| 1 | package hostclaim |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "fmt" |
| 7 | "slices" |
| 8 | "strconv" |
| 9 | |
| 10 | "github.com/carlosframework/platform/internal/regionname" |
| 11 | "github.com/carlosframework/platform/internal/release" |
| 12 | ) |
| 13 | |
| 14 | const KindRelocating = "relocating" |
| 15 | |
| 16 | // PreFlipGuard verifies the durable pre-flip conversion decision in the |
| 17 | // authoritative instance record. That decision's record CAS competes with |
| 18 | // MoveFlip and makes the move terminal for every cutting/flip transition. |
| 19 | // It preserves the old move journal for replay, and binds this op, incarnation, |
| 20 | // source and exact fence reference. A snapshot of an undecided record is not |
| 21 | // an implementation of this interface. While the claim is moving the source |
| 22 | // is its "source" and the region is the record's; once relocating they are |
| 23 | // its "source_box" and "region". The returned region is the verified engine |
| 24 | // identity of the record placement, never an empty placement code. |
| 25 | type PreFlipGuard interface { |
| 26 | CheckMovingBeforeFlip(context.Context, Observed, string, FenceRef) (string, error) |
| 27 | } |
| 28 | |
| 29 | // RelocationRecoveryOwner reports replay ownership, never execution authority. |
| 30 | func RelocationRecoveryOwner(c *Claim) (string, error) { |
| 31 | if c == nil || c.Op.Kind != KindRelocating || c.Op.Intent["region"] == "" { |
| 32 | return "", ErrWrongState |
| 33 | } |
| 34 | return "region:" + c.Op.Intent["region"], nil |
| 35 | } |
| 36 | |
| 37 | func validateRelocating(c *Claim) error { |
| 38 | if c.Op.Kind != KindRelocating { |
| 39 | return nil |
| 40 | } |
| 41 | if c.Holder == nil || c.Op.Intent["region"] == "" || !validBoxReference(c.Op.Intent["source_box"]) || c.Op.Intent["source_box"] == "none" { |
| 42 | return ErrMalformed |
| 43 | } |
| 44 | switch c.Op.Phase { |
| 45 | case "starting", "releasing": |
| 46 | if c.Op.Exec != c.Op.Intent["source_box"] && c.Op.Exec != "none" { |
| 47 | return ErrMalformed |
| 48 | } |
| 49 | case "pinning": |
| 50 | if c.Op.Exec != "none" { |
| 51 | return ErrMalformed |
| 52 | } |
| 53 | case "installing", "installed", "probing", "probe-failed", "serving": |
| 54 | if !validBoxReference(c.Op.Exec) || c.Op.Exec == "none" || c.Op.Exec != c.Op.Intent["target_box"] { |
| 55 | return ErrMalformed |
| 56 | } |
| 57 | default: |
| 58 | return ErrMalformed |
| 59 | } |
| 60 | return nil |
| 61 | } |
| 62 | |
| 63 | // ConvertMovingToRelocating performs exactly one claim CAS. It deliberately |
| 64 | // exposes no transition capable of granting relocation execution. Callers must |
| 65 | // be claim-enabled and first persist the record conversion decision. |
| 66 | func ConvertMovingToRelocating(ctx context.Context, s release.Store, observed Observed, opID string, ref FenceRef, authority FenceAuthority, guard PreFlipGuard) (Observed, error) { |
| 67 | if guard == nil || observed.Absent || observed.Claim == nil || observed.ETag == "" { |
| 68 | return Observed{}, ErrWrongState |
| 69 | } |
| 70 | before := observed.Claim |
| 71 | if opID == "" || before.Op.ID != opID { |
| 72 | return Observed{}, ErrStaleOp |
| 73 | } |
| 74 | // The move's own intent, as BeginAdmittedMove writes it: "source" and |
| 75 | // "target", the source enrollment from the signed roster, and the app |
| 76 | // admission the request ran under (claim Plan 5 Task 7). |
| 77 | source := before.Op.Intent["source"] |
| 78 | if before.Holder == nil || !validBoxReference(source) || source == "none" || before.Op.Intent["source_enrollment"] == "" || before.Op.Intent["app_id"] == "" || before.Op.Intent["admission_id"] != before.Op.ID || before.Holder.Family != "instance" || before.Op.Kind != KindMoving || before.Op.Phase == "landing" || before.Op.Phase == "committed" { |
