| 1 | package hostops |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "crypto/rand" |
| 6 | "crypto/sha256" |
| 7 | "encoding/hex" |
| 8 | "encoding/json" |
| 9 | "errors" |
| 10 | "fmt" |
| 11 | "os" |
| 12 | "sort" |
| 13 | "strconv" |
| 14 | "strings" |
| 15 | |
| 16 | "github.com/carlosframework/platform/internal/domains" |
| 17 | "github.com/carlosframework/platform/internal/hostactors" |
| 18 | "github.com/carlosframework/platform/internal/hostclaim" |
| 19 | "github.com/carlosframework/platform/internal/hostfloor" |
| 20 | "github.com/carlosframework/platform/internal/park" |
| 21 | "github.com/carlosframework/platform/internal/provision" |
| 22 | ) |
| 23 | |
| 24 | func (a ProductionAuthority) admission(ctx context.Context, account, app string) (AdmissionObserved, error) { |
| 25 | raw, _, err := a.Store.Get(ctx, "apps/"+account+"/"+app+"/admission.json") |
| 26 | if err != nil { |
| 27 | return AdmissionObserved{}, err |
| 28 | } |
| 29 | var id struct { |
| 30 | AppID string `json:"app_id"` |
| 31 | } |
| 32 | if err = json.Unmarshal(raw, &id); err != nil { |
| 33 | return AdmissionObserved{}, err |
| 34 | } |
| 35 | return ReadAdmission(ctx, a.Store, account, app, id.AppID) |
| 36 | } |
| 37 | func (a ProductionAuthority) Create(ctx context.Context, before hostclaim.Observed, r Record, id string) error { |
| 38 | if err := CheckCreateFreeze(ctx, a.Store, a.Directory.PublicKey); err != nil { |
| 39 | return err |
| 40 | } |
| 41 | adm, err := a.admission(ctx, r.Holder.Account, r.Holder.App) |
| 42 | if err != nil { |
| 43 | return err |
| 44 | } |
| 45 | reg, ok := adm.Value.Registrations[id] |
| 46 | inc, err := r.Incarnation() |
| 47 | if err != nil { |
| 48 | return err |
| 49 | } |
| 50 | if !ok || adm.Value.State != "open" || adm.Value.Op != nil || reg.Kind != "create" || reg.Host != r.Holder.Host || reg.AppID != adm.Value.AppID || reg.Epoch != adm.Value.Epoch || reg.Owner.Console != a.ConsoleID || reg.Owner.Enrollment != a.Enrollment || reg.Claim != SavedObservation(before) || reg.Incarnation != inc || !reg.CreatedAt.Equal(r.Holder.CreatedAt) || !RegistrationLive(reg, a.Now()) { |
| 51 | return ErrPending |
| 52 | } |
| 53 | owner, ownerErr := domains.GetClaim(ctx, a.Store, r.Holder.Host) |
| 54 | if ownerErr != nil && !errors.Is(ownerErr, os.ErrNotExist) { |
| 55 | return ownerErr |
| 56 | } |
| 57 | if ownerErr == nil && owner != nil { |
| 58 | legacy := owner.Acquisition == "" && owner.RecordIncarnation == "" |
| 59 | if legacy && owner.Pending && !r.VerifiedOwnership { |
| 60 | return ErrPending |
| 61 | } |
| 62 | bound := owner.Acquisition == r.Holder.Acquisition && owner.RecordIncarnation == inc |
| 63 | if owner.Account != r.Holder.Account || owner.App != r.Holder.App || (!legacy && !bound) { |
| 64 | return ErrPending |
| 65 | } |
| 66 | if legacy { |
| 67 | if _, _, e := a.Store.Get(ctx, r.Holder.Key); !errors.Is(e, os.ErrNotExist) { |
| 68 | if e != nil { |
| 69 | return e |
| 70 | } |
| 71 | return ErrPending |
| 72 | } |
| 73 | } |
| 74 | } |
| 75 | _, _, err = a.Store.Get(ctx, "apps/"+r.Holder.Account+"/"+r.Holder.App+"/purge.json") |
