| 1 | package steering |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "fmt" |
| 7 | "reflect" |
| 8 | "sort" |
| 9 | "strings" |
| 10 | "testing" |
| 11 | |
| 12 | "github.com/aws/aws-sdk-go-v2/aws" |
| 13 | "github.com/aws/aws-sdk-go-v2/service/route53" |
| 14 | r53types "github.com/aws/aws-sdk-go-v2/service/route53/types" |
| 15 | ) |
| 16 | |
| 17 | // --------------------------------------------------------------------------- |
| 18 | // the fake |
| 19 | // --------------------------------------------------------------------------- |
| 20 | |
| 21 | // fakeR53 is an in-memory Route53 that records every call and APPLIES the |
| 22 | // changes it is handed, so a second converge against the same fake sees the |
| 23 | // state the first one left. No network, ever — the whole point of route53API |
| 24 | // being an interface. |
| 25 | // |
| 26 | // It deliberately reproduces the real API's refusals that matter to this |
| 27 | // converger: DELETE of a record that is not present errors (Route53's |
| 28 | // InvalidChangeBatch), a name+type may not hold BOTH a plain record and a |
| 29 | // routing-policy record set (also InvalidChangeBatch, and it takes the whole |
| 30 | // atomic batch with it), ListTagsForResources refuses more than 10 ids, and |
| 31 | // DeleteHealthCheck of a check still referenced by a record errors |
| 32 | // (HealthCheckInUse). A fake that is more permissive than the service would |
| 33 | // let a bug through — the plain-vs-policy rule is in here precisely because |
| 34 | // an earlier version of this file was more permissive and enshrined a zone |
| 35 | // state Route53 forbids. |
| 36 | type fakeR53 struct { |
| 37 | records []r53types.ResourceRecordSet |
| 38 | checks []r53types.HealthCheck |
| 39 | tags map[string][]r53types.Tag // health check id -> tags |
| 40 | |
| 41 | pageSize int // 0 = one page for everything |
| 42 | nextID int |
| 43 | |
| 44 | // call counters — the idempotency assertions live on these |
| 45 | nListRecords int |
| 46 | nChangeRecords int |
| 47 | nListChecks int |
| 48 | nListTags int |
| 49 | nCreateCheck int |
| 50 | nDeleteCheck int |
| 51 | nChangeTags int |
| 52 | |
| 53 | submitted []r53types.Change // every change submitted, in order |
| 54 | createdChecks []*route53.CreateHealthCheckInput |
| 55 | deletedChecks []string |
| 56 | taggedIDs []string |
| 57 | |
| 58 | failChange error // ChangeResourceRecordSets returns this |
| 59 | failTags error // ChangeTagsForResource returns this |
| 60 | } |
| 61 | |
| 62 | func newFake() *fakeR53 { return &fakeR53{tags: map[string][]r53types.Tag{}} } |
| 63 | |
| 64 | // writes is the total number of state-changing calls — the number an |
| 65 | // idempotent second pass must leave at zero. |
| 66 | func (f *fakeR53) writes() int { |
| 67 | return f.nChangeRecords + f.nCreateCheck + f.nDeleteCheck + f.nChangeTags |
| 68 | } |
| 69 | |
| 70 | // validateZone enforces the three Route53 rules this converger can violate, |
| 71 | // each of which rejects the WHOLE batch: |
| 72 | // |
| 73 | // 1. a name+type holds either one simple record OR a routing-policy record |
| 74 | // set, never both; |
| 75 | // 2. a name holding a CNAME holds no other record type; |
| 76 | // 3. one routing policy per name+type — a weighted set and a latency set |
| 77 | // cannot share a name. |
| 78 | // |
| 79 | // Modelled on the projected post-batch state, so a batch is judged by what it |
| 80 | // would produce, atomically. |
| 81 | func validateZone(recs []r53types.ResourceRecordSet) error { |
| 82 | type nt struct{ name, typ string } |
| 83 | plain, policy := map[nt]bool{}, map[nt]bool{} |
| 84 | policies := map[nt]map[string]bool{} |
| 85 | names := map[string]map[string]bool{} // name -> set of types |
| 86 | for _, rr := range recs { |
| 87 | name := strings.ToLower(aws.ToString(rr.Name)) |
| 88 | k := nt{name, string(rr.Type)} |
| 89 | if names[name] == nil { |
| 90 | names[name] = map[string]bool{} |
| 91 | } |
| 92 | names[name][string(rr.Type)] = true |
| 93 | if aws.ToString(rr.SetIdentifier) == "" { |
| 94 | plain[k] = true |
| 95 | continue |
| 96 | } |
| 97 | policy[k] = true |
| 98 | if policies[k] == nil { |
| 99 | policies[k] = map[string]bool{} |
| 100 | } |
| 101 | policies[k][fakePolicyKind(rr)] = true |
| 102 | } |
| 103 | for k := range plain { |
| 104 | if policy[k] { |
| 105 | return fmt.Errorf("fake: InvalidChangeBatch: RRSet with DNS name %s and type %s cannot be both a simple record and a routing-policy record set", k.name, k.typ) |
| 106 | } |
| 107 | } |
| 108 | for name, types := range names { |
| 109 | if types["CNAME"] && len(types) > 1 { |
| 110 | return fmt.Errorf("fake: InvalidChangeBatch: RRSet with DNS name %s has a CNAME, which cannot coexist with any other record type", name) |
| 111 | } |
| 112 | } |
| 113 | for k, kinds := range policies { |
| 114 | if len(kinds) > 1 { |
| 115 | return fmt.Errorf("fake: InvalidChangeBatch: RRSet with DNS name %s and type %s mixes routing policies (%v) — only one policy per name and type", k.name, k.typ, sortedKeys(kinds)) |
| 116 | } |
| 117 | } |
| 118 | return nil |
| 119 | } |
| 120 | |
| 121 | func fakePolicyKind(rr r53types.ResourceRecordSet) string { |
| 122 | switch { |
| 123 | case rr.Region != "": |
| 124 | return "latency" |
| 125 | case rr.Weight != nil: |
| 126 | return "weighted" |
| 127 | case rr.Failover != "": |
| 128 | return "failover" |
| 129 | case rr.GeoLocation != nil: |
| 130 | return "geolocation" |
| 131 | case rr.GeoProximityLocation != nil: |
| 132 | return "geoproximity" |
| 133 | case rr.CidrRoutingConfig != nil: |
| 134 | return "cidr" |
| 135 | case aws.ToBool(rr.MultiValueAnswer): |
| 136 | return "multivalue" |
| 137 | } |
| 138 | return "unknown" |
| 139 | } |
| 140 | |
| 141 | func sortedKeys(m map[string]bool) []string { |
| 142 | out := make([]string, 0, len(m)) |
| 143 | for k := range m { |
| 144 | out = append(out, k) |
| 145 | } |
| 146 | sort.Strings(out) |
| 147 | return out |
| 148 | } |
| 149 | |
| 150 | func rrKey(rr r53types.ResourceRecordSet) string { |
| 151 | return strings.ToLower(aws.ToString(rr.Name)) + "|" + string(rr.Type) + "|" + aws.ToString(rr.SetIdentifier) |
| 152 | } |
| 153 | |
| 154 | func (f *fakeR53) ListResourceRecordSets(ctx context.Context, in *route53.ListResourceRecordSetsInput, _ ...func(*route53.Options)) (*route53.ListResourceRecordSetsOutput, error) { |
| 155 | f.nListRecords++ |
| 156 | if aws.ToString(in.HostedZoneId) == "" { |
| 157 | return nil, errors.New("fake: no hosted zone id") |
| 158 | } |
| 159 | start := 0 |
| 160 | if in.StartRecordName != nil { |
| 161 | for i, rr := range f.records { |
| 162 | if strings.EqualFold(aws.ToString(rr.Name), aws.ToString(in.StartRecordName)) && |
| 163 | rr.Type == in.StartRecordType && |
| 164 | aws.ToString(rr.SetIdentifier) == aws.ToString(in.StartRecordIdentifier) { |
| 165 | start = i |
| 166 | break |
| 167 | } |
| 168 | } |
| 169 | } |
| 170 | end := len(f.records) |
| 171 | if f.pageSize > 0 && start+f.pageSize < end { |
| 172 | end = start + f.pageSize |
| 173 | } |
| 174 | out := &route53.ListResourceRecordSetsOutput{ |
