| 1 | package console |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "encoding/json" |
| 6 | "errors" |
| 7 | "os" |
| 8 | "time" |
| 9 | |
| 10 | "github.com/carlosframework/platform/internal/domains" |
| 11 | "github.com/carlosframework/platform/internal/hostclaim" |
| 12 | "github.com/carlosframework/platform/internal/hostops" |
| 13 | "github.com/carlosframework/platform/internal/provision" |
| 14 | "github.com/carlosframework/platform/internal/release" |
| 15 | ) |
| 16 | |
| 17 | // retireClaimHostRecord runs one delete as an exclusive "deleting" |
| 18 | // operation. The order is fixed: |
| 19 | // |
| 20 | // 1. Begin (or resume) the operation and drain its registrations. |
| 21 | // 2. Under the operation's own lifecycle grant, and with no effect in |
| 22 | // flight, CAS the host claim from serving to deleting. The production |
| 23 | // authority refuses that CAS while any effect of the app is unsettled, |
| 24 | // so it never runs inside an effect. |
| 25 | // 3. Delete the record and park the claim inside one record effect: the |
| 26 | // typed record delete needs that effect's permit. |
| 27 | // 4. Free the claim once parking has no obligations left, release the |
| 28 | // wake and ownership markers (one effect) and reopen (FinishDelete). |
| 29 | // |
| 30 | // From step 2 on the delete intent is durable and owned by this |
| 31 | // operation: a step that is only waiting (parking, a box acknowledgement) |
| 32 | // answers claimRetireAccepted, and recovery or a retry carries it on. |
| 33 | func (s *Server) retireClaimHostRecord(ctx context.Context, ref provision.RetireRef) error { |
| 34 | c := s.cfg.HostClaimOps |
| 35 | if c == nil || c.Control == nil || s.claimOwner == (hostops.RegistrationOwner{}) { |
| 36 | return provision.ErrClaimRetirer |
| 37 | } |
| 38 | acct, err := s.store.AccountBySqid(ref.Account) |
| 39 | if err != nil { |
| 40 | return err |
| 41 | } |
| 42 | if acct == nil { |
| 43 | return hostops.ErrIdentity |
| 44 | } |
| 45 | appID, err := s.store.ClaimAppID(acct.ID, ref.App) |
| 46 | if err != nil { |
| 47 | return err |
| 48 | } |
| 49 | admission, err := hostops.ReadAdmission(ctx, s.cfg.Releases, ref.Account, ref.App, appID) |
| 50 | if err != nil { |
| 51 | return err |
| 52 | } |
| 53 | var h *hostops.OperationHandle |
| 54 | if op := admission.Value.Op; op == nil { |
| 55 | key, err := claimRetireRecordKey(ref) |
| 56 | if err != nil { |
| 57 | return err |
| 58 | } |
| 59 | obs, err := hostclaim.Read(ctx, s.cfg.Releases, ref.Host) |
| 60 | if err != nil { |
| 61 | return err |
| 62 | } |
| 63 | if obs.Absent { |
| 64 | // Nothing holds the host: "not there" is reached when no record |
| 65 | // exists either. A record without a claim waits for backfill. |
| 66 | if _, _, gerr := s.cfg.Releases.Get(ctx, key); errors.Is(gerr, os.ErrNotExist) { |
| 67 | return nil |
| 68 | } else if gerr != nil { |
| 69 | return gerr |
| 70 | } |
| 71 | return hostops.ErrPending |
| 72 | } |
| 73 | holder := obs.Claim.Holder |
| 74 | if holder == nil { |
| 75 | if obs.Claim.Free() { |
| 76 | return nil |
| 77 | } |
| 78 | return hostops.ErrPending |
| 79 | } |
| 80 | if !claimRetireHolderIs(holder, ref, key) { |
| 81 | return hostops.ErrIdentity |
| 82 | } |
| 83 | id, err := claimEffectID() |
| 84 | if err != nil { |
| 85 | return err |
| 86 | } |
| 87 | begin := hostops.AppOperation{ID: id, Kind: hostops.OpDeleting, AppID: appID, Phase: "draining", Intent: map[string]string{"host": ref.Host, "incarnation": holder.Incarnation, "acquisition": holder.Acquisition, "family": ref.Family, "name": ref.Name}} |
