| 1 | package provision |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "log" |
| 8 | "os" |
| 9 | |
| 10 | "github.com/carlosframework/platform/internal/registry" |
| 11 | "github.com/carlosframework/platform/internal/release" |
| 12 | "github.com/carlosframework/platform/internal/seeds" |
| 13 | ) |
| 14 | |
| 15 | // rollbackImport is spec §5.6's rollback, driven by the journal and not the |
| 16 | // record (codex R2-7). Each step is idempotent, and the journal's state is |
| 17 | // where a later pass resumes: |
| 18 | // |
| 19 | // swapped 1. close the gate (journal rolling-back, with a fence generation) |
| 20 | // rolling-back 2. fence and stop the host under the journal's generation |
| 21 | // 3. retire the instance key the import created, then |
| 22 | // remove the import's local data |
| 23 | // 4. put the original route back (journal cleanup-pending) |
| 24 | // 5. lift the fence |
| 25 | // cleanup-pending 6. delete this seed's snapshot and the host's replica |
| 26 | // 7. delete the journal row |
| 27 | // 8. CAS the record to rolled-back, if it names this seed |
| 28 | // |
| 29 | // The route steps act only on the import's own row (importRowOf). Any other |
| 30 | // row at the host belongs to someone else, a later instance when a stale |
| 31 | // journal row came back from a snapshot, and the rollback holds there, |
| 32 | // touching nothing, until an operator looks. A host with no row at all is |
| 33 | // rolled back without a restore: whatever removed the import's row decided |
| 34 | // the host should not be served. Its local database at the host's default |
| 35 | // path is still removed when the database's <db>.seed marker names this |
| 36 | // seed and the host still has no row. |
| 37 | // |
| 38 | // Step 3 retires the import's instance key first (merge review S1). The |
| 39 | // key the import's first launch minted into the host's own env file would |
| 40 | // otherwise outlive the rollback and outrank the key a second import of |
| 41 | // the host configures. The journal's EnvBefore, recorded at the swap, says |
| 42 | // what was there before the import: only what the import created goes |
| 43 | // (SeedHooks.RetireKey). With the row gone the key is retired under the |
| 44 | // same marker rule as the database, and before it, so the marker that |
| 45 | // proves the installation is the import's is still there if the key's |
| 46 | // removal has to be tried again. |
| 47 | // |
| 48 | // rec is the pass's view of the journal's record (nil when gone). It chose |
| 49 | // the rollback; the destructive cleanup reads the record again (step 6). |
| 50 | func rollbackImport(ctx context.Context, s release.Store, reg *registry.Registry, cfg ReconcileConfig, j registry.ImportJournal, rec *Record) { |
| 51 | h := cfg.Seeds |
| 52 | if cfg.DryRun { |
| 53 | log.Printf("provision: dry-run: would roll back %s's import %s", j.Host, j.SeedID) |
| 54 | return |
| 55 | } |
| 56 | if !rollbackArmed(h) { |
| 57 | log.Printf("provision: %s: import %s cannot roll back on this box: the activator is not wired; it stays journaled and gated", j.Host, j.SeedID) |
| 58 | return |
| 59 | } |
| 60 | switch j.State { |
| 61 | case registry.JournalSwapped, registry.JournalRollingBack: |
| 62 | var ok bool |
| 63 | if j, ok = rollBackRoute(ctx, s, reg, cfg, j); !ok { |
| 64 | return |
| 65 | } |
| 66 | case registry.JournalCleanupPending: |
| 67 | default: |
| 68 | log.Printf("provision: %s: import %s is journaled in an unknown state %q; leaving it alone", j.Host, j.SeedID, j.State) |
| 69 | return |
| 70 | } |
