| 1 | package domains |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "os" |
| 7 | "reflect" |
| 8 | "testing" |
| 9 | "time" |
| 10 | |
| 11 | "github.com/carlosframework/platform/internal/release" |
| 12 | ) |
| 13 | |
| 14 | // countingStore counts the GETs the fold spends. That count IS the property |
| 15 | // under test — the whole point of the cache is that a claim whose bucket |
| 16 | // stamp has not moved costs nothing to fold again. |
| 17 | type countingStore struct { |
| 18 | release.Store |
| 19 | gets int |
| 20 | } |
| 21 | |
| 22 | func (c *countingStore) Get(ctx context.Context, key string) ([]byte, string, error) { |
| 23 | c.gets++ |
| 24 | return c.Store.Get(ctx, key) |
| 25 | } |
| 26 | |
| 27 | func (c *countingStore) ListMod(ctx context.Context, prefix string) ([]release.KeyMod, error) { |
| 28 | return c.Store.(release.ModLister).ListMod(ctx, prefix) |
| 29 | } |
| 30 | |
| 31 | func seedClaim(t *testing.T, s release.Store, host, app string) { |
| 32 | t.Helper() |
| 33 | if _, _, err := AttachClaim(context.Background(), s, Claim{Host: host, Account: "acct", App: app}); err != nil { |
| 34 | t.Fatal(err) |
| 35 | } |
| 36 | } |
| 37 | |
| 38 | // backdate moves a claim's bucket stamp out of the modSettle window, which |
| 39 | // is what a claim written more than a couple of seconds ago looks like. Real |
| 40 | // claims change on human timescales; a test cannot afford to sleep for them. |
| 41 | func backdate(t *testing.T, dir, host string) { |
| 42 | t.Helper() |
| 43 | old := time.Now().Add(-time.Hour) |
| 44 | if err := os.Chtimes(dir+"/"+ClaimKey(host), old, old); err != nil { |
| 45 | t.Fatal(err) |
| 46 | } |
| 47 | } |
| 48 | |
| 49 | func TestClaimCacheReReadsNothingWhenNothingChanged(t *testing.T) { |
| 50 | dir := t.TempDir() |
| 51 | s := &countingStore{Store: release.NewDirStore(dir)} |
| 52 | ctx := context.Background() |
| 53 | for _, h := range []string{"a.customer.test", "b.customer.test", "c.customer.test"} { |
| 54 | seedClaim(t, s, h, "kass") |
| 55 | backdate(t, dir, h) |
| 56 | } |
| 57 | s.gets = 0 |
| 58 | |
| 59 | var c ClaimCache |
| 60 | first, err := c.List(ctx, s) |
| 61 | if err != nil { |
| 62 | t.Fatal(err) |
| 63 | } |
| 64 | if len(first) != 3 { |
| 65 | t.Fatalf("first fold = %+v, want 3 claims", first) |
| 66 | } |
| 67 | cold := s.gets |
| 68 | if cold != 3 { |
| 69 | t.Fatalf("cold fold spent %d gets, want one per claim", cold) |
| 70 | } |
| 71 | |
| 72 | s.gets = 0 |
| 73 | second, err := c.List(ctx, s) |
| 74 | if err != nil { |
| 75 | t.Fatal(err) |
| 76 | } |
| 77 | if len(second) != 3 { |
| 78 | t.Fatalf("warm fold = %+v, want the same 3 claims", second) |
| 79 | } |
| 80 | if s.gets != 0 { |
| 81 | t.Errorf("warm fold spent %d gets, want 0 — an unmoved stamp must cost nothing", s.gets) |
| 82 | } |
| 83 | } |
| 84 | |
| 85 | // The cache must be invisible: what it returns has to be what the uncached |
| 86 | // fold returns, claim for claim, or a hostname appears on one read path and |
| 87 | // vanishes on another. |
| 88 | func TestClaimCacheAgreesWithTheUncachedFold(t *testing.T) { |
| 89 | dir := t.TempDir() |
| 90 | s := release.NewDirStore(dir) |
| 91 | ctx := context.Background() |
| 92 | for _, h := range []string{"a.customer.test", "b.customer.test"} { |
| 93 | seedClaim(t, s, h, "kass") |
| 94 | backdate(t, dir, h) |
| 95 | } |
| 96 | var c ClaimCache |
| 97 | if _, err := c.List(ctx, s); err != nil { // prime |
| 98 | t.Fatal(err) |
| 99 | } |
| 100 | cached, err := c.List(ctx, s) |
| 101 | if err != nil { |
| 102 | t.Fatal(err) |
| 103 | } |
| 104 | plain, err := ListClaims(ctx, s) |
| 105 | if err != nil { |
| 106 | t.Fatal(err) |
| 107 | } |
| 108 | if len(cached) != len(plain) { |
| 109 | t.Fatalf("cached fold has %d claims, uncached has %d", len(cached), len(plain)) |
| 110 | } |
| 111 | byHost := map[string]Claim{} |
| 112 | for _, p := range plain { |
| 113 | byHost[p.Host] = p |
| 114 | } |
| 115 | for _, got := range cached { |
| 116 | want, ok := byHost[got.Host] |
| 117 | if !ok { |
| 118 | t.Errorf("cached fold invented %s", got.Host) |
| 119 | continue |
| 120 | } |
| 121 | if !reflect.DeepEqual(got, want) { |
| 122 | t.Errorf("%s: cached %+v, uncached %+v", got.Host, got, want) |
| 123 | } |
| 124 | } |
| 125 | } |
| 126 | |
| 127 | // A detached claim leaves the listing, and leaving the listing IS the |
