| 1 | package park |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "fmt" |
| 6 | "reflect" |
| 7 | "sort" |
| 8 | "strings" |
| 9 | "time" |
| 10 | |
| 11 | "github.com/carlosframework/platform/internal/hostclaim" |
| 12 | "github.com/carlosframework/platform/internal/lease" |
| 13 | ) |
| 14 | |
| 15 | // View is a moves-owned, completely read snapshot bound to one claim ETag. |
| 16 | // GenerationOf is a pure classifier for any exact reference, including a |
| 17 | // version already deleted by a previous attempt. It does no storage I/O. |
| 18 | // Pinned includes the exact noncurrent versions needed by the installed root. |
| 19 | // MarkerKnown means its read succeeded; an absent marker may denote @root. |
| 20 | type View struct { |
| 21 | HomeRegion string |
| 22 | // ArchivedIncarnations labels proven retired locations independently of |
| 23 | // the moves-sweep proof's incarnation. Missing entries never invent labels. |
| 24 | ArchivedIncarnations map[string]string |
| 25 | RefusedLocations map[string]error |
| 26 | ClaimETag, Region, AuthoritativeRegion string |
| 27 | MarkerKnown bool |
| 28 | MarkerGeneration string |
| 29 | Marker Ref |
| 30 | Pinned map[Ref]bool |
| 31 | FilePins map[Ref]bool |
| 32 | GenerationOf func(Ref) (string, bool) |
| 33 | } |
| 34 | type Alert struct{ Location, Reason string } |
| 35 | type Candidate struct{ Location, Incarnation, Generation string } |
| 36 | type Decision struct { |
| 37 | Candidates []Candidate |
| 38 | Alerts []Alert |
| 39 | } |
| 40 | |
| 41 | func generation(s string) string { |
| 42 | if s == "" { |
| 43 | return hostclaim.RootGeneration |
| 44 | } |
| 45 | return s |
| 46 | } |
| 47 | |
| 48 | func protected(c *hostclaim.Claim, v View, location string) (map[string]bool, error) { |
| 49 | live := map[string]bool{} |
| 50 | if c.Holder == nil || c.Op.Kind == hostclaim.KindParking { |
| 51 | return live, nil |
| 52 | } |
| 53 | if v.AuthoritativeRegion == "" { |
| 54 | return nil, ErrUnsafe |
| 55 | } |
| 56 | if !v.sameRegion(v.Region, v.AuthoritativeRegion) { |
| 57 | return live, nil |
| 58 | } |
| 59 | if !v.MarkerKnown || v.GenerationOf == nil { |
| 60 | return nil, ErrUnsafe |
| 61 | } |
| 62 | for r, on := range v.Pinned { |
| 63 | if on { |
| 64 | g, ok := v.GenerationOf(r) |
| 65 | if !ok { |
| 66 | return nil, ErrUnsafe |
| 67 | } |
| 68 | if v.FilePins[r] && strings.HasPrefix(r.Key, c.Holder.Location+"/files/") { |
| 69 | continue |
| 70 | } |
| 71 | live[generation(g)] = true |
| 72 | } |
| 73 | } |
| 74 | h := c.Holder |
| 75 | if location != "" && location != h.Location && !strings.HasPrefix(location, h.Location+"/") && !strings.HasPrefix(h.Location, location+"/") { |
| 76 | return live, nil |
| 77 | } |
| 78 | current := generation(h.Generation) |
| 79 | marker := generation(v.MarkerGeneration) |
| 80 | live[current] = true |
| 81 | live[marker] = true |
| 82 | if current == marker { |
| 83 | return live, nil |
| 84 | } |
| 85 | // Walk the recorded chronological predecessor chain backwards. Warm |
| 86 | // admissions do not form cold unfinished links and cannot bridge a gap. |
| 87 | want, region := current, v.AuthoritativeRegion |
| 88 | for i := len(h.Admissions) - 1; i >= 0; i-- { |
| 89 | a := h.Admissions[i] |
| 90 | if generation(a.Generation) != want || !v.sameRegion(a.Region, region) { |
| 91 | continue |
| 92 | } |
| 93 | if a.Mode != "cold" { |
| 94 | return nil, ErrUnsafe |
| 95 | } |
| 96 | pred := generation(a.Predecessor) |
| 97 | if pred == want { |
| 98 | return nil, ErrUnsafe |
| 99 | } |
| 100 | want = pred |
| 101 | region = a.PredecessorRegion |
| 102 | if region == "" || v.sameRegion(region, v.AuthoritativeRegion) { |