| 88 | if h, err = c.Control.BeginExclusive(ctx, ref.Account, ref.App, appID, begin, s.claimOwner); err != nil { |
| 89 | return err |
| 90 | } |
| 91 | if admission, err = hostops.ReadAdmission(ctx, s.cfg.Releases, ref.Account, ref.App, appID); err != nil { |
| 92 | return err |
| 93 | } |
| 94 | } else { |
| 95 | // Only this host's own delete is resumed. Any other exclusive |
| 96 | // operation (a foreign mutation, another host's delete, a purge or a |
| 97 | // placement) refuses here, before any effect. |
| 98 | if op.Kind != hostops.OpDeleting || op.Intent["host"] != ref.Host || (ref.Incarnation != "" && op.Intent["incarnation"] != ref.Incarnation) { |
| 99 | return hostops.ErrPending |
| 100 | } |
| 101 | if h, err = c.Control.ResumeExclusive(ctx, ref.Account, ref.App, appID, op.ID, hostops.OpDeleting, s.claimOwner); err != nil { |
| 102 | return err |
| 103 | } |
| 104 | } |
| 105 | op := admission.Value.Op |
| 106 | if op == nil || op.ID != h.ID() || op.Intent["host"] != ref.Host || (ref.Incarnation != "" && op.Intent["incarnation"] != ref.Incarnation) { |
| 107 | return hostops.ErrPending |
| 108 | } |
| 109 | ref.Incarnation = op.Intent["incarnation"] |
| 110 | if ref.Family == "" { |
| 111 | ref.Family, ref.Name = op.Intent["family"], op.Intent["name"] |
| 112 | } |
| 113 | key, err := claimRetireRecordKey(ref) |
| 114 | if err != nil { |
| 115 | return err |
| 116 | } |
| 117 | // Every claim write below carries this operation's lifecycle grant. |
| 118 | ctx = h.Lifecycle(ctx) |
| 119 | // The operation is durable from here: a step that is only waiting |
| 120 | // answers claimRetireAccepted with this operation's ID and phase. |
| 121 | if err = h.Drain(ctx, c.Now(), c.RecoverCreate); err != nil { |
| 122 | return claimRetireWaiting(h, "draining", ref.Host, err) |
| 123 | } |
| 124 | if err = s.advanceClaimRetire(ctx, c, h, ref, key); err != nil { |
| 125 | return err |
| 126 | } |
| 127 | // The claim is free: the incarnation is retired, and its move log |
| 128 | // outbox rows go with it (Plan 6 Task 3; the drainer's drop of a row |
| 129 | // whose claim is gone is the backstop). |
| 130 | if err = s.store.PurgeMoveLogIncarnation(ref.Host, ref.Incarnation); err != nil { |
| 131 | return claimRetireWaiting(h, "releasing", ref.Host, err) |
| 132 | } |
| 133 | if err = s.releaseClaimRetireMarkers(ctx, h, ref, appID, op.Intent["acquisition"]); err != nil { |
| 134 | return claimRetireWaiting(h, "releasing", ref.Host, err) |
| 135 | } |
| 136 | return claimRetireWaiting(h, "releasing", ref.Host, h.FinishDelete(ctx)) |
| 137 | } |
| 138 | |
| 139 | // advanceClaimRetire moves the host claim from wherever this operation |
| 140 | // left it to free. nil means free; claimRetireAccepted means the owned |
| 141 | // delete is waiting on boxes or parking. |
| 142 | func (s *Server) advanceClaimRetire(ctx context.Context, c *hostops.Controller, h *hostops.OperationHandle, ref provision.RetireRef, key string) error { |
| 143 | runner := h.Runner(time.Now) |
| 144 | owned := false |
| 145 | // Each pass makes at most one transition, so four passes cover |
| 146 | // serving, deleting, parking and free. |
| 147 | for range 4 { |
| 148 | obs, err := hostclaim.Read(ctx, s.cfg.Releases, ref.Host) |
| 149 | if err != nil { |
| 150 | return err |
| 151 | } |
| 152 | if obs.Absent { |
| 153 | return hostops.ErrPending |
| 154 | } |
| 155 | if obs.Claim.Free() { |
| 156 | return nil |