| 79 | return Observed{}, ErrWrongState |
| 80 | } |
| 81 | // Check the instance record even while the claim still says moving: the |
| 82 | // instance flip is earlier than final commit and can precede claim landing. |
| 83 | region, err := guard.CheckMovingBeforeFlip(ctx, observed, opID, ref) |
| 84 | if err != nil { |
| 85 | return Observed{}, err |
| 86 | } |
| 87 | fence, err := VerifyFence(ctx, s, authority, ref) |
| 88 | if err != nil { |
| 89 | return Observed{}, err |
| 90 | } |
| 91 | // The region is the guard's (the authoritative record) and must equal the |
| 92 | // fence's roster binding; the claim names only the target region. |
| 93 | if region == "" || region != fence.binding.Region || source != fence.record.Box || before.Op.Intent["source_enrollment"] != ref.Enrollment || (before.Op.Exec != fence.record.Box && before.Op.Exec != "none" && before.Op.Exec != "") { |
| 94 | return Observed{}, ErrFence |
| 95 | } |
| 96 | raw, err := json.Marshal(ref) |
| 97 | if err != nil { |
| 98 | return Observed{}, err |
| 99 | } |
| 100 | decisionDigest := "" |
| 101 | if bound, ok := guard.(interface { |
| 102 | ConversionDecisionDigest(context.Context, Observed, string, FenceRef) (string, error) |
| 103 | }); ok { |
| 104 | decisionDigest, err = bound.ConversionDecisionDigest(ctx, observed, opID, ref) |
| 105 | if err != nil { |
| 106 | return Observed{}, err |
| 107 | } |
| 108 | if !isHex64(decisionDigest) { |
| 109 | return Observed{}, ErrWrongState |
| 110 | } |
| 111 | } |
| 112 | return Update(ctx, s, observed, func(c *Claim) error { |
| 113 | if c.Op.ID != opID || c.Op.Kind != KindMoving || c.Op.Phase != before.Op.Phase { |
| 114 | return ErrStaleOp |
| 115 | } |
| 116 | key := "fence." + fence.record.Box |
| 117 | if old, ok := c.Op.Journal[key]; ok && old != string(raw) { |
| 118 | return fmt.Errorf("%w: fence reference already bound", ErrFence) |
| 119 | } |
| 120 | // Recheck the durable record decision immediately before the claim CAS. |
| 121 | // Its terminal state prevents any intervening flip. |
| 122 | currentRegion, err := guard.CheckMovingBeforeFlip(ctx, observed, opID, ref) |
| 123 | if err != nil { |
| 124 | return err |
| 125 | } |
| 126 | if currentRegion != region { |
| 127 | return ErrWrongState |
| 128 | } |
| 129 | if c.Op.Journal == nil { |
| 130 | c.Op.Journal = map[string]string{} |
| 131 | } |
| 132 | c.Op.Journal[key] = string(raw) |
| 133 | c.Op.Journal["reason"] = "source-terminated-before-flip" |
| 134 | c.Op.Journal["source_stamp"] = strconv.FormatUint(c.Holder.Stamp, 10) |
| 135 | c.Op.Journal["source_generation"] = c.Holder.Generation |
| 136 | c.Op.Journal["roster_revision"] = strconv.FormatUint(fence.binding.RosterRevision, 10) |
| 137 | c.Op.Journal["roster_etag"] = fence.binding.RosterETag |
| 138 | c.Op.Journal["app_id"] = before.Op.Intent["app_id"] |
| 139 | c.Op.Journal["admission_id"] = before.Op.Intent["admission_id"] |
| 140 | c.Op.Journal["conversion_claim_gen"] = strconv.FormatUint(before.Gen, 10) |
| 141 | c.Op.Journal["conversion_claim_etag"] = observed.ETag |
| 142 | if decisionDigest != "" { |
| 143 | c.Op.Journal["conversion_decision_sha256"] = decisionDigest |
| 144 | } |
| 145 | c.Op.Kind, c.Op.Phase, c.Op.Exec = KindRelocating, "pinning", "none" |
| 146 | if c.Op.Intent == nil { |
| 147 | c.Op.Intent = map[string]string{} |
| 148 | } |
| 149 | c.Op.Intent["region"] = region |
| 150 | c.Op.Intent["source_box"] = source |
| 151 | // The original move target is in a different region. It is not a selected |
| 152 | // same-region replacement. Keep the move history without authorizing it. |