| 103 | if a.Covered { |
| 104 | return nil, ErrUnsafe |
| 105 | } |
| 106 | live[want] = true |
| 107 | } |
| 108 | if want == marker { |
| 109 | if region != "" && !v.sameRegion(region, v.AuthoritativeRegion) { |
| 110 | return nil, ErrUnsafe |
| 111 | } |
| 112 | return live, nil |
| 113 | } |
| 114 | } |
| 115 | return nil, ErrUnsafe |
| 116 | } |
| 117 | |
| 118 | // LiveGenerations is the region-qualified live generation set the engine |
| 119 | // protects at location: the holder's generation, the marker's, every |
| 120 | // generation the pinned versions belong to, and the unfinished cold |
| 121 | // admission chain walked back to the marker in the authoritative region. |
| 122 | // It is protected itself, exported so a view that needs the live set (the |
| 123 | // moves files live set) enumerates the same generations rather than a |
| 124 | // parallel walk. An error is protected's refusal. |
| 125 | func LiveGenerations(c *hostclaim.Claim, v View, location string) (map[string]bool, error) { |
| 126 | if c == nil { |
| 127 | return nil, ErrUnsafe |
| 128 | } |
| 129 | return protected(c, v, location) |
| 130 | } |
| 131 | |
| 132 | // Decide uses a complete namespace version listing supplied by Discover. |
| 133 | // LegacyIncarnation is a caller-supplied, evidenced archive label only. It is |
| 134 | // never used to discover locations, and its absence refuses legacy cleanup. |
| 135 | func Decide(obs hostclaim.Observed, v View, versions []lease.ObjectVersion, legacyIncarnation string) (Decision, error) { |
| 136 | var d Decision |
| 137 | if obs.Absent || obs.Claim == nil || obs.ETag == "" || v.ClaimETag != obs.ETag || v.Region == "" { |
| 138 | return d, ErrUnsafe |
| 139 | } |
| 140 | c := obs.Claim |
| 141 | if err := c.Validate(); err != nil { |
| 142 | return d, err |
| 143 | } |
| 144 | joining := c.Holder == nil && c.Op.Kind == hostclaim.KindParking && c.Op.Intent["park_sweep"] == "v1" |
| 145 | if !c.Free() && c.Op.Kind != hostclaim.KindServing && !joining { |
| 146 | return d, nil |
| 147 | } |
| 148 | for _, o := range c.Obligations { |
| 149 | if v.sameRegion(o.Region, v.Region) { |
| 150 | return d, nil |
| 151 | } |
| 152 | } |
| 153 | live, liveErr := protected(c, v, "") |
| 154 | if liveErr != nil && c.Holder != nil { |
| 155 | d.Alerts = append(d.Alerts, Alert{c.Holder.Location, "marker does not match the claim or an unfinished admission run"}) |
| 156 | } |
| 157 | groups := map[Candidate]bool{} |
| 158 | refused := map[string]bool{} |
| 159 | seen := map[Ref]bool{} |
| 160 | for _, x := range versions { |
| 161 | loc, err := locationOf(c.Host, x.Key) |
| 162 | if err != nil || x.VersionID == "" || seen[ref(x)] { |
| 163 | return Decision{}, ErrUnsafe |
| 164 | } |
| 165 | seen[ref(x)] = true |
| 166 | if isLock(c.Host, loc, x.Key) { |
| 167 | continue |
| 168 | } |
| 169 | if refusal := v.RefusedLocations[loc]; refusal != nil && !(c.Holder != nil && loc == c.Holder.Location && v.sameRegion(v.Region, v.AuthoritativeRegion)) { |
| 170 | if !refused[loc] { |
| 171 | d.Alerts = append(d.Alerts, Alert{loc, refusal.Error()}) |
| 172 | refused[loc] = true |
| 173 | } |
| 174 | continue |
| 175 | } |
| 176 | candidate := Candidate{Location: loc} |
| 177 | if tail := strings.TrimPrefix(loc, "instances/"+c.Host+"/"); tail != loc { |
| 178 | candidate.Incarnation = tail |
| 179 | } else { |
| 180 | candidate.Incarnation = legacyIncarnation |
| 181 | if loc == "instances/"+c.Host && v.ArchivedIncarnations[loc] != "" { |
| 182 | candidate.Incarnation = v.ArchivedIncarnations[loc] |
| 183 | } |
| 184 | } |
| 185 | if c.Holder != nil && loc == c.Holder.Location && v.sameRegion(v.Region, v.AuthoritativeRegion) { |
| 186 | if liveErr != nil { |
