string]string{"CARLOS_REGION": h.src, "CARLOS_CONTROL_SOCKET": "/run/carlos/" + p1Host + ".sock.control", "APP_MODE": "prod"} |
| 536 | h.probe, h.strict = moves.Check{OK: true}, moves.Check{OK: true} |
| 537 | ctx := context.Background() |
| 538 | |
| 539 | h.cfg.Enabled = true |
| 540 | h.cfg.Box = h.srcBox |
| 541 | h.cfg.Now = h.clock.Now |
| 542 | h.cfg.OptDir = h.optDir |
| 543 | h.cfg.DeployBucket = "deploy" |
| 544 | h.cfg.StagingLease = func(context.Context, string, string) (moves.SafeUntilGuard, func(context.Context) error, error) { |
| 545 | h.lease.mu.Lock() |
| 546 | h.lease.acquired++ |
| 547 | h.lease.mu.Unlock() |
| 548 | return h.lease, func(context.Context) error { |
| 549 | h.lease.mu.Lock() |
| 550 | h.lease.released++ |
| 551 | h.lease.mu.Unlock() |
| 552 | return nil |
| 553 | }, nil |
| 554 | } |
| 555 | h.cfg.Staging = func(context.Context, string) (lease.StagingWriter, error) { return h.f, nil } |
| 556 | h.cfg.Versioned = func(context.Context, string, string) (lease.VersionedStore, error) { return h.vs, nil } |
| 557 | h.cfg.SecretsKey, h.cfg.SecretsPriv, h.cfg.SecretsPub = privB64, priv, priv.PublicKey() |
| 558 | h.cfg.LaunchEnv = func(string) (map[string]string, map[string]bool, error) { return h.env, nil, nil } |
| 559 | if h.realEnv { |
| 560 | src := &activator.ExecBackend{DataDir: h.cfg.DataDir, Region: h.src} |
| 561 | h.cfg.LaunchEnv = func(string) (map[string]string, map[string]bool, error) { |
| 562 | env, instance, _, err := src.LaunchLayers(h.route) |
| 563 | return env, instance, err |
| 564 | } |
| 565 | } |
| 566 | h.cfg.FilesDir = func(host string) (string, string) { |
| 567 | return filepath.Join(h.cfg.DataDir, host), filepath.Join(h.cfg.DataDir, host+".env") |
| 568 | } |
| 569 | h.cfg.ReadyProbe = func(_ context.Context, _, path string) moves.Check { |
| 570 | h.probePaths = append(h.probePaths, path) |
| 571 | return h.probe |
| 572 | } |
| 573 | h.cfg.StrictNext = func(string) moves.Check { return h.strict } |
| 574 | h.cfg.Route = func(host string) (*registry.Route, error) { |
| 575 | if host != p1Host { |
| 576 | return nil, nil |
| 577 | } |
| 578 | rt := h.route |
| 579 | return &rt, nil |
| 580 | } |
| 581 | |
| 582 | // The source's files and carried .env, its live release v7, and its |
| 583 | // replica: two generations under the scoped location (g-old and the |
| 584 | // live g-cur), plus a generation.json and a lock.json beside them. |
| 585 | files := filepath.Join(h.cfg.DataDir, p1Host) |
| 586 | if err := os.MkdirAll(files, 0o755); err != nil { |
| 587 | t.Fatal(err) |
| 588 | } |
| 589 | if err := os.WriteFile(filepath.Join(files, "upload.txt"), []byte("hello"), 0o644); err != nil { |
| 590 | t.Fatal(err) |
| 591 | } |
| 592 | if err := os.WriteFile(filepath.Join(h.cfg.DataDir, p1Host+".env"), []byte(h.srcEnvFile), 0o600); err != nil { |
| 593 | t.Fatal(err) |
| 594 | } |
| 595 | pointLiveForTest(t, h.optDir, p1Host, "v7") |
| 596 | loc := "instances/" + p1Host + "/" + inc + "/" |
| 597 | h.f.store(h.srcBucket, loc+"g-cur/0000/0000000000000001-0000000000000002.ltx", |
| 598 | fakeMoveObject{lease.MirrorObject{ETag: "cur-a", Size: 2 << 20}, string(ltxForTest(t, 4096, 4, 1, 2))}) |