| 71 | cleanUpImport(ctx, s, reg, cfg, j) |
| 72 | } |
| 73 | |
| 74 | // importRowOf reports whether rt is j's import row as the swap made it: |
| 75 | // the registry's test (gated, provisioned, naming the seed, whose ids are |
| 76 | // never reused), at the nested database path the swap gave it. The path is |
| 77 | // what step 3 removes at, so a row pointing anywhere else is not acted on. |
| 78 | func importRowOf(cfg ReconcileConfig, rt registry.Route, j registry.ImportJournal) bool { |
| 79 | return registry.IsImportRow(rt, j) && rt.DBPath == NestedDBPath(cfg.DataDir, j.Host) |
| 80 | } |
| 81 | |
| 82 | // rollBackRoute is steps 1 to 5. It returns the journal as it now stands |
| 83 | // and whether the route steps are done (the journal is cleanup-pending). |
| 84 | func rollBackRoute(ctx context.Context, s release.Store, reg *registry.Registry, cfg ReconcileConfig, j registry.ImportJournal) (registry.ImportJournal, bool) { |
| 85 | h := cfg.Seeds |
| 86 | // Nothing is closed, fenced or removed for a row that is not the |
| 87 | // import's. The registry refuses the gate close and the restore for |
| 88 | // one as well; this check keeps the fence and the removal off it. |
| 89 | row, err := reg.Get(j.Host) |
| 90 | if err != nil { |
| 91 | log.Printf("provision: %s: import %s rollback deferred: the route could not be read: %v", j.Host, j.SeedID, err) |
| 92 | return j, false |
| 93 | } |
| 94 | if row != nil && !importRowOf(cfg, *row, j) { |
| 95 | // A fence an earlier pass kept for this rollback must not hold the |
| 96 | // host for whoever owns it now. LiftFence lifts only this |
| 97 | // generation. |
| 98 | if j.State == registry.JournalRollingBack { |
| 99 | h.LiftFence(j.Host, j.FenceGen) |
| 100 | } |
| 101 | log.Printf("provision: %s: import %s is not rolled back: the route at the host is not the import's, so nothing there is touched", j.Host, j.SeedID) |
| 102 | return j, false |
| 103 | } |
| 104 | |
| 105 | // Step 1. The generation an earlier close stored is the one the fence |
| 106 | // is known by, so the journal is read back rather than trusted. |
| 107 | if j.State == registry.JournalSwapped { |
| 108 | gen, err := h.NextFenceGen() |
| 109 | if err == nil && gen == 0 { |
| 110 | err = errors.New("the generation source answered 0") |
| 111 | } |
| 112 | if err != nil { |
| 113 | log.Printf("provision: %s: import %s rollback deferred: no fence generation: %v", j.Host, j.SeedID, err) |
| 114 | return j, false |
| 115 | } |
| 116 | if err := reg.CloseImportGate(j.Host, j.SeedID, gen); err != nil { |
| 117 | log.Printf("provision: %s: import %s: close the import gate: %v", j.Host, j.SeedID, err) |
| 118 | return j, false |
| 119 | } |
| 120 | cur, err := reg.ImportJournal(j.Host) |
| 121 | if err != nil || cur == nil || cur.SeedID != j.SeedID || cur.State != registry.JournalRollingBack { |
| 122 | log.Printf("provision: %s: import %s: the journal after the gate closed is %+v (%v)", j.Host, j.SeedID, cur, err) |
| 123 | return j, false |
| 124 | } |
| 125 | j = *cur |
| 126 | } |
| 127 | |
| 128 | // Step 2. The host is fenced and stopped before anything is removed, |
| 129 | // so no wake can write the database or the replica afterwards. The |
| 130 | // hook lifts any fence it installs when it does not proceed. |
| 131 | proceed, err := h.FenceForRollback(ctx, j.Host, j.FenceGen) |
| 132 | if err != nil || !proceed { |
| 133 | if err != nil { |
| 134 | h.LiftFence(j.Host, j.FenceGen) // a hook that failed may have fenced first |
| 135 | } |
| 136 | log.Printf("provision: %s: import %s rollback deferred: the host is not fenced and stopped yet (%v)", j.Host, j.SeedID, err) |