| 187 | continue |
| 188 | } |
| 189 | if x.Key == v.Marker.Key || isBuildState(c.Holder.Location, x.Key) || v.Pinned[ref(x)] { |
| 190 | continue |
| 191 | } |
| 192 | g, ok := v.GenerationOf(ref(x)) |
| 193 | if !ok { |
| 194 | return Decision{}, ErrUnsafe |
| 195 | } |
| 196 | g = generation(g) |
| 197 | if live[g] { |
| 198 | continue |
| 199 | } |
| 200 | candidate.Generation = g |
| 201 | candidate.Incarnation = c.Holder.Incarnation |
| 202 | } |
| 203 | if !hex64(candidate.Incarnation) { |
| 204 | return Decision{}, ErrUnsafe |
| 205 | } |
| 206 | groups[candidate] = true |
| 207 | } |
| 208 | for c := range groups { |
| 209 | d.Candidates = append(d.Candidates, c) |
| 210 | } |
| 211 | sort.Slice(d.Candidates, func(i, j int) bool { |
| 212 | a, b := d.Candidates[i], d.Candidates[j] |
| 213 | if a.Location == b.Location { |
| 214 | return a.Generation < b.Generation |
| 215 | } |
| 216 | return a.Location < b.Location |
| 217 | }) |
| 218 | return d, nil |
| 219 | } |
| 220 | |
| 221 | // Discover expands both namespaces across every version page. This is a |
| 222 | // conservative superset of delimiter discovery: deriving first child prefixes |
| 223 | // from the complete version listing also discovers marker-only locations. |
| 224 | func (e Engine) Discover(ctx context.Context, host string) ([]lease.ObjectVersion, error) { |
| 225 | var out []lease.ObjectVersion |
| 226 | for _, p := range []string{"instances/" + host + "/", "moves/" + host + "/"} { |
| 227 | vs, err := e.Store.ListVersions(ctx, e.Bucket, p) |
| 228 | if err != nil { |
| 229 | return nil, err |
| 230 | } |
| 231 | out = append(out, vs...) |
| 232 | } |
| 233 | return out, nil |
| 234 | } |
| 235 | |
| 236 | func parkedGenerations(c *hostclaim.Claim, v View, chores []hostclaim.Obligation, versions []lease.ObjectVersion) ([]string, error) { |
| 237 | if c.Holder == nil || len(c.Holder.Admissions) == 0 { |
| 238 | return nil, nil |
| 239 | } |
| 240 | seen := map[string]bool{} |
| 241 | for _, o := range chores { |
| 242 | if o.Park.Location != c.Holder.Location || o.Park.Incarnation != c.Holder.Incarnation { |
| 243 | continue |
| 244 | } |
| 245 | for _, x := range versions { |
| 246 | // The holder location's build state belongs to no generation |
| 247 | // (ruling R-PARK), so it names none parked, whichever region |
| 248 | // parked it. |
| 249 | if !strings.HasPrefix(x.Key, c.Holder.Location+"/") || isLock(c.Host, o.Park.Location, x.Key) || x.Key == v.Marker.Key || isBuildState(c.Holder.Location, x.Key) { |
| 250 | continue |
| 251 | } |
| 252 | inScope, err := scopeContains(c.Host, o.Park.Location, x.Key) |
| 253 | if err != nil { |
| 254 | return nil, err |
| 255 | } |
| 256 | if !inScope { |
| 257 | continue |
| 258 | } |
| 259 | if v.GenerationOf == nil { |
| 260 | return nil, ErrUnsafe |
| 261 | } |
| 262 | g, ok := v.GenerationOf(ref(x)) |
| 263 | if !ok { |
| 264 | return nil, ErrUnsafe |
| 265 | } |
| 266 | g = generation(g) |
| 267 | if o.Kind != "park-generation" || o.Generation == g { |
| 268 | seen[g] = true |
| 269 | } |
| 270 | } |
| 271 | } |
| 272 | var generations []string |
| 273 | for g := range seen { |
| 274 | generations = append(generations, g) |
| 275 | } |
| 276 | sort.Strings(generations) |
| 277 | return generations, nil |
| 278 | } |
| 279 | |
| 280 | // Sweep queues every decision in one claim CAS; Run is the common executor. |
| 281 | // EvidenceAt is genuine prior stop/fence evidence for the residue. A sweep |
| 282 | // uses listing-time evidence for moves/<host> only after VerifyMovesSweep |
| 283 | // proves that the newest retired move was fenced or released before listing. |