| 157 | } |
| 158 | holder := obs.Claim.Holder |
| 159 | if holder == nil || !claimRetireHolderIs(holder, ref, key) || holder.Incarnation != ref.Incarnation { |
| 160 | return hostops.ErrIdentity |
| 161 | } |
| 162 | claimOp := obs.Claim.Op |
| 163 | switch claimOp.Kind { |
| 164 | case hostclaim.KindServing: |
| 165 | if owned { |
| 166 | return hostops.ErrPending |
| 167 | } |
| 168 | // Step 2, the durable delete intent. No effect is issued here. |
| 169 | if _, err = c.BeginRetire(ctx, key, ref.By, ref.Incarnation); err != nil { |
| 170 | return claimRetireWaiting(h, "starting", ref.Host, err) |
| 171 | } |
| 172 | case hostclaim.KindDeleting: |
| 173 | owned = true |
| 174 | refreshed, err := s.refreshClaimRetireRecord(ctx, c, obs, key) |
| 175 | if err != nil { |
| 176 | return claimRetireWaiting(h, claimOp.Kind, ref.Host, err) |
| 177 | } |
| 178 | if refreshed { |
| 179 | // Re-read the re-bound claim before deleting. |
| 180 | continue |
| 181 | } |
| 182 | effectID, err := claimEffectID() |
| 183 | if err != nil { |
| 184 | return err |
| 185 | } |
| 186 | // Step 3: the typed record delete runs under this effect's permit. |
| 187 | _, err = runner.Run(ctx, hostops.Effect{ID: effectID, Resource: "record:" + ref.Host, Action: "retire"}, func(ctx context.Context) (string, error) { |
| 188 | rerr := c.Retire(ctx, ref.Host, claimOp.ID) |
| 189 | if rerr == nil { |
| 190 | return "retired", nil |
| 191 | } |
| 192 | if errors.Is(rerr, hostops.ErrPending) { |
| 193 | fresh, ferr := hostclaim.Read(ctx, s.cfg.Releases, ref.Host) |
| 194 | if ferr == nil && !fresh.Absent && fresh.Claim.Op.ID == claimOp.ID && fresh.Claim.Op.Kind == hostclaim.KindParking { |
| 195 | // Retire answers pending after a successful park. |
| 196 | return "parked", nil |
| 197 | } |
| 198 | // The record was not deleted (boxes not ready yet). |
| 199 | return "", errors.Join(hostops.ErrNotApplied, rerr) |
| 200 | } |
| 201 | return "", rerr |
| 202 | }) |
| 203 | if err != nil { |
| 204 | return claimRetireWaiting(h, claimOp.Kind, ref.Host, err) |
| 205 | } |
| 206 | case hostclaim.KindParking: |
| 207 | owned = true |
| 208 | if len(obs.Claim.Obligations) != 0 { |
| 209 | return &claimRetireAccepted{operation: h.ID(), phase: hostclaim.KindParking, host: ref.Host} |
| 210 | } |
| 211 | // A claim-only transition under the lifecycle grant. |
| 212 | if err = c.Free(ctx, ref.Host, claimOp.ID); err != nil { |
| 213 | return claimRetireWaiting(h, claimOp.Kind, ref.Host, err) |
| 214 | } |
| 215 | default: |
| 216 | // Purging, placing and moving claims belong to their own operations. |
| 217 | return hostops.ErrPending |
| 218 | } |
| 219 | } |
| 220 | return &claimRetireAccepted{operation: h.ID(), phase: hostclaim.KindDeleting, host: ref.Host} |
| 221 | } |
| 222 | |
| 223 | // refreshClaimRetireRecord re-binds the deleting claim to a record whose |
| 224 | // ETag moved since the claim was written. It runs outside any effect: the |
| 225 | // production refresh refuses while an effect is unsettled. |
| 226 | func (s *Server) refreshClaimRetireRecord(ctx context.Context, c *hostops.Controller, obs hostclaim.Observed, key string) (bool, error) { |
| 227 | rec, err := hostops.ReadRecord(ctx, s.cfg.Releases, key) |
| 228 | if errors.Is(err, os.ErrNotExist) { |
| 229 | return false, nil |
| 230 | } |
| 231 | if err != nil { |
| 232 | return false, err |
| 233 | } |
| 234 | if rec.ETag == obs.Claim.Op.Intent["etag"] { |
| 235 | return false, nil |
| 236 | } |
| 237 | refresher, ok := c.Authority.(interface { |