| 175 | ResourceRecordSets: append([]r53types.ResourceRecordSet(nil), f.records[start:end]...), |
| 176 | } |
| 177 | if end < len(f.records) { |
| 178 | next := f.records[end] |
| 179 | out.IsTruncated = true |
| 180 | out.NextRecordName = next.Name |
| 181 | out.NextRecordType = next.Type |
| 182 | out.NextRecordIdentifier = next.SetIdentifier |
| 183 | } |
| 184 | return out, nil |
| 185 | } |
| 186 | |
| 187 | func (f *fakeR53) ChangeResourceRecordSets(ctx context.Context, in *route53.ChangeResourceRecordSetsInput, _ ...func(*route53.Options)) (*route53.ChangeResourceRecordSetsOutput, error) { |
| 188 | f.nChangeRecords++ |
| 189 | if f.failChange != nil { |
| 190 | return nil, f.failChange |
| 191 | } |
| 192 | if in.ChangeBatch == nil || len(in.ChangeBatch.Changes) == 0 { |
| 193 | return nil, errors.New("fake: empty change batch") |
| 194 | } |
| 195 | |
| 196 | // A batch is ATOMIC: project it onto a copy, validate the result, and |
| 197 | // commit only if the whole thing is legal. |
| 198 | projected := append([]r53types.ResourceRecordSet(nil), f.records...) |
| 199 | for _, c := range in.ChangeBatch.Changes { |
| 200 | rr := *c.ResourceRecordSet |
| 201 | key := rrKey(rr) |
| 202 | idx := -1 |
| 203 | for i, have := range projected { |
| 204 | if rrKey(have) == key { |
| 205 | idx = i |
| 206 | break |
| 207 | } |
| 208 | } |
| 209 | switch c.Action { |
| 210 | case r53types.ChangeActionUpsert: |
| 211 | if idx >= 0 { |
| 212 | projected[idx] = rr |
| 213 | } else { |
| 214 | projected = append(projected, rr) |
| 215 | } |
| 216 | case r53types.ChangeActionDelete: |
| 217 | if idx < 0 { |
| 218 | return nil, fmt.Errorf("fake: InvalidChangeBatch: no such record %s", key) |
| 219 | } |
| 220 | projected = append(projected[:idx], projected[idx+1:]...) |
| 221 | default: |
| 222 | return nil, fmt.Errorf("fake: unsupported action %q", c.Action) |
| 223 | } |
| 224 | } |
| 225 | |
| 226 | if err := validateZone(projected); err != nil { |
| 227 | return nil, err |
| 228 | } |
| 229 | |
| 230 | f.submitted = append(f.submitted, in.ChangeBatch.Changes...) |
| 231 | f.records = projected |
| 232 | sort.SliceStable(f.records, func(i, j int) bool { return rrKey(f.records[i]) < rrKey(f.records[j]) }) |
| 233 | return &route53.ChangeResourceRecordSetsOutput{}, nil |
| 234 | } |
| 235 | |
| 236 | func (f *fakeR53) ListHealthChecks(ctx context.Context, in *route53.ListHealthChecksInput, _ ...func(*route53.Options)) (*route53.ListHealthChecksOutput, error) { |
| 237 | f.nListChecks++ |
| 238 | start := 0 |
| 239 | if in.Marker != nil { |
| 240 | for i, hc := range f.checks { |
| 241 | if aws.ToString(hc.Id) == aws.ToString(in.Marker) { |
| 242 | start = i |
| 243 | break |
| 244 | } |
| 245 | } |
| 246 | } |
| 247 | end := len(f.checks) |
| 248 | if f.pageSize > 0 && start+f.pageSize < end { |
| 249 | end = start + f.pageSize |
| 250 | } |
| 251 | out := &route53.ListHealthChecksOutput{ |
| 252 | HealthChecks: append([]r53types.HealthCheck(nil), f.checks[start:end]...), |
| 253 | } |
| 254 | if end < len(f.checks) { |
| 255 | out.IsTruncated = true |
| 256 | out.NextMarker = f.checks[end].Id |
| 257 | } |
| 258 | return out, nil |
| 259 | } |
| 260 | |
| 261 | func (f *fakeR53) ListTagsForResources(ctx context.Context, in *route53.ListTagsForResourcesInput, _ ...func(*route53.Options)) (*route53.ListTagsForResourcesOutput, error) { |
| 262 | f.nListTags++ |
| 263 | if in.ResourceType != r53types.TagResourceTypeHealthcheck { |
| 264 | return nil, fmt.Errorf("fake: unexpected resource type %q", in.ResourceType) |
| 265 | } |
| 266 | if len(in.ResourceIds) == 0 || len(in.ResourceIds) > 10 { |
| 267 | return nil, fmt.Errorf("fake: ListTagsForResources takes 1-10 ids, got %d", len(in.ResourceIds)) |
| 268 | } |
| 269 | out := &route53.ListTagsForResourcesOutput{} |
| 270 | for _, id := range in.ResourceIds { |
| 271 | out.ResourceTagSets = append(out.ResourceTagSets, r53types.ResourceTagSet{ |
| 272 | ResourceId: aws.String(id), |
| 273 | ResourceType: r53types.TagResourceTypeHealthcheck, |
| 274 | Tags: f.tags[id], |
| 275 | }) |
| 276 | } |
| 277 | return out, nil |
| 278 | } |
| 279 | |
| 280 | func (f *fakeR53) CreateHealthCheck(ctx context.Context, in *route53.CreateHealthCheckInput, _ ...func(*route53.Options)) (*route53.CreateHealthCheckOutput, error) { |
| 281 | f.nCreateCheck++ |
| 282 | f.createdChecks = append(f.createdChecks, in) |
| 283 | f.nextID++ |
| 284 | id := fmt.Sprintf("hc-%d", f.nextID) |
| 285 | hc := r53types.HealthCheck{ |
| 286 | Id: aws.String(id), |
| 287 | CallerReference: in.CallerReference, |
| 288 | HealthCheckConfig: in.HealthCheckConfig, |
| 289 | } |
| 290 | f.checks = append(f.checks, hc) |
| 291 | return &route53.CreateHealthCheckOutput{HealthCheck: &hc, Location: aws.String("https://route53/" + id)}, nil |
| 292 | } |
| 293 | |
| 294 | func (f *fakeR53) DeleteHealthCheck(ctx context.Context, in *route53.DeleteHealthCheckInput, _ ...func(*route53.Options)) (*route53.DeleteHealthCheckOutput, error) { |
| 295 | f.nDeleteCheck++ |
| 296 | id := aws.ToString(in.HealthCheckId) |
| 297 | for _, rr := range f.records { |
| 298 | if aws.ToString(rr.HealthCheckId) == id { |
| 299 | return nil, fmt.Errorf("fake: HealthCheckInUse: %s", id) |
| 300 | } |
| 301 | } |
| 302 | f.deletedChecks = append(f.deletedChecks, id) |
| 303 | for i, hc := range f.checks { |
| 304 | if aws.ToString(hc.Id) == id { |
| 305 | f.checks = append(f.checks[:i], f.checks[i+1:]...) |
| 306 | break |
| 307 | } |
| 308 | } |
| 309 | delete(f.tags, id) |
| 310 | return &route53.DeleteHealthCheckOutput{}, nil |
| 311 | } |
| 312 | |
| 313 | func (f *fakeR53) ChangeTagsForResource(ctx context.Context, in *route53.ChangeTagsForResourceInput, _ ...func(*route53.Options)) (*route53.ChangeTagsForResourceOutput, error) { |
| 314 | f.nChangeTags++ |
| 315 | if f.failTags != nil { |
| 316 | return nil, f.failTags |
| 317 | } |
| 318 | id := aws.ToString(in.ResourceId) |
| 319 | f.taggedIDs = append(f.taggedIDs, id) |
| 320 | f.tags[id] = append(f.tags[id], in.AddTags...) |
| 321 | return &route53.ChangeTagsForResourceOutput{}, nil |
| 322 | } |
| 323 | |
| 324 | // --------------------------------------------------------------------------- |
| 325 | // fixtures + helpers |
| 326 | // --------------------------------------------------------------------------- |
| 327 | |
| 328 | const testZone = "Z0STEERINGTEST" |
| 329 | |
| 330 | var ( |
| 331 | dublinRec = desiredRecord{Name: "edge.example.com.", EIP: "1.2.3.4", Box: "dublin", Region: "eu-west-1"} |
| 332 | oregonRec = desiredRecord{Name: "edge.example.com.", EIP: "5.6.7.8", Box: "oregon", Region: "us-west-2"} |
| 333 | ) |
| 334 | |
| 335 | // hostRecs mints the per-host half of a desired set: one record per edge. |
| 336 | func hostRecs(name string, edges ...desiredRecord) []desiredRecord { |
| 337 | out := make([]desiredRecord, 0, len(edges)) |
| 338 | for _, e := range edges { |
| 339 | out = append(out, desiredRecord{Name: name, EIP: e.EIP, Box: e.Box, Region: e.Region}) |
| 340 | } |
| 341 | return out |
| 342 | } |
| 343 | |
| 344 | func staticA(name, ip string) r53types.ResourceRecordSet { |
| 345 | return r53types.ResourceRecordSet{ |