| 137 | return j, false |
| 138 | } |
| 139 | |
| 140 | // Step 3, under the fence. A failure from here on keeps the fence: the |
| 141 | // next pass finds it its own by the journal's generation and resumes. |
| 142 | row, err = reg.Get(j.Host) |
| 143 | if err != nil { |
| 144 | log.Printf("provision: %s: import %s: read the route under the fence: %v", j.Host, j.SeedID, err) |
| 145 | return j, false |
| 146 | } |
| 147 | switch { |
| 148 | case row == nil: |
| 149 | if !removeOrphanedLocal(reg, cfg, j) { |
| 150 | return j, false |
| 151 | } |
| 152 | case !importRowOf(cfg, *row, j) || row.ImportGate != registry.ImportGateClosed: |
| 153 | h.LiftFence(j.Host, j.FenceGen) |
| 154 | log.Printf("provision: %s: import %s is not rolled back: the route changed under the fence, so nothing there is touched", j.Host, j.SeedID) |
| 155 | return j, false |
| 156 | default: |
| 157 | if err := h.RetireKey(row.DBPath, j.Host, j.EnvBefore); err != nil { |
| 158 | log.Printf("provision: %s: import %s: retire its instance key: %v", j.Host, j.SeedID, err) |
| 159 | return j, false |
| 160 | } |
| 161 | if err := h.RemoveLocal(*row); err != nil { |
| 162 | log.Printf("provision: %s: import %s: remove its local data: %v", j.Host, j.SeedID, err) |
| 163 | return j, false |
| 164 | } |
| 165 | } |
| 166 | |
| 167 | // Step 4: the original route back, and the journal at cleanup-pending, |
| 168 | // in one transaction. Step 5: the fence goes. |
| 169 | if err := reg.RestoreImportRoute(j.Host, j.SeedID); err != nil { |
| 170 | log.Printf("provision: %s: import %s: restore the original route: %v", j.Host, j.SeedID, err) |
| 171 | return j, false |
| 172 | } |
| 173 | h.LiftFence(j.Host, j.FenceGen) |
| 174 | j.State = registry.JournalCleanupPending |
| 175 | if row != nil { |
| 176 | log.Printf("provision: %s: import %s rolled back to its original route", j.Host, j.SeedID) |
| 177 | } |
| 178 | return j, true |
| 179 | } |
| 180 | |
| 181 | // cleanUpImport is steps 6 to 8, run from cleanup-pending. It never fences |
| 182 | // or stops anything: the route is the original row by now (or none), and |
| 183 | // its process is the app's own. |
| 184 | func cleanUpImport(ctx context.Context, s release.Store, reg *registry.Registry, cfg ReconcileConfig, j registry.ImportJournal) { |
| 185 | h := cfg.Seeds |
| 186 | valid := seeds.Check(j.Host, j.SeedID) == nil |
| 187 | if !valid { |
| 188 | log.Printf("provision: %s: import %q names no snapshot and no replica; only its journal row is cleaned up", j.Host, j.SeedID) |
| 189 | } else if err := h.Data.Delete(ctx, seeds.SnapshotKey(j.Host, j.SeedID)); err != nil && !errors.Is(err, os.ErrNotExist) { |
| 190 | log.Printf("provision: %s: import %s: delete its snapshot: %v", j.Host, j.SeedID, err) |
| 191 | return |
| 192 | } |
| 193 | |
| 194 | // The replica is keyed by host, not by seed, so the record is read |
| 195 | // again right before it: the pass's copy may be stale. |
| 196 | rec, err := journalRecordByKey(ctx, s, j) |
| 197 | if err != nil { |
| 198 | log.Printf("provision: %s: import %s cleanup deferred: its record cannot be read: %v", j.Host, j.SeedID, err) |
| 199 | return |
| 200 | } |
| 201 | if valid { |
| 202 | ours, why, err := replicaIsTheImports(ctx, s, j, rec) |
| 203 | switch { |
| 204 | case err != nil: |
| 205 | log.Printf("provision: %s: import %s cleanup deferred: %v", j.Host, j.SeedID, err) |
| 206 | return |
| 207 | case !ours: |
| 208 | log.Printf("provision: %s: import %s: the host's replica is left alone: %s", j.Host, j.SeedID, why) |
| 209 | default: |