| 238 | RefreshDeleteRecord(context.Context, hostclaim.Observed, hostops.Record) error |
| 239 | }) |
| 240 | if !ok { |
| 241 | return true, hostops.ErrPending |
| 242 | } |
| 243 | return true, refresher.RefreshDeleteRecord(ctx, obs, rec) |
| 244 | } |
| 245 | |
| 246 | // releaseClaimRetireMarkers removes the sidecar wake object and the |
| 247 | // ownership marker, both bound to this incarnation, as one effect. It runs |
| 248 | // only after the claim is free. |
| 249 | func (s *Server) releaseClaimRetireMarkers(ctx context.Context, h *hostops.OperationHandle, ref provision.RetireRef, appID, acquisition string) error { |
| 250 | cleanupID, err := claimEffectID() |
| 251 | if err != nil { |
| 252 | return err |
| 253 | } |
| 254 | _, err = h.Runner(time.Now).Run(ctx, hostops.Effect{ID: cleanupID, Resource: "ownership:" + ref.Host, Action: "cleanup"}, func(ctx context.Context) (string, error) { |
| 255 | if ref.Family == "sidecar" { |
| 256 | key := release.SidecarWakeKey(ref.Account, ref.App, ref.Name) |
| 257 | raw, etag, err := s.cfg.Releases.Get(ctx, key) |
| 258 | if err != nil && !errors.Is(err, os.ErrNotExist) { |
| 259 | return "", err |
| 260 | } |
| 261 | if err == nil { |
| 262 | var wake struct { |
| 263 | AppID string `json:"app_id"` |
| 264 | Incarnation string `json:"incarnation"` |
| 265 | RequestID string `json:"request_id"` |
| 266 | } |
| 267 | if json.Unmarshal(raw, &wake) != nil { |
| 268 | return "", hostops.ErrIdentity |
| 269 | } |
| 270 | if (wake.AppID != "" || wake.Incarnation != "" || wake.RequestID != "") && (wake.AppID != appID || wake.Incarnation != ref.Incarnation || wake.RequestID == "") { |
| 271 | return "", hostops.ErrIdentity |
| 272 | } |
| 273 | if err = release.DeleteIfMatch(ctx, s.cfg.Releases, key, etag); err != nil { |
| 274 | return "", err |
| 275 | } |
| 276 | } |
| 277 | } |
| 278 | key := domains.ClaimKey(ref.Host) |
| 279 | raw, etag, err := s.cfg.Releases.Get(ctx, key) |
| 280 | if err != nil && !errors.Is(err, os.ErrNotExist) { |
| 281 | return "", err |
| 282 | } |
| 283 | if err == nil { |
| 284 | var owner domains.Claim |
| 285 | if json.Unmarshal(raw, &owner) != nil { |
| 286 | return "", hostops.ErrIdentity |
| 287 | } |
| 288 | if owner.Account != ref.Account || owner.App != ref.App || owner.Acquisition != acquisition || owner.RecordIncarnation != ref.Incarnation { |
| 289 | return "", hostops.ErrIdentity |
| 290 | } |
| 291 | if err = release.DeleteIfMatch(ctx, s.cfg.Releases, key, etag); err != nil { |
| 292 | return "", err |
| 293 | } |
| 294 | } |
| 295 | return "joined", nil |
| 296 | }) |
| 297 | return err |
| 298 | } |
| 299 | |
| 300 | func (s *Server) retireInstanceRecord(ctx context.Context, account, app, host, by, incarnation string) error { |
| 301 | return s.retireHostRecord(ctx, provision.RetireRef{Family: "instance", Account: account, App: app, Host: host, By: by, Incarnation: incarnation}) |
| 302 | } |
| 303 | |
| 304 | func mustCreateIncarnation(account, app, host string, created time.Time) string { |
| 305 | inc, err := hostclaim.Incarnation(account, app, host, created) |
| 306 | if err != nil { |
| 307 | return "invalid-incarnation" |
| 308 | } |
| 309 | return inc |
| 310 | } |
| 311 | |
| 312 | func (s *Server) deleteLegacySidecarWake(ctx context.Context, account, app, name string) error { |
| 313 | if s.cfg.HostClaim { |
| 314 | return nil |
| 315 | } |
| 316 | return s.cfg.Releases.Delete(ctx, release.SidecarWakeKey(account, app, name)) |
| 317 | } |
| 318 | |
amadan