| 1 | package host |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "reflect" |
| 8 | "sync" |
| 9 | "sync/atomic" |
| 10 | "testing" |
| 11 | "time" |
| 12 | |
| 13 | "github.com/carlosframework/platform/internal/activator" |
| 14 | "github.com/carlosframework/platform/internal/layout" |
| 15 | "github.com/carlosframework/platform/internal/registry" |
| 16 | ) |
| 17 | |
| 18 | type fakeKeyBackfiller struct { |
| 19 | got map[string]activator.InstanceKeyOutcome |
| 20 | err map[string]error |
| 21 | asked []string |
| 22 | } |
| 23 | |
| 24 | func (f *fakeKeyBackfiller) BackfillInstanceKey(_ context.Context, rt registry.Route) (activator.InstanceKeyOutcome, error) { |
| 25 | f.asked = append(f.asked, rt.Host) |
| 26 | return f.got[rt.Host], f.err[rt.Host] |
| 27 | } |
| 28 | |
| 29 | func TestBackfillInstanceKeysWritesOnlyReadyActivatorInstances(t *testing.T) { |
| 30 | ready := func(host string) registry.Route { |
| 31 | return registry.Route{Host: host, Kind: "instance", App: "shop", Status: "ready", Socket: "/run/carlos/" + host + ".sock"} |
| 32 | } |
| 33 | held := ready("moving.example.com") |
| 34 | held.Status = "moving" |
| 35 | landing := ready("landing.example.com") |
| 36 | landing.Status = "landing" |
| 37 | provisioning := ready("new.example.com") |
| 38 | provisioning.Status = "provisioning" |
| 39 | unit := ready("unit.example.com") |
| 40 | unit.Backing = registry.BackingUnit |
| 41 | alias := ready("alias.example.com") |
| 42 | alias.Platform = true |
| 43 | sidecar := ready("worker.example.com") |
| 44 | sidecar.Kind = "sidecar" |
| 45 | addr := ready("addr.example.com") |
| 46 | addr.Socket, addr.Addr = "", "127.0.0.1:8444" |
| 47 | replica := ready("shop-au.example.com") |
| 48 | replica.Replica = true |
| 49 | f := &fakeKeyBackfiller{ |
| 50 | got: map[string]activator.InstanceKeyOutcome{"shop.example.com": activator.InstanceKeyMinted, "old.example.com": activator.InstanceKeyPresent, "cfg.example.com": activator.InstanceKeyFromConfig}, |
| 51 | err: map[string]error{"bad.example.com": errors.New("read: permission denied")}, |
| 52 | } |
| 53 | st := backfillInstanceKeys(t.Context(), []registry.Route{ready("shop.example.com"), ready("old.example.com"), ready("cfg.example.com"), ready("bad.example.com"), held, landing, provisioning, unit, alias, sidecar, addr, replica}, f) |
| 54 | if len(f.asked) != 4 { |
| 55 | t.Fatalf("asked %v, want only the four ready activator instances", f.asked) |
| 56 | } |
| 57 | if st != (instanceKeyBackfill{Minted: 1, Present: 1, FromConfig: 1, Failed: 1, Skipped: 8}) { |
| 58 | t.Fatalf("stats = %+v", st) |
| 59 | } |
| 60 | } |
| 61 | |
| 62 | // A refusal (a fenced or conflicting row) answers InstanceKeyPresent |
| 63 | // together with its error (activator.BackfillInstanceKey). The error |
| 64 | // decides: the row is a failure, never "present", or the rollout check |
| 65 | // would read a host whose key was never written as ready. |
| 66 | func TestBackfillInstanceKeysCountsARefusalAsFailedNotPresent(t *testing.T) { |
| 67 | rt := registry.Route{Host: "shop.example.com", Kind: "instance", Status: "ready", Socket: "/run/carlos/shop.example.com.sock"} |
| 68 | f := &fakeKeyBackfiller{ |
| 69 | got: map[string]activator.InstanceKeyOutcome{rt.Host: activator.InstanceKeyPresent}, |
| 70 | err: map[string]error{rt.Host: errors.New("not backfilled: layout migration in flight")}, |
| 71 | } |
| 72 | st := backfillInstanceKeys(t.Context(), []registry.Route{rt}, f) |