| 76 | if !errors.Is(err, os.ErrNotExist) { |
| 77 | if err != nil { |
| 78 | return err |
| 79 | } |
| 80 | return ErrPending |
| 81 | } |
| 82 | return nil |
| 83 | } |
| 84 | |
| 85 | func operationToken() (string, error) { |
| 86 | var b [16]byte |
| 87 | if _, err := rand.Read(b[:]); err != nil { |
| 88 | return "", err |
| 89 | } |
| 90 | return hex.EncodeToString(b[:]), nil |
| 91 | } |
| 92 | |
| 93 | type executorCandidate struct { |
| 94 | box, enrollment, hash string |
| 95 | sameApp bool |
| 96 | hosts int |
| 97 | } |
| 98 | |
| 99 | // selectExecutor is claim Plan 5's create-time selection over this |
| 100 | // console's namespace, the record's region and pool, active and unfenced. |
| 101 | // With a chooser (ProductionAuthority.Executors, CARLOS_REGION_HA on) the |
| 102 | // one choice is the chooser's, over its edges evidence; this scope still |
| 103 | // binds it, so a chosen box outside it refuses the create. |
| 104 | func (a ProductionAuthority) selectExecutor(ctx context.Context, before hostclaim.Observed, r Record) (string, string, string, error) { |
| 105 | var shape struct { |
| 106 | Pool string `json:"pool"` |
| 107 | Replicas []string `json:"replicas"` |
| 108 | } |
| 109 | if err := json.Unmarshal(r.Body, &shape); err != nil { |
| 110 | return "", "", "", err |
| 111 | } |
| 112 | if len(shape.Replicas) != 0 { |
| 113 | return "", "", "", ErrContract |
| 114 | } |
| 115 | everywhere := false |
| 116 | if r.Holder.Family == "instance" { |
| 117 | var rec provision.Record |
| 118 | if err := json.Unmarshal(r.Body, &rec); err != nil { |
| 119 | return "", "", "", err |
| 120 | } |
| 121 | everywhere = provision.ClaimEverywhere(rec) |
| 122 | } |
| 123 | pool := shape.Pool |
| 124 | if pool == "" { |
| 125 | pool = "default" |
| 126 | } |
| 127 | roster, err := a.Directory.Read(ctx) |
| 128 | if err != nil { |
| 129 | return "", "", "", err |
| 130 | } |
| 131 | if !roster.Roster.MatchesHome(a.HomeRegion) { |
| 132 | return "", "", "", ErrIdentity |
| 133 | } |
| 134 | candidates := map[string]*executorCandidate{} |
| 135 | servingScope := 0 |
| 136 | for ref, b := range roster.Roster.Boxes { |
| 137 | if b.Namespace == a.Namespace && b.Pool == pool && b.State == "active" { |
| 138 | servingScope++ |
| 139 | } |
| 140 | if b.Namespace != a.Namespace || !park.SameRegion(a.HomeRegion, b.Region, r.Holder.Region) || b.Pool != pool || b.State != "active" { |
| 141 | continue |
| 142 | } |
| 143 | fenced, err := a.Directory.TerminationFence(ctx, roster, ref, a.Now()) |
| 144 | if err != nil { |
| 145 | return "", "", "", err |
| 146 | } |
| 147 | if fenced { |
| 148 | continue |
| 149 | } |
| 150 | sum := sha256.Sum256([]byte(r.Holder.Host + "\x00" + ref)) |
| 151 | candidates[ref] = &executorCandidate{box: ref, enrollment: b.Enrollment, hash: hex.EncodeToString(sum[:])} |
| 152 | } |
| 153 | if everywhere && servingScope != 1 { |
| 154 | return "", "", "", ErrContract |
| 155 | } |
| 156 | if sel, selected, err := a.SelectExecutor(ctx, before, r); err != nil { |
| 157 | return "", "", "", err |
| 158 | } else if selected { |