| 599 | h.f.store(h.srcBucket, loc+"g-cur/0001/0000000000000001-0000000000000002.ltx", |
| 600 | fakeMoveObject{lease.MirrorObject{ETag: "cur-b", Size: 4096}, string(ltxForTest(t, 4096, 4, 1, 2))}) |
| 601 | h.f.putSized(h.srcBucket, loc+"g-old/0000/0000000000000001-0000000000000001.ltx", "old-a", 4096) |
| 602 | h.f.put(h.srcBucket, loc+"generation.json", "gen") |
| 603 | h.f.put(h.srcBucket, loc+"g-cur/lock.json", "lock") |
| 604 | |
| 605 | // The source's edges file: ie publishes its data bucket too (T10). |
| 606 | if _, err := h.s.Put(ctx, "control/edges/"+h.srcBox+".json", |
| 607 | []byte(`{"v":1,"box":"`+h.srcBox+`","eip":"1.2.3.5","placement":"`+h.src+`","data_bucket":"`+h.srcBucket+`","data_bucket_region":"`+h.srcAWS+`","hosts":{}}`), ""); err != nil { |
| 608 | t.Fatal(err) |
| 609 | } |
| 610 | // The target's edges file, when it is not newMoveHarness's uk-box: a |
| 611 | // home box publishes its data bucket and no placement. |
| 612 | if h.dstBox != "uk-box" { |
| 613 | if err := h.s.Delete(ctx, "control/edges/uk-box.json"); err != nil { |
| 614 | t.Fatal(err) |
| 615 | } |
| 616 | placement := "" |
| 617 | if h.dst != "" { |
| 618 | placement = `"placement":"` + h.dst + `",` |
| 619 | } |
| 620 | if _, err := h.s.Put(ctx, "control/edges/"+h.dstBox+".json", |
| 621 | []byte(`{"v":1,"box":"`+h.dstBox+`","eip":"1.2.3.4",`+placement+`"data_bucket":"`+h.dstBucket+`","data_bucket_region":"`+h.dstAWS+`","hosts":{}}`), ""); err != nil { |
| 622 | t.Fatal(err) |
| 623 | } |
| 624 | } |
| 625 | // The app's instances manifest: it declares no readiness path. |
| 626 | if err := provision.PutManifest(ctx, h.s, "bab", "app", provision.Manifest{Schema: 1}); err != nil { |
| 627 | t.Fatal(err) |
| 628 | } |
| 629 | // The console's record (T2). |
| 630 | if _, err := h.s.Put(ctx, moves.ConsoleKey, []byte(`{"v":1,"protocol":1,"secrets_key_fp":"`+appconfig.KeyFingerprint(priv.PublicKey())+`"}`), ""); err != nil { |
| 631 | t.Fatal(err) |
| 632 | } |
| 633 | return h |
| 634 | } |
| 635 | |
| 636 | // pointLiveForTest makes optDir/live/<host> point at version v, the shape |
| 637 | // release.LiveVersion reads. |
| 638 | func pointLiveForTest(t *testing.T, optDir, host, v string) { |
| 639 | t.Helper() |
| 640 | target := filepath.Join(optDir, "versions", v, "bab", "app") |
| 641 | if err := os.MkdirAll(target, 0o755); err != nil { |
| 642 | t.Fatal(err) |
| 643 | } |
| 644 | if err := os.MkdirAll(filepath.Join(optDir, "live"), 0o755); err != nil { |
| 645 | t.Fatal(err) |
| 646 | } |
| 647 | link := filepath.Join(optDir, "live", host) |
| 648 | os.Remove(link) |
| 649 | if err := os.Symlink(target, link); err != nil { |
| 650 | t.Fatal(err) |
| 651 | } |
| 652 | } |
| 653 | |
| 654 | // start writes the record (MoveStart'ed, id p1ID, cut mode cut) and the |
| 655 | // claim the door's CAS leaves: serving under exec ie-box, then moving with |
| 656 | // target uk-box. The claim has touched ie and ie-box only. Under intoHome |
| 657 | // the source is ca-box, the target home-box and the move's region the |
| 658 | // placement-derived @home, begun with the admission as the claim-mode door |
| 659 | // begins it (BeginAdmittedMove, home_region ie). The retained options add |