| 73 | if st != (instanceKeyBackfill{Failed: 1}) { |
| 74 | t.Fatalf("stats = %+v, want the refusal counted as failed only", st) |
| 75 | } |
| 76 | } |
| 77 | |
| 78 | // A host with no own env file has never been started on this box: nothing |
| 79 | // is created, and the pass counts it on its own line. |
| 80 | func TestBackfillInstanceKeysCountsAHostNeverStartedHere(t *testing.T) { |
| 81 | rt := registry.Route{Host: "shop.example.com", Kind: "instance", Status: "ready", Socket: "/run/carlos/shop.example.com.sock"} |
| 82 | f := &fakeKeyBackfiller{got: map[string]activator.InstanceKeyOutcome{rt.Host: activator.InstanceKeyNoEnvFile}} |
| 83 | st := backfillInstanceKeys(t.Context(), []registry.Route{rt}, f) |
| 84 | if st != (instanceKeyBackfill{NoEnvFile: 1}) { |
| 85 | t.Fatalf("stats = %+v", st) |
| 86 | } |
| 87 | } |
| 88 | |
| 89 | // A row whose configuration is on its way (activator.ErrInstanceKeyDeferred) |
| 90 | // is counted apart from a failure: not minted by design, and found at its |
| 91 | // next start. The whole pass runs under one budget, so a bucket that is |
| 92 | // down costs the pass at most instanceKeyBackfillBudget, not one read's |
| 93 | // timeout per row. |
| 94 | func TestBackfillInstanceKeysCountsADeferredRowApartWithinOneBudget(t *testing.T) { |
| 95 | ready := func(host string) registry.Route { |
| 96 | return registry.Route{Host: host, Kind: "instance", Status: "ready", Socket: "/run/carlos/" + host + ".sock"} |
| 97 | } |
| 98 | var deadlines []time.Time |
| 99 | f := keyBackfillFunc(func(ctx context.Context, rt registry.Route) (activator.InstanceKeyOutcome, error) { |
| 100 | d, ok := ctx.Deadline() |
| 101 | if !ok { |
| 102 | t.Fatalf("%s: the backfill's context has no deadline", rt.Host) |
| 103 | } |
| 104 | deadlines = append(deadlines, d) |
| 105 | if rt.Host == "waiting.example.com" { |
| 106 | return activator.InstanceKeyFailed, fmt.Errorf("%w: its configuration sets CARLOS_INSTANCE_KEY", activator.ErrInstanceKeyDeferred) |
| 107 | } |
| 108 | return activator.InstanceKeyMinted, nil |
| 109 | }) |
| 110 | st := backfillInstanceKeys(context.Background(), []registry.Route{ready("waiting.example.com"), ready("shop.example.com")}, f) |
| 111 | if st != (instanceKeyBackfill{Minted: 1, Deferred: 1}) { |
| 112 | t.Fatalf("stats = %+v, want one minted and one waiting", st) |
| 113 | } |
| 114 | if len(deadlines) != 2 || !deadlines[0].Equal(deadlines[1]) || time.Until(deadlines[0]) > instanceKeyBackfillBudget { |
| 115 | t.Fatalf("deadlines %v, want one budget shared by both rows", deadlines) |
| 116 | } |
| 117 | } |
| 118 | |
| 119 | // keyBackfillFunc is a backfiller made of one function. |
| 120 | type keyBackfillFunc func(context.Context, registry.Route) (activator.InstanceKeyOutcome, error) |
| 121 | |
| 122 | func (f keyBackfillFunc) BackfillInstanceKey(ctx context.Context, rt registry.Route) (activator.InstanceKeyOutcome, error) { |
| 123 | return f(ctx, rt) |
| 124 | } |
| 125 | |
| 126 | // openSlot is a preparation slot with no fence held: it runs prepare. |
| 127 | func openSlot(_ string, prepare func() (string, error)) (string, error) { return prepare() } |
| 128 | |
| 129 | // routeOf answers the registry's Get from a fixed list. |
| 130 | func routeOf(routes ...registry.Route) func(string) (*registry.Route, error) { |
| 131 | return func(host string) (*registry.Route, error) { |
| 132 | for _, rt := range routes { |
| 133 | if rt.Host == host { |
