| 1 | package console |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/rand" |
| 6 | "crypto/sha256" |
| 7 | "encoding/hex" |
| 8 | "encoding/json" |
| 9 | "errors" |
| 10 | "fmt" |
| 11 | "log" |
| 12 | "net/http" |
| 13 | "os" |
| 14 | "slices" |
| 15 | "sort" |
| 16 | "strings" |
| 17 | "time" |
| 18 | |
| 19 | "github.com/carlosframework/platform/internal/domains" |
| 20 | "github.com/carlosframework/platform/internal/hostclaim" |
| 21 | "github.com/carlosframework/platform/internal/hostops" |
| 22 | "github.com/carlosframework/platform/internal/moves" |
| 23 | "github.com/carlosframework/platform/internal/provision" |
| 24 | "github.com/carlosframework/platform/internal/release" |
| 25 | "github.com/carlosframework/platform/internal/status" |
| 26 | ) |
| 27 | |
| 28 | // The offline operator move's doors (docs/superpowers/plans/2026-10-20- |
| 29 | // offline-operator-move.md, Task 6). One hibernating instance moves between |
| 30 | // two regions with planned downtime, an operator at each step: |
| 31 | // |
| 32 | // POST .../instances/{host}/move/offline/stage {"region":"uk"} pre-copy: -> staging |
| 33 | // POST .../instances/{host}/move/offline {"region":"uk"} start (staging -> stopping) |
| 34 | // POST .../instances/{host}/move/offline/seal stopping -> sealing |
| 35 | // POST .../instances/{host}/move/offline/flip sealed -> landing |
| 36 | // POST .../instances/{host}/move/offline/open landed -> open |
| 37 | // POST .../instances/{host}/move/offline/finish open -> finishing -> cleared |
| 38 | // POST .../instances/{host}/move/offline/abort back to the source |
| 39 | // GET .../instances/{host}/move/offline the status view |
| 40 | // |
| 41 | // Every door is moveOperatorGate (the byte-identical 404 for anyone but an |
| 42 | // operator's own CLI token), then MV-O20 when this console runs host claims |
| 43 | // or Phase 1 moves, then its own checks, then exactly one SetRecordMove |
| 44 | // transition and a serial bump, and answers 202 with the status view. |
| 45 | // |
| 46 | // The doors decide from evidence the console reads and never from what the |
| 47 | // operator says: the record, the edges files, the boxes' status reports and |
| 48 | // the reports each box writes under control/offline-moves/. The console |
| 49 | // writes nothing but the record. |
| 50 | |
| 51 | // offlineMoveCapability is the edges-file capability a box publishes only |
| 52 | // when hibernation is configured and host claims, Phase 1 and an armed or |
| 53 | // enforced guard are all off (internal/host, Task 3). |
| 54 | const offlineMoveCapability = "offline-move-1" |
| 55 | |
| 56 | // offlineStageCapability is the one a box publishes beside it when it can |
| 57 | // stage (internal/host movesoffline_stage.go): run the pre-copy as the |
| 58 | // stager and remove it on abort. A box without it treats a staging record |
| 59 | // as nothing to do, so the stage door refuses a region holding one. |
| 60 | const offlineStageCapability = "offline-stage-1" |
| 61 | |
| 62 | // offlineReportFresh is how recently a box's report must be written for a |
| 63 | // door to believe it. Boxes rewrite their reports on every moves pass, and a |
| 64 | // source or sealer box every 15 s or so while its report is one a door reads |
| 65 | // (internal/host offlineReassertEvery), so a report older than this is from a |
| 66 | // box that has stopped reporting, and the door says so (MV-O178). |
| 67 | const offlineReportFresh = 2 * time.Minute |
| 68 | |
| 69 | // Every sentence a door speaks has its row in the copy record (MV-O20 to |
| 70 | // MV-O31, MV-O50, MV-O51) or is an existing approved one, written out whole |
| 71 | // where it is used so the copy gate reads it. |
| 72 | |
| 73 | // offlineName is a region code as the operator reads it: the home's |
| 74 | // configured name, or the word home when none is configured. |
| 75 | func (s *Server) offlineName(code string) string { |
| 76 | if name := s.homeRegionOut(code); name != "" { |
| 77 | return name |
| 78 | } |
| 79 | return "home" |
| 80 | } |
| 81 | |
| 82 | // offlineDoor is what every door has by the time its own checks begin. |
| 83 | type offlineDoor struct { |
| 84 | acct *Account |
| 85 | app string |
| 86 | host string |
| 87 | rec *provision.Record |
| 88 | addr string |
| 89 | } |
| 90 | |
| 91 | // offlineGate is the front half every door shares: the move family's gate |
| 92 | // and operator check (a 404 for anyone else), then MV-O20. |
| 93 | func (s *Server) offlineGate(w http.ResponseWriter, r *http.Request) (offlineDoor, bool) { |
| 94 | acct, app, host, rec, addr, ok := s.moveOperatorGate(w, r) |
| 95 | if !ok { |
| 96 | return offlineDoor{}, false |
| 97 | } |
| 98 | if s.cfg.HostClaim || s.cfg.MovesEnabled { |
| 99 | jsonError(w, http.StatusConflict, "offline moves are not available while host claims or Phase 1 moves are on") |
| 100 | return offlineDoor{}, false |
| 101 | } |
| 102 | return offlineDoor{acct: acct, app: app, host: host, rec: rec, addr: addr}, true |
| 103 | } |
| 104 | |
| 105 | // offlineInFlight is the move of a door that acts on one: an offline move |
| 106 | // the record carries, else MV-O24. |
| 107 | func (s *Server) offlineInFlight(w http.ResponseWriter, d offlineDoor) (*provision.MoveState, bool) { |
| 108 | if d.rec.Move == nil || !d.rec.Move.Offline() { |
| 109 | jsonError(w, http.StatusConflict, fmt.Sprintf("%s has no offline move in flight", d.host)) |
| 110 | return nil, false |
| 111 | } |
| 112 | return d.rec.Move, true |
| 113 | } |
| 114 | |
| 115 | // offlineStateRefusal is MV-O25. |
| 116 | func offlineStateRefusal(state, host string) *offlineRefusal { |
| 117 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("this move is %s. run carlos move offline status --host %s to see what it waits for", state, host)} |
| 118 | } |
| 119 | |
| 120 | // offlineWaitRefusal is M-C10 with <from> filled in: a source box has not |
| 121 | // confirmed the stop. |
| 122 | func (s *Server) offlineWaitRefusal(box, from, host string) *offlineRefusal { |
| 123 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("Waiting for %s in %s to confirm it has stopped %s. %s is offline.", box, from, host, host)} |
| 124 | } |
| 125 | |
| 126 | // offlineStaleRefusal is MV-O178: box's report is there but older than |
| 127 | // offlineReportFresh. That says nothing about the box's process or its stop, |
| 128 | // only that the door cannot believe the report yet; a box taking part in a |
| 129 | // move rewrites it every few seconds, so the step is retried, not looked into |
| 130 | // (harness Run 9, finding 7: flip answered MV-O31 for a stale report, and an |
| 131 | // operator went looking for a process that was not there). |
| 132 | func offlineStaleRefusal(box string) *offlineRefusal { |
| 133 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("the report from %s is not current. it reports again within a minute. try again then", box)} |
| 134 | } |
| 135 | |
| 136 | // offlineRefusal is what a door refuses with: the status and the sentence. |
| 137 | type offlineRefusal struct { |
| 138 | status int |
| 139 | msg string |
| 140 | } |
| 141 | |
| 142 | func writeRefusal(w http.ResponseWriter, ref *offlineRefusal) { |
| 143 | jsonError(w, ref.status, ref.msg) |
| 144 | } |
| 145 | |
| 146 | // offlineConflict is the sentence for a transition's refusal. Never the |
| 147 | // error's own text: phase1Conflict passes only recorded sentences and |
| 148 | // answers a fixed one for anything else. |
| 149 | func offlineConflict(err error) string { |
| 150 | msg := strings.TrimPrefix(err.Error(), provision.ErrMoveConflict.Error()+": ") |
| 151 | switch { |
| 152 | // The offline transitions' own recorded sentences (MV-O34 to MV-O43). |
| 153 | // Each is a whole sentence or one format filled with a state, a box or a |
| 154 | // region the record named, so it may go out as it is. |
| 155 | case msg == "this is not an offline move", |
| 156 | msg == "the record carries a different offline move", |
| 157 | msg == "this is an offline move. use carlos move offline", |
| 158 | strings.HasPrefix(msg, "a seal names the box that seals and the boxes that stopped the instance."), |
| 159 | strings.HasPrefix(msg, "this offline move is ") && strings.Contains(msg, ". this step follows "), |
| 160 | strings.HasPrefix(msg, "this offline move is ") && strings.Contains(msg, " is sealed by "), |
| 161 | strings.HasPrefix(msg, "this offline move is ") && strings.HasSuffix(msg, ". it cannot be aborted once it is open"), |
| 162 | strings.HasPrefix(msg, "this offline move is ") && strings.HasSuffix(msg, ". it finishes from finishing or reverting"), |
| 163 | strings.HasPrefix(msg, "the record's region is ") && strings.HasSuffix(msg, " the instance changed under the move"), |
| 164 | strings.HasPrefix(msg, "the instance is in ") && strings.Contains(msg, ", not "), |