| 128 | // eviction — there is no separate invalidation call to forget to make. |
| 129 | func TestClaimCacheDropsADetachedClaim(t *testing.T) { |
| 130 | dir := t.TempDir() |
| 131 | s := release.NewDirStore(dir) |
| 132 | ctx := context.Background() |
| 133 | seedClaim(t, s, "a.customer.test", "kass") |
| 134 | seedClaim(t, s, "b.customer.test", "kass") |
| 135 | backdate(t, dir, "a.customer.test") |
| 136 | backdate(t, dir, "b.customer.test") |
| 137 | |
| 138 | var c ClaimCache |
| 139 | if got, err := c.List(ctx, s); err != nil || len(got) != 2 { |
| 140 | t.Fatalf("first fold = %+v, %v", got, err) |
| 141 | } |
| 142 | if _, err := DetachClaim(ctx, s, "b.customer.test", "acct", "kass"); err != nil { |
| 143 | t.Fatal(err) |
| 144 | } |
| 145 | got, err := c.List(ctx, s) |
| 146 | if err != nil { |
| 147 | t.Fatal(err) |
| 148 | } |
| 149 | if len(got) != 1 || got[0].Host != "a.customer.test" { |
| 150 | t.Fatalf("after detach = %+v, want only a.customer.test", got) |
| 151 | } |
| 152 | } |
| 153 | |
| 154 | // A changed claim is re-read, and only it. This is the case a cache gets |
| 155 | // wrong: serving a stale body because the listing said the key is still |
| 156 | // there. |
| 157 | func TestClaimCacheReReadsOnlyWhatChanged(t *testing.T) { |
| 158 | dir := t.TempDir() |
| 159 | s := &countingStore{Store: release.NewDirStore(dir)} |
| 160 | ctx := context.Background() |
| 161 | seedClaim(t, s, "a.customer.test", "kass") |
| 162 | seedClaim(t, s, "b.customer.test", "kass") |
| 163 | backdate(t, dir, "a.customer.test") |
| 164 | backdate(t, dir, "b.customer.test") |
| 165 | |
| 166 | var c ClaimCache |
| 167 | if _, err := c.List(ctx, s); err != nil { |
| 168 | t.Fatal(err) |
| 169 | } |
| 170 | if _, err := SetClaimCapabilities(ctx, s, "b.customer.test", "acct", "kass", true, false); err != nil { |
| 171 | t.Fatal(err) |
| 172 | } |
| 173 | backdate(t, dir, "b.customer.test") // settled, but at a DIFFERENT stamp |
| 174 | s.gets = 0 |
| 175 | got, err := c.List(ctx, s) |
| 176 | if err != nil { |
| 177 | t.Fatal(err) |
| 178 | } |
| 179 | if s.gets != 1 { |
| 180 | t.Errorf("spent %d gets, want exactly 1 — only the changed claim", s.gets) |
| 181 | } |
| 182 | for _, cl := range got { |
| 183 | if cl.Host == "b.customer.test" && !cl.Wildcard { |
| 184 | t.Error("the changed claim came back stale — the cache served a body the bucket no longer holds") |
| 185 | } |
| 186 | } |
| 187 | } |
| 188 | |
| 189 | // THE lost update the modSettle bound exists for. S3 stamps have one-second |
| 190 | // granularity, so a claim read and then rewritten inside the same second |
| 191 | // carries an unchanged-looking stamp. Caching that read would serve the old |
| 192 | // body forever, and "forever" is not an exaggeration: nothing else would |
| 193 | // ever move the stamp. |
| 194 | func TestClaimCacheRefusesToTrustAStampItJustSaw(t *testing.T) { |
| 195 | dir := t.TempDir() |
| 196 | s := &countingStore{Store: release.NewDirStore(dir)} |
| 197 | ctx := context.Background() |
| 198 | seedClaim(t, s, "a.customer.test", "kass") // written just now: inside modSettle |
| 199 | |
| 200 | var c ClaimCache |
| 201 | if _, err := c.List(ctx, s); err != nil { |
| 202 | t.Fatal(err) |
| 203 | } |
| 204 | s.gets = 0 |
| 205 | if _, err := c.List(ctx, s); err != nil { |
| 206 | t.Fatal(err) |
| 207 | } |
| 208 | if s.gets != 1 { |
| 209 | t.Errorf("spent %d gets, want 1 — a stamp from the last two seconds must not be trusted as unchanged", s.gets) |
| 210 | } |
| 211 | } |
| 212 | |
| 213 | // A store that cannot answer ListMod — the fleet store door's client, every |
| 214 | // test fake — degrades to the plain fold. Not to a weaker cache: a customer |
| 215 | // fleet box is the hardest place to notice a stale claim set. |
| 216 | func TestClaimCacheDegradesOnAStoreWithNoStamps(t *testing.T) { |
| 217 | s := stamplessStore{release.NewDirStore(t.TempDir())} |
| 218 | ctx := context.Background() |
| 219 | seedClaim(t, s, "a.customer.test", "kass") |
| 220 | var c ClaimCache |
| 221 | got, err := c.List(ctx, s) |
| 222 | if err != nil { |
| 223 | t.Fatal(err) |
| 224 | } |
| 225 | if len(got) != 1 { |
| 226 | t.Fatalf("fold = %+v, want the claim", got) |
| 227 | } |
| 228 | // Nothing was cached, so a second fold still sees a change made between |
| 229 | // the two — which is the whole point of degrading rather than guessing. |
| 230 | seