| 1 | package hostops |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "os" |
| 7 | "slices" |
| 8 | "time" |
| 9 | |
| 10 | "github.com/carlosframework/platform/internal/hostclaim" |
| 11 | "github.com/carlosframework/platform/internal/provision" |
| 12 | "github.com/carlosframework/platform/internal/release" |
| 13 | ) |
| 14 | |
| 15 | // These callbacks are authority providers, not optional best-effort hints. |
| 16 | // Production construction waits for the decisions named in this plan. |
| 17 | type Authority interface { |
| 18 | // SelectExecutor chooses the new host's executor once, before the |
| 19 | // create CAS (claim P5-D2, region HA §5.2). For a non-production |
| 20 | // authority, selected false means no choice at create: Attempt runs |
| 21 | // today's BeginCreate and Executor names the box at completion. |
| 22 | // Production is ExecutorChooser.Choose behind CARLOS_REGION_HA, |
| 23 | // consulted inside StartCreate's own selection; with it off, claim |
| 24 | // Plan 5's selectExecutor chooses as before. |
| 25 | SelectExecutor(context.Context, hostclaim.Observed, Record) (sel hostclaim.ExecutorSelection, selected bool, err error) |
| 26 | Create(context.Context, hostclaim.Observed, Record, string) error |
| 27 | Executor(context.Context, hostclaim.Observed, Record) (string, error) |
| 28 | StartDelete(context.Context, hostclaim.Observed, Record) error |
| 29 | FirstPhase(context.Context, hostclaim.Observed) error |
| 30 | Inventory(context.Context, hostclaim.Observed) ([]string, []string, error) |
| 31 | Adoption(context.Context, Record) (region, box string, otherRegions, otherBoxes []string, err error) |
| 32 | PurgeRetirement(context.Context, hostclaim.Observed) error |
| 33 | } |
| 34 | |
| 35 | type Controller struct { |
| 36 | Store release.Store |
| 37 | // Control is the admission engine's typed writer (Task 7). nil in the |
| 38 | // C5-only slice; every admission/journal transition requires it. |
| 39 | Control *ControlWriter |
| 40 | Authority Authority |
| 41 | Now func() time.Time |
| 42 | NewID func() string |
| 43 | } |
| 44 | |
| 45 | func (c Controller) ready() error { |
| 46 | if c.Store == nil || c.Authority == nil || c.Now == nil || c.NewID == nil { |
| 47 | return ErrContract |
| 48 | } |
| 49 | return nil |
| 50 | } |
| 51 | |
| 52 | // Attempt consumes the claim observation taken BEFORE app registration. |
| 53 | // Create's caller owns registration renewal/removal. A failed CAS is never retried here. |
| 54 | func (c Controller) Attempt(ctx context.Context, before hostclaim.Observed, r Record, id, by string) (hostclaim.Observed, error) { |
| 55 | if err := c.ready(); err != nil { |
| 56 | return hostclaim.Observed{}, err |
| 57 | } |
| 58 | if before.Host != r.Holder.Host || r.Holder.CreatedAt.IsZero() || id == "" { |
| 59 | return hostclaim.Observed{}, ErrIdentity |
| 60 | } |
| 61 | // With region HA on, the executor is selected once, before the CAS, |
| 62 | // and recorded in it; off, the create is today's. The production |
| 63 | // authority makes that one selection inside StartCreate. |
| 64 | production, isProduction := c.Authority.(productionCreateAuthority) |
| 65 | var sel hostclaim.ExecutorSelection |
| 66 | var selected bool |
| 67 | if !isProduction { |
| 68 | var err error |
| 69 | if sel, selected, err = c.Authority.SelectExecutor(ctx, before, r); err != nil { |
| 70 | return hostclaim.Observed{}, err |
| 71 | } |
| 72 | } |
| 73 | // Last pre-CAS read of registration, open admission and purge reservation. |
| 74 | if err := c.Authority.Create(ctx, before, r, id); err != nil { |
| 75 | return hostclaim.Observed{}, err |
| 76 | } |
| 77 | var obs hostclaim.Observed |
| 78 | var err error |
| 79 | switch { |
| 80 | case isProduction: |
| 81 | obs, err = production.StartCreate(ctx, before, r, id, by) |
| 82 | case selected: |
| 83 | obs, err = hostclaim.Update(ctx, c.Store, before, hostclaim.BeginCreateOn(r.Holder, r.Body, id, by, c.Now(), sel)) |
| 84 | default: |
| 85 | obs, err = hostclaim.Update(ctx, c.Store, before, hostclaim.BeginCreate(r.Holder, r.Body, id, by, c.Now())) |
| 86 | } |
| 87 | if err != nil { |
| 88 | return hostclaim.Observed{}, err |
| 89 | } |
| 90 | if isProduction { |
| 91 | if err := production.RecordEffects(ctx, obs, r); err != nil { |
| 92 | return obs, err |
| 93 | } |
| 94 | return c.Complete(ctx, obs.Host, id) |
| 95 | } |
| 96 | // No raw record effect precedes the durable claim intent. |
| 97 | if _, _, err = provision.WriteRecord(provision.WithRecordSequence(ctx), c.Store, r.Holder.Key, []byte(`{}`), r.Body, ""); err != nil { |
| 98 | if !errors.Is(err, release.ErrCASConflict) { |
| 99 | return obs, err |
| 100 | } |
| 101 | existing, rerr := ReadRecord(ctx, c.Store, r.Holder.Key) |
| 102 | if rerr != nil { |
| 103 | return obs, rerr |
| 104 | } |
| 105 | if !matches(existing, obs.Claim.Holder) { |
| 106 | removed, rerr := RemoveOrphan(ctx, c.Store, existing) |
| 107 | if rerr != nil { |
| 108 | return obs, rerr |
| 109 | } |
| 110 | if !removed { |
| 111 | return obs, ErrPending |
| 112 | } |
| 113 | if _, _, err = provision.WriteRecord(provision.WithRecordSequence(ctx), c.Store, r.Holder.Key, []byte(`{}`), r.Body, ""); err != nil { |
| 114 | return obs, err |
| 115 | } |
| 116 | } |
| 117 | } |
| 118 | return c.Complete(ctx, obs.Host, id) |
| 119 | } |
| 120 | |
| 121 | func (c Controller) Complete(ctx context.Context, host, id string) (hostclaim.Observed, error) { |
| 122 | if err := c.ready(); err != nil { |
| 123 | return hostclaim.Observed{}, err |
| 124 | } |
| 125 | obs, err := hostclaim.Read(ctx, c.Store, host) |
| 126 | if err != nil { |
| 127 | return obs, err |
| 128 | } |
| 129 | if obs.Absent || obs.Claim.Op.Kind != hostclaim.KindCreating || obs.Claim.Op.ID != id { |
| 130 | return obs, ErrPending |
| 131 | } |
| 132 | r, err := ReadRecord(ctx, c.Store, obs.Claim.Holder.Key) |
| 133 | if err != nil { |
| 134 | return obs, err |
| 135 | } |
| 136 | if !matches(r, obs.Claim.Holder) { |
| 137 | return obs, ErrIdentity |
| 138 | } |
| 139 | box, err := c.Authority.Executor(ctx, obs, r) |
| 140 | if err != nil { |
| 141 | return obs, err |
| 142 | } |
| 143 | return hostclaim.Update(ctx, c.Store, obs, hostclaim.CompleteCreate(id, box)) |
| 144 | } |
| 145 | |
| 146 | func (c Controller) RecoverCreate(ctx context.Context, host string) error { |
| 147 | if err := c.ready(); err != nil { |
| 148 | return err |
| 149 | } |
| 150 | obs, err := hostclaim.Read(ctx, c.Store, host) |
| 151 | if err != nil { |
| 152 | return err |
| 153 | } |
| 154 | if obs.Absent || obs.Claim.Op.Kind != hostclaim.KindCreating { |
| 155 | return nil |
| 156 | } |
| 157 | if obs.Claim.Op.StartedAt.IsZero() || !c.Now().After(obs.Claim.Op.StartedAt.Add(5*time.Minute)) { |
| 158 | return ErrPending |
| 159 | } |
| 160 | r, err := ReadRecord(ctx, c.Store, obs.Claim.Holder.Key) |
| 161 | if errors.Is(err, os.ErrNotExist) { |
| 162 | if _, production := c.Authority.(productionCreateAuthority); production { |
| 163 | // The five-minute age is inspection eligibility. An absent record |
| 164 | // cannot settle a delayed ownership, store, or record effect. |
| 165 | return ErrPending |
| 166 | } |
| 167 | _, err = hostclaim.Update(ctx, c.Store, obs, hostclaim.CancelCreate(obs.Claim.Op.ID, c.Now())) |
| 168 | return err |
| 169 | } |
| 170 | if err != nil { |
| 171 | return err |
| 172 | } |
| 173 | if !matches(r, obs.Claim.Holder) { |