| 165 | // The move family's own recorded refusals, as phase1Conflict passes |
| 166 | // them. |
| 167 | msg == "no move in flight", |
| 168 | msg == "malformed move start", |
| 169 | strings.HasPrefix(msg, "a move to ") && strings.Contains(msg, " is already "), |
| 170 | strings.HasPrefix(msg, "the instance is already in "), |
| 171 | strings.HasPrefix(msg, "a ") && strings.HasSuffix(msg, " does not move. delete it and create one in the new region"): |
| 172 | return msg |
| 173 | } |
| 174 | return "this move's state does not allow that. read its state and try again" |
| 175 | } |
| 176 | |
| 177 | // ---- evidence ---- |
| 178 | |
| 179 | // offlineReport is one report read from the bucket, with the bytes its |
| 180 | // digest is taken over. |
| 181 | type offlineReport struct { |
| 182 | moves.OfflineReport |
| 183 | raw []byte |
| 184 | } |
| 185 | |
| 186 | // offlineEvidence is every report of one move, as read, and its seal: the |
| 187 | // immutable record of what the sealer decided (moves.OfflineSealKey), nil |
| 188 | // when absent or not a valid seal of this move. |
| 189 | type offlineEvidence struct { |
| 190 | reports []offlineReport |
| 191 | seal *offlineReport |
| 192 | } |
| 193 | |
| 194 | func (e offlineEvidence) get(role, box string) (offlineReport, bool) { |
| 195 | for _, r := range e.reports { |
| 196 | if r.Role == role && r.Box == box { |
| 197 | return r, true |
| 198 | } |
| 199 | } |
| 200 | return offlineReport{}, false |
| 201 | } |
| 202 | |
| 203 | func (e offlineEvidence) byRole(role string) []offlineReport { |
| 204 | var out []offlineReport |
| 205 | for _, r := range e.reports { |
| 206 | if r.Role == role { |
| 207 | out = append(out, r) |
| 208 | } |
| 209 | } |
| 210 | return out |
| 211 | } |
| 212 | |
| 213 | // readOfflineEvidence lists and reads every report of the move. A report |
| 214 | // that does not parse, fails moves.ValidateOfflineReport, or does not sit at |
| 215 | // the key its own fields name is not evidence and is skipped, which can only |
| 216 | // make a door refuse. A listing or a read that fails is the error. |
| 217 | func (s *Server) readOfflineEvidence(ctx context.Context, host, id string) (offlineEvidence, error) { |
| 218 | prefix := moves.OfflineReportPrefix(host, id) |
| 219 | if prefix == "" { |
| 220 | return offlineEvidence{}, nil |
| 221 | } |
| 222 | keys, err := s.cfg.Releases.List(ctx, prefix) |
| 223 | if err != nil { |
| 224 | return offlineEvidence{}, err |
| 225 | } |
| 226 | var ev offlineEvidence |
| 227 | sealKey := moves.OfflineSealKey(host, id) |
| 228 | for _, key := range keys { |
| 229 | if key == sealKey { |
| 230 | continue // read below, as the seal and not as a report |
| 231 | } |
| 232 | raw, _, err := s.cfg.Releases.Get(ctx, key) |
| 233 | if errors.Is(err, os.ErrNotExist) { |
| 234 | continue |
| 235 | } |
| 236 | if err != nil { |
| 237 | return offlineEvidence{}, err |
| 238 | } |
| 239 | var rep moves.OfflineReport |
| 240 | if err := json.Unmarshal(raw, &rep); err != nil || moves.ValidateOfflineReport(rep) != nil || |
| 241 | rep.Host != host || rep.Op != id || moves.OfflineReportKey(rep.Host, rep.Op, rep.Role, rep.Box) != key { |
| 242 | log.Printf("console: offline move %s: ignoring %s: not a report of this move", host, key) |
| 243 | continue |
| 244 | } |
| 245 | ev.reports = append(ev.reports, offlineReport{OfflineReport: rep, raw: raw}) |
| 246 | } |
| 247 | sort.Slice(ev.reports, func(i, j int) bool { |
| 248 | if ev.reports[i].Role != ev.reports[j].Role { |
| 249 | return ev.reports[i].Role < ev.reports[j].Role |
| 250 | } |
| 251 | return ev.reports[i].Box < ev.reports[j].Box |
| 252 | }) |
| 253 | raw, _, err := s.cfg.Releases.Get(ctx, sealKey) |
| 254 | switch { |
| 255 | case errors.Is(err, os.ErrNotExist): |
| 256 | case err != nil: |
| 257 | return offlineEvidence{}, err |
| 258 | default: |
| 259 | var rep moves.OfflineReport |
| 260 | if err := json.Unmarshal(raw, &rep); err != nil || moves.ValidateOfflineSeal(rep, host, id) != nil { |
| 261 | log.Printf("console: offline move %s: ignoring %s: not the seal of this move", host, sealKey) |
| 262 | } else { |
| 263 | ev.seal = &offlineReport{OfflineReport: rep, raw: raw} |
| 264 | } |
| 265 | } |
| 266 | return ev, nil |
| 267 | } |
| 268 | |
| 269 | func offlineFresh(r moves.OfflineReport, now time.Time) bool { |
| 270 | age := now.Sub(r.At) |
| 271 | return !r.At.IsZero() && age <= offlineReportFresh && age >= -time.Minute |
| 272 | } |
| 273 | |
| 274 | func offlineStopped(r moves.OfflineReport) bool { return r.Held && r.Fenced && r.Quiesced } |
| 275 | |
| 276 | // offlineSealItems and offlineLandingItems are the one place the doors ask |
| 277 | // the manifest which items a report must carry OK: flip needs the sealer's |
| 278 | // report OK for every item the source side reports (Source "seal"), open the |
| 279 | // target's for every item the landing owns. |
| 280 | func offlineSealItems(r moves.OfflineReport) (bool, []string) { |
| 281 | return r.SourceItemsOK() |
| 282 | } |
| 283 | |
| 284 | func offlineLandingItems(r moves.OfflineReport) (bool, []string) { |
| 285 | return r.ItemsOK(moves.OfflineOwnerLanding) |
| 286 | } |
| 287 | |
| 288 | // offlineEvidenceMatches says whether the record's source.evidence names |
| 289 | // this report: the report's sha256 as hex, bare or after the report's key |
| 290 | // and one separator. |
| 291 | func offlineEvidenceMatches(evidence, key string, raw []byte) bool { |
| 292 | sum := sha256.Sum256(raw) |
| 293 | h := hex.EncodeToString(sum[:]) |
| 294 | tail, ok := strings.CutPrefix(evidence, key) |
| 295 | if ok && tail != "" && strings.Contains("@: #", tail[:1]) { |
| 296 | evidence = tail[1:] |
| 297 | } |
| 298 | return evidence == h || evidence == "sha256:"+h |
| 299 | } |
| 300 | |
| 301 | // offlineFleet is the edges files and each box's status report, read |
| 302 | // together. Incomplete evidence is not an answer: ok is false. |
| 303 | type offlineFleet struct { |
| 304 | snap edgesSnapshot |
| 305 | metrics map[string]*status.BoxMetrics |
| 306 | } |
| 307 | |
| 308 | func (s *Server) readOfflineFleet(ctx context.Context) (offlineFleet, bool) { |
| 309 | snap, err := s.liveEdges(ctx) |
| 310 | if err != nil || !snap.EvidenceComplete() { |
| 311 | log.Printf("console: offline move: the edges files are not complete evidence (error %v)", err) |
| 312 | return offlineFleet{}, false |
| 313 | } |
| 314 | boxes, complete, err := status.LoadBoxesComplete(ctx, s.cfg.Releases) |
| 315 | if err != nil || !complete { |
| 316 | log.Printf("console: offline move: the boxes' status reports are not complete evidence (complete %v, error %v)", complete, err) |
| 317 | return offlineFleet{}, false |
| 318 | } |
| 319 | f := offlineFleet{snap: snap, metrics: map[string]*status.BoxMetrics{}} |
| 320 | for _, b := range boxes { |
| 321 | f.metrics[b.Label] = b.Metrics |
| 322 | } |
| 323 | sort.Slice(f.snap.Files, func(i, j int) bool { return f.snap.Files[i].Box < f.snap.Files[j].Box }) |
| 324 | return f, true |
| 325 | } |
| 326 | |
| 327 | // offlineCanHold says whether a box's published state lets it hold an |
| 328 | // ordinary instance: hibernation not configured off, and the shared pool not |
| 329 | // served off. Absent is unknown, and unknown could hold. |
| 330 | func offlineCanHold(f edgesFilePlacementView) bool { |
| 331 | return (f.HibernationConfigured == nil || *f.HibernationConfigured) && |
| 332 | (f.ServesSharedPool == nil || *f.ServesSharedPool) |
| 333 | } |
| 334 | |
| 335 | func offlineCapable(f edgesFilePlacementView) bool { |
| 336 | return slices.Contains(f.Capabilities, offlineMoveCapability) |
| 337 | } |
| 338 | |
| 339 | func offlineServes(f edgesFilePlacementView, host string) bool { |
| 340 | _, ok := f.Hosts[host] |
| 341 | return ok |
| 342 | } |
| 343 | |
| 344 | // offlineClaimsOnBox is what a claims-on box has to show the start door |
| 345 | // (ruling 2026-10-20, point 3). Every field is published state; none is the |
| 346 | // operator's word. Census and SharedPool are pointers: nil is "the box does |
| 347 | // not say", and an unsaid fact refuses. |
| 348 | type offlineClaimsOnBox struct { |
| 349 | HostRow bool // its edges file lists the host |
| 350 | Census *bool // its report says root's census.json is present |
| 351 | SharedPool *bool // its edges file says whether it serves the shared pool |
| 352 | } |
| 353 | |
| 354 | // ok: no row for the host, census.json reported present, and pools that do |
| 355 | // not match the record's placement (the record is in the shared pool, so a |
| 356 | // box that serves it matches). |
| 357 | func (b offlineClaimsOnBox) ok() bool { |
| 358 | return !b.HostRow && b.Census != nil && *b.Census && b.SharedPool != nil && !*b.SharedPool |
| 359 | } |
| 360 | |
| 361 | // offlineCensusReported is whether a box's status report says root's |
| 362 | // census.json is present. No report field carries it today, and the move |