| 284 | // Other namespaces still require their original stop/fence evidence. |
| 285 | func (e Engine) Sweep(ctx context.Context, host, legacyIncarnation string, evidenceAt time.Time, requestLimit time.Duration) (Decision, error) { |
| 286 | return e.sweep(ctx, hostclaim.Observed{Host: host}, nil, legacyIncarnation, evidenceAt, requestLimit, false) |
| 287 | } |
| 288 | |
| 289 | // SweepObserved binds retirement evidence to the exact claim observation and |
| 290 | // complete version census from which it was verified. Any claim or namespace |
| 291 | // change requires new evidence; it is never folded into the old timestamp. |
| 292 | func (e Engine) SweepObserved(ctx context.Context, expected hostclaim.Observed, versions []lease.ObjectVersion, legacyIncarnation string, evidenceAt time.Time, requestLimit time.Duration) (Decision, error) { |
| 293 | if expected.Host == "" || expected.Absent || expected.Claim == nil || expected.ETag == "" { |
| 294 | return Decision{}, ErrUnsafe |
| 295 | } |
| 296 | return e.sweep(ctx, expected, versions, legacyIncarnation, evidenceAt, requestLimit, true) |
| 297 | } |
| 298 | |
| 299 | func sameCensus(a, b []lease.ObjectVersion) bool { |
| 300 | if len(a) != len(b) { |
| 301 | return false |
| 302 | } |
| 303 | indexed := make(map[Ref]lease.ObjectVersion, len(a)) |
| 304 | for _, version := range a { |
| 305 | r := ref(version) |
| 306 | if r.Key == "" || r.VersionID == "" { |
| 307 | return false |
| 308 | } |
| 309 | if _, duplicate := indexed[r]; duplicate { |
| 310 | return false |
| 311 | } |
| 312 | indexed[r] = version |
| 313 | } |
| 314 | seen := make(map[Ref]bool, len(b)) |
| 315 | for _, version := range b { |
| 316 | r := ref(version) |
| 317 | want, ok := indexed[r] |
| 318 | if !ok || seen[r] || !reflect.DeepEqual(want, version) { |
| 319 | return false |
| 320 | } |
| 321 | seen[r] = true |
| 322 | } |
| 323 | return true |
| 324 | } |
| 325 | |
| 326 | func (e Engine) sweep(ctx context.Context, expected hostclaim.Observed, expectedVersions []lease.ObjectVersion, legacyIncarnation string, evidenceAt time.Time, requestLimit time.Duration, bound bool) (Decision, error) { |
| 327 | var empty Decision |
| 328 | if e.View == nil || e.NewID == nil || e.Authorize == nil || e.Now == nil { |
| 329 | return empty, ErrUnsafe |
| 330 | } |
| 331 | host := expected.Host |
| 332 | obs, err := hostclaim.Read(ctx, e.Claims, host) |
| 333 | if err != nil { |
| 334 | return empty, err |
| 335 | } |
| 336 | if bound && (obs.ETag != expected.ETag || !reflect.DeepEqual(obs.Claim, expected.Claim)) { |
| 337 | return empty, ErrChanged |
| 338 | } |
| 339 | if obs.Absent || obs.Claim == nil { |
| 340 | return empty, ErrUnsafe |
| 341 | } |
| 342 | v, err := e.View(ctx, obs) |
| 343 | if err != nil { |
| 344 | return empty, err |
| 345 | } |
| 346 | home, err := BoundHome(e.HomeRegion, obs.Claim.Op.Intent["home_region"]) |
| 347 | if err != nil { |
| 348 | return empty, err |
| 349 | } |
| 350 | v.HomeRegion = home |
| 351 | listingStarted := e.Now() |
| 352 | vs, err := e.Discover(ctx, host) |
| 353 | listedAt := e.Now() |
| 354 | if err != nil { |
| 355 | return empty, err |
| 356 | } |
| 357 | if bound && !sameCensus(vs, expectedVersions) { |
| 358 | return empty, ErrChanged |
| 359 | } |
| 360 | d, err := Decide(obs, v, vs, legacyIncarnation) |
| 361 | if err != nil || len(d.Candidates) == 0 { |
| 362 | return d, err |
| 363 | } |
| 364 | var movesEvidence time.Time |
| 365 | candidates := d.Candidates[:0] |
| 366 | for _, candidate := range d.Candidates { |
| 367 | if candidate.Location == "moves/"+host { |
| 368 | if e.VerifyMovesSweep == nil || listedAt.Before(listingStarted) { |
| 369 | continue |
| 370 | } |