| 159 | if c := candidates[sel.Box]; c == nil || c.enrollment != sel.Enrollment { |
| 160 | return "", "", "", fmt.Errorf("%w: %s is not in this console's namespace and the record's pool", ErrNoEligibleBox, sel.Box) |
| 161 | } |
| 162 | return sel.Box, sel.Enrollment, pool, nil |
| 163 | } |
| 164 | hosts, err := ListClaims(ctx, a.Store) |
| 165 | if err != nil { |
| 166 | return "", "", "", err |
| 167 | } |
| 168 | for _, host := range hosts { |
| 169 | obs, err := hostclaim.Read(ctx, a.Store, host) |
| 170 | if err != nil { |
| 171 | return "", "", "", err |
| 172 | } |
| 173 | if obs.Absent || obs.Claim.Holder == nil { |
| 174 | continue |
| 175 | } |
| 176 | box := obs.Claim.Op.Exec |
| 177 | if box == "" { |
| 178 | box = obs.Claim.Op.Intent["selected_executor"] |
| 179 | } |
| 180 | c := candidates[box] |
| 181 | if c == nil { |
| 182 | continue |
| 183 | } |
| 184 | c.hosts++ |
| 185 | h := obs.Claim.Holder |
| 186 | if h.Account == r.Holder.Account && h.App == r.Holder.App { |
| 187 | c.sameApp = true |
| 188 | } |
| 189 | } |
| 190 | var ordered []*executorCandidate |
| 191 | for _, c := range candidates { |
| 192 | ordered = append(ordered, c) |
| 193 | } |
| 194 | sort.Slice(ordered, func(i, j int) bool { |
| 195 | x, y := ordered[i], ordered[j] |
| 196 | if x.sameApp != y.sameApp { |
| 197 | return !x.sameApp |
| 198 | } |
| 199 | if x.hosts != y.hosts { |
| 200 | return x.hosts < y.hosts |
| 201 | } |
| 202 | if x.hash != y.hash { |
| 203 | return x.hash > y.hash |
| 204 | } |
| 205 | return x.box < y.box |
| 206 | }) |
| 207 | if len(ordered) == 0 { |
| 208 | return "", "", "", ErrPending |
| 209 | } |
| 210 | return ordered[0].box, ordered[0].enrollment, pool, nil |
| 211 | } |
| 212 | |
| 213 | func (a ProductionAuthority) StartCreate(ctx context.Context, before hostclaim.Observed, r Record, id, by string) (hostclaim.Observed, error) { |
| 214 | box, enrollment, pool, err := a.selectExecutor(ctx, before, r) |
| 215 | if err != nil { |
| 216 | return hostclaim.Observed{}, err |
| 217 | } |
| 218 | adm, err := a.admission(ctx, r.Holder.Account, r.Holder.App) |
| 219 | if err != nil { |
| 220 | return hostclaim.Observed{}, err |
| 221 | } |
| 222 | region := r.Holder.Region |
| 223 | if region == park.HomeRegion { |
| 224 | region = a.HomeRegion |
| 225 | } |
| 226 | scope := hostactors.Scope{Namespace: a.Namespace, Regions: []string{region}, Pools: []string{pool}, Participants: []string{box}, ConsoleID: a.ConsoleID} |
| 227 | // Final admission check is after census/reservation, immediately before |
| 228 | // the one CAS against the observation saved before registration. |
| 229 | if err = a.Create(ctx, before, r, id); err != nil { |
| 230 | return hostclaim.Observed{}, err |
| 231 | } |
| 232 | ownershipEffect, err := operationToken() |
| 233 | if err != nil { |
| 234 | return hostclaim.Observed{}, err |
| 235 | } |
| 236 | recordEffect, err := operationToken() |
| 237 | if err != nil { |
| 238 | return hostclaim.Observed{}, err |
| 239 | } |
| 240 | transition := hostclaim.BeginSelectedCreate(r.Holder, r.Body, id, by, a.Now(), box, enrollment, adm.Value.AppID, adm.Value.Epoch, ownershipEffect, recordEffect) |