| 153 | for from, to := range map[string]string{"target": "move_target_box", "to": "move_to_region", "target_enrollment": "move_target_enrollment"} { |
| 154 | if v := c.Op.Intent[from]; v != "" { |
| 155 | c.Op.Intent[to] = v |
| 156 | } |
| 157 | delete(c.Op.Intent, from) |
| 158 | } |
| 159 | delete(c.Op.Intent, "source") |
| 160 | participants := make([]string, 0, len(c.Op.Participants)) |
| 161 | for _, box := range c.Op.Participants { |
| 162 | if box != fence.record.Box { |
| 163 | participants = append(participants, box) |
| 164 | } |
| 165 | } |
| 166 | c.Op.Participants = participants |
| 167 | return nil |
| 168 | }) |
| 169 | } |
| 170 | |
| 171 | // CommittedLanding is the authoritative instance record's committed move |
| 172 | // as a CommittedLandingGuard read it: the regions it moved between, as |
| 173 | // engine regions (park.PlacementRegion of the record's placement codes, so |
| 174 | // the home is park.HomeRegion, exactly the regions moves' own commit |
| 175 | // queues its parks in), and the two parks the commit queues, built by |
| 176 | // moves.CommitScopes from the record's flip time and the claim's seal (the one builder moves' own |
| 177 | // commit uses). The scopes are used only when the claim has not yet |
| 178 | // followed the commit. |
| 179 | type CommittedLanding struct { |
| 180 | FromRegion, ToRegion string |
| 181 | SourceScope, StagingScope ParkScope |
| 182 | } |
| 183 | |
| 184 | // CommittedLandingGuard verifies in the authoritative instance record that |
| 185 | // move opID has committed (moves §7.5: the record's commit CAS, record |
| 186 | // first, terminal) for this claim's incarnation. A record still landing, |
| 187 | // rolling back, or naming another move or incarnation refuses with |
| 188 | // ErrWrongState: before the commit the target's fence takes moves' |
| 189 | // declared-loss rollback (moves §7.10), never this conversion. A snapshot |
| 190 | // of an undecided record is not an implementation of this interface. |
| 191 | type CommittedLandingGuard interface { |
| 192 | CheckLandingCommitted(context.Context, Observed, string) (CommittedLanding, error) |
| 193 | } |
| 194 | |
| 195 | // landingConvertible reports the landing phases at which a committed |
| 196 | // record can stand: the record commits only after its probed, so the |
| 197 | // claim is at probing or probed (lagging, the record-commit race) or has |
| 198 | // followed it to committed. |
| 199 | func landingConvertible(phase string) bool { |
| 200 | return phase == PhaseLandingProbing || phase == PhaseLandingProbed || phase == PhaseLandingCommitted |
| 201 | } |
| 202 | |
| 203 | // ConvertCommittedLanding is J3 (Paul's ruling J3/F5 on amadan #236; the |
| 204 | // model's fence-converts-committed-landing): a landing whose move record |
| 205 | // has committed and whose target is then terminated becomes a same-region |
| 206 | // relocating failover in the target's region, in exactly one claim CAS. |
| 207 | // The record's commit is the evidence (guard), never the claim's phase |
| 208 | // alone; the claim may lag the record at probing or probed, and then the |
| 209 | // same CAS first follows the commit exactly as moves does (probed, then |
| 210 | // Commit under KeepWriterSession, queueing the source region's location |
| 211 | // park, the source's park-local and the staging lane's park), so every |
| 212 | // cleanup obligation the commit owes exists before the op stops being a |
| 213 | // landing. Before the commit the guard refuses and the fence is moves' |
| 214 | // (declared-loss rollback). |
| 215 | // |
| 216 | // The result is the failover StartRelocation starts for a fenced |
| 217 | // executor, without a target: relocating in starting, op.exec none, the |