| 346 | Name: aws.String(name), |
| 347 | Type: r53types.RRTypeA, |
| 348 | TTL: aws.Int64(300), |
| 349 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String(ip)}}, |
| 350 | } |
| 351 | } |
| 352 | |
| 353 | func recordNames(recs []r53types.ResourceRecordSet) []string { |
| 354 | out := make([]string, 0, len(recs)) |
| 355 | for _, rr := range recs { |
| 356 | out = append(out, rrKey(rr)) |
| 357 | } |
| 358 | sort.Strings(out) |
| 359 | return out |
| 360 | } |
| 361 | |
| 362 | func find(t *testing.T, f *fakeR53, name, setID string) r53types.ResourceRecordSet { |
| 363 | t.Helper() |
| 364 | for _, rr := range f.records { |
| 365 | if strings.EqualFold(aws.ToString(rr.Name), name) && aws.ToString(rr.SetIdentifier) == setID { |
| 366 | return rr |
| 367 | } |
| 368 | } |
| 369 | t.Fatalf("no record %s / setid %q in zone; have %v", name, setID, recordNames(f.records)) |
| 370 | return r53types.ResourceRecordSet{} |
| 371 | } |
| 372 | |
| 373 | func mustConverge(t *testing.T, f *fakeR53, want []desiredRecord) { |
| 374 | t.Helper() |
| 375 | if err := converge(context.Background(), f, testZone, want); err != nil { |
| 376 | t.Fatalf("converge: %v", err) |
| 377 | } |
| 378 | } |
| 379 | |
| 380 | // --------------------------------------------------------------------------- |
| 381 | // tests |
| 382 | // --------------------------------------------------------------------------- |
| 383 | |
| 384 | // TestConvergeCreateFromEmpty is the brief's create-from-empty case: an |
| 385 | // empty zone, two armed edges, one steered host. |
| 386 | func TestConvergeCreateFromEmpty(t *testing.T) { |
| 387 | f := newFake() |
| 388 | want := append([]desiredRecord{dublinRec, oregonRec}, hostRecs("app.example.com.", dublinRec, oregonRec)...) |
| 389 | mustConverge(t, f, want) |
| 390 | |
| 391 | if len(f.records) != 4 { |
| 392 | t.Fatalf("want 4 records, got %d: %v", len(f.records), recordNames(f.records)) |
| 393 | } |
| 394 | if f.nCreateCheck != 2 { |
| 395 | t.Fatalf("want one health check per distinct EIP (2), got %d", f.nCreateCheck) |
| 396 | } |
| 397 | if f.nChangeTags != 2 { |
| 398 | t.Fatalf("want every created check tagged (2), got %d", f.nChangeTags) |
| 399 | } |
| 400 | |
| 401 | // health check shape |
| 402 | byIP := map[string]*route53.CreateHealthCheckInput{} |
| 403 | for _, in := range f.createdChecks { |
| 404 | byIP[aws.ToString(in.HealthCheckConfig.IPAddress)] = in |
| 405 | } |
| 406 | for _, ip := range []string{"1.2.3.4", "5.6.7.8"} { |
| 407 | in, ok := byIP[ip] |
| 408 | if !ok { |
| 409 | t.Fatalf("no health check created for %s", ip) |
| 410 | } |
| 411 | cfg := in.HealthCheckConfig |
| 412 | if cfg.Type != r53types.HealthCheckTypeHttps { |
| 413 | t.Errorf("%s: type = %q, want HTTPS", ip, cfg.Type) |
| 414 | } |
| 415 | if aws.ToInt32(cfg.Port) != 443 { |
| 416 | t.Errorf("%s: port = %d, want 443", ip, aws.ToInt32(cfg.Port)) |
| 417 | } |
| 418 | if !aws.ToBool(cfg.EnableSNI) { |
| 419 | t.Errorf("%s: EnableSNI false, want true (the SNI name is the whole point)", ip) |
| 420 | } |
| 421 | // FQDN = the steering hostname, dot-trimmed (it is an SNI/Host value, |
| 422 | // not a Route53 record name). |
| 423 | if got := aws.ToString(cfg.FullyQualifiedDomainName); got != "edge.example.com" { |
| 424 | t.Errorf("%s: FQDN = %q, want the steering hostname edge.example.com", ip, got) |
| 425 | } |
| 426 | if !strings.HasPrefix(aws.ToString(in.CallerReference), "carlos-steering-") { |
| 427 | t.Errorf("%s: CallerReference = %q, want a carlos-steering- prefix", ip, aws.ToString(in.CallerReference)) |
| 428 | } |
| 429 | if n := len(aws.ToString(in.CallerReference)); n > 64 { |
| 430 | t.Errorf("%s: CallerReference is %d chars, Route53 caps it at 64", ip, n) |
| 431 | } |
| 432 | } |
| 433 | |
| 434 | // every created check carries the zone-scoped tag |
| 435 | for id, tags := range f.tags { |
| 436 | var found bool |
| 437 | for _, tg := range tags { |
| 438 | if aws.ToString(tg.Key) == steeringTagKey && aws.ToString(tg.Value) == testZone { |
| 439 | found = true |
| 440 | } |
| 441 | } |
| 442 | if !found { |
| 443 | t.Errorf("check %s not tagged %s=%s", id, steeringTagKey, testZone) |
| 444 | } |
| 445 | } |
| 446 | |
| 447 | // record shape |
| 448 | rr := find(t, f, "app.example.com.", "dublin") |
| 449 | if rr.Type != r53types.RRTypeA { |
| 450 | t.Errorf("type = %q, want A", rr.Type) |
| 451 | } |
| 452 | if aws.ToInt64(rr.TTL) != 60 { |
| 453 | t.Errorf("TTL = %d, want 60", aws.ToInt64(rr.TTL)) |
| 454 | } |
| 455 | if string(rr.Region) != "eu-west-1" { |
| 456 | t.Errorf("Region = %q, want eu-west-1", rr.Region) |
| 457 | } |
| 458 | if rr.AliasTarget != nil { |
| 459 | t.Errorf("AliasTarget set; these are plain latency records to EIPs") |
| 460 | } |
| 461 | if len(rr.ResourceRecords) != 1 || aws.ToString(rr.ResourceRecords[0].Value) != "1.2.3.4" { |
| 462 | t.Errorf("value = %v, want 1.2.3.4", rr.ResourceRecords) |
| 463 | } |
| 464 | if aws.ToString(rr.HealthCheckId) == "" { |
| 465 | t.Errorf("record not bound to a health check") |
| 466 | } |
| 467 | // the record for the same EIP under a different name shares the check |
| 468 | edgeRR := find(t, f, "edge.example.com.", "dublin") |
| 469 | if aws.ToString(edgeRR.HealthCheckId) != aws.ToString(rr.HealthCheckId) { |
| 470 | t.Errorf("one check per EIP violated: %q vs %q", aws.ToString(edgeRR.HealthCheckId), aws.ToString(rr.HealthCheckId)) |
| 471 | } |
| 472 | } |
| 473 | |
| 474 | // TestConvergeIdempotentZeroWrites is the brief's headline safety property: |
| 475 | // a second pass with an identical want makes NO write calls at all. |
| 476 | func TestConvergeIdempotentZeroWrites(t *testing.T) { |
| 477 | f := newFake() |
| 478 | want := append([]desiredRecord{dublinRec, oregonRec}, hostRecs("app.example.com.", dublinRec, oregonRec)...) |
| 479 | mustConverge(t, f, want) |
| 480 | |
| 481 | before := f.writes() |
| 482 | if before == 0 { |
| 483 | t.Fatalf("first pass made no writes; the test proves nothing") |
| 484 | } |
| 485 | mustConverge(t, f, want) |
| 486 | if got := f.writes() - before; got != 0 { |
| 487 | t.Fatalf("second pass made %d write calls, want 0 (records=%d creates=%d deletes=%d tags=%d)", |
| 488 | got, f.nChangeRecords, f.nCreateCheck, f.nDeleteCheck, f.nChangeTags) |
| 489 | } |
| 490 | // a third pass, for good measure — no slow drift |
| 491 | mustConverge(t, f, want) |
| 492 | if got := f.writes() - before; got != 0 { |
| 493 | t.Fatalf("third pass made %d write calls, want 0", got) |
| 494 | } |
| 495 | } |
| 496 | |
| 497 | // TestConvergeAddEdge: a second edge arms. Only the new records are written; |
| 498 | // the existing ones are left alone. |
| 499 | func TestConvergeAddEdge(t *testing.T) { |
| 500 | f := newFake() |
| 501 | one := append([]desiredRecord{dublinRec}, hostRecs("app.example.com.", dublinRec)...) |
| 502 | mustConverge(t, f, one) |
| 503 | base := f.writes() |
| 504 | f.submitted = nil |
| 505 | |
| 506 | two := append([]desiredRecord{dublinRec, oregonRec}, hostRecs("app.example.com.", dublinRec, oregonRec)...) |
| 507 | mustConverge(t, f, two) |
| 508 | |
| 509 | if f.nCreateCheck != 2 { |