| 363 | // adds none, so every claims-on box fails this condition until one does. |
| 364 | func offlineCensusReported(*status.BoxMetrics) *bool { return nil } |
| 365 | |
| 366 | // offlineSourceBoxes are the edges files of the record's source placement |
| 367 | // that could hold the host, in box order. |
| 368 | func (f offlineFleet) boxesAt(code string) []edgesFilePlacementView { |
| 369 | var out []edgesFilePlacementView |
| 370 | for _, b := range f.snap.Files { |
| 371 | if b.Placement == code && offlineCanHold(b) { |
| 372 | out = append(out, b) |
| 373 | } |
| 374 | } |
| 375 | return out |
| 376 | } |
| 377 | |
| 378 | // ---- start ---- |
| 379 | |
| 380 | // offlineSelect is the start door's box rules, from the fleet read and |
| 381 | // nothing else: exactly one target box, at least one source box, the |
| 382 | // capability on each, no claims-on box in either region that could touch the |
| 383 | // host, an empty target, the deployment's secrets key on the target and a |
| 384 | // published data bucket there. |
| 385 | func (s *Server) offlineSelect(ctx context.Context, f offlineFleet, host, from, to string) *offlineRefusal { |
| 386 | fromName, toName := s.offlineName(from), s.offlineName(to) |
| 387 | src, tgt := f.boxesAt(from), f.boxesAt(to) |
| 388 | if len(tgt) != 1 { |
| 389 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s has %d boxes. an offline move needs exactly one box in the region it moves to", toName, len(tgt))} |
| 390 | } |
| 391 | if len(src) == 0 { |
| 392 | return &offlineRefusal{status: http.StatusBadRequest, msg: "no eligible box in " + fromName + " publishes an edges file"} |
| 393 | } |
| 394 | for _, b := range append(slices.Clone(src), tgt...) { |
| 395 | if !offlineCapable(b) { |
| 396 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s cannot take part in an offline move. it runs an older release or has host claims on", b.Box)} |
| 397 | } |
| 398 | } |
| 399 | // Ruling 2026-10-20, point 3: a box in either region that runs claims, or |
| 400 | // whose report does not say it does not, must publish no row for the |
| 401 | // host, report census.json present and serve pools the record's |
| 402 | // placement does not match. A box with the capability has vouched for |
| 403 | // itself (it publishes it only with claims off) unless its report says |
| 404 | // claims are on. |
| 405 | for _, b := range f.snap.Files { |
| 406 | if b.Placement != from && b.Placement != to { |
| 407 | continue |
| 408 | } |
| 409 | m := f.metrics[b.Box] |
| 410 | if (offlineCapable(b) && (m == nil || !m.HostClaimEnabled)) || (m != nil && !m.HostClaimEnabled) { |
| 411 | continue |
| 412 | } |
| 413 | if !(offlineClaimsOnBox{HostRow: offlineServes(b, host), Census: offlineCensusReported(m), SharedPool: b.ServesSharedPool}).ok() { |
| 414 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s cannot take part in an offline move. it runs an older release or has host claims on", b.Box)} |
| 415 | } |
| 416 | } |
| 417 | t := tgt[0] |
| 418 | if offlineServes(t, host) { |
| 419 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s already serves %s. an offline move needs the region it moves to to be empty of it", t.Box, host)} |
| 420 | } |
| 421 | if fp := s.expectedSecretsFP(); fp == "" || t.SecretsKeyFP != fp { |
| 422 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s does not hold this deployment's secrets key", t.Box)} |
| 423 | } |
| 424 | if t.DataBucket == "" { |
| 425 | caps, _, _ := movePlacements(ctx, s.cfg.Releases, s.moveHomeName()) |
| 426 | return &offlineRefusal{status: http.StatusBadRequest, msg: "region " + toName + " cannot take a move in this deployment. it publishes no staging bucket. Regions that can: " + s.stagingNames(caps)} |
| 427 | } |
| 428 | return nil |
| 429 | } |
| 430 | |
| 431 | // offlineClaimEvidence refuses a host that has, anywhere, a claim: the |
| 432 | // record's acquisition, control/hosts/<host>.json, a domains claim carrying |
| 433 | // one, or the app's admission (MV-O29); and a deployment that has begun the |
| 434 | // claim backfill (MV-O20). A read that fails is a 502 and never an absence. |
| 435 | func (s *Server) offlineClaimEvidence(ctx context.Context, d offlineDoor) *offlineRefusal { |
| 436 | has := &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s has a host claim. offline moves are only for hosts without one", d.host)} |
| 437 | unreadable := &offlineRefusal{status: http.StatusBadGateway, msg: "could not read the host claim"} |
| 438 | if d.rec.Acquisition != "" { |
| 439 | return has |
| 440 | } |
| 441 | obs, err := hostclaim.Read(ctx, s.cfg.Releases, d.host) |
| 442 | if err != nil { |
| 443 | log.Printf("console: offline move %s: claim read: %v", d.host, err) |
| 444 | return unreadable |
| 445 | } |
| 446 | if !obs.Absent { |
| 447 | return has |
| 448 | } |
| 449 | if c, err := domains.GetClaim(ctx, s.cfg.Releases, d.host); err == nil && c.Acquisition != "" { |
| 450 | return has |
| 451 | } else if err != nil && !errors.Is(err, os.ErrNotExist) { |
| 452 | log.Printf("console: offline move %s: domains claim read: %v", d.host, err) |
| 453 | return unreadable |
| 454 | } |
| 455 | key, err := hostops.AdmissionKey(d.acct.Sqid, d.app) |
| 456 | if err != nil { |
| 457 | log.Printf("console: offline move %s: admission key: %v", d.host, err) |
| 458 | return unreadable |
| 459 | } |
| 460 | if _, _, err := s.cfg.Releases.Get(ctx, key); err == nil { |
| 461 | return has |
| 462 | } else if !errors.Is(err, os.ErrNotExist) { |
| 463 | log.Printf("console: offline move %s: admission read: %v", d.host, err) |
| 464 | return unreadable |
| 465 | } |
| 466 | // A plan at done is allowed; one that has begun, or that cannot be |
| 467 | // verified, is not. |
| 468 | switch err := hostops.CheckCreateFreeze(ctx, s.cfg.Releases, s.cfg.ReleasePub); { |
| 469 | case err == nil: |
| 470 | case errors.Is(err, hostops.ErrPending), errors.Is(err, hostops.ErrIdentity): |
| 471 | return &offlineRefusal{status: http.StatusConflict, msg: "offline moves are not available while host claims or Phase 1 moves are on"} |
| 472 | default: |
| 473 | log.Printf("console: offline move %s: backfill plan read: %v", d.host, err) |
| 474 | return unreadable |
| 475 | } |
| 476 | return nil |
| 477 | } |
| 478 | |
| 479 | // offlineBeginChecks is what start and stage check before either writes: |
| 480 | // the instance's class, its claims, the region asked for, the app's |
| 481 | // regions and the fleet's box rules (offlineSelect). It answers the region |
| 482 | // code and the fleet it read, or false once it has answered the request. |
| 483 | // A staged move is checked again in full when it starts: the fleet and the |
| 484 | // claims may have changed since the stage. |
| 485 | func (s *Server) offlineBeginChecks(w http.ResponseWriter, r *http.Request, d offlineDoor) (string, offlineFleet, bool) { |
| 486 | rec, host := d.rec, d.host |
| 487 | if s.refuseSeedMove(w, rec) { |
| 488 | return "", offlineFleet{}, false |
| 489 | } |
| 490 | if rec.Role != "" { |
| 491 | jsonError(w, http.StatusBadRequest, "a replica does not move. delete it and create one in the new region") |
| 492 | return "", offlineFleet{}, false |
| 493 | } |
| 494 | if len(rec.Replicas) > 0 { |
| 495 | jsonError(w, http.StatusBadRequest, "a home with replicas does not move. remove its replicas first") |
| 496 | return "", offlineFleet{}, false |
| 497 | } |
| 498 | if rec.Steering != "" || rec.Unit != "" { |
| 499 | jsonError(w, http.StatusBadRequest, fmt.Sprintf("%s is steered or runs under a unit. offline moves cannot carry either yet", host)) |
| 500 | return "", offlineFleet{}, false |
| 501 | } |
| 502 | ctx := r.Context() |
| 503 | if ref := s.offlineClaimEvidence(ctx, d); ref != nil { |
| 504 | writeRefusal(w, ref) |
| 505 | return "", offlineFleet{}, false |
| 506 | } |
| 507 | var req struct { |
| 508 | Region string `json:"region"` |
| 509 | } |
| 510 | if !readJSON(w, r, &req) { |
| 511 | return "", offlineFleet{}, false |
| 512 | } |
| 513 | region := strings.TrimSpace(req.Region) |
| 514 | to := s.homeRegionIn(region) |
| 515 | if region == "" || !provision.ValidRegion(to) { |
| 516 | jsonError(w, http.StatusBadRequest, "invalid region") |
| 517 | return "", offlineFleet{}, false |
| 518 | } |
| 519 | if to == rec.Region { |
| 520 | jsonError(w, http.StatusConflict, "the instance is already in "+region) |
| 521 | return "", offlineFleet{}, false |
| 522 | } |
| 523 | m, err := provision.GetManifest(ctx, s.cfg.Releases, d.acct.Sqid, d.app) |
| 524 | if err != nil && !errors.Is(err, os.ErrNotExist) { |
| 525 | log.Printf("console: offline move %s: manifest: %v", host, err) |
| 526 | jsonError(w, http.StatusServiceUnavailable, "could not read the app's instances manifest. try again") |
| 527 | return "", offlineFleet{}, false |
| 528 | } |
| 529 | if m != nil && !s.moveRegionAllowed(m.Regions, to) { |