| 218 | // dead target as source_box (on lost, its fence journaled with its |
| 219 | // observation time and the loss retained in evidence_refs by fenceBox, as |
| 220 | // FenceSource does), the op id, holder and every obligation kept. The |
| 221 | // target is named in starting (NameTarget) and acknowledges before |
| 222 | // PinFailover, as an initial failover does. The move's own intent is kept |
| 223 | // as history under move_* keys and its journal is kept whole. It |
| 224 | // deliberately exposes no transition capable of granting execution. |
| 225 | func ConvertCommittedLanding(ctx context.Context, s release.Store, observed Observed, opID, targetBox string, ref FenceRef, authority FenceAuthority, guard CommittedLandingGuard, d *LocalDischarge, newID func() string) (Observed, error) { |
| 226 | if guard == nil || newID == nil || observed.Absent || observed.Claim == nil || observed.ETag == "" { |
| 227 | return Observed{}, ErrWrongState |
| 228 | } |
| 229 | before := observed.Claim |
| 230 | if opID == "" || before.Op.ID != opID { |
| 231 | return Observed{}, ErrStaleOp |
| 232 | } |
| 233 | // The landing's own intent, as BeginAdmittedMove wrote it: "source", |
| 234 | // "target" and "to", the target's enrollment from the signed roster, |
| 235 | // and the app admission the move ran under. |
| 236 | in := before.Op.Intent |
| 237 | if before.Holder == nil || before.Holder.Family != "instance" || before.Op.Kind != KindLanding || !landingConvertible(before.Op.Phase) || |
| 238 | !validBoxReference(targetBox) || targetBox == "none" || in["target"] != targetBox || before.Op.Exec != targetBox || slices.Contains(before.Lost, targetBox) || |
| 239 | !validBoxReference(in["source"]) || in["source"] == "none" || in["source"] == targetBox || in["to"] == "" || |
| 240 | !validEnrollment(in["target_enrollment"]) || in["app_id"] == "" || in["admission_id"] != opID { |
| 241 | return Observed{}, ErrWrongState |
| 242 | } |
| 243 | committed, err := guard.CheckLandingCommitted(ctx, observed, opID) |
| 244 | if err != nil { |
| 245 | return Observed{}, err |
| 246 | } |
| 247 | // The committed regions are engine regions, and so is the move's "to" |
| 248 | // (BeginClaimMove's park.PlacementRegion); the home's two names are one |
| 249 | // region only under the move's own home binding (main's hostclaim reads |
| 250 | // it from the intent the same way). |
| 251 | home := in["home_region"] |
| 252 | if committed.FromRegion == "" || !regionname.SameRegion(home, committed.ToRegion, in["to"]) || regionname.SameRegion(home, committed.FromRegion, committed.ToRegion) { |
| 253 | return Observed{}, ErrWrongState |
| 254 | } |
| 255 | fence, err := VerifyFence(ctx, s, authority, ref) |
| 256 | if err != nil { |
| 257 | return Observed{}, err |
| 258 | } |
| 259 | // The fence is the target's, bound to the target's region in the |
| 260 | // roster, which is the record's destination. The binding is the roster |
| 261 | // name, so a landing in the home matches its park.HomeRegion only under |
| 262 | // the move's home binding. |
| 263 | if fence.record.Box != targetBox || fence.record.Enrollment != in["target_enrollment"] || !regionname.SameRegion(home, fence.binding.Region, committed.ToRegion) { |
| 264 | return Observed{}, ErrFence |
| 265 | } |
| 266 | return Update(ctx, s, observed, func(c *Claim) error { |
| 267 | if c.Op.ID != opID || c.Op.Kind != KindLanding || c.Op.Phase != before.Op.Phase || c.Op.Exec != targetBox || slices.Contains(c.Lost, targetBox) { |
| 268 | return ErrStaleOp |
| 269 | } |
| 270 | // Recheck the record immediately before the claim CAS. |
| 271 | again, err := guard.CheckLandingCommitted(ctx, observed, opID) |
| 272 | if err != nil { |