| 510 | t.Fatalf("want exactly one NEW check (2 total), got %d total", f.nCreateCheck) |
| 511 | } |
| 512 | if f.nDeleteCheck != 0 { |
| 513 | t.Fatalf("adding an edge deleted %d checks", f.nDeleteCheck) |
| 514 | } |
| 515 | if len(f.submitted) != 2 { |
| 516 | t.Fatalf("want 2 changes (the new edge's records only), got %d: %v", len(f.submitted), f.submitted) |
| 517 | } |
| 518 | for _, c := range f.submitted { |
| 519 | if c.Action != r53types.ChangeActionUpsert { |
| 520 | t.Errorf("adding an edge submitted a %s", c.Action) |
| 521 | } |
| 522 | if aws.ToString(c.ResourceRecordSet.SetIdentifier) != "oregon" { |
| 523 | t.Errorf("rewrote an untouched record: %s", rrKey(*c.ResourceRecordSet)) |
| 524 | } |
| 525 | } |
| 526 | if len(f.records) != 4 { |
| 527 | t.Fatalf("want 4 records, got %v", recordNames(f.records)) |
| 528 | } |
| 529 | _ = base |
| 530 | } |
| 531 | |
| 532 | // TestConvergeRemoveEdge: an edge disarms. Its records go, and the health |
| 533 | // check it alone used is GC'd; the surviving edge's check is retained. |
| 534 | func TestConvergeRemoveEdge(t *testing.T) { |
| 535 | f := newFake() |
| 536 | two := append([]desiredRecord{dublinRec, oregonRec}, hostRecs("app.example.com.", dublinRec, oregonRec)...) |
| 537 | mustConverge(t, f, two) |
| 538 | oregonCheck := aws.ToString(find(t, f, "edge.example.com.", "oregon").HealthCheckId) |
| 539 | dublinCheck := aws.ToString(find(t, f, "edge.example.com.", "dublin").HealthCheckId) |
| 540 | f.submitted = nil |
| 541 | |
| 542 | one := append([]desiredRecord{dublinRec}, hostRecs("app.example.com.", dublinRec)...) |
| 543 | mustConverge(t, f, one) |
| 544 | |
| 545 | if len(f.records) != 2 { |
| 546 | t.Fatalf("want 2 records left, got %v", recordNames(f.records)) |
| 547 | } |
| 548 | for _, rr := range f.records { |
| 549 | if aws.ToString(rr.SetIdentifier) == "oregon" { |
| 550 | t.Errorf("oregon record survived: %s", rrKey(rr)) |
| 551 | } |
| 552 | } |
| 553 | for _, c := range f.submitted { |
| 554 | if c.Action != r53types.ChangeActionDelete { |
| 555 | t.Errorf("removing an edge submitted a %s for %s", c.Action, rrKey(*c.ResourceRecordSet)) |
| 556 | } |
| 557 | } |
| 558 | if len(f.deletedChecks) != 1 || f.deletedChecks[0] != oregonCheck { |
| 559 | t.Fatalf("want the orphaned check %s GC'd, got %v", oregonCheck, f.deletedChecks) |
| 560 | } |
| 561 | // the surviving check is untouched and still bound |
| 562 | if got := aws.ToString(find(t, f, "app.example.com.", "dublin").HealthCheckId); got != dublinCheck { |
| 563 | t.Errorf("dublin's check changed: %q -> %q", dublinCheck, got) |
| 564 | } |
| 565 | } |
| 566 | |
| 567 | // TestConvergeRemoveSteering: a host stops being steered. Its records go, |
| 568 | // but the checks stay — another host (and the steering hostname) still uses |
| 569 | // the same EIPs. |
| 570 | func TestConvergeRemoveSteering(t *testing.T) { |
| 571 | f := newFake() |
| 572 | both := append([]desiredRecord{dublinRec, oregonRec}, hostRecs("app.example.com.", dublinRec, oregonRec)...) |
| 573 | both = append(both, hostRecs("other.example.com.", dublinRec, oregonRec)...) |
| 574 | mustConverge(t, f, both) |
| 575 | if len(f.records) != 6 { |
| 576 | t.Fatalf("setup: want 6 records, got %v", recordNames(f.records)) |
| 577 | } |
| 578 | checksBefore := len(f.checks) |
| 579 | f.submitted = nil |
| 580 | |
| 581 | one := append([]desiredRecord{dublinRec, oregonRec}, hostRecs("other.example.com.", dublinRec, oregonRec)...) |
| 582 | mustConverge(t, f, one) |
| 583 | |
| 584 | if len(f.records) != 4 { |
| 585 | t.Fatalf("want 4 records left, got %v", recordNames(f.records)) |
| 586 | } |
| 587 | for _, rr := range f.records { |
| 588 | if strings.EqualFold(aws.ToString(rr.Name), "app.example.com.") { |
| 589 | t.Errorf("unsteered host's record survived: %s", rrKey(rr)) |
| 590 | } |
| 591 | } |
| 592 | if len(f.deletedChecks) != 0 { |
| 593 | t.Fatalf("checks still in use were GC'd: %v", f.deletedChecks) |
| 594 | } |
| 595 | if len(f.checks) != checksBefore { |
| 596 | t.Fatalf("check count changed %d -> %d", checksBefore, len(f.checks)) |
| 597 | } |
| 598 | } |
| 599 | |
| 600 | // TestConvergeStaticRecordsUntouched is the deletion-safety test. The zone is |
| 601 | // seeded with everything an operator's zone really holds; converge must |
| 602 | // submit no change that touches any of it. |
| 603 | func TestConvergeStaticRecordsUntouched(t *testing.T) { |
| 604 | f := newFake() |
| 605 | f.records = []r53types.ResourceRecordSet{ |
| 606 | staticA("example.com.", "9.9.9.9"), // apex A, no SetIdentifier |
| 607 | staticA("www.example.com.", "9.9.9.9"), // plain A |
| 608 | staticA("app.example.com.", "9.9.9.9"), // plain A sharing a STEERED name |
| 609 | {Name: aws.String("example.com."), Type: r53types.RRTypeNs, TTL: aws.Int64(172800)}, |
| 610 | {Name: aws.String("example.com."), Type: r53types.RRTypeSoa, TTL: aws.Int64(900)}, |
| 611 | {Name: aws.String("mail.example.com."), Type: r53types.RRTypeCname, TTL: aws.Int64(300)}, |
| 612 | { // a WEIGHTED A with a SetIdentifier — has an identifier but is not latency |
| 613 | Name: aws.String("weighted.example.com."), Type: r53types.RRTypeA, |
| 614 | SetIdentifier: aws.String("dublin"), Weight: aws.Int64(10), TTL: aws.Int64(60), |
| 615 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("8.8.8.8")}}, |
| 616 | }, |
| 617 | { // a LATENCY A bound to a health check that is not ours |
| 618 | Name: aws.String("legacy.example.com."), Type: r53types.RRTypeA, |
| 619 | SetIdentifier: aws.String("dublin"), Region: "eu-west-1", TTL: aws.Int64(60), |
| 620 | HealthCheckId: aws.String("hc-someone-else"), |
| 621 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("7.7.7.7")}}, |
| 622 | }, |
| 623 | { // a LATENCY ALIAS record — never a shape this converger writes |
| 624 | Name: aws.String("alias.example.com."), Type: r53types.RRTypeA, |
| 625 | SetIdentifier: aws.String("dublin"), Region: "eu-west-1", |
| 626 | AliasTarget: &r53types.AliasTarget{DNSName: aws.String("elb.example.com.")}, |
| 627 | }, |
| 628 | } |
| 629 | before := recordNames(f.records) |
| 630 | |
| 631 | buf := captureLog(t) |
| 632 | want := append([]desiredRecord{dublinRec}, hostRecs("app.example.com.", dublinRec)...) |
| 633 | mustConverge(t, f, want) |
| 634 | |
| 635 | for _, c := range f.submitted { |
| 636 | if c.Action == r53types.ChangeActionDelete { |
| 637 | t.Errorf("submitted a DELETE for %s — static/foreign records are out of bounds", rrKey(*c.ResourceRecordSet)) |
| 638 | } |
| 639 | if aws.ToString(c.ResourceRecordSet.SetIdentifier) == "" { |
| 640 | t.Errorf("submitted a change for a record with no SetIdentifier: %s", rrKey(*c.ResourceRecordSet)) |
| 641 | } |
| 642 | } |
| 643 | for _, name := range before { |
| 644 | var still bool |
| 645 | for _, have := range recordNames(f.records) { |
| 646 | if have == name { |
| 647 | still = true |
| 648 | } |
| 649 | } |
| 650 | if !still { |
| 651 | t.Errorf("record %s disappeared", name) |
| 652 | } |
| 653 | } |
| 654 | |
| 655 | // app.example.com. already holds a PLAIN A record. Route53 forbids a |
| 656 | // name from holding both a simple record and a latency record set, and |