| 134 | return &rt, nil |
| 135 | } |
| 136 | } |
| 137 | return nil, nil |
| 138 | } |
| 139 | } |
| 140 | |
| 141 | // Task 6 ruling (K1), as the v0.26.1 hotfix changed it: the start sequence |
| 142 | // is activatorStart's, so the two guards runAgent cannot be tested for are |
| 143 | // pinned here. A claim box starts and nothing else: no sweep, no recovery, |
| 144 | // no backfill (ruling 7). A legacy box sweeps, recovers, publishes the |
| 145 | // pass and starts, in that order, and the backfill runs after Start, once |
| 146 | // every journal is settled. A route list that fails skips the backfill |
| 147 | // only. |
| 148 | func TestActivatorStartBackfillsOnlyALegacyBoxAfterRecoveryAndStart(t *testing.T) { |
| 149 | ready := registry.Route{Host: "shop.example.com", Kind: "instance", Status: "ready", Socket: "/run/carlos/shop.example.com.sock"} |
| 150 | for _, tc := range []struct { |
| 151 | name string |
| 152 | claims bool |
| 153 | listErr error |
| 154 | want []string |
| 155 | }{ |
| 156 | {"legacy box", false, nil, []string{"configure", "sweep", "recover", "publish", "start", "ready", "backfill"}}, |
| 157 | {"legacy box, routes unreadable", false, errors.New("db locked"), []string{"configure", "sweep", "recover", "publish", "start", "ready"}}, |
| 158 | {"claim box", true, nil, []string{"configure", "start", "ready"}}, |
| 159 | } { |
| 160 | t.Run(tc.name, func(t *testing.T) { |
| 161 | var mu sync.Mutex |
| 162 | var events []string |
| 163 | note := func(e string) { mu.Lock(); events = append(events, e); mu.Unlock() } |
| 164 | done := activatorStart{ |
| 165 | Configure: func(context.Context) { note("configure") }, |
| 166 | Ready: func() { note("ready") }, |
| 167 | Claims: tc.claims, |
| 168 | Sweep: func(context.Context) error { note("sweep"); return nil }, |
| 169 | Recover: func(context.Context, bool) *layout.Pass { note("recover"); return nil }, |
| 170 | Routes: func() ([]registry.Route, error) { return []registry.Route{ready}, tc.listErr }, |
| 171 | Route: routeOf(ready), |
| 172 | Keys: keyBackfillFunc(func(context.Context, registry.Route) (activator.InstanceKeyOutcome, error) { |
| 173 | note("backfill") |
| 174 | return activator.InstanceKeyMinted, nil |
| 175 | }), |
| 176 | Slot: openSlot, |
| 177 | Publish: func(*layout.Pass) { note("publish") }, |
| 178 | Start: func() { note("start") }, |
| 179 | }.Run(context.Background()) |
| 180 | if tc.claims { |
| 181 | select { |
| 182 | case <-done: |
| 183 | default: |
| 184 | t.Fatal("claim-box startup became asynchronous") |
| 185 | } |
| 186 | } |
| 187 | select { |
| 188 | case <-done: |
| 189 | case <-time.After(10 * time.Second): |
| 190 | t.Fatal("the backfill never finished") |
| 191 | } |
| 192 | mu.Lock() |
| 193 | defer mu.Unlock() |
| 194 | if !reflect.DeepEqual(events, tc.want) { |
| 195 | t.Fatalf("events = %v, want %v", events, tc.want) |
| 196 | } |
| 197 | }) |
| 198 | } |
| 199 | } |
| 200 | |
| 201 | // fenceCtx is a context with the deadline FenceAndStop requires. |
| 202 | func fenceCtx(t *testing.T, d time.Duration) context.Context { |
| 203 | ctx, cancel := context.WithTimeout(t.Context(), d) |
| 204 | t.Cleanup(cancel) |
| 205 | return ctx |
| 206 | } |
| 207 | |
| 208 | // Off the startup path, the backfill shares the box with the layout pass |
| 209 | // and moves, which fence a host before they move its files. A host that |
| 210 | // is fenced when its turn comes is skipped and nothing is asked of it, and |