| 273 | return err |
| 274 | } |
| 275 | if again.FromRegion != committed.FromRegion || again.ToRegion != committed.ToRegion { |
| 276 | return ErrWrongState |
| 277 | } |
| 278 | phase := c.Op.Phase |
| 279 | // The claim half of the record's commit, as moves writes it |
| 280 | // (moveslandcommit.go commitClaim): a claim lagging at probing is |
| 281 | // stepped to probed first, then Commit, on the same session. |
| 282 | if c.Op.Phase == PhaseLandingProbing { |
| 283 | if err := KeepWriterSession(opID, SetMovePhase(opID, KindLanding, PhaseLandingProbing, PhaseLandingProbed))(c); err != nil { |
| 284 | return err |
| 285 | } |
| 286 | } |
| 287 | if c.Op.Phase == PhaseLandingProbed { |
| 288 | if err := KeepWriterSession(opID, Commit(opID, committed.FromRegion, in["source"], committed.ToRegion, committed.SourceScope, committed.StagingScope, newID))(c); err != nil { |
| 289 | return err |
| 290 | } |
| 291 | } |
| 292 | // Defensive: SetMovePhase and Commit above leave every accepted |
| 293 | // phase at committed, so this cannot fire today. It keeps the |
| 294 | // conversion from running over a landing the commit did not reach |
| 295 | // if either helper ever changes. |
| 296 | if c.Op.Phase != PhaseLandingCommitted { |
| 297 | return ErrWrongState |
| 298 | } |
| 299 | ensureJournal(c) |
| 300 | c.Op.Journal["reason"] = ReasonFailover |
| 301 | c.Op.Journal["conversion"] = "committed-landing" |
| 302 | c.Op.Journal["conversion_landing_phase"] = phase |
| 303 | c.Op.Journal["conversion_claim_gen"] = strconv.FormatUint(before.Gen, 10) |
| 304 | c.Op.Journal["conversion_claim_etag"] = observed.ETag |
| 305 | c.Op.Journal["move_from_region"] = committed.FromRegion |
| 306 | c.Op.Journal["roster_revision"] = strconv.FormatUint(fence.binding.RosterRevision, 10) |
| 307 | c.Op.Journal["roster_etag"] = fence.binding.RosterETag |
| 308 | c.Op.Kind, c.Op.Phase, c.Op.Exec = KindRelocating, PhaseStarting, "none" |
| 309 | intent := make(map[string]string, len(c.Op.Intent)+4) |
| 310 | for k, v := range c.Op.Intent { |
| 311 | intent[k] = v |
| 312 | } |
| 313 | // The move's other side is in another region and is no participant: |
| 314 | // its finalization chores are obligations, kept as they are. Keep |
| 315 | // the move's history without authorizing it. |
| 316 | for from, to := range map[string]string{"source": "move_source_box", "source_enrollment": "move_source_enrollment", "to": "move_to_region"} { |
| 317 | if v := intent[from]; v != "" { |
| 318 | intent[to] = v |
| 319 | } |
| 320 | delete(intent, from) |
| 321 | } |
| 322 | delete(intent, "target") |
| 323 | delete(intent, "target_enrollment") |
| 324 | // The relocation's region is the verified roster identity, as |
| 325 | // ConvertMovingToRelocating's is (the fence binding, which matched |
| 326 | // the destination above): the region worker's own name, never |
| 327 | // park.HomeRegion. |
| 328 | intent["region"] = fence.binding.Region |
| 329 | intent["source_box"] = targetBox |
| 330 | intent["source_enrollment"] = in["target_enrollment"] |
| 331 | intent["stamp_from"] = strconv.FormatUint(c.Holder.Stamp, 10) |
| 332 | intent["generation_from"] = c.Holder.Generation |
| 333 | c.Op.Intent = intent |
| 334 | c.Op.Participants = nil |
| 335 | return fenceBox(c, targetBox, ref, fence.record.ObservedAt, d) |
| 336 | }) |
| 337 | } |
| 338 | |
| 339 | // Box and Region expose a verified fence's terminated box and its roster |
| 340 | // region binding, so a caller binds them to the authoritative record rather |
| 341 | // than to a decoded reference. |
| 342 | func (v VerifiedFence) Box() string { return v.record.Box } |
| 343 | func (v VerifiedFence) Region() string { return v.binding.Region } |
| 344 | |
amadan