| 657 | // change batches are atomic — so writing a latency record there would |
| 658 | // not merely fail for that name, it would take every other steering |
| 659 | // record in the zone down with it, on every tick, forever. The name is |
| 660 | // skipped entirely: no upsert, no delete, and loudly. |
| 661 | for _, c := range f.submitted { |
| 662 | if strings.EqualFold(aws.ToString(c.ResourceRecordSet.Name), "app.example.com.") { |
| 663 | t.Errorf("submitted a change for the plain-record-conflicted name: %s %s", c.Action, rrKey(*c.ResourceRecordSet)) |
| 664 | } |
| 665 | } |
| 666 | if !strings.Contains(buf.String(), "SKIPPING app.example.com.") { |
| 667 | t.Errorf("the conflict was not named in the log; log = %q", buf.String()) |
| 668 | } |
| 669 | if !strings.Contains(buf.String(), "9.9.9.9") { |
| 670 | t.Errorf("the log did not name the conflicting record's value; log = %q", buf.String()) |
| 671 | } |
| 672 | plain := find(t, f, "app.example.com.", "") |
| 673 | if aws.ToInt64(plain.TTL) != 300 || aws.ToString(plain.ResourceRecords[0].Value) != "9.9.9.9" { |
| 674 | t.Errorf("the operator's app.example.com. A record was changed: TTL %d value %v", aws.ToInt64(plain.TTL), plain.ResourceRecords) |
| 675 | } |
| 676 | |
| 677 | // ...and the OTHER name's record still lands. One conflicting name must |
| 678 | // not wedge the whole zone. |
| 679 | edgeRR := find(t, f, "edge.example.com.", "dublin") |
| 680 | if aws.ToString(edgeRR.ResourceRecords[0].Value) != "1.2.3.4" { |
| 681 | t.Errorf("the unconflicted name did not converge: %v", edgeRR.ResourceRecords) |
| 682 | } |
| 683 | if f.nChangeRecords != 1 { |
| 684 | t.Errorf("want 1 change batch (the unconflicted name), got %d", f.nChangeRecords) |
| 685 | } |
| 686 | } |
| 687 | |
| 688 | // TestFakeRejectsConflictingBatches proves the fake's refusals are real, so |
| 689 | // the skips are fixes and not tautologies: submitting the batch the converger |
| 690 | // used to submit does fail, atomically, taking an unrelated and perfectly |
| 691 | // legal change in the same batch down with it. One row per conflict shape. |
| 692 | func TestFakeRejectsConflictingBatches(t *testing.T) { |
| 693 | tests := []struct { |
| 694 | name string |
| 695 | seeded r53types.ResourceRecordSet |
| 696 | wantErr string |
| 697 | }{ |
| 698 | { |
| 699 | name: "plain A", |
| 700 | seeded: staticA("app.example.com.", "9.9.9.9"), |
| 701 | wantErr: "cannot be both a simple record and a routing-policy record set", |
| 702 | }, |
| 703 | { |
| 704 | name: "CNAME", |
| 705 | seeded: r53types.ResourceRecordSet{ |
| 706 | Name: aws.String("app.example.com."), Type: r53types.RRTypeCname, TTL: aws.Int64(300), |
| 707 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("elsewhere.example.net.")}}, |
| 708 | }, |
| 709 | wantErr: "cannot coexist with any other record type", |
| 710 | }, |
| 711 | { |
| 712 | name: "weighted A set", |
| 713 | seeded: r53types.ResourceRecordSet{ |
| 714 | Name: aws.String("app.example.com."), Type: r53types.RRTypeA, |
| 715 | SetIdentifier: aws.String("legacy"), Weight: aws.Int64(10), TTL: aws.Int64(60), |
| 716 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("8.8.8.8")}}, |
| 717 | }, |
| 718 | wantErr: "mixes routing policies", |
| 719 | }, |
| 720 | } |
| 721 | for _, tt := range tests { |
| 722 | t.Run(tt.name, func(t *testing.T) { |
| 723 | f := newFake() |
| 724 | f.records = []r53types.ResourceRecordSet{tt.seeded} |
| 725 | |
| 726 | latency := r53types.ResourceRecordSet{ |
| 727 | Name: aws.String("app.example.com."), Type: r53types.RRTypeA, |
| 728 | SetIdentifier: aws.String("dublin"), Region: "eu-west-1", TTL: aws.Int64(60), |
| 729 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("1.2.3.4")}}, |
| 730 | } |
| 731 | innocent := r53types.ResourceRecordSet{ |
| 732 | Name: aws.String("edge.example.com."), Type: r53types.RRTypeA, |
| 733 | SetIdentifier: aws.String("dublin"), Region: "eu-west-1", TTL: aws.Int64(60), |
| 734 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("1.2.3.4")}}, |
| 735 | } |
| 736 | _, err := f.ChangeResourceRecordSets(context.Background(), &route53.ChangeResourceRecordSetsInput{ |
| 737 | HostedZoneId: aws.String(testZone), |
| 738 | ChangeBatch: &r53types.ChangeBatch{Changes: []r53types.Change{ |
| 739 | {Action: r53types.ChangeActionUpsert, ResourceRecordSet: &innocent}, |
| 740 | {Action: r53types.ChangeActionUpsert, ResourceRecordSet: &latency}, |
| 741 | }}, |
| 742 | }) |
| 743 | if err == nil { |
| 744 | t.Fatalf("the fake accepted the batch — it is more permissive than Route53") |
| 745 | } |
| 746 | if !strings.Contains(err.Error(), "InvalidChangeBatch") || !strings.Contains(err.Error(), tt.wantErr) { |
| 747 | t.Errorf("wrong error: %v", err) |
| 748 | } |
| 749 | // atomic: the innocent change did NOT land either |
| 750 | if len(f.records) != 1 { |
| 751 | t.Fatalf("the batch was not atomic: %v", recordNames(f.records)) |
| 752 | } |
| 753 | }) |
| 754 | } |
| 755 | } |
| 756 | |
| 757 | // TestConvergeSkipsConflictingNames is FINDING A: the plain-A wedge has two |
| 758 | // siblings with the identical failure mode. Each must be skipped by name, and |
| 759 | // each must leave the OTHER steered names converging normally. |
| 760 | func TestConvergeSkipsConflictingNames(t *testing.T) { |
| 761 | tests := []struct { |
| 762 | name string |
| 763 | seeded r53types.ResourceRecordSet |
| 764 | wantLogged string |
| 765 | }{ |
| 766 | { |
| 767 | name: "a CNAME at a steered name", |
| 768 | seeded: r53types.ResourceRecordSet{ |
| 769 | Name: aws.String("app.example.com."), Type: r53types.RRTypeCname, TTL: aws.Int64(300), |
| 770 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("elsewhere.example.net.")}}, |
| 771 | }, |
| 772 | wantLogged: "a name holding a CNAME may hold no other record type", |
| 773 | }, |
| 774 | { |
| 775 | name: "a weighted A set at a steered name", |
| 776 | seeded: r53types.ResourceRecordSet{ |
| 777 | Name: aws.String("app.example.com."), Type: r53types.RRTypeA, |
| 778 | SetIdentifier: aws.String("legacy"), Weight: aws.Int64(10), TTL: aws.Int64(60), |
| 779 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("8.8.8.8")}}, |
| 780 | }, |
| 781 | wantLogged: "only one routing policy per name and type", |
| 782 | }, |
| 783 | { |
| 784 | name: "a failover A set at a steered name", |
| 785 | seeded: r53types.ResourceRecordSet{ |
| 786 | Name: aws.String("app.example.com."), Type: r53types.RRTypeA, |
| 787 | SetIdentifier: aws.String("primary"), Failover: r53types.ResourceRecordSetFailoverPrimary, |
| 788 | TTL: aws.Int64(60), |
| 789 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("8.8.8.8")}}, |
| 790 | }, |
| 791 | wantLogged: "only one routing policy per name and type", |
| 792 | }, |
| 793 | } |
| 794 | for _, tt := range tests { |
| 795 | t.Run(tt.name, func(t *testing.T) { |
| 796 | f := newFake() |
| 797 | f.records = []r53types.ResourceRecordSet{tt.seeded} |
| 798 | seededKey := rrKey(tt.seeded) |
| 799 | |
| 800 | buf := captureLog(t) |
| 801 | want := append([]desiredRecord{dublinRec}, hostRecs("app.example.com.", dublinRec)...) |
| 802 | mustConverge(t, f, want) |
| 803 | |