| 174 | return ErrPending |
| 175 | } |
| 176 | _, err = c.Complete(ctx, host, obs.Claim.Op.ID) |
| 177 | return err |
| 178 | } |
| 179 | |
| 180 | // Adopt is the shared IO wrapper around the one hostclaim.Adopt transition. |
| 181 | func (c Controller) Adopt(ctx context.Context, r Record) (hostclaim.Observed, error) { |
| 182 | if err := c.ready(); err != nil { |
| 183 | return hostclaim.Observed{}, err |
| 184 | } |
| 185 | obs, err := hostclaim.Read(ctx, c.Store, r.Holder.Host) |
| 186 | if err != nil { |
| 187 | return obs, err |
| 188 | } |
| 189 | if !obs.Absent { |
| 190 | if matches(r, obs.Claim.Holder) { |
| 191 | return obs, nil |
| 192 | } |
| 193 | if obs.Claim.Holder == nil || obs.Claim.Holder.Acquisition != "" || r.Holder.Acquisition == "" { |
| 194 | return obs, ErrIdentity |
| 195 | } |
| 196 | // #232 also repairs already-claimed legacy holders while globally |
| 197 | // frozen. Adoption authority verifies the exact manifest/ownership. |
| 198 | prior := *obs.Claim.Holder |
| 199 | prior.Acquisition = r.Holder.Acquisition |
| 200 | if !matches(r, &prior) { |
| 201 | return obs, ErrIdentity |
| 202 | } |
| 203 | _, box, _, _, err := c.Authority.Adoption(ctx, r) |
| 204 | if err != nil { |
| 205 | return obs, err |
| 206 | } |
| 207 | if obs.Claim.Op.Kind != hostclaim.KindServing || obs.Claim.Op.Exec != box || !slices.Contains(obs.Claim.Touched.Boxes, box) { |
| 208 | return obs, ErrIdentity |
| 209 | } |
| 210 | fresh, err := ReadRecord(ctx, c.Store, r.Holder.Key) |
| 211 | if err != nil { |
| 212 | return obs, err |
| 213 | } |
| 214 | if fresh.ETag != r.ETag { |
| 215 | return obs, ErrPending |
| 216 | } |
| 217 | return hostclaim.Update(ctx, c.Store, obs, func(next *hostclaim.Claim) error { |
| 218 | if next.Holder == nil || next.Holder.Acquisition != "" { |
| 219 | return ErrPending |
| 220 | } |
| 221 | next.Holder.Acquisition = r.Holder.Acquisition |
| 222 | return nil |
| 223 | }) |
| 224 | } |
| 225 | region, box, regions, boxes, err := c.Authority.Adoption(ctx, r) |
| 226 | if err != nil { |
| 227 | return obs, err |
| 228 | } |
| 229 | // Re-read the exact record after discovery; never adopt a stale listing body. |
| 230 | fresh, err := ReadRecord(ctx, c.Store, r.Holder.Key) |
| 231 | if err != nil { |
| 232 | return obs, err |
| 233 | } |
| 234 | if fresh.ETag != r.ETag { |
| 235 | return obs, ErrPending |
| 236 | } |
| 237 | n := fresh.Holder |
| 238 | n.Region = region |
| 239 | return hostclaim.Update(ctx, c.Store, obs, hostclaim.Adopt(n, box, regions, boxes, c.NewID(), c.NewID)) |
| 240 | } |
| 241 | |
| 242 | // BeginRetire only starts the claim operation. Its successful return does not |
| 243 | // mean the record is gone or the host is free; the caller must run/retry Retire. |
| 244 | func (c Controller) BeginRetire(ctx context.Context, key, by, expectedInc string) (hostclaim.Observed, error) { |
| 245 | if err := c.ready(); err != nil { |
| 246 | return hostclaim.Observed{}, err |
| 247 | } |
| 248 | r, err := ReadRecord(ctx, c.Store, key) |
| 249 | if err != nil { |
| 250 | return hostclaim.Observed{}, err |
| 251 | } |
| 252 | inc, ierr := r.Incarnation() |
| 253 | if ierr != nil { |
| 254 | return hostclaim.Observed{}, ierr |
| 255 | } |
| 256 | if expectedInc == "" || inc != expectedInc { |
| 257 | return hostclaim.Observed{}, ErrIdentity |
| 258 | } |
| 259 | obs, err := hostclaim.Read(ctx, c.Store, r.Holder.Host) |
| 260 | if err != nil { |
| 261 | return obs, err |
| 262 | } |
| 263 | if obs.Absent { |
| 264 | return obs, ErrPending |
| 265 | } |
| 266 | if !matches(r, obs.Claim.Holder) { |
| 267 | return obs, ErrIdentity |
| 268 | } |
| 269 | if err := c.Authority.StartDelete(ctx, obs, r); err != nil { |