| 241 | return a.startHeldClaim(ctx, before, id, scope, func() error { return a.Create(ctx, before, r, id) }, func(c *hostclaim.Claim) error { |
| 242 | if err := transition(c); err != nil { |
| 243 | return err |
| 244 | } |
| 245 | if r.VerifiedOwnership { |
| 246 | c.Op.Intent["ownership_verified"] = "true" |
| 247 | } |
| 248 | return nil |
| 249 | }) |
| 250 | } |
| 251 | |
| 252 | func (a ProductionAuthority) Executor(ctx context.Context, obs hostclaim.Observed, r Record) (string, error) { |
| 253 | if obs.Absent || obs.Claim.Holder == nil || !matches(r, obs.Claim.Holder) { |
| 254 | return "", ErrIdentity |
| 255 | } |
| 256 | op := obs.Claim.Op |
| 257 | box, enrollment := op.Intent["selected_executor"], op.Intent["executor_enrollment"] |
| 258 | if box == "" || enrollment == "" { |
| 259 | return "", ErrContract |
| 260 | } |
| 261 | if err := a.Directory.CheckHold(ctx, obs.Claim, box, enrollment); err != nil { |
| 262 | return "", err |
| 263 | } |
| 264 | actor, err := a.Directory.Lookup(ctx, box) |
| 265 | if err != nil { |
| 266 | return "", err |
| 267 | } |
| 268 | if actor.State != "active" || actor.Enrollment != enrollment { |
| 269 | return "", ErrPending |
| 270 | } |
| 271 | adm, err := ReadAdmission(ctx, a.Store, r.Holder.Account, r.Holder.App, op.Intent["app_id"]) |
| 272 | if err != nil { |
| 273 | return "", err |
| 274 | } |
| 275 | epoch, err := strconv.ParseUint(op.Intent["admission_epoch"], 10, 64) |
| 276 | if err != nil || epoch != adm.Value.Epoch { |
| 277 | return "", ErrIdentity |
| 278 | } |
| 279 | foundRecord := false |
| 280 | for _, e := range adm.Value.Effects { |
| 281 | if e.OperationID != op.ID { |
| 282 | continue |
| 283 | } |
| 284 | if e.AppID != adm.Value.AppID || e.State != "complete" { |
| 285 | return "", ErrPending |
| 286 | } |
| 287 | if e.Resource == r.Holder.Key && e.Action == "record-put" { |
| 288 | foundRecord = true |
| 289 | } |
| 290 | } |
| 291 | if !foundRecord { |
| 292 | return "", ErrPending |
| 293 | } |
| 294 | if err = a.FirstPhase(ctx, obs); err != nil { |
| 295 | return "", err |
| 296 | } |
| 297 | return box, nil |
| 298 | } |
| 299 | |
| 300 | func (a ProductionAuthority) FirstPhase(ctx context.Context, obs hostclaim.Observed) error { |
| 301 | if obs.Absent || obs.Claim == nil || obs.Claim.Op.ID == "" { |
| 302 | return ErrIdentity |
| 303 | } |
| 304 | roster, err := a.Directory.Read(ctx) |
| 305 | if err != nil { |
| 306 | return err |
| 307 | } |
| 308 | hold, ok := roster.Roster.Holds[obs.Claim.Op.Intent["actors_hold"]] |
| 309 | if !ok || hold.Host != obs.Host || hold.OpID != obs.Claim.Op.ID { |
| 310 | return ErrPending |
| 311 | } |
| 312 | floor, err := hostfloor.Read(ctx, a.Store, a.Directory.PublicKey, a.Directory.Bucket) |
| 313 | if err != nil { |
| 314 | return err |
| 315 | } |
| 316 | if floor.Digest != hold.FloorSHA256 || roster.Roster.FloorSHA256 != floor.Digest { |
| 317 | return ErrPending |
| 318 | } |
| 319 | for actor, enrollment := range hold.Actors { |
| 320 | report, ok := hold.Releases[actor] |