| 660 | // the claim state an earlier stay left. |
| 661 | func (h *phase1Harness) start(t *testing.T, cut string) provision.Record { |
| 662 | t.Helper() |
| 663 | ctx := context.Background() |
| 664 | rec := provision.Record{Schema: 1, Host: p1Host, Slug: "s", App: "app", Account: "bab", Region: h.src, Channel: "stable", CreatedAt: p1Created} |
| 665 | if err := provision.CreateRecord(ctx, h.s, rec); err != nil { |
| 666 | t.Fatal(err) |
| 667 | } |
| 668 | if _, err := provision.SetRecordMove(ctx, h.s, "bab", "app", p1Host, provision.MoveStartTransition(provision.MoveStart{ |
| 669 | ID: p1ID, Incarnation: h.inc, From: h.src, To: h.dst, Cut: cut, By: "op", Now: h.clock.Now()})); err != nil { |
| 670 | t.Fatal(err) |
| 671 | } |
| 672 | obs, err := hostclaim.Update(ctx, h.s, hostclaim.Observed{Host: p1Host, Absent: true}, func(c *hostclaim.Claim) error { |
| 673 | c.Holder = &hostclaim.Holder{Family: "instance", Key: provision.RecordKey("bab", "app", p1Host), Account: "bab", App: "app", |
| 674 | Incarnation: h.inc, CreatedAt: p1Created, Location: "instances/" + p1Host + "/" + h.inc, Generation: "cur", Stamp: 1} |
| 675 | c.Op = hostclaim.Op{Kind: hostclaim.KindServing, Exec: h.srcBox} |
| 676 | c.Touched = hostclaim.Touched{Complete: true, Regions: []string{park.PlacementRegion(h.src)}, Boxes: []string{h.srcBox}} |
| 677 | if h.retain.touched != "" { |
| 678 | c.Touched.Regions = append(c.Touched.Regions, h.retain.touched) |
| 679 | } |
| 680 | if h.retain.quiet != "" { |
| 681 | c.QuietAfter = map[string]time.Time{h.retain.quiet: h.clock.Now().Add(-time.Hour).UTC()} |
| 682 | } |
| 683 | return nil |
| 684 | }) |
| 685 | if err != nil { |
| 686 | t.Fatal(err) |
| 687 | } |
| 688 | to := park.PlacementRegion(h.dst) |
| 689 | begin := hostclaim.BeginMove(p1ID, "op", h.srcBox, h.dstBox, to, h.clock.Now()) |
| 690 | if h.admitted { |
| 691 | begin = hostclaim.BeginAdmittedMove(p1ID, "op", h.srcBox, h.dstBox, to, h.clock.Now(), hostclaim.MoveAdmission{HomeRegion: h.homeRegion, |
| 692 | AppID: "app_" + p1Host, AdmissionID: p1ID, SourceEnrollment: h.roster[h.srcBox].Enrollment, TargetEnrollment: h.roster[h.dstBox].Enrollment}) |
| 693 | } |
| 694 | if _, err := hostclaim.Update(ctx, h.s, obs, begin); err != nil { |
| 695 | t.Fatal(err) |
| 696 | } |
| 697 | if h.retain.chore != "" { |
| 698 | // An earlier stay's residue chore in the home region, owed under |
| 699 | // the spelling the earlier op wrote: one generation of the earlier |
| 700 | // stay, parked under the location's scope, as QueueResidueParks |
| 701 | // queues a landing's residue. Seeded after the start: the start |
| 702 | // guard refuses a move into a region with a chore owed. |
| 703 | rewriteP1Claim(t, h.s, func(c *hostclaim.Claim) { |
| 704 | c.Obligations = append(c.Obligations, hostclaim.Obligation{ID: "chore-earlier", Op: retainedChoreOp, Kind: "park-generation", Region: h.retain.chore, |
| 705 | Generation: retainedChoreGeneration, Owner: "region:" + h.retain.chore, Retirement: "retirement-earlier", |
| 706 | Park: &hostclaim.ParkScope{Location: c.Holder.Location, Incarnation: h.inc, EvidenceAt: h.clock.Now().Add(-time.Hour).UTC(), RequestLimit: moves.ParkRequestLimit}}) |
| 707 | }) |
| 708 | } |
| 709 | return h.read(t, p1Host) |