| 210 | if err := h.DeleteReplica(ctx, j.Host); err != nil { |
| 211 | log.Printf("provision: %s: import %s: delete the host's replica: %v", j.Host, j.SeedID, err) |
| 212 | return |
| 213 | } |
| 214 | } |
| 215 | } |
| 216 | |
| 217 | if err := reg.DeleteImportJournal(j.Host, j.SeedID); err != nil { |
| 218 | log.Printf("provision: %s: import %s: retire its journal row: %v", j.Host, j.SeedID, err) |
| 219 | return |
| 220 | } |
| 221 | log.Printf("provision: %s: import %s is cleaned up and its journal row retired", j.Host, j.SeedID) |
| 222 | |
| 223 | // Step 8. Only from aborted (SeedRolledBack): a record already |
| 224 | // refused or rolled back, or gone, is done. An aborted record whose |
| 225 | // CAS fails is finished by the no-journal path on a later pass. |
| 226 | if rec == nil || rec.Seed == nil || rec.Seed.ID != j.SeedID { |
| 227 | return |
| 228 | } |
| 229 | if _, err := SetRecordSeed(ctx, s, rec.Account, rec.App, rec.Host, SeedRolledBack(j.SeedID)); err != nil && |
| 230 | !errors.Is(err, ErrSeedConflict) && !errors.Is(err, os.ErrNotExist) { |
| 231 | log.Printf("provision: %s: mark import %s rolled back: %v", j.Host, j.SeedID, err) |
| 232 | } |
| 233 | } |
| 234 | |
| 235 | // removeOrphanedLocal is step 3 for an import whose row something removed, |
| 236 | // run under the fence. Its local database is still at the host's default |
| 237 | // nested path (routeFor's), where a later instance at the host would open |
| 238 | // it as a warm cache and replicate it as its own. It is decided from the |
| 239 | // file, whoever holds the host's record: while the journal row exists no |
| 240 | // other row can be minted at the host, so under the fence nothing else |
| 241 | // has that file open. The database is removed only when its <db>.seed |
| 242 | // marker names this seed and the host still has no row. A marker that is |
| 243 | // absent or names another seed does not prove the data is the import's, |
| 244 | // so the file is left. The import's instance key is retired by the same |
| 245 | // rule, just before the database, and left with it. ok false defers the |
| 246 | // rollback, with the fence lifted. |
| 247 | func removeOrphanedLocal(reg *registry.Registry, cfg ReconcileConfig, j registry.ImportJournal) (ok bool) { |
| 248 | h := cfg.Seeds |
| 249 | db := NestedDBPath(cfg.DataDir, j.Host) |
| 250 | id, err := h.LocalSeedID(db) |
| 251 | if err != nil { |
| 252 | h.LiftFence(j.Host, j.FenceGen) |
| 253 | log.Printf("provision: %s: import %s rollback deferred: the seed marker beside %s cannot be read: %v", j.Host, j.SeedID, db, err) |
| 254 | return false |
| 255 | } |
| 256 | if id != j.SeedID { |
| 257 | log.Printf("provision: %s: import %s has no route on this box; the database at %s is left: its seed marker names %q, not this seed", j.Host, j.SeedID, db, id) |
| 258 | return true |
| 259 | } |
| 260 | row, err := reg.Get(j.Host) |
| 261 | if err != nil { |
| 262 | h.LiftFence(j.Host, j.FenceGen) |
| 263 | log.Printf("provision: %s: import %s rollback deferred: the route could not be read: %v", j.Host, j.SeedID, err) |
| 264 | return false |
| 265 | } |
| 266 | if row != nil { |
| 267 | // Not expected while journaled. The next pass judges the row. |
| 268 | h.LiftFence(j.Host, j.FenceGen) |
| 269 | log.Printf("provision: %s: import %s: a route appeared at the host, so its local database is left and the rollback waits", j.Host, j.SeedID) |
| 270 | return false |
| 271 | } |
| 272 | // The key before the database: the marker just read is what says this |
| 273 | // installation is the import's, and RemoveLocal removes it. |