| 371 | detached, copyErr := moveSweepObservation(obs) |
| 372 | if copyErr != nil { |
| 373 | return d, copyErr |
| 374 | } |
| 375 | proof, verifyErr := e.VerifyMovesSweep(ctx, detached, append([]lease.ObjectVersion(nil), vs...)) |
| 376 | if verifyErr != nil || proof.Check(obs, candidate.Incarnation, listingStarted) != nil { |
| 377 | continue |
| 378 | } |
| 379 | movesEvidence = listedAt |
| 380 | } |
| 381 | candidates = append(candidates, candidate) |
| 382 | } |
| 383 | d.Candidates = candidates |
| 384 | if len(d.Candidates) == 0 { |
| 385 | return d, nil |
| 386 | } |
| 387 | var op string |
| 388 | if obs.Claim.Holder != nil { |
| 389 | op = fmt.Sprintf("session:%d", obs.Claim.Holder.Session) |
| 390 | } else if obs.Claim.Op.Kind == hostclaim.KindParking { |
| 391 | op = obs.Claim.Op.ID |
| 392 | } else { |
| 393 | op = e.NewID() |
| 394 | if !token(op) { |
| 395 | return d, ErrUnsafe |
| 396 | } |
| 397 | } |
| 398 | chores := make([]hostclaim.Obligation, 0, len(d.Candidates)) |
| 399 | for _, x := range d.Candidates { |
| 400 | boundAt := evidenceAt |
| 401 | if x.Location == "moves/"+host { |
| 402 | boundAt = movesEvidence |
| 403 | } |
| 404 | o := hostclaim.Obligation{ID: e.NewID(), Op: op, Kind: "park-replica", Region: e.Region, Generation: x.Generation, Retirement: e.NewID(), Owner: "region:" + e.Region, Park: &hostclaim.ParkScope{Location: x.Location, Incarnation: x.Incarnation, EvidenceAt: boundAt, RequestLimit: requestLimit}} |
| 405 | if x.Generation != "" { |
| 406 | o.Kind = "park-generation" |
| 407 | } |
| 408 | if !token(o.ID) || validateScope(host, o) != nil { |
| 409 | return d, ErrUnsafe |
| 410 | } |
| 411 | if err := e.Authorize(ctx, obs, o); err != nil { |
| 412 | return d, err |
| 413 | } |
| 414 | chores = append(chores, o) |
| 415 | } |
| 416 | generations, err := parkedGenerations(obs.Claim, v, chores, vs) |
| 417 | if err != nil { |
| 418 | return d, err |
| 419 | } |
| 420 | if bound { |
| 421 | fresh, discoverErr := e.Discover(ctx, host) |
| 422 | if discoverErr != nil { |
| 423 | return d, discoverErr |
| 424 | } |
| 425 | if !sameCensus(fresh, expectedVersions) { |
| 426 | return d, ErrChanged |
| 427 | } |
| 428 | } |
| 429 | _, err = hostclaim.Update(ctx, e.Claims, obs, func(c *hostclaim.Claim) error { |
| 430 | if c.Holder != nil { |
| 431 | if err := hostclaim.CoverParkedPredecessors(obs.Claim.Op.ID, obs.Claim.Op.Phase, e.Region, generations)(c); err != nil { |
| 432 | return err |
| 433 | } |
| 434 | } |
| 435 | if c.Free() { |
| 436 | c.Op = hostclaim.Op{Kind: hostclaim.KindParking, ID: op, Intent: map[string]string{"park_sweep": "v1"}} |
| 437 | } |
| 438 | c.Obligations = append(c.Obligations, chores...) |
| 439 | return nil |
| 440 | }) |
| 441 | if err != nil { |
| 442 | return d, err |
| 443 | } |
| 444 | for _, o := range chores { |
| 445 | if err := e.Run(ctx, Work{host, o.ID, o.Op}); err != nil { |
| 446 | return d, fmt.Errorf("sweep %s: %w", o.ID, err) |
| 447 | } |
| 448 | } |
| 449 | return d, nil |
| 450 | } |
| 451 | |
| 452 | // ResidueLocations uses the same namespace and lock classifier as Decide. |
| 453 | func ResidueLocations(host string, versions []lease.ObjectVersion) ([]string, error) { |
| 454 | seen := map[string]bool{} |
| 455 | for _, v := range versions { |
| 456 | loc, err := locationOf(host, v.Key) |
| 457 | if err != nil || v.VersionID == "" { |
| 458 | return nil, ErrUnsafe |
| 459 | } |
| 460 | if !isLock(host, loc, v.Key) { |
| 461 | seen[loc] = true |
| 462 | } |
| 463 | } |
| 464 | out := make([]string, 0, len(seen)) |
| 465 | for loc := range seen { |
| 466 | out = append(out, loc) |
| 467 | } |
| 468 | sort.Strings(out) |
| 469 | return out, nil |
| 470 | } |
| 471 | |
amadan