| 530 | jsonError(w, http.StatusBadRequest, "region "+region+" is not one of this app's regions: "+strings.Join(m.Regions, ", ")) |
| 531 | return "", offlineFleet{}, false |
| 532 | } |
| 533 | fleet, ok := s.readOfflineFleet(ctx) |
| 534 | if !ok { |
| 535 | jsonError(w, http.StatusConflict, "the fleet's edges files could not all be read. try again") |
| 536 | return "", offlineFleet{}, false |
| 537 | } |
| 538 | if to == "" || rec.Region == "" { |
| 539 | // A box that sets CARLOS_REGION to the home's name is home to this |
| 540 | // console and not to itself, so it cannot take a move into home, and |
| 541 | // out of home the placement it publishes is not the record's. |
| 542 | for _, f := range fleet.snap.Files { |
| 543 | if f.Placement != "" && homePlacementCode(s.moveHomeName(), f.Placement) == "" { |
| 544 | jsonError(w, http.StatusConflict, f.Box+" sets CARLOS_REGION, so it cannot take a move into the home region") |
| 545 | return "", offlineFleet{}, false |
| 546 | } |
| 547 | } |
| 548 | } |
| 549 | if ref := s.offlineSelect(ctx, fleet, host, rec.Region, to); ref != nil { |
| 550 | writeRefusal(w, ref) |
| 551 | return "", offlineFleet{}, false |
| 552 | } |
| 553 | return to, fleet, true |
| 554 | } |
| 555 | |
| 556 | // offlineStagedFor is the refusal of a start or a stage that names another |
| 557 | // region than the one the instance is staged for (MV-O179). |
| 558 | func (s *Server) offlineStagedFor(host, to string) *offlineRefusal { |
| 559 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s is staged for %s. abort the move to move it to another region", host, s.offlineName(to))} |
| 560 | } |
| 561 | |
| 562 | // offlineStaged is the record's offline move when it is staging. |
| 563 | func offlineStaged(rec *provision.Record) (*provision.MoveState, bool) { |
| 564 | if mv := rec.Move; mv.Offline() && mv.State == provision.OfflineStaging { |
| 565 | return mv, true |
| 566 | } |
| 567 | return nil, false |
| 568 | } |
| 569 | |
| 570 | // offlineStagerGone is MV-O190: the pinned stager of a staged move no |
| 571 | // longer publishes an edges file in the source region. Its disappearance |
| 572 | // is never proof that it stopped copying, so start and seal refuse. |
| 573 | func (s *Server) offlineStagerGone(box, host, from string) *offlineRefusal { |
| 574 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s, which staged %s, no longer publishes an edges file in %s. bring it back or abort the move", box, host, from)} |
| 575 | } |
| 576 | |
| 577 | // offlineStageCoords refuses a staged move whose target publishes other |
| 578 | // staging coordinates than the ones the stage fixed on the record (MV-O192, |
| 579 | // Astra round 2, finding 2): the staged copy is in the stored bucket, and |
| 580 | // nothing is ever retargeted. offlineSelect has already required exactly |
| 581 | // one target box. |
| 582 | func (s *Server) offlineStageCoords(f offlineFleet, mv *provision.MoveState, host string) *offlineRefusal { |
| 583 | if mv.StagedAt == nil { |
| 584 | return nil |
| 585 | } |
| 586 | tgt := f.boxesAt(mv.To) |
| 587 | if len(tgt) != 1 || tgt[0].DataBucket != mv.StageBucket || tgt[0].DataBucketRegion != mv.StageRegion { |
| 588 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s now publishes a different data bucket or region from what was recorded when %s was staged. abort the move and stage it again", s.offlineName(mv.To), host)} |
| 589 | } |
| 590 | return nil |
| 591 | } |
| 592 | |
| 593 | // offlineStagerPresent refuses a staged move whose pinned stager is not in |
| 594 | // the source region's fleet now (MV-O190). |
| 595 | func (s *Server) offlineStagerPresent(f offlineFleet, mv *provision.MoveState, host string) *offlineRefusal { |
| 596 | if mv.StagedAt == nil { |
| 597 | return nil |
| 598 | } |
| 599 | for _, b := range f.boxesAt(mv.From) { |
| 600 | if b.Box == mv.Stager { |
| 601 | return nil |
| 602 | } |
| 603 | } |
| 604 | return s.offlineStagerGone(mv.Stager, host, s.offlineName(mv.From)) |
| 605 | } |
| 606 | |
| 607 | func (s *Server) handleAPIInstanceOfflineStart(w http.ResponseWriter, r *http.Request) { |
| 608 | d, ok := s.offlineGate(w, r) |
| 609 | if !ok { |
| 610 | return |
| 611 | } |
| 612 | to, fleet, ok := s.offlineBeginChecks(w, r, d) |
| 613 | if !ok { |
| 614 | return |
| 615 | } |
| 616 | // A staged instance starts the move it is staged for: staging to |
| 617 | // stopping, the instance offline from now. The lock is the stage's. |
| 618 | if mv, staged := offlineStaged(d.rec); staged { |
| 619 | if to != mv.To { |
| 620 | writeRefusal(w, s.offlineStagedFor(d.host, mv.To)) |
| 621 | return |
| 622 | } |
| 623 | if ref := s.offlineStagerPresent(fleet, mv, d.host); ref != nil { |
| 624 | writeRefusal(w, ref) |
| 625 | return |
| 626 | } |
| 627 | if ref := s.offlineStageCoords(fleet, mv, d.host); ref != nil { |
| 628 | writeRefusal(w, ref) |
| 629 | return |
| 630 | } |
| 631 | s.offlineAdvance(w, r, d, provision.OfflineStartStaged(mv.ID, d.addr, time.Now())) |
| 632 | return |
| 633 | } |
| 634 | s.offlineBegin(w, r, d, func(id string) func(*provision.Record) error { |
| 635 | return provision.OfflineStart(id, d.rec.Region, to, d.addr, time.Now()) |
| 636 | }) |
| 637 | } |
| 638 | |
| 639 | // handleAPIInstanceOfflineStage is the pre-copy door. On a record at rest it |
| 640 | // creates the move in offline-staging, checked exactly as start is and |
| 641 | // under the same one-move lock, and the source region's stager runs one |
| 642 | // round on the serial bump. On a record already staging for that region it |
| 643 | // moves staged_at forward (OfflineRestage), so the stager runs one more |
| 644 | // round; each round copies only what is new (the receipts). |
| 645 | func (s *Server) handleAPIInstanceOfflineStage(w http.ResponseWriter, r *http.Request) { |
| 646 | d, ok := s.offlineGate(w, r) |
| 647 | if !ok { |
| 648 | return |
| 649 | } |
| 650 | to, fleet, ok := s.offlineBeginChecks(w, r, d) |
| 651 | if !ok { |
| 652 | return |
| 653 | } |
| 654 | if mv, staged := offlineStaged(d.rec); staged { |
| 655 | if to != mv.To { |
| 656 | writeRefusal(w, s.offlineStagedFor(d.host, mv.To)) |
| 657 | return |
| 658 | } |
| 659 | if ref := s.offlineStageCoords(fleet, mv, d.host); ref != nil { |
| 660 | writeRefusal(w, ref) |
| 661 | return |
| 662 | } |
| 663 | s.offlineAdvance(w, r, d, provision.OfflineRestage(mv.ID, time.Now())) |
| 664 | return |
| 665 | } |
| 666 | for _, b := range append(fleet.boxesAt(d.rec.Region), fleet.boxesAt(to)...) { |
| 667 | if !slices.Contains(b.Capabilities, offlineStageCapability) { |
| 668 | jsonError(w, http.StatusConflict, fmt.Sprintf("%s cannot stage an offline move. it runs an older release", b.Box)) |
| 669 | return |
| 670 | } |
| 671 | } |
| 672 | // The stager is pinned here, for the whole move: the one box that ever |
| 673 | // writes this move's staging prefix, and the one whose cleanup an abort |
| 674 | // waits for (Astra round 1, finding 2). offlineSelect has already |
| 675 | // required at least one source box. |
| 676 | var names []string |
| 677 | for _, b := range fleet.boxesAt(d.rec.Region) { |
| 678 | names = append(names, b.Box) |
| 679 | } |
| 680 | // The staging destination is fixed here too, before any copy (Astra |
| 681 | // round 2, finding 2): the one target box's data bucket and region, as |
| 682 | // offlineSelect found it. A target that publishes a source box's own |
| 683 | // bucket is refused: the activator there would promote the staged copy |
| 684 | // over the serving replica on its next wake (MV-O191). |
| 685 | tgt := fleet.boxesAt(to)[0] |
| 686 | for _, b := range fleet.boxesAt(d.rec.Region) { |
| 687 | if b.DataBucket == tgt.DataBucket { |
| 688 | jsonError(w, http.StatusConflict, fmt.Sprintf("%s publishes the same data bucket as %s. a staged copy there would replace the serving one", s.offlineName(to), s.offlineName(d.rec.Region))) |
| 689 | return |
| 690 | } |
| 691 | } |
| 692 | st := provision.StageTarget{Stager: moves.OfflineStager(names), Bucket: tgt.DataBucket, Region: tgt.DataBucketRegion} |
| 693 | s.offlineBegin(w, r, d, func(id string) func(*provision.Record) error { |
| 694 | return provision.OfflineStage(id, d.rec.Region, to, d.addr, st, time.Now()) |
| 695 | }) |
| 696 | } |
| 697 | |
| 698 | // offlineBegin mints the move's id and writes the transition begin makes |
| 699 | // with it (OfflineStart or OfflineStage) under the one-move lock, then |
| 700 | // bumps the serial and answers 202 with the status view. |
| 701 | func (s *Server) offlineBegin(w http.ResponseWriter, r *http.Request, d offlineDoor, begin func(id string) func(*provision.Record) error) { |
| 702 | ctx := r.Context() |
| 703 | host := d.host |
| 704 | id, err := moves.NewMoveID(rand.Reader) |
| 705 | if err != nil { |
| 706 | log.Printf("console: offline move %s: move id: %v", host, err) |