| 804 | // the conflicted name is skipped ENTIRELY — no upsert, no delete |
| 805 | for _, c := range f.submitted { |
| 806 | if strings.EqualFold(aws.ToString(c.ResourceRecordSet.Name), "app.example.com.") { |
| 807 | t.Errorf("submitted a change for the conflicted name: %s %s", c.Action, rrKey(*c.ResourceRecordSet)) |
| 808 | } |
| 809 | } |
| 810 | // the operator's record is untouched |
| 811 | var found bool |
| 812 | for _, rr := range f.records { |
| 813 | if rrKey(rr) == seededKey { |
| 814 | found = true |
| 815 | } |
| 816 | } |
| 817 | if !found { |
| 818 | t.Errorf("the operator's record disappeared; zone = %v", recordNames(f.records)) |
| 819 | } |
| 820 | // the log names the conflict and why |
| 821 | logged := buf.String() |
| 822 | if !strings.Contains(logged, "SKIPPING app.example.com.") { |
| 823 | t.Errorf("the conflict was not named; log = %q", logged) |
| 824 | } |
| 825 | if !strings.Contains(logged, tt.wantLogged) { |
| 826 | t.Errorf("the log did not explain the conflict (%q); log = %q", tt.wantLogged, logged) |
| 827 | } |
| 828 | // ...and the OTHER name still converges in the same pass |
| 829 | edgeRR := find(t, f, "edge.example.com.", "dublin") |
| 830 | if aws.ToString(edgeRR.ResourceRecords[0].Value) != "1.2.3.4" { |
| 831 | t.Errorf("the unconflicted name did not converge: %v", edgeRR.ResourceRecords) |
| 832 | } |
| 833 | }) |
| 834 | } |
| 835 | } |
| 836 | |
| 837 | // TestConvergeSNIDowngradeIsLogged: when the synthetic edge.<apps-domain> |
| 838 | // name is itself dropped for a conflict, every health check created |
| 839 | // afterwards silently falls back to a member hostname for SNI. That |
| 840 | // downgrade must be visible. |
| 841 | func TestConvergeSNIDowngradeIsLogged(t *testing.T) { |
| 842 | f := newFake() |
| 843 | f.records = []r53types.ResourceRecordSet{staticA("edge.example.com.", "9.9.9.9")} |
| 844 | |
| 845 | buf := captureLog(t) |
| 846 | want := append([]desiredRecord{dublinRec}, hostRecs("app.example.com.", dublinRec)...) |
| 847 | mustConverge(t, f, want) |
| 848 | |
| 849 | if len(f.createdChecks) != 1 { |
| 850 | t.Fatalf("want 1 check, got %d", len(f.createdChecks)) |
| 851 | } |
| 852 | if got := aws.ToString(f.createdChecks[0].HealthCheckConfig.FullyQualifiedDomainName); got != "app.example.com" { |
| 853 | t.Fatalf("SNI = %q, want the member hostname fallback", got) |
| 854 | } |
| 855 | logged := buf.String() |
| 856 | if !strings.Contains(logged, "fall back to a member hostname") { |
| 857 | t.Errorf("the SNI downgrade was silent; log = %q", logged) |
| 858 | } |
| 859 | if !strings.Contains(logged, "SKIPPING edge.example.com.") { |
| 860 | t.Errorf("the cause was not logged alongside it; log = %q", logged) |
| 861 | } |
| 862 | } |
| 863 | |
| 864 | // TestConvergeShrinkGuard is FINDING 1: the empty-desired-set refusal only |
| 865 | // catches a read that returned NOTHING. A SHORT read has a non-empty want and |
| 866 | // would delete every record for the members that went missing — NXDOMAIN for |
| 867 | // a live steered host. A pass that would remove most of what it manages must |
| 868 | // refuse and describe itself instead. |
| 869 | func TestConvergeShrinkGuard(t *testing.T) { |
| 870 | // 2 names x 2 edges = 4 managed records. |
| 871 | full := append([]desiredRecord{dublinRec, oregonRec}, hostRecs("app.example.com.", dublinRec, oregonRec)...) |
| 872 | |
| 873 | t.Run("deleting 3 of 4 refuses with zero writes and names them", func(t *testing.T) { |
| 874 | f := newFake() |
| 875 | mustConverge(t, f, full) |
| 876 | before := f.writes() |
| 877 | state := recordNames(f.records) |
| 878 | |
| 879 | buf := captureLog(t) |
| 880 | // A short read that saw only one edge and only the synthetic name. |
| 881 | mustConverge(t, f, []desiredRecord{dublinRec}) |
| 882 | |
| 883 | if got := f.writes() - before; got != 0 { |
| 884 | t.Fatalf("the refused pass made %d write calls, want 0", got) |
| 885 | } |
| 886 | if len(f.deletedChecks) != 0 { |
| 887 | t.Errorf("the refused pass GC'd checks: %v", f.deletedChecks) |
| 888 | } |
| 889 | if got := recordNames(f.records); !reflect.DeepEqual(got, state) { |
| 890 | t.Fatalf("the zone changed: %v -> %v", state, got) |
| 891 | } |
| 892 | logged := buf.String() |
| 893 | if !strings.Contains(logged, "refusing to converge") { |
| 894 | t.Errorf("the refusal was not logged; log = %q", logged) |
| 895 | } |
| 896 | // all three intended deletions named by (name, setid) |
| 897 | for _, want := range []string{ |
| 898 | "would delete edge.example.com. (set oregon", |
| 899 | "would delete app.example.com. (set dublin", |
| 900 | "would delete app.example.com. (set oregon", |
| 901 | } { |
| 902 | if !strings.Contains(logged, want) { |
| 903 | t.Errorf("log does not name %q; log = %q", want, logged) |
| 904 | } |
| 905 | } |
| 906 | }) |
| 907 | |
| 908 | t.Run("deleting 1 of 4 proceeds", func(t *testing.T) { |
| 909 | f := newFake() |
| 910 | mustConverge(t, f, full) |
| 911 | |
| 912 | // drop exactly one record: app.example.com. on oregon |
| 913 | var want []desiredRecord |
| 914 | for _, rec := range full { |
| 915 | if rec.Name == "app.example.com." && rec.Box == "oregon" { |
| 916 | continue |
| 917 | } |
| 918 | want = append(want, rec) |
| 919 | } |
| 920 | mustConverge(t, f, want) |
| 921 | |
| 922 | if len(f.records) != 3 { |
| 923 | t.Fatalf("want 3 records left, got %v", recordNames(f.records)) |
| 924 | } |
| 925 | for _, rr := range f.records { |
| 926 | if strings.EqualFold(aws.ToString(rr.Name), "app.example.com.") && aws.ToString(rr.SetIdentifier) == "oregon" { |
| 927 | t.Errorf("the ordinary deletion did not happen") |
| 928 | } |
| 929 | } |
| 930 | }) |
| 931 | |
| 932 | t.Run("the floor lets a small zone shrink", func(t *testing.T) { |
| 933 | // 2 managed records; half of 2 is 1, but the floor allows 2. |
| 934 | f := newFake() |
| 935 | two := []desiredRecord{dublinRec, oregonRec} |
| 936 | mustConverge(t, f, two) |
| 937 | mustConverge(t, f, append(hostRecs("app.example.com.", dublinRec), dublinRec)) |
| 938 | if len(f.records) != 2 { |
| 939 | t.Fatalf("want 2 records, got %v", recordNames(f.records)) |
| 940 | } |
| 941 | }) |
| 942 | } |
| 943 | |
| 944 | // TestShrinkLimit pins the threshold arithmetic on its own — the zone-level |
| 945 | // test above can only exercise the sizes it happens to build. |
| 946 | func TestShrinkLimit(t *testing.T) { |
| 947 | tests := []struct{ managed, want int }{ |
| 948 | {0, shrinkFloor}, {1, shrinkFloor}, {2, shrinkFloor}, {4, shrinkFloor}, |
| 949 | {6, 3}, {10, 5}, {11, 5}, {100, 50}, |
| 950 | } |
| 951 | for _, tt := range tests { |
| 952 | if got := shrinkLimit(tt.managed); got != tt.want { |
| 953 | t.Errorf("shrinkLimit(%d) = %d, want %d", tt.managed, got, tt.want) |
| 954 | } |
| 955 | } |
| 956 | } |
| 957 | |
| 958 | // TestConvergeHealthCheckGCIsTagScoped: a check must carry OUR tag, with OUR |
| 959 | // zone id, before it can ever be deleted. Everything else in the account is |
| 960 | // invisible to the GC. |
| 961 | func TestConvergeHealthCheckGCIsTagScoped(t *testing.T) { |
| 962 | f := newFake() |
| 963 | // three checks on an EIP nobody wants any more |