| 270 | return obs, err |
| 271 | } |
| 272 | regions, boxes, err := c.Authority.Inventory(ctx, obs) |
| 273 | if err != nil { |
| 274 | return obs, err |
| 275 | } |
| 276 | if production, ok := c.Authority.(interface { |
| 277 | StartDeleteClaim(context.Context, hostclaim.Observed, Record, string) (hostclaim.Observed, error) |
| 278 | }); ok { |
| 279 | return production.StartDeleteClaim(ctx, obs, r, by) |
| 280 | } |
| 281 | return hostclaim.Update(ctx, c.Store, obs, hostclaim.BeginDelete(c.NewID(), by, r.ETag, c.Now(), regions, boxes, "")) |
| 282 | } |
| 283 | |
| 284 | func (c Controller) Retire(ctx context.Context, host, id string) error { |
| 285 | if err := c.ready(); err != nil { |
| 286 | return err |
| 287 | } |
| 288 | obs, err := hostclaim.Read(ctx, c.Store, host) |
| 289 | if err != nil { |
| 290 | return err |
| 291 | } |
| 292 | if obs.Absent || obs.Claim.Op.ID != id { |
| 293 | return ErrPending |
| 294 | } |
| 295 | op := obs.Claim.Op |
| 296 | if op.Kind != hostclaim.KindDeleting && op.Kind != hostclaim.KindPurging { |
| 297 | return ErrPending |
| 298 | } |
| 299 | if obs.Claim.Holder == nil || op.Intent["key"] != obs.Claim.Holder.Key || op.Intent["etag"] == "" { |
| 300 | return ErrIdentity |
| 301 | } |
| 302 | if err := c.Authority.FirstPhase(ctx, obs); err != nil { |
| 303 | return err |
| 304 | } |
| 305 | if op.Kind == hostclaim.KindPurging { |
| 306 | if err := c.Authority.PurgeRetirement(ctx, obs); err != nil { |
| 307 | return err |
| 308 | } |
| 309 | } |
| 310 | regions, boxes, err := c.Authority.Inventory(ctx, obs) |
| 311 | if err != nil { |
| 312 | return err |
| 313 | } |
| 314 | // Validate record identity before deleting even the stored version. |
| 315 | r, rerr := ReadRecord(ctx, c.Store, op.Intent["key"]) |
| 316 | if rerr != nil && !errors.Is(rerr, os.ErrNotExist) { |
| 317 | return rerr |
| 318 | } |
| 319 | if rerr == nil { |
| 320 | if !matches(r, obs.Claim.Holder) { |
| 321 | return ErrIdentity |
| 322 | } |
| 323 | if r.ETag != op.Intent["etag"] { |
| 324 | if authority, ok := c.Authority.(interface { |
| 325 | RefreshDeleteRecord(context.Context, hostclaim.Observed, Record) error |
| 326 | }); ok { |
| 327 | return authority.RefreshDeleteRecord(ctx, obs, r) |
| 328 | } |
| 329 | return ErrPending |
| 330 | } |
| 331 | if err := release.DeleteIfMatch(ctx, c.Store, r.Holder.Key, r.ETag); err != nil { |
| 332 | return err |
| 333 | } |
| 334 | } |
| 335 | // A success/412 alone never certifies absence. Re-read before parking. |
| 336 | _, _, err = c.Store.Get(ctx, op.Intent["key"]) |
| 337 | if !errors.Is(err, os.ErrNotExist) { |
| 338 | if err != nil { |
| 339 | return err |
| 340 | } |
| 341 | return ErrPending |
| 342 | } |
| 343 | // The old observation binds the entire sequence to this operation. |
| 344 | transition := hostclaim.ToParking(id, regions, boxes, c.NewID) |
| 345 | if _, production := c.Authority.(productionCreateAuthority); production { |
| 346 | transition = hostclaim.ToScopedParking(id, regions, boxes, c.NewID) |
| 347 | } |
| 348 | _, err = hostclaim.Update(ctx, c.Store, obs, transition) |
| 349 | if err != nil { |
| 350 | return err |
| 351 | } |
| 352 | return ErrPending |
| 353 | } |
| 354 | |
| 355 | func (c Controller) Free(ctx context.Context, host, id string) error { |
| 356 | if err := c.ready(); err != nil { |
| 357 | return err |
| 358 | } |
| 359 | obs, err := hostclaim.Read(ctx, c.Store, host) |
| 360 | if err != nil { |
| 361 | return err |
| 362 | } |
| 363 | if obs.Absent { |
| 364 | return ErrPending |
| 365 | } |
| 366 | _, err = hostclaim.Update(ctx, c.Store, obs, hostclaim.FreeAfterPark(id)) |
| 367 | return err |
| 368 | } |
| 369 | |
amadan