| 710 | } |
| 711 | |
| 712 | func (h *phase1Harness) claim(t *testing.T) *hostclaim.Claim { |
| 713 | t.Helper() |
| 714 | obs, err := hostclaim.Read(context.Background(), h.s, p1Host) |
| 715 | if err != nil || obs.Absent { |
| 716 | t.Fatalf("claim: %v absent=%v", err, obs.Absent) |
| 717 | } |
| 718 | return obs.Claim |
| 719 | } |
| 720 | |
| 721 | // pass runs one source pass. |
| 722 | func (h *phase1Harness) pass(t *testing.T) { |
| 723 | t.Helper() |
| 724 | if err := movesPass(context.Background(), h.s, h.w()); err != nil { |
| 725 | t.Fatal(err) |
| 726 | } |
| 727 | } |
| 728 | |
| 729 | // w is one worker kept across the harness's passes, as one agent keeps it. |
| 730 | func (h *phase1Harness) w() *movesWorker { |
| 731 | if h.worker1 == nil { |
| 732 | h.worker1 = &movesWorker{cfg: h.cfg, mirrors: map[string]lease.MirrorStore{}} |
| 733 | } |
| 734 | h.worker1.cfg = h.cfg |
| 735 | return h.worker1 |
| 736 | } |
| 737 | |
| 738 | // targetCfg is the uk box's configuration over the same store and buckets, |
| 739 | // its hooks stubbed to pass (each test turns one off). Under intoHome it is |
| 740 | // home-box's: placement "", ParkHomeRegion ie, and a held environment with |
| 741 | // no CARLOS_REGION. |
| 742 | func (h *phase1Harness) targetCfg() moveCfg { |
| 743 | c := h.cfg |
| 744 | c.Region, c.Box, c.OwnBucket, c.OwnBucketRegion = h.dst, h.dstBox, h.dstBucket, h.dstAWS |
| 745 | if h.dst == "" { |
| 746 | c.ParkHomeRegion = h.homeRegion |
| 747 | } |
| 748 | c.OwnsRoute = func(string) bool { return false } |
| 749 | c.Route = func(string) (*registry.Route, error) { return nil, nil } |
| 750 | c.ComposeHeldEnv = func(ctx context.Context, s release.Store, rec provision.Record, carried []byte) (map[string]string, map[string]bool, []string, provision.MoveInputs, error) { |
| 751 | env, instance, err := h.heldEnv(ctx, s, rec, carried) |
| 752 | return env, instance, nil, provision.MoveInputs{BundleETags: map[string]string{"default": "e1", "environment": ""}}, err |
| 753 | } |
| 754 | c.ProbeStore = func(context.Context, map[string]string, string) (string, error) { return "", nil } |
| 755 | c.DiskFree = func(string) (uint64, uint64, error) { return 100 << 30, 90 << 30, nil } |
| 756 | c.ReleaseReady = func(context.Context, release.Store, provision.Record, string) moves.Check { |
| 757 | return moves.Check{OK: true} |
| 758 | } |
| 759 | c.LaneProbe = func(context.Context, string, string, string) error { return nil } |
| 760 | c.MeasureS3 = func(context.Context, string) (moves.MeasuredRate, error) { |
| 761 | return moves.MeasuredRate{MiBPerS: 250, At: h.clock.Now(), Bytes: 32 << 20}, nil |
| 762 | } |
| 763 | c.MeasureDisk = func(context.Context) (moves.MeasuredRate, moves.FilesRate, error) { |
| 764 | now := h.clock.Now() |
| 765 | return moves.MeasuredRate{MiBPerS: 500, At: now, Bytes: 128 << 20}, moves.FilesRate{PerS: 1600, At: now, Files: 512}, nil |
| 766 | } |
| 767 | // The cut guard's T13: the target states it can land and roll back, |
| 768 | // as every box will once the wake-source gate opens (armStandIns). |
| 769 | armStandIns(&c, h.s, roleTarget) |
| 770 | return c |
| 771 | } |
| 772 | |