| 274 | if err := h.RetireKey(db, j.Host, j.EnvBefore); err != nil { |
| 275 | log.Printf("provision: %s: import %s: retire its instance key: %v", j.Host, j.SeedID, err) |
| 276 | return false |
| 277 | } |
| 278 | if err := h.RemoveLocal(registry.Route{Host: j.Host, Kind: "instance", DBPath: db}); err != nil { |
| 279 | log.Printf("provision: %s: import %s: remove its local data: %v", j.Host, j.SeedID, err) |
| 280 | return false |
| 281 | } |
| 282 | log.Printf("provision: %s: import %s has no route on this box: its local data is removed and no route is put back", j.Host, j.SeedID) |
| 283 | return true |
| 284 | } |
| 285 | |
| 286 | // replicaIsTheImports decides step 6's replica delete. The host's replica |
| 287 | // is the import's to delete only while the journal's record, if there is |
| 288 | // one, still names this seed in an abandoned state, and no other record |
| 289 | // holds the host. A record gone with no other record for the host counts. |
| 290 | // A record naming another seed, or none, or another record for the host, |
| 291 | // means the replica is someone else's: ours is false, and why says so. |
| 292 | // Anything this pass cannot tell is an error, and the cleanup waits. |
| 293 | func replicaIsTheImports(ctx context.Context, s release.Store, j registry.ImportJournal, rec *Record) (ours bool, why string, err error) { |
| 294 | if rec != nil { |
| 295 | switch { |
| 296 | case rec.Seed == nil || rec.Seed.ID != j.SeedID: |
| 297 | return false, "its record no longer names this seed", nil |
| 298 | case rec.Seed.State != seeds.StateAborted && rec.Seed.State != seeds.StateRefused && rec.Seed.State != seeds.StateRolledBack: |
| 299 | return false, "", fmt.Errorf("its record names this seed as %s, not abandoned", rec.Seed.State) |
| 300 | } |
| 301 | } |
| 302 | known, failed, err := knownAccounts(ctx, s) |
| 303 | if err != nil { |
| 304 | return false, "", err |
| 305 | } |
| 306 | if failed { |
| 307 | return false, "", errors.New("an account could not be read, so another record for the host cannot be ruled out") |
| 308 | } |
| 309 | keys, partial, err := listAppSubtree(ctx, s, known, "instances/") |
| 310 | if err != nil { |
| 311 | return false, "", err |
| 312 | } |
| 313 | if partial { |
| 314 | return false, "", errors.New("the records could not all be listed, so another record for the host cannot be ruled out") |
| 315 | } |
| 316 | for _, key := range keys { |
| 317 | if _, _, host, ok := ParseRecordKey(key, known); ok && host == j.Host && key != j.RecordKey { |
| 318 | return false, "the record " + key + " holds the host", nil |
| 319 | } |
| 320 | } |
| 321 | return true, "", nil |
| 322 | } |
| 323 | |
| 324 | // journalRecordByKey reads the record under the journal's own key: nil when |
| 325 | // it is gone, an error when it cannot be read, does not decode, or is not |
| 326 | // the journal's (another host or key in its body). |
| 327 | func journalRecordByKey(ctx context.Context, s release.Store, j registry.ImportJournal) (*Record, error) { |
| 328 | raw, _, err := s.Get(ctx, j.RecordKey) |
| 329 | if errors.Is(err, os.ErrNotExist) { |
| 330 | return nil, nil |
| 331 | } |
| 332 | if err != nil { |
| 333 | return nil, err |
| 334 | } |
| 335 | r, err := DecodeRecord(raw) |
| 336 | if err != nil { |
| 337 | return nil, err |
| 338 | } |
| 339 | if r.Host != j.Host || RecordKey(r.Account, r.App, r.Host) != j.RecordKey { |
| 340 | return nil, fmt.Errorf("the record under %s is %s's", j.RecordKey, RecordKey(r.Account, r.App, r.Host)) |
| 341 | } |
| 342 | return &r, nil |
| 343 | } |
| 344 | |
amadan