| 707 | jsonError(w, http.StatusBadGateway, "could not mint a move id") |
| 708 | return |
| 709 | } |
| 710 | start := begin(id) |
| 711 | probe := *d.rec |
| 712 | if err := start(&probe); err != nil { |
| 713 | jsonError(w, http.StatusConflict, offlineConflict(err)) |
| 714 | return |
| 715 | } |
| 716 | // One offline move at a time: the lock is taken before the record is |
| 717 | // written, and released if the write fails. |
| 718 | if ref := s.offlineTakeLock(ctx, d, id); ref != nil { |
| 719 | writeRefusal(w, ref) |
| 720 | return |
| 721 | } |
| 722 | // The write is bounded well under offlineLockGrace, so the lock cannot be |
| 723 | // judged stale while it is still in flight. |
| 724 | wctx, cancel := context.WithTimeout(ctx, offlineStartWrite) |
| 725 | defer cancel() |
| 726 | got, err := s.offlineApply(wctx, d, start) |
| 727 | if err != nil { |
| 728 | // A failed write does not prove nothing was written (a commit whose |
| 729 | // reply was lost). The lock is never released here: if the record |
| 730 | // carries this move the start succeeded, and if it does not the |
| 731 | // lock goes stale and frees itself. |
| 732 | rctx, rcancel := context.WithTimeout(context.WithoutCancel(ctx), 10*time.Second) |
| 733 | defer rcancel() |
| 734 | if cur, rerr := provision.GetRecord(rctx, s.cfg.Releases, d.acct.Sqid, d.app, host); rerr == nil && cur.Move.Offline() && cur.Move.ID == id { |
| 735 | got, err = cur, nil |
| 736 | } |
| 737 | } |
| 738 | if err != nil { |
| 739 | s.offlineWriteFailed(w, d, err) |
| 740 | return |
| 741 | } |
| 742 | s.bumpSerial(ctx) |
| 743 | writeJSON(w, http.StatusAccepted, s.offlineViewOf(ctx, got)) |
| 744 | } |
| 745 | |
| 746 | // offlineStartWrite bounds the start door's record write. |
| 747 | const offlineStartWrite = 30 * time.Second |
| 748 | |
| 749 | // offlineApply is one transition through the record writer. |
| 750 | func (s *Server) offlineApply(ctx context.Context, d offlineDoor, mut func(*provision.Record) error) (*provision.Record, error) { |
| 751 | return provision.SetRecordMove(ctx, s.cfg.Releases, d.acct.Sqid, d.app, d.host, mut) |
| 752 | } |
| 753 | |
| 754 | // offlineWriteFailed answers a transition's failure: a refusal with its |
| 755 | // recorded sentence, anything else as unwritten. |
| 756 | func (s *Server) offlineWriteFailed(w http.ResponseWriter, d offlineDoor, err error) { |
| 757 | if errors.Is(err, provision.ErrMoveConflict) { |
| 758 | jsonError(w, http.StatusConflict, offlineConflict(err)) |
| 759 | return |
| 760 | } |
| 761 | log.Printf("console: offline move %s/%s/%s: %v", d.acct.Sqid, d.app, d.host, err) |
| 762 | jsonError(w, http.StatusBadGateway, "could not write the move. try again") |
| 763 | } |
| 764 | |
| 765 | // offlineWrite applies one transition and bumps the serial, or answers the |
| 766 | // refusal itself. When the move it leaves is cleared (finish, or the finish |
| 767 | // that ends an abort) it releases the lock. |
| 768 | func (s *Server) offlineWrite(w http.ResponseWriter, r *http.Request, d offlineDoor, mut func(*provision.Record) error) (*provision.Record, bool) { |
| 769 | ctx := r.Context() |
| 770 | rec, err := s.offlineApply(ctx, d, mut) |
| 771 | if err != nil { |
| 772 | s.offlineWriteFailed(w, d, err) |
| 773 | return nil, false |
| 774 | } |
| 775 | s.bumpSerial(ctx) |
| 776 | if rec.Move == nil && d.rec.Move != nil { |
| 777 | s.offlineReleaseLock(ctx, d.host, d.rec.Move.ID) |
| 778 | } |
| 779 | return rec, true |
| 780 | } |
| 781 | |
| 782 | // offlineAdvance applies one transition and answers 202 with the status |
| 783 | // view of the record it left. |
| 784 | func (s *Server) offlineAdvance(w http.ResponseWriter, r *http.Request, d offlineDoor, mut func(*provision.Record) error) { |
| 785 | if rec, ok := s.offlineWrite(w, r, d, mut); ok { |
| 786 | writeJSON(w, http.StatusAccepted, s.offlineViewOf(r.Context(), rec)) |
| 787 | } |
| 788 | } |
| 789 | |
| 790 | // offlineLockGrace is how long a lock is believed whatever its record says: |
| 791 | // the start door takes the lock, then writes the record. |
| 792 | const offlineLockGrace = 2 * time.Minute |
| 793 | |
| 794 | // offlineTakeLock takes control/offline-moves/active.json for this host and |
| 795 | // move, create-only. A lock already there is live while the record it names |
| 796 | // still carries that move, or is younger than offlineLockGrace (MV-O30 naming |
| 797 | // its host); otherwise it is stale (a crashed start, a failed release) and is |
| 798 | // replaced under its etag. A |
| 799 | // lock that cannot be read is not guessed at. |
| 800 | func (s *Server) offlineTakeLock(ctx context.Context, d offlineDoor, id string) *offlineRefusal { |
| 801 | unreadable := &offlineRefusal{status: http.StatusBadGateway, msg: "could not read the instance"} |
| 802 | body, err := json.Marshal(moves.OfflineActive{Host: d.host, Op: id, Account: d.acct.Sqid, App: d.app, At: time.Now().UTC()}) |
| 803 | if err != nil { |
| 804 | return unreadable |
| 805 | } |
| 806 | _, err = s.cfg.Releases.Put(ctx, moves.OfflineActiveKey, body, "") |
| 807 | if err == nil { |
| 808 | return nil |
| 809 | } |
| 810 | if !errors.Is(err, release.ErrCASConflict) { |
| 811 | log.Printf("console: offline move %s: taking the lock: %v", d.host, err) |
| 812 | return unreadable |
| 813 | } |
| 814 | raw, etag, err := s.cfg.Releases.Get(ctx, moves.OfflineActiveKey) |
| 815 | var held moves.OfflineActive |
| 816 | if err != nil || json.Unmarshal(raw, &held) != nil || held.Host == "" { |
| 817 | log.Printf("console: offline move %s: reading the lock: %v", d.host, err) |
| 818 | return unreadable |
| 819 | } |
| 820 | busy := &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s is moving offline. move one instance at a time", held.Host)} |
| 821 | // A lock is taken before its record is written, so a record without the |
| 822 | // move is only a stale lock once the holder has had time to write it. |
| 823 | if time.Since(held.At) < offlineLockGrace { |
| 824 | return busy |
| 825 | } |
| 826 | other, err := provision.GetRecord(ctx, s.cfg.Releases, held.Account, held.App, held.Host) |
| 827 | switch { |
| 828 | case errors.Is(err, os.ErrNotExist): |
| 829 | case err != nil: |
| 830 | log.Printf("console: offline move %s: reading the lock holder %s: %v", d.host, held.Host, err) |
| 831 | return unreadable |
| 832 | case other.Move.Offline() && other.Move.ID == held.Op: |
| 833 | return busy |
| 834 | } |
| 835 | if _, err := s.cfg.Releases.Put(ctx, moves.OfflineActiveKey, body, etag); err != nil { |
| 836 | if errors.Is(err, release.ErrCASConflict) { |
| 837 | return busy |
| 838 | } |
| 839 | log.Printf("console: offline move %s: replacing a stale lock: %v", d.host, err) |
| 840 | return unreadable |
| 841 | } |
| 842 | return nil |
| 843 | } |
| 844 | |
| 845 | // offlineReleaseLock deletes the lock under the etag it read, when it is |
| 846 | // this move's. A failure is logged and left: the next start finds the lock |
| 847 | // stale (its record no longer carries the move) and replaces it. |
| 848 | func (s *Server) offlineReleaseLock(ctx context.Context, host, id string) { |
| 849 | ctx = context.WithoutCancel(ctx) |
| 850 | raw, etag, err := s.cfg.Releases.Get(ctx, moves.OfflineActiveKey) |
| 851 | if errors.Is(err, os.ErrNotExist) { |
| 852 | return |
| 853 | } |
| 854 | var held moves.OfflineActive |
| 855 | if err != nil || json.Unmarshal(raw, &held) != nil { |
| 856 | log.Printf("console: offline move %s: reading the lock to release it: %v", host, err) |
| 857 | return |
| 858 | } |
| 859 | if held.Host != host || held.Op != id { |
| 860 | return |
| 861 | } |
| 862 | if err := release.DeleteIfMatch(ctx, s.cfg.Releases, moves.OfflineActiveKey, etag); err != nil { |
| 863 | log.Printf("console: offline move %s: releasing the lock: %v", host, err) |
| 864 | } |
| 865 | } |
| 866 | |
| 867 | // ---- the step each state waits for ---- |
| 868 | |
| 869 | // The verbs. |
| 870 | const ( |
| 871 | offlineVerbStart = "start" |
| 872 | offlineVerbSeal = "seal" |
| 873 | offlineVerbFlip = "flip" |
| 874 | offlineVerbOpen = "open" |
| 875 | offlineVerbFinish = "finish" |
| 876 | ) |
| 877 | |
| 878 | // offlineVerbOf is the one door a state is waiting on. sealing and landing |
| 879 | // wait for a box, not for the operator. |
| 880 | func offlineVerbOf(state string) string { |
| 881 | switch state { |
| 882 | case provision.OfflineStaging: |
| 883 | return offlineVerbStart |
| 884 | case provision.OfflineStopping: |
| 885 | return offlineVerbSeal |
| 886 | case provision.OfflineSealed: |
| 887 | return offlineVerbFlip |
| 888 | case provision.OfflineLanded: |
| 889 | return offlineVerbOpen |
| 890 | case provision.OfflineOpen, provision.OfflineFinishing, provision.OfflineReverting: |
| 891 | return offlineVerbFinish |
| 892 | } |
| 893 | return "" |
| 894 | } |
| 895 | |
| 896 | // offlineStep is a door's checks passed: the transition to write. |