| 321 | if !ok { |
| 322 | return ErrPending |
| 323 | } |
| 324 | instance := "" |
| 325 | if ref, ok := strings.CutPrefix(actor, "box:"); ok { |
| 326 | box := roster.Roster.Boxes[ref] |
| 327 | if box.State != "active" || box.Enrollment != enrollment { |
| 328 | return ErrPending |
| 329 | } |
| 330 | fenced, err := a.Directory.TerminationFence(ctx, roster, ref, a.Now()) |
| 331 | if err != nil { |
| 332 | return err |
| 333 | } |
| 334 | if fenced { |
| 335 | return ErrPending |
| 336 | } |
| 337 | instance = box.Cloud.InstanceID |
| 338 | } else if ref, ok := strings.CutPrefix(actor, "console:"); ok { |
| 339 | console := roster.Roster.Consoles[ref] |
| 340 | if console.Enrollment != enrollment { |
| 341 | return ErrPending |
| 342 | } |
| 343 | instance = console.InstanceID |
| 344 | } else { |
| 345 | return ErrIdentity |
| 346 | } |
| 347 | if err = hostactors.VerifyReport(ctx, a.Evidence, a.Directory.PublicKey, floor, report, enrollment, instance, a.Now()); err != nil { |
| 348 | return err |
| 349 | } |
| 350 | } |
| 351 | if obs.Claim.Op.Kind == hostclaim.KindPurging { |
| 352 | if err := a.StopBeforeTeardown(ctx, obs); err != nil { |
| 353 | return err |
| 354 | } |
| 355 | } |
| 356 | for _, box := range obs.Claim.Op.Participants { |
| 357 | var ack hostclaim.ExecutionAcknowledgement |
| 358 | if json.Unmarshal([]byte(obs.Claim.Op.Journal["exec_ack."+box]), &ack) != nil || ack.Op != obs.Claim.Op.ID || ack.Enrollment != hold.Actors["box:"+box] || ack.BootID == "" || ack.ControllerID == "" || ack.LeaseETag == "" || !a.Now().Before(ack.Expires.Add(-RegistrationSkew)) { |
| 359 | return ErrPending |
| 360 | } |
| 361 | } |
| 362 | return nil |
| 363 | } |
| 364 | func (a ProductionAuthority) StartDelete(ctx context.Context, obs hostclaim.Observed, r Record) error { |
| 365 | if obs.Absent || obs.Claim.Holder == nil || !matches(r, obs.Claim.Holder) { |
| 366 | return ErrIdentity |
| 367 | } |
| 368 | if obs.Claim.Op.Kind != hostclaim.KindServing { |
| 369 | return ErrPending |
| 370 | } |
| 371 | adm, err := a.admission(ctx, r.Holder.Account, r.Holder.App) |
| 372 | if err != nil { |
| 373 | return err |
| 374 | } |
| 375 | if adm.Value.State != "closed" || adm.Value.Op == nil || adm.Value.Op.Kind != "deleting" || len(adm.Value.Registrations) != 0 { |
| 376 | return ErrPending |
| 377 | } |
| 378 | for _, effect := range adm.Value.Effects { |
| 379 | if effect.State != "complete" { |
| 380 | return ErrPending |
| 381 | } |
| 382 | } |
| 383 | return nil |
| 384 | } |
| 385 | func (a ProductionAuthority) PurgeRetirement(ctx context.Context, obs hostclaim.Observed) error { |
| 386 | if obs.Absent || obs.Claim.Holder == nil || obs.Claim.Op.Kind != hostclaim.KindPurging { |
| 387 | return ErrIdentity |
| 388 | } |
| 389 | h := obs.Claim.Holder |
| 390 | journal, _, err := ReadPurge(ctx, a.Store, h.Account, h.App) |
| 391 | if err != nil { |
| 392 | return err |
| 393 | } |
| 394 | if journal.ID != obs.Claim.Op.ID || journal.Phase != "torn-down" { |
| 395 | return ErrPending |
| 396 | } |
| 397 | return nil |
| 398 | } |
| 399 | |
amadan