| 211 | // a row whose status changed after the list (it began a move) is skipped |
| 212 | // the same way; its next start mints the key. The rest are written from |
| 213 | // the row as it stands, not as it was listed. |
| 214 | func TestBackfillInSlotSkipsAFencedHostAndReadsTheRowAgain(t *testing.T) { |
| 215 | ready := func(host string) registry.Route { |
| 216 | return registry.Route{Host: host, Kind: "instance", Status: "ready", Socket: "/run/carlos/" + host + ".sock"} |
| 217 | } |
| 218 | fenced, moved, shop := ready("fenced.example.com"), ready("moved.example.com"), ready("shop.example.com") |
| 219 | svc := &activator.Service{} |
| 220 | if _, err := svc.FenceAndStop(fenceCtx(t, time.Second), fenced.Host, 1); err != nil && !errors.Is(err, activator.ErrStopUnproven) { |
| 221 | t.Fatalf("fence: %v", err) |
| 222 | } |
| 223 | nowMoving := moved |
| 224 | nowMoving.Status = "moving" |
| 225 | nowShop := shop |
| 226 | nowShop.Socket = "/run/carlos/nested/shop.example.com.sock" |
| 227 | f := &fakeKeyBackfiller{got: map[string]activator.InstanceKeyOutcome{shop.Host: activator.InstanceKeyMinted}} |
| 228 | var sockets []string |
| 229 | g := keyBackfillInSlot{ |
| 230 | keys: keyBackfillFunc(func(ctx context.Context, rt registry.Route) (activator.InstanceKeyOutcome, error) { |
| 231 | sockets = append(sockets, rt.Socket) |
| 232 | return f.BackfillInstanceKey(ctx, rt) |
| 233 | }), |
| 234 | slot: svc.PrepareScheduledToken, |
| 235 | route: routeOf(fenced, nowMoving, nowShop), |
| 236 | } |
| 237 | st := backfillInstanceKeys(t.Context(), []registry.Route{fenced, moved, shop}, g) |
| 238 | if !reflect.DeepEqual(f.asked, []string{shop.Host}) { |
| 239 | t.Fatalf("asked %v, want only %s", f.asked, shop.Host) |
| 240 | } |
| 241 | if !reflect.DeepEqual(sockets, []string{nowShop.Socket}) { |
| 242 | t.Fatalf("written from %v, want the row as it stands now", sockets) |
| 243 | } |
| 244 | if st != (instanceKeyBackfill{Minted: 1, Skipped: 2}) { |
| 245 | t.Fatalf("stats = %+v, want one minted and two skipped", st) |
| 246 | } |
| 247 | } |
| 248 | |
| 249 | // The other half of "nothing moves a file under it": a fence asked for |
| 250 | // while a row is being written (the layout pass or a move about to move |
| 251 | // the host's files) is installed only once the write has returned. |
| 252 | func TestBackfillInSlotHoldsOffAFenceUntilTheWriteReturns(t *testing.T) { |
| 253 | t.Parallel() |
| 254 | rt := registry.Route{Host: "shop.example.com", Kind: "instance", Status: "ready", Socket: "/run/carlos/shop.example.com.sock"} |
| 255 | svc := &activator.Service{} |
| 256 | writing, release := make(chan struct{}), make(chan struct{}) |
| 257 | var wrote atomic.Bool |
| 258 | g := keyBackfillInSlot{ |
| 259 | keys: keyBackfillFunc(func(context.Context, registry.Route) (activator.InstanceKeyOutcome, error) { |
| 260 | close(writing) |
| 261 | <-release |
| 262 | wrote.Store(true) |
| 263 | return activator.InstanceKeyMinted, nil |
| 264 | }), |
| 265 | slot: svc.PrepareScheduledToken, |
| 266 | route: routeOf(rt), |
| 267 | } |
| 268 | done := make(chan instanceKeyBackfill) |
| 269 | go func() { done <- backfillInstanceKeys(t.Context(), []registry.Route{rt}, g) }() |
| 270 | <-writing |
| 271 | fenced := make(chan bool) |
| 272 | go func() { |
| 273 | _, err := svc.FenceAndStopIfUnfenced(fenceCtx(t, 10*time.Second), rt.Host, 1) |
| 274 | if err != nil && !errors.Is(err, activator.ErrStopUnproven) { |