| 773 | // setHeldRegion is the agent's rule for the target's held environment |
| 774 | // (agentOwnedVars): CARLOS_REGION is the box's placement when it has one, |
| 775 | // and absent on a home box. |
| 776 | func setHeldRegion(env map[string]string, placement string) { |
| 777 | if placement == "" { |
| 778 | delete(env, "CARLOS_REGION") |
| 779 | return |
| 780 | } |
| 781 | env["CARLOS_REGION"] = placement |
| 782 | } |
| 783 | |
| 784 | // targetPass runs one pass as the target box. |
| 785 | func (h *phase1Harness) targetPass(t *testing.T, c moveCfg) { |
| 786 | t.Helper() |
| 787 | if h.worker2 == nil { |
| 788 | h.worker2 = &movesWorker{mirrors: map[string]lease.MirrorStore{}} |
| 789 | } |
| 790 | h.worker2.cfg = c |
| 791 | if err := movesPass(context.Background(), h.s, h.worker2); err != nil { |
| 792 | t.Fatal(err) |
| 793 | } |
| 794 | } |
| 795 | |
| 796 | func (h *phase1Harness) targetReport(t *testing.T) *moves.TargetReport { |
| 797 | t.Helper() |
| 798 | r, err := readTargetReport(context.Background(), h.s, h.read(t, p1Host)) |
| 799 | if err != nil { |
| 800 | t.Fatal(err) |
| 801 | } |
| 802 | return r |
| 803 | } |
| 804 | |
| 805 | func (h *phase1Harness) sourceReport(t *testing.T) *moves.SourceReport { |
| 806 | t.Helper() |
| 807 | r, err := readSourceReport(context.Background(), h.s, h.read(t, p1Host)) |
| 808 | if err != nil { |
| 809 | t.Fatal(err) |
| 810 | } |
| 811 | return r |
| 812 | } |
| 813 | |
| 814 | func (h *phase1Harness) ledger(t *testing.T) moveLedger { |
| 815 | t.Helper() |
| 816 | l, err := readMoveLedger(context.Background(), h.f, h.srcBucket, moves.LedgerKey(p1Host, p1ID)) |
| 817 | if err != nil { |
| 818 | t.Fatal(err) |
| 819 | } |
| 820 | return l |
| 821 | } |
| 822 | |
| 823 | // ltxForTest is a real LTX file (lz4-compressed pages) for a database of |
| 824 | // commit pages of pageSize bytes, covering minTX to maxTX. A snapshot |
| 825 | // (minTX 1) carries every page; an increment carries page 1 alone. Pages |
| 826 | // are zeros, so the file is far smaller than the database it describes. |
| 827 | func ltxForTest(t *testing.T, pageSize, commit uint32, minTX, maxTX uint64) []byte { |
| 828 | t.Helper() |
| 829 | var buf bytes.Buffer |
| 830 | enc, err := ltx.NewEncoder(&buf) |
| 831 | if err != nil { |
| 832 | t.Fatal(err) |
| 833 | } |
| 834 | hdr := ltx.Header{Version: ltx.Version, PageSize: pageSize, Commit: commit, MinTXID: ltx.TXID(minTX), MaxTXID: ltx.TXID(maxTX), Timestamp: 1000} |
| 835 | if minTX > 1 { |
| 836 | hdr.PreApplyChecksum = ltx.ChecksumFlag | 1 |
| 837 | } |
| 838 | if err := enc.EncodeHeader(hdr); err != nil { |
| 839 | t.Fatal(err) |
| 840 | } |
| 841 | pages := uint32(1) |
| 842 | if minTX == 1 { |
| 843 | pages = commit |
| 844 | } |
| 845 | page := make([]byte, pageSize) |
| 846 | for pgno := uint32(1); pgno <= pages; pgno++ { |
| 847 | if pgno == ltx.LockPgno(pageSize) { |
| 848 | continue |
| 849 | } |
| 850 | if err := enc.EncodePage(ltx.PageHeader{Pgno: pgno}, page); err != nil { |
| 851 | t.Fatal(err) |
| 852 | } |
| 853 | } |
| 854 | enc.SetPostApplyChecksum(ltx.ChecksumFlag | 2) |
| 855 | if err := enc.Close(); err != nil { |
| 856 | t.Fatal(err) |
| 857 | } |
| 858 | return buf.Bytes() |
| 859 | } |
| 860 | |
amadan