| 897 | type offlineStep struct { |
| 898 | mut func(*provision.Record) error |
| 899 | } |
| 900 | |
| 901 | // offlineOpenMut is the open transition. The one place the door names it. |
| 902 | func offlineOpenMut(id string, now time.Time) func(*provision.Record) error { |
| 903 | return provision.OfflineOpenTransition(id, now) |
| 904 | } |
| 905 | |
| 906 | // offlineStepFor runs the checks of the verb the move's state waits for, |
| 907 | // from the evidence. A refusal is the sentence the door answers and the |
| 908 | // status view shows as what the move waits for. |
| 909 | func (s *Server) offlineStepFor(ctx context.Context, d offlineDoor, mv *provision.MoveState, ev offlineEvidence, now time.Time) (offlineStep, *offlineRefusal) { |
| 910 | host, from := d.host, s.offlineName(mv.From) |
| 911 | switch mv.State { |
| 912 | case provision.OfflineStaging: |
| 913 | // What start would check of the fleet now; its request checks |
| 914 | // (the class, the claims, the region) are the start door's. |
| 915 | fleet, ok := s.readOfflineFleet(ctx) |
| 916 | if !ok { |
| 917 | return offlineStep{}, &offlineRefusal{status: http.StatusConflict, msg: "the fleet's edges files could not all be read. try again"} |
| 918 | } |
| 919 | if ref := s.offlineSelect(ctx, fleet, host, mv.From, mv.To); ref != nil { |
| 920 | return offlineStep{}, ref |
| 921 | } |
| 922 | if ref := s.offlineStagerPresent(fleet, mv, host); ref != nil { |
| 923 | return offlineStep{}, ref |
| 924 | } |
| 925 | if ref := s.offlineStageCoords(fleet, mv, host); ref != nil { |
| 926 | return offlineStep{}, ref |
| 927 | } |
| 928 | return offlineStep{mut: provision.OfflineStartStaged(mv.ID, d.addr, now)}, nil |
| 929 | case provision.OfflineStopping: |
| 930 | fleet, ok := s.readOfflineFleet(ctx) |
| 931 | if !ok { |
| 932 | return offlineStep{}, &offlineRefusal{status: http.StatusConflict, msg: "the fleet's edges files could not all be read. try again"} |
| 933 | } |
| 934 | sealer, boxes, ref := s.offlineSealCheck(fleet, ev, mv, host, from, now) |
| 935 | if ref != nil { |
| 936 | return offlineStep{}, ref |
| 937 | } |
| 938 | return offlineStep{mut: provision.OfflineSeal(mv.ID, sealer, boxes)}, nil |
| 939 | case provision.OfflineSealed: |
| 940 | if ref := s.offlineFlipCheck(ev, mv, host, from, now); ref != nil { |
| 941 | return offlineStep{}, ref |
| 942 | } |
| 943 | return offlineStep{mut: provision.OfflineFlip(mv.ID)}, nil |
| 944 | case provision.OfflineLanded: |
| 945 | if ref := s.offlineOpenCheck(ev, mv, host, from, now); ref != nil { |
| 946 | return offlineStep{}, ref |
| 947 | } |
| 948 | return offlineStep{mut: offlineOpenMut(mv.ID, now)}, nil |
| 949 | case provision.OfflineOpen: |
| 950 | if ref := offlineFinishCheck(ev, mv, host, moves.OfflinePhaseOpened, "", moves.OfflinePhaseParked, mv.Boxes); ref != nil { |
| 951 | return offlineStep{}, ref |
| 952 | } |
| 953 | return offlineStep{mut: provision.OfflineFinishingTransition(mv.ID)}, nil |
| 954 | case provision.OfflineFinishing: |
| 955 | if ref := offlineFinishCheck(ev, mv, host, "", moves.OfflinePhaseLifted, moves.OfflinePhaseGone, mv.Boxes); ref != nil { |
| 956 | return offlineStep{}, ref |
| 957 | } |
| 958 | return offlineStep{mut: provision.OfflineFinish(mv.ID, now)}, nil |
| 959 | case provision.OfflineReverting: |
| 960 | // Every box the seal named, and the sealer. A move aborted before |
| 961 | // the seal named neither: the boxes are the source region's as the |
| 962 | // edges files publish them now, and there is no sealer (it never |
| 963 | // took a tombstone). |
| 964 | boxes, sealerPhase := mv.Boxes, moves.OfflinePhaseReverted |
| 965 | if mv.Sealer == "" { |
| 966 | fleet, ok := s.readOfflineFleet(ctx) |
| 967 | if !ok { |
| 968 | return offlineStep{}, &offlineRefusal{status: http.StatusConflict, msg: "the fleet's edges files could not all be read. try again"} |
| 969 | } |
| 970 | boxes, sealerPhase = nil, "" |
| 971 | for _, b := range fleet.boxesAt(mv.From) { |
| 972 | boxes = append(boxes, b.Box) |
| 973 | } |
| 974 | } |
| 975 | if ref := offlineFinishCheck(ev, mv, host, "", sealerPhase, moves.OfflinePhaseReverted, boxes); ref != nil { |
| 976 | return offlineStep{}, ref |
| 977 | } |
| 978 | // A staged move also waits for its pinned stager to say the staged |
| 979 | // copy is gone, whatever its last round reported (Astra round 1, |
| 980 | // findings 1 and 2). |
| 981 | if mv.StagedAt != nil { |
| 982 | if rep, ok := ev.get(moves.OfflineRoleStager, mv.Stager); !ok || rep.Phase != moves.OfflinePhaseReverted { |
| 983 | return offlineStep{}, offlineStateRefusal(mv.State, host) |
| 984 | } |
| 985 | } |
| 986 | return offlineStep{mut: provision.OfflineFinish(mv.ID, now)}, nil |
| 987 | } |
| 988 | return offlineStep{}, offlineStateRefusal(mv.State, host) |
| 989 | } |
| 990 | |
| 991 | // offlineSealCheck: every source box reports this move held, fenced and |
| 992 | // quiesced, and its edges file no longer lists the host. It names the sealer |
| 993 | // and the boxes the later doors wait for. |
| 994 | func (s *Server) offlineSealCheck(f offlineFleet, ev offlineEvidence, mv *provision.MoveState, host, from string, now time.Time) (sealer string, boxes []string, ref *offlineRefusal) { |
| 995 | src := f.boxesAt(mv.From) |
| 996 | if len(src) == 0 { |
| 997 | return "", nil, &offlineRefusal{status: http.StatusBadRequest, msg: "no eligible box in " + from + " publishes an edges file"} |
| 998 | } |
| 999 | // A staged move's pinned stager is part of the barrier whatever the |
| 1000 | // fleet says (Astra round 2, finding 1): it must still be in the |
| 1001 | // source region's fleet (so it is one of the boxes below, and of the |
| 1002 | // boxes the seal names), and it must have said, after the move left |
| 1003 | // staging and under its pass lock, that no stage round is in flight |
| 1004 | // and none will start (its stager report, phase stopped, fresh). The |
| 1005 | // staging destination must still be the one the stage fixed. |
| 1006 | if mv.StagedAt != nil { |
| 1007 | if ref := s.offlineStagerPresent(f, mv, host); ref != nil { |
| 1008 | return "", nil, ref |
| 1009 | } |
| 1010 | if ref := s.offlineStageCoords(f, mv, host); ref != nil { |
| 1011 | return "", nil, ref |
| 1012 | } |
| 1013 | ack, ok := ev.get(moves.OfflineRoleStager, mv.Stager) |
| 1014 | if ok && ack.Phase == moves.OfflinePhaseStopped && !offlineFresh(ack.OfflineReport, now) { |
| 1015 | return "", nil, offlineStaleRefusal(mv.Stager) |
| 1016 | } |
| 1017 | if !ok || ack.Phase != moves.OfflinePhaseStopped { |
| 1018 | return "", nil, s.offlineWaitRefusal(mv.Stager, from, host) |
| 1019 | } |
| 1020 | } |
| 1021 | var data []string |
| 1022 | for _, b := range src { |
| 1023 | rep, ok := ev.get(moves.OfflineRoleSource, b.Box) |
| 1024 | if ok && !offlineFresh(rep.OfflineReport, now) { |
| 1025 | return "", nil, offlineStaleRefusal(b.Box) |
| 1026 | } |
| 1027 | if !ok || !offlineStopped(rep.OfflineReport) || offlineServes(b, host) { |
| 1028 | return "", nil, s.offlineWaitRefusal(b.Box, from, host) |
| 1029 | } |
| 1030 | boxes = append(boxes, b.Box) |
| 1031 | if rep.LocalDB || rep.Files || rep.Env { |
| 1032 | data = append(data, b.Box) |
| 1033 | } |
| 1034 | } |
| 1035 | for _, b := range boxes { |
| 1036 | if rep, _ := ev.get(moves.OfflineRoleSource, b); rep.Sidecar { |
| 1037 | return "", nil, &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s keeps a sidecar file for %s. instances with sidecars cannot move yet", b, host)} |
| 1038 | } |
| 1039 | } |
| 1040 | if len(data) > 1 { |
| 1041 | return "", nil, &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s keeps local data on more than one box: %s. remove the stale copy first", host, strings.Join(data, ", "))} |
| 1042 | } |
| 1043 | sealer = boxes[0] |
| 1044 | if len(data) == 1 { |
| 1045 | sealer = data[0] |
| 1046 | } |
| 1047 | return sealer, boxes, nil |
| 1048 | } |
| 1049 | |
| 1050 | // offlineSealHolds says the record's source.evidence names the seal on disk, |
| 1051 | // byte for byte. The seal is written once, so the hash holds for the whole |
| 1052 | // move whatever the sealer's living report does each pass. |
| 1053 | func offlineSealHolds(ev offlineEvidence, mv *provision.MoveState, host string) bool { |
| 1054 | return ev.seal != nil && ev.seal.Box == mv.Sealer && mv.Source != nil && |
| 1055 | offlineEvidenceMatches(mv.Source.Evidence, moves.OfflineSealKey(host, mv.ID), ev.seal.raw) |
| 1056 | } |
| 1057 | |
| 1058 | // offlineFlipCheck: the seal is the one the record names and every item the |
| 1059 | // source side reports is OK in it; every source box has stopped and left no |
| 1060 | // process or layout journal, on a report this fresh. |
| 1061 | func (s *Server) offlineFlipCheck(ev offlineEvidence, mv *provision.MoveState, host, from string, now time.Time) *offlineRefusal { |
| 1062 | if !offlineSealHolds(ev, mv, host) { |