| 275 | t.Errorf("fence: %v", err) |
| 276 | } |
| 277 | fenced <- wrote.Load() |
| 278 | }() |
| 279 | select { |
| 280 | case <-fenced: |
| 281 | t.Fatal("the fence was installed while the backfill was writing the host") |
| 282 | case <-time.After(200 * time.Millisecond): |
| 283 | } |
| 284 | close(release) |
| 285 | if !<-fenced { |
| 286 | t.Fatal("the fence was installed before the write returned") |
| 287 | } |
| 288 | if st := <-done; st != (instanceKeyBackfill{Minted: 1}) { |
| 289 | t.Fatalf("stats = %+v", st) |
| 290 | } |
| 291 | } |
| 292 | |
| 293 | // v0.26.1 hotfix: the v0.26.0 roll to a 127-instance box failed its update |
| 294 | // gate because the backfill ran serially before Start, one bucket read per |
| 295 | // keyless row, and the edge never listened inside the gate's 20 seconds. |
| 296 | // On a box with hundreds of keyless rows and a slow bucket, Start is |
| 297 | // called, and Run returns (runAgent goes on to the agent socket and the |
| 298 | // edge), without waiting for the backfill. |
| 299 | func TestActivatorStartDoesNotWaitForTheBackfillOnALargeBox(t *testing.T) { |
| 300 | const rows, perRow, bound = 300, 30 * time.Millisecond, 2 * time.Second |
| 301 | var routes []registry.Route |
| 302 | for i := range rows { |
| 303 | host := fmt.Sprintf("shop%03d.example.com", i) |
| 304 | routes = append(routes, registry.Route{Host: host, Kind: "instance", Status: "ready", Socket: "/run/carlos/" + host + ".sock"}) |
| 305 | } |
| 306 | ctx, cancel := context.WithCancel(context.Background()) |
| 307 | defer cancel() |
| 308 | slowBucket := keyBackfillFunc(func(ctx context.Context, rt registry.Route) (activator.InstanceKeyOutcome, error) { |
| 309 | select { |
| 310 | case <-time.After(perRow): |
| 311 | return activator.InstanceKeyMinted, nil |
| 312 | case <-ctx.Done(): |
| 313 | return activator.InstanceKeyFailed, fmt.Errorf("%w: %w", activator.ErrInstanceKeyDeferred, ctx.Err()) |
| 314 | } |
| 315 | }) |
| 316 | began := time.Now() |
| 317 | started := make(chan time.Duration, 1) |
| 318 | done := activatorStart{ |
| 319 | Sweep: func(context.Context) error { return nil }, |
| 320 | Recover: func(context.Context, bool) *layout.Pass { return nil }, |
| 321 | Routes: func() ([]registry.Route, error) { return routes, nil }, |
| 322 | Route: routeOf(routes...), |
| 323 | Keys: slowBucket, |
| 324 | Slot: (&activator.Service{}).PrepareScheduledToken, |
| 325 | Publish: func(*layout.Pass) {}, |
| 326 | Start: func() { started <- time.Since(began) }, |
| 327 | }.Run(ctx) |
| 328 | returned := time.Since(began) |
| 329 | select { |
| 330 | case startedAfter := <-started: |
| 331 | if startedAfter > bound { |
| 332 | t.Errorf("Start took %v, want within %v", startedAfter, bound) |
| 333 | } |
| 334 | case <-time.After(bound): |
| 335 | t.Fatal("Start waited for the backfill") |
| 336 | } |
| 337 | if returned > bound { |
| 338 | t.Errorf("Run returned after %v, want within %v: the edge waits for Run", returned, bound) |
| 339 | } |
| 340 | select { |
| 341 | case <-done: |
| 342 | t.Fatal("the backfill had finished when Run returned; it should still be running") |
| 343 | default: |
| 344 | } |
| 345 | // Agent shutdown is not this test's: cancel the pass and let it end. |
| 346 | cancel() |
| 347 | select { |
| 348 | case <-done: |
| 349 | case <-time.After(30 * time.Second): |
| 350 | t.Fatal("the backfill did not end once its context was cancelled") |
| 351 | } |
| 352 | } |
| 353 | |
amadan