| 964 | f.checks = []r53types.HealthCheck{ |
| 965 | {Id: aws.String("hc-untagged"), HealthCheckConfig: &r53types.HealthCheckConfig{ |
| 966 | Type: r53types.HealthCheckTypeHttps, Port: aws.Int32(443), IPAddress: aws.String("10.0.0.1")}}, |
| 967 | {Id: aws.String("hc-other-zone"), HealthCheckConfig: &r53types.HealthCheckConfig{ |
| 968 | Type: r53types.HealthCheckTypeHttps, Port: aws.Int32(443), IPAddress: aws.String("10.0.0.2")}}, |
| 969 | {Id: aws.String("hc-ours"), HealthCheckConfig: &r53types.HealthCheckConfig{ |
| 970 | Type: r53types.HealthCheckTypeHttps, Port: aws.Int32(443), IPAddress: aws.String("10.0.0.3")}}, |
| 971 | {Id: aws.String("hc-other-service"), HealthCheckConfig: &r53types.HealthCheckConfig{ |
| 972 | Type: r53types.HealthCheckTypeHttp, Port: aws.Int32(80), IPAddress: aws.String("10.0.0.4")}}, |
| 973 | } |
| 974 | f.tags["hc-other-zone"] = []r53types.Tag{{Key: aws.String(steeringTagKey), Value: aws.String("ZSOMEOTHERZONE")}} |
| 975 | f.tags["hc-ours"] = []r53types.Tag{{Key: aws.String(steeringTagKey), Value: aws.String(testZone)}} |
| 976 | f.tags["hc-other-service"] = []r53types.Tag{{Key: aws.String("Name"), Value: aws.String("someone else's check")}} |
| 977 | |
| 978 | mustConverge(t, f, []desiredRecord{dublinRec}) |
| 979 | |
| 980 | if len(f.deletedChecks) != 1 || f.deletedChecks[0] != "hc-ours" { |
| 981 | t.Fatalf("GC deleted %v; only the tagged-for-THIS-zone check may ever go", f.deletedChecks) |
| 982 | } |
| 983 | } |
| 984 | |
| 985 | // TestConvergeEmptyWantRefusesTeardown: an empty desired set is a degraded |
| 986 | // read (zero armed edges), not an instruction to delete production DNS. |
| 987 | func TestConvergeEmptyWantRefusesTeardown(t *testing.T) { |
| 988 | f := newFake() |
| 989 | want := append([]desiredRecord{dublinRec}, hostRecs("app.example.com.", dublinRec)...) |
| 990 | mustConverge(t, f, want) |
| 991 | before := f.writes() |
| 992 | recs := len(f.records) |
| 993 | |
| 994 | buf := captureLog(t) |
| 995 | if err := converge(context.Background(), f, testZone, nil); err != nil { |
| 996 | t.Fatalf("empty want should not error: %v", err) |
| 997 | } |
| 998 | if f.writes() != before { |
| 999 | t.Fatalf("empty want made %d write calls", f.writes()-before) |
| 1000 | } |
| 1001 | if len(f.records) != recs { |
| 1002 | t.Fatalf("empty want removed records: %v", recordNames(f.records)) |
| 1003 | } |
| 1004 | if len(f.deletedChecks) != 0 { |
| 1005 | t.Fatalf("empty want GC'd checks: %v", f.deletedChecks) |
| 1006 | } |
| 1007 | if !strings.Contains(buf.String(), "refusing") { |
| 1008 | t.Errorf("the refusal was silent; log = %q", buf.String()) |
| 1009 | } |
| 1010 | } |
| 1011 | |
| 1012 | // TestConvergeRejectsBadZoneID: a malformed or empty zone id never reaches |
| 1013 | // the API — a typo must not become a call against some other zone. |
| 1014 | func TestConvergeRejectsBadZoneID(t *testing.T) { |
| 1015 | for _, zone := range []string{"", " ", "z0abc", "/hostedzone/Z0ABC", "Z0abc", "NOTAZONE"} { |
| 1016 | f := newFake() |
| 1017 | err := converge(context.Background(), f, zone, []desiredRecord{dublinRec}) |
| 1018 | if err == nil { |
| 1019 | t.Errorf("zone %q accepted", zone) |
| 1020 | } |
| 1021 | if f.nListRecords+f.nListChecks+f.writes() != 0 { |
| 1022 | t.Errorf("zone %q reached the API", zone) |
| 1023 | } |
| 1024 | } |
| 1025 | } |
| 1026 | |
| 1027 | // TestConvergeHealthCheckFQDNFallback: with no edge.<apps-domain> record in |
| 1028 | // want, the check's SNI name falls back to a record's own name. |
| 1029 | func TestConvergeHealthCheckFQDNFallback(t *testing.T) { |
| 1030 | f := newFake() |
| 1031 | mustConverge(t, f, hostRecs("app.example.com.", dublinRec)) |
| 1032 | |
| 1033 | if len(f.createdChecks) != 1 { |
| 1034 | t.Fatalf("want 1 check, got %d", len(f.createdChecks)) |
| 1035 | } |
| 1036 | if got := aws.ToString(f.createdChecks[0].HealthCheckConfig.FullyQualifiedDomainName); got != "app.example.com" { |
| 1037 | t.Fatalf("FQDN = %q, want the record's own name app.example.com", got) |
| 1038 | } |
| 1039 | } |
| 1040 | |
| 1041 | // TestConvergePagination: both listings paginate, and a converger that |
| 1042 | // stopped at page one would see an empty zone and rewrite everything. |
| 1043 | func TestConvergePagination(t *testing.T) { |
| 1044 | f := newFake() |
| 1045 | want := append([]desiredRecord{dublinRec, oregonRec}, hostRecs("app.example.com.", dublinRec, oregonRec)...) |
| 1046 | want = append(want, hostRecs("b.example.com.", dublinRec, oregonRec)...) |
| 1047 | want = append(want, hostRecs("c.example.com.", dublinRec, oregonRec)...) |
| 1048 | mustConverge(t, f, want) |
| 1049 | |
| 1050 | f.pageSize = 1 // force many pages on the second pass |
| 1051 | before := f.writes() |
| 1052 | mustConverge(t, f, want) |
| 1053 | if got := f.writes() - before; got != 0 { |
| 1054 | t.Fatalf("paginated pass made %d writes, want 0 — a page was missed", got) |
| 1055 | } |
| 1056 | if f.nListRecords < 3 || f.nListChecks < 2 { |
| 1057 | t.Fatalf("pagination not exercised: %d record pages, %d check pages", f.nListRecords, f.nListChecks) |
| 1058 | } |
| 1059 | } |
| 1060 | |
| 1061 | // TestConvergeTagFailureRemovesTheOrphan: a check created but not tagged can |
| 1062 | // never be reused (untagged = not ours) and never GC'd — an unbounded leak, |
| 1063 | // one per pass. It is deleted again before the pass fails. |
| 1064 | func TestConvergeTagFailureRemovesTheOrphan(t *testing.T) { |
| 1065 | f := newFake() |
| 1066 | f.failTags = errors.New("boom") |
| 1067 | err := converge(context.Background(), f, testZone, []desiredRecord{dublinRec}) |
| 1068 | if err == nil { |
| 1069 | t.Fatalf("tagging failure did not fail the pass") |
| 1070 | } |
| 1071 | if len(f.deletedChecks) != 1 { |
| 1072 | t.Fatalf("the untaggable check was leaked: created %d, deleted %v", f.nCreateCheck, f.deletedChecks) |
| 1073 | } |
| 1074 | if f.nChangeRecords != 0 { |
| 1075 | t.Errorf("records were written despite the failed pass") |
| 1076 | } |
| 1077 | } |
| 1078 | |
| 1079 | // TestConvergeChangeFailureIsLoud: a failed batch fails the pass. The |
| 1080 | // housekeeper retries next tick; it must not be swallowed. |
| 1081 | func TestConvergeChangeFailureIsLoud(t *testing.T) { |
| 1082 | f := newFake() |
| 1083 | f.failChange = errors.New("InvalidChangeBatch") |
| 1084 | err := converge(context.Background(), f, testZone, []desiredRecord{dublinRec}) |
| 1085 | if err == nil { |
| 1086 | t.Fatalf("a failed change batch did not fail the pass") |
| 1087 | } |
| 1088 | if !strings.Contains(err.Error(), "InvalidChangeBatch") { |
| 1089 | t.Errorf("the underlying error was not wrapped: %v", err) |
| 1090 | } |
| 1091 | // nothing may be GC'd on a pass whose record changes did not land |
| 1092 | if len(f.deletedChecks) != 0 { |
| 1093 | t.Errorf("checks GC'd after a failed record batch: %v", f.deletedChecks) |
| 1094 | } |
| 1095 | } |
| 1096 | |
| 1097 | // TestConvergeBatchesChanges: a large desired set is chunked, and every |
| 1098 | // chunk is within the batch limit. |
| 1099 | func TestConvergeBatchesChanges(t *testing.T) { |
| 1100 | f := newFake() |
| 1101 | var want []desiredRecord |