| 1063 | return offlineStateRefusal(mv.State, host) |
| 1064 | } |
| 1065 | if items, _ := offlineSealItems(ev.seal.OfflineReport); !items { |
| 1066 | return offlineStateRefusal(mv.State, host) |
| 1067 | } |
| 1068 | for _, box := range mv.Boxes { |
| 1069 | rep, ok := ev.get(moves.OfflineRoleSource, box) |
| 1070 | if !ok { |
| 1071 | return s.offlineWaitRefusal(box, from, host) |
| 1072 | } |
| 1073 | if !offlineFresh(rep.OfflineReport, now) { |
| 1074 | return offlineStaleRefusal(box) |
| 1075 | } |
| 1076 | if !offlineStopped(rep.OfflineReport) { |
| 1077 | return s.offlineWaitRefusal(box, from, host) |
| 1078 | } |
| 1079 | if rep.Procs != 0 || rep.LayoutJournal { |
| 1080 | return processRefusal(box, host) |
| 1081 | } |
| 1082 | } |
| 1083 | return nil |
| 1084 | } |
| 1085 | |
| 1086 | func processRefusal(box, host string) *offlineRefusal { |
| 1087 | return &offlineRefusal{status: http.StatusConflict, msg: fmt.Sprintf("%s still runs a process for %s. wait for it to stop, then flip", box, host)} |
| 1088 | } |
| 1089 | |
| 1090 | // offlineOpenCheck is the re-check of the stop: the target's landing report |
| 1091 | // has every landing item OK, the seal is still the one the record names, the |
| 1092 | // sealer's living report is from this pass and holds the tombstone for this |
| 1093 | // move, and every source box is still held, fenced and quiesced. |
| 1094 | func (s *Server) offlineOpenCheck(ev offlineEvidence, mv *provision.MoveState, host, from string, now time.Time) *offlineRefusal { |
| 1095 | tgt := ev.byRole(moves.OfflineRoleTarget) |
| 1096 | if len(tgt) != 1 { |
| 1097 | return offlineStateRefusal(mv.State, host) |
| 1098 | } |
| 1099 | if items, _ := offlineLandingItems(tgt[0].OfflineReport); !items { |
| 1100 | return offlineStateRefusal(mv.State, host) |
| 1101 | } |
| 1102 | if !offlineSealHolds(ev, mv, host) { |
| 1103 | return offlineStateRefusal(mv.State, host) |
| 1104 | } |
| 1105 | seal, ok := ev.get(moves.OfflineRoleSealer, mv.Sealer) |
| 1106 | if ok && !offlineFresh(seal.OfflineReport, now) { |
| 1107 | return offlineStaleRefusal(mv.Sealer) |
| 1108 | } |
| 1109 | if !ok || seal.TombstoneOwner != moves.TombstoneOwner(mv.ID) || !seal.TombstoneUntil.After(now) { |
| 1110 | return offlineStateRefusal(mv.State, host) |
| 1111 | } |
| 1112 | for _, box := range mv.Boxes { |
| 1113 | rep, ok := ev.get(moves.OfflineRoleSource, box) |
| 1114 | if ok && !offlineFresh(rep.OfflineReport, now) { |
| 1115 | return offlineStaleRefusal(box) |
| 1116 | } |
| 1117 | if !ok || !offlineStopped(rep.OfflineReport) { |
| 1118 | return s.offlineWaitRefusal(box, from, host) |
| 1119 | } |
| 1120 | } |
| 1121 | return nil |
| 1122 | } |
| 1123 | |
| 1124 | // offlineFinishCheck: the target's phase (when targetPhase is set), the |
| 1125 | // sealer's phase (when sealerPhase is set) and every source box's phase, |
| 1126 | // each as the box's own report says. Otherwise MV-O25. Open to finishing |
| 1127 | // wants the target opened and every source parked; finishing to cleared the |
| 1128 | // sealer lifted and every source gone; reverting to cleared the sealer and |
| 1129 | // every source reverted (the target is not asked: a target box that died |
| 1130 | // after the flip must not hold the abort). |
| 1131 | func offlineFinishCheck(ev offlineEvidence, mv *provision.MoveState, host, targetPhase, sealerPhase, sourcePhase string, boxes []string) *offlineRefusal { |
| 1132 | refuse := offlineStateRefusal(mv.State, host) |
| 1133 | if targetPhase != "" { |
| 1134 | tgt := ev.byRole(moves.OfflineRoleTarget) |
| 1135 | if len(tgt) != 1 || tgt[0].Phase != targetPhase { |
| 1136 | return refuse |
| 1137 | } |
| 1138 | } |
| 1139 | if sealerPhase != "" { |
| 1140 | if rep, ok := ev.get(moves.OfflineRoleSealer, mv.Sealer); !ok || rep.Phase != sealerPhase { |
| 1141 | return refuse |
| 1142 | } |
| 1143 | } |
| 1144 | if len(boxes) == 0 { |
| 1145 | return refuse |
| 1146 | } |
| 1147 | for _, box := range boxes { |
| 1148 | rep, ok := ev.get(moves.OfflineRoleSource, box) |
| 1149 | if !ok || rep.Phase != sourcePhase || !offlineRetiredTerminal(rep.Retired) { |
| 1150 | return refuse |
| 1151 | } |
| 1152 | } |
| 1153 | return nil |
| 1154 | } |
| 1155 | |
| 1156 | // offlineRetiredTerminal says every path a parked report lists is settled: |
| 1157 | // parked, absent, or left with the reason it could not be parked (a |
| 1158 | // root-owned directory, a view mount). A left path is terminal; the move |
| 1159 | // goes on and the status view shows it. An unknown action, or a left with no |
| 1160 | // reason, is not a settled path. |
| 1161 | func offlineRetiredTerminal(retired []moves.RetiredPath) bool { |
| 1162 | for _, p := range retired { |
| 1163 | switch p.Action { |
| 1164 | case moves.RetiredParked, moves.RetiredAbsent: |
| 1165 | case moves.RetiredLeft: |
| 1166 | if p.Reason == "" { |
| 1167 | return false |
| 1168 | } |
| 1169 | default: |
| 1170 | return false |
| 1171 | } |
| 1172 | if p.Path == "" { |
| 1173 | return false |
| 1174 | } |
| 1175 | } |
| 1176 | return true |
| 1177 | } |
| 1178 | |
| 1179 | // offlineVerbDoor is the shape of seal, flip, open and finish: the gate, an |
| 1180 | // offline move in flight, the state the verb follows, the evidence, the |
| 1181 | // verb's checks, one transition. |
| 1182 | func (s *Server) offlineVerbDoor(verb string) http.HandlerFunc { |
| 1183 | return func(w http.ResponseWriter, r *http.Request) { |
| 1184 | d, ok := s.offlineGate(w, r) |
| 1185 | if !ok { |
| 1186 | return |
| 1187 | } |
| 1188 | mv, ok := s.offlineInFlight(w, d) |
| 1189 | if !ok { |
| 1190 | return |
| 1191 | } |
| 1192 | if offlineVerbOf(mv.State) != verb { |
| 1193 | writeRefusal(w, offlineStateRefusal(mv.State, d.host)) |
| 1194 | return |
| 1195 | } |
| 1196 | ctx := r.Context() |
| 1197 | ev, err := s.readOfflineEvidence(ctx, d.host, mv.ID) |
| 1198 | if err != nil { |
| 1199 | log.Printf("console: offline move %s: reports: %v", d.host, err) |
| 1200 | jsonError(w, http.StatusBadGateway, "could not read the move's reports. try again") |
| 1201 | return |
| 1202 | } |
| 1203 | step, ref := s.offlineStepFor(ctx, d, mv, ev, time.Now()) |
| 1204 | if ref != nil { |
| 1205 | writeRefusal(w, ref) |
| 1206 | return |
| 1207 | } |
| 1208 | s.offlineAdvance(w, r, d, step.mut) |
| 1209 | } |
| 1210 | } |
| 1211 | |
| 1212 | func (s *Server) handleAPIInstanceOfflineAbort(w http.ResponseWriter, r *http.Request) { |
| 1213 | d, ok := s.offlineGate(w, r) |
| 1214 | if !ok { |
| 1215 | return |
| 1216 | } |
| 1217 | mv, ok := s.offlineInFlight(w, d) |
| 1218 | if !ok { |
| 1219 | return |
| 1220 | } |
| 1221 | switch mv.State { |
| 1222 | case provision.OfflineOpen, provision.OfflineFinishing: |
| 1223 | jsonError(w, http.StatusConflict, fmt.Sprintf("%s is already served from %s. move it back with a new move", d.host, s.offlineName(mv.To))) |
| 1224 | return |
| 1225 | case provision.OfflineReverting: |
| 1226 | // Already reverting: nothing to write. |
| 1227 | writeJSON(w, http.StatusAccepted, s.offlineViewOf(r.Context(), d.rec)) |
| 1228 | return |
| 1229 | } |
| 1230 | s.offlineAdvance(w, r, d, provision.OfflineAbort(mv.ID)) |
| 1231 | } |
| 1232 | |
| 1233 | // ---- status ---- |
| 1234 | |
| 1235 | // offlineReportView is a report as the status view carries it: every field |
| 1236 | // the box wrote, and whether it is current enough for a door to believe. |
| 1237 | type offlineReportView struct { |
| 1238 | moves.OfflineReport |
| 1239 | Fresh bool `json:"fresh"` |
| 1240 | } |
| 1241 | |
| 1242 | // offlineLeftView is a path a source box could not park and left in place, |
| 1243 | // with the reason it gave (the parked report's inventory). |
| 1244 | type offlineLeftView struct { |
| 1245 | Box string `json:"box"` |
| 1246 | Path string `json:"path"` |
| 1247 | Reason string `json:"reason"` |
| 1248 | } |
| 1249 | |
| 1250 | // offlineItemView is one row of the manifest with the latest check of it. |
| 1251 | // Status is ok, waiting, failed or unchecked (no report decides the item). |
| 1252 | type offlineItemView struct { |
| 1253 | ID string `json:"id"` |
| 1254 | Item string `json:"item"` |
| 1255 | Disposition string `json:"disposition"` |
| 1256 | Owner string `json:"owner"` |
| 1257 | Source string `json:"source,omitempty"` |
| 1258 | Status string `json:"status"` |
| 1259 | Detail string `json:"detail,omitempty"` |
| 1260 | } |
| 1261 | |
| 1262 | // offlineView is the status answer, and the 202 every mutating door gives. |
| 1263 | // State is "none" when no offline move is in flight (Last then carries the |
| 1264 | // outcome of the one that ended). Next is the verb that moves the move on |
| 1265 | // right now; empty while a box has work to do. Waiting is the sentence that |
| 1266 | // verb would be refused with. CanAbort says abort is legal. Failed says the |
| 1267 | // sealer or the target reported it could not do its step: abort is the way on. |
| 1268 | type offlineView struct { |
| 1269 | Host string `json:"host"` |
| 1270 | State string `json:"state"` |
| 1271 | ID string `json:"id,omitempty"` |
| 1272 | From string `json:"from,omitempty"` |
| 1273 | To string `json:"to,omitempty"` |
| 1274 | Sealer string `json:"sealer,omitempty"` |
| 1275 | Stager string `json:"stager,omitempty"` |
| 1276 | Boxes []string `json:"boxes,omitempty"` |
| 1277 | RequestedBy string `json:"requested_by,omitempty"` |
| 1278 | RequestedAt int64 `json:"requested_at,omitempty"` |
| 1279 | UpdatedAt int64 `json:"updated_at,omitempty"` |
| 1280 | OpenedAt int64 `json:"opened_at,omitempty"` |
| 1281 | Next string `json:"next,omitempty"` |
| 1282 | Waiting string `json:"waiting,omitempty"` |
| 1283 | CanAbort bool `json:"can_abort,omitempty"` |
| 1284 | // NeverOffline: the move was staged and never started, so its |
| 1285 | // instance has kept serving throughout (an abort from staging). |
| 1286 | NeverOffline bool `json:"never_offline,omitempty"` |
| 1287 | // StagedAt is the record's staged_at in Unix nanoseconds: the latest |
| 1288 | // stage request. Stage.Request equals it once the stager has answered |
| 1289 | // that request. |
| 1290 | StagedAt int64 `json:"staged_at,omitempty"` |
| 1291 | Failed bool `json:"failed,omitempty"` |
| 1292 | Left []offlineLeftView `json:"left,omitempty"` |
| 1293 | Stage *offlineStageView `json:"stage,omitempty"` |
| 1294 | Reports []offlineReportView `json:"reports,omitempty"` |
| 1295 | Items []offlineItemView `json:"items,omitempty"` |
| 1296 | Last *provision.MoveOutcome `json:"last,omitempty"` |
| 1297 | } |
| 1298 | |
| 1299 | // offlineStageView is the pre-copy's progress, from the latest stager |
| 1300 | // report: when its last round ended, the replica as that round listed it, |
| 1301 | // how much of it the target's staging prefix holds, and what the round |
| 1302 | // copied. Phase failed carries the box's own diagnostic in Error. Stager |
| 1303 | // reports are rewritten each moves pass, not every few seconds, so their |
| 1304 | // age is shown and never judged. |
| 1305 | type offlineStageView struct { |
| 1306 | Box string `json:"box"` |
| 1307 | Phase string `json:"phase"` |
| 1308 | At int64 `json:"at"` |
| 1309 | ReplicaObjects int `json:"replica_objects"` |
| 1310 | ReplicaBytes int64 `json:"replica_bytes"` |
| 1311 | StagedObjects int `json:"staged_objects"` |
| 1312 | StagedBytes int64 `json:"staged_bytes"` |
| 1313 | RoundObjects int `json:"round_objects"` |
| 1314 | RoundBytes int64 `json:"round_bytes"` |
| 1315 | Error string `json:"error,omitempty"` |
| 1316 | // Request is the staged_at (Unix nanoseconds) the round answered. |
| 1317 | Request int64 `json:"request,omitempty"` |
| 1318 | } |
| 1319 | |
| 1320 | // offlineStageOf is the stage view of the newest stager report, nil when no |
| 1321 | // box has staged this move. |
| 1322 | func offlineStageOf(ev offlineEvidence) *offlineStageView { |
| 1323 | var latest *offlineReport |
| 1324 | for _, r := range ev.byRole(moves.OfflineRoleStager) { |
| 1325 | if latest == nil || r.At.After(latest.At) { |
| 1326 | latest = &r |
| 1327 | } |
| 1328 | } |
| 1329 | if latest == nil { |
| 1330 | return nil |
| 1331 | } |
| 1332 | r := latest.OfflineReport |
| 1333 | v := &offlineStageView{Box: r.Box, Phase: r.Phase, At: r.At.Unix(), |
| 1334 | ReplicaObjects: r.ReplicaObjects, ReplicaBytes: r.ReplicaBytes, StagedObjects: r.StagedObjects, StagedBytes: r.StagedBytes, |
| 1335 | RoundObjects: r.RoundObjects, RoundBytes: r.RoundBytes, Error: r.StageError} |
| 1336 | if !r.StageRequest.IsZero() { |
| 1337 | v.Request = r.StageRequest.UnixNano() |
| 1338 | } |
| 1339 | return v |
| 1340 | } |
| 1341 | |
| 1342 | // offlineViewOf builds the status view of a record. Evidence it cannot read |
| 1343 | // leaves the reports out and says so in Waiting; it never fails the answer. |
| 1344 | func (s *Server) offlineViewOf(ctx context.Context, rec *provision.Record) offlineView { |
| 1345 | v := offlineView{Host: rec.Host, State: "none", Last: rec.LastMove} |
| 1346 | mv := rec.Move |
| 1347 | if mv == nil || !mv.Offline() { |
| 1348 | return v |
| 1349 | } |
| 1350 | now := time.Now() |
| 1351 | v.State, v.ID = mv.State, mv.ID |
| 1352 | v.From, v.To = s.offlineName(mv.From), s.offlineName(mv.To) |
| 1353 | v.Sealer, v.Boxes, v.Stager = mv.Sealer, mv.Boxes, mv.Stager |
| 1354 | v.NeverOffline = !mv.StartedOffline() |
| 1355 | if mv.StagedAt != nil { |
| 1356 | v.StagedAt = mv.StagedAt.UnixNano() |
| 1357 | } |
| 1358 | v.RequestedBy, v.RequestedAt, v.UpdatedAt = mv.RequestedBy, mv.RequestedAt.Unix(), mv.UpdatedAt.Unix() |
| 1359 | if mv.OpenedAt != nil { |
| 1360 | v.OpenedAt = mv.OpenedAt.Unix() |
| 1361 | } |
| 1362 | switch mv.State { |
| 1363 | case provision.OfflineStaging, provision.OfflineStopping, provision.OfflineSealing, provision.OfflineSealed, provision.OfflineLanding, provision.OfflineLanded: |
| 1364 | v.CanAbort = true |
| 1365 | } |
| 1366 | ev, err := s.readOfflineEvidence(ctx, rec.Host, mv.ID) |
| 1367 | if err != nil { |
| 1368 | log.Printf("console: offline move %s: reading the reports: %v", rec.Host, err) |
| 1369 | v.Waiting = "could not read the move's reports. try again" |
| 1370 | return v |
| 1371 | } |
| 1372 | for _, r := range ev.reports { |
| 1373 | v.Reports = append(v.Reports, offlineReportView{OfflineReport: r.OfflineReport, Fresh: offlineFresh(r.OfflineReport, now)}) |
| 1374 | } |
| 1375 | v.Items = offlineItems(ev, mv) |
| 1376 | v.Stage = offlineStageOf(ev) |
| 1377 | for _, r := range ev.byRole(moves.OfflineRoleSource) { |
| 1378 | for _, p := range r.Retired { |
| 1379 | if p.Action == moves.RetiredLeft { |
| 1380 | v.Left = append(v.Left, offlineLeftView{Box: r.Box, Path: p.Path, Reason: p.Reason}) |
| 1381 | } |
| 1382 | } |
| 1383 | } |
| 1384 | if verb := offlineVerbOf(mv.State); verb != "" { |
| 1385 | d := offlineDoor{host: rec.Host} |
| 1386 | if _, ref := s.offlineStepFor(ctx, d, mv, ev, now); ref != nil { |
| 1387 | v.Waiting = ref.msg |
| 1388 | } else { |
| 1389 | v.Next = verb |
| 1390 | } |
| 1391 | } |
| 1392 | // A box that says it failed (the sealer, or the target that cannot |
| 1393 | // land) means abort is the way on. The CLI says it with MV-O6. |
| 1394 | if v.CanAbort { |
| 1395 | for _, r := range ev.reports { |
| 1396 | if r.Phase == moves.OfflinePhaseFailed && (r.Role == moves.OfflineRoleSealer || r.Role == moves.OfflineRoleTarget) { |
| 1397 | v.Failed = true |
| 1398 | } |
| 1399 | } |
| 1400 | } |
| 1401 | return v |
| 1402 | } |
| 1403 | |
| 1404 | // offlineItems is the manifest with each item's latest check, from the |
| 1405 | // report of the step that owns it: the sealer's for a seal item, the |
| 1406 | // target's landing report for a landing item. A check present and not OK is |
| 1407 | // failed when the report that holds it says failed, else waiting (the box |
| 1408 | // retries it). An item no report decides is unchecked, and a door-owned one |
| 1409 | // was decided at start. |
| 1410 | func offlineItems(ev offlineEvidence, mv *provision.MoveState) []offlineItemView { |
| 1411 | owners := map[string][]offlineReport{moves.OfflineOwnerLanding: ev.byRole(moves.OfflineRoleTarget)} |
| 1412 | if r, ok := ev.get(moves.OfflineRoleSealer, mv.Sealer); ok { |
| 1413 | owners[moves.OfflineOwnerSeal] = []offlineReport{r} |
| 1414 | } |
| 1415 | if ev.seal != nil { |
| 1416 | // The seal is what the sealer decided; the living report is the |
| 1417 | // latest word on the same items, and is asked first. |
| 1418 | owners[moves.OfflineOwnerSeal] = append(owners[moves.OfflineOwnerSeal], *ev.seal) |
| 1419 | } |
| 1420 | out := make([]offlineItemView, 0, len(moves.OfflineManifest)) |
| 1421 | for _, it := range moves.OfflineManifest { |
| 1422 | row := offlineItemView{ID: it.ID, Item: it.Item, Disposition: it.Disposition, Owner: it.Owner, Source: it.Source, Status: "waiting"} |
| 1423 | switch it.Owner { |
| 1424 | case moves.OfflineOwnerDoor: |
| 1425 | row.Status = "ok" |
| 1426 | case moves.OfflineOwnerNone: |
| 1427 | row.Status = "unchecked" |
| 1428 | default: |
| 1429 | for _, r := range owners[it.Owner] { |
| 1430 | c, found := r.Items[it.ID] |
| 1431 | if !found { |
| 1432 | continue |
| 1433 | } |
| 1434 | row.Detail = c.Reason |
| 1435 | switch { |
| 1436 | case c.OK: |
| 1437 | row.Status = "ok" |
| 1438 | case r.Phase == moves.OfflinePhaseFailed: |
| 1439 | row.Status = "failed" |
| 1440 | } |
| 1441 | break |
| 1442 | } |
| 1443 | } |
| 1444 | out = append(out, row) |
| 1445 | } |
| 1446 | return out |
| 1447 | } |
| 1448 | |
| 1449 | func (s *Server) handleAPIInstanceOfflineStatus(w http.ResponseWriter, r *http.Request) { |
| 1450 | d, ok := s.offlineGate(w, r) |
| 1451 | if !ok { |
| 1452 | return |
| 1453 | } |
| 1454 | writeJSON(w, http.StatusOK, s.offlineViewOf(r.Context(), d.rec)) |
| 1455 | } |
| 1456 | |
amadan