| 1102 | for i := 0; i < changeBatchLimit+5; i++ { |
| 1103 | want = append(want, desiredRecord{ |
| 1104 | Name: fmt.Sprintf("h%03d.example.com.", i), EIP: "1.2.3.4", Box: "dublin", Region: "eu-west-1", |
| 1105 | }) |
| 1106 | } |
| 1107 | mustConverge(t, f, want) |
| 1108 | if f.nChangeRecords != 2 { |
| 1109 | t.Fatalf("want 2 batches for %d changes, got %d", len(want), f.nChangeRecords) |
| 1110 | } |
| 1111 | if len(f.records) != len(want) { |
| 1112 | t.Fatalf("want %d records, got %d", len(want), len(f.records)) |
| 1113 | } |
| 1114 | } |
| 1115 | |
| 1116 | // TestConvergeAdoptsForeignLatencyRecordUnderOurKey: a latency record already |
| 1117 | // sitting at a (name, SetIdentifier) we want is taken over by UPSERT — it is |
| 1118 | // exactly the record we would otherwise create, and CREATE would collide. |
| 1119 | func TestConvergeAdoptsForeignLatencyRecordUnderOurKey(t *testing.T) { |
| 1120 | f := newFake() |
| 1121 | f.records = []r53types.ResourceRecordSet{{ |
| 1122 | Name: aws.String("edge.example.com."), Type: r53types.RRTypeA, |
| 1123 | SetIdentifier: aws.String("dublin"), Region: "eu-west-1", TTL: aws.Int64(300), |
| 1124 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("99.99.99.99")}}, |
| 1125 | }} |
| 1126 | buf := captureLog(t) |
| 1127 | mustConverge(t, f, []desiredRecord{dublinRec}) |
| 1128 | |
| 1129 | rr := find(t, f, "edge.example.com.", "dublin") |
| 1130 | if aws.ToString(rr.ResourceRecords[0].Value) != "1.2.3.4" || aws.ToInt64(rr.TTL) != 60 { |
| 1131 | t.Fatalf("record not converged: %v TTL=%d", rr.ResourceRecords, aws.ToInt64(rr.TTL)) |
| 1132 | } |
| 1133 | if !strings.Contains(buf.String(), "adopting") { |
| 1134 | t.Errorf("the takeover was silent; log = %q", buf.String()) |
| 1135 | } |
| 1136 | // ...and it must NOT also claim to be leaving that same record alone. |
| 1137 | // The two lines describe opposite fates for one record; only one of them |
| 1138 | // is true. (Regression: the "leaving alone" log lost its wanted-key |
| 1139 | // guard when the loop was split out of recordChanges in review round 1.) |
| 1140 | if strings.Contains(buf.String(), "leaving latency record") { |
| 1141 | t.Errorf("logged BOTH adoption and 'leaving alone' for the same record; log = %q", buf.String()) |
| 1142 | } |
| 1143 | } |
| 1144 | |
| 1145 | // TestConvergeLeavingAloneLogNamesOnlyUnwantedKeys: the "leaving alone" line |
| 1146 | // is still owed for a foreign latency record this pass genuinely does not |
| 1147 | // touch — the FINDING B fix must not silence it everywhere. |
| 1148 | func TestConvergeLeavingAloneLogNamesOnlyUnwantedKeys(t *testing.T) { |
| 1149 | f := newFake() |
| 1150 | f.records = []r53types.ResourceRecordSet{ |
| 1151 | { // a foreign latency record at a name/identifier we do NOT want |
| 1152 | Name: aws.String("legacy.example.com."), Type: r53types.RRTypeA, |
| 1153 | SetIdentifier: aws.String("dublin"), Region: "eu-west-1", TTL: aws.Int64(60), |
| 1154 | HealthCheckId: aws.String("hc-someone-else"), |
| 1155 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("7.7.7.7")}}, |
| 1156 | }, |
| 1157 | } |
| 1158 | buf := captureLog(t) |
| 1159 | mustConverge(t, f, []desiredRecord{dublinRec}) |
| 1160 | |
| 1161 | if !strings.Contains(buf.String(), "leaving latency record legacy.example.com.") { |
| 1162 | t.Errorf("the untouched foreign record was not reported; log = %q", buf.String()) |
| 1163 | } |
| 1164 | if len(f.records) != 2 { |
| 1165 | t.Errorf("the foreign record was touched; zone = %v", recordNames(f.records)) |
| 1166 | } |
| 1167 | } |
| 1168 | |
| 1169 | // TestIsManagedGates pins the deletion boundary GATE BY GATE. |
| 1170 | // |
| 1171 | // The zone-level test above (TestConvergeStaticRecordsUntouched) proves the |
| 1172 | // boundary holds for the records a real zone contains, but its fixtures fail |
| 1173 | // several gates at once — so it cannot tell which gate is doing the work, and |
| 1174 | // stays green if one of them is deleted. This table starts from a record that |
| 1175 | // passes EVERY gate and breaks exactly one at a time. Each row must flip the |
| 1176 | // verdict to unmanaged on its own. |
| 1177 | func TestIsManagedGates(t *testing.T) { |
| 1178 | owned := map[string]bool{"hc-ours": true} |
| 1179 | managed := func() r53types.ResourceRecordSet { |
| 1180 | return r53types.ResourceRecordSet{ |
| 1181 | Name: aws.String("app.example.com."), Type: r53types.RRTypeA, |
| 1182 | SetIdentifier: aws.String("dublin"), Region: "eu-west-1", TTL: aws.Int64(60), |
| 1183 | HealthCheckId: aws.String("hc-ours"), |
| 1184 | ResourceRecords: []r53types.ResourceRecord{{Value: aws.String("1.2.3.4")}}, |
| 1185 | } |
| 1186 | } |
| 1187 | if !isManaged(managed(), owned) { |
| 1188 | t.Fatalf("the baseline record is not managed; the table below proves nothing") |
| 1189 | } |
| 1190 | |
| 1191 | tests := []struct { |
| 1192 | name string |
| 1193 | break_ func(*r53types.ResourceRecordSet) |
| 1194 | }{ |
| 1195 | {"not an A record", func(rr *r53types.ResourceRecordSet) { rr.Type = r53types.RRTypeAaaa }}, |
| 1196 | {"no SetIdentifier (static/operator territory)", func(rr *r53types.ResourceRecordSet) { rr.SetIdentifier = nil }}, |
| 1197 | {"empty SetIdentifier", func(rr *r53types.ResourceRecordSet) { rr.SetIdentifier = aws.String("") }}, |
| 1198 | {"not a latency record", func(rr *r53types.ResourceRecordSet) { rr.Region = "" }}, |
| 1199 | {"an alias record", func(rr *r53types.ResourceRecordSet) { |
| 1200 | rr.AliasTarget = &r53types.AliasTarget{DNSName: aws.String("elb.example.com.")} |
| 1201 | }}, |
| 1202 | {"health check we do not own", func(rr *r53types.ResourceRecordSet) { rr.HealthCheckId = aws.String("hc-theirs") }}, |
| 1203 | {"no health check at all", func(rr *r53types.ResourceRecordSet) { rr.HealthCheckId = nil }}, |
| 1204 | } |
| 1205 | for _, tt := range tests { |
| 1206 | t.Run(tt.name, func(t *testing.T) { |
| 1207 | rr := managed() |
| 1208 | tt.break_(&rr) |
| 1209 | if isManaged(rr, owned) { |
| 1210 | t.Fatalf("record is still managed — this gate does not stand on its own") |
| 1211 | } |
| 1212 | }) |
| 1213 | } |
| 1214 | |
| 1215 | // and the empty-ownership case: if the tag listing found nothing ours, |
| 1216 | // NOTHING in the zone is deletable. |
| 1217 | if isManaged(managed(), map[string]bool{}) { |
| 1218 | t.Errorf("a record is managed with no owned health checks at all") |
| 1219 | } |
| 1220 | } |
| 1221 | |
| 1222 | // TestSteeringZoneIDFromEnv pins the dormancy contract Task 6 consumes. |
| 1223 | func TestSteeringZoneIDFromEnv(t *testing.T) { |
| 1224 | tests := []struct { |
| 1225 | set string |
| 1226 | want string |
| 1227 | }{ |
| 1228 | {"", ""}, |
| 1229 | {" ", ""}, |
| 1230 | {"Z0STEERINGTEST", "Z0STEERINGTEST"}, |
| 1231 | {" Z0STEERINGTEST ", "Z0STEERINGTEST"}, |
| 1232 | {"/hostedzone/Z0STEERINGTEST", ""}, |
| 1233 | {"z0steeringtest", ""}, |
| 1234 | } |
| 1235 | for _, tt := range tests { |
| 1236 | t.Setenv(steeringZoneEnv, tt.set) |
| 1237 | if got := SteeringZoneID(); got != tt.want { |
| 1238 | t.Errorf("SteeringZoneID() with %q = %q, want %q", tt.set, got, tt.want) |
| 1239 | } |
| 1240 | } |
| 1241 | } |
| 1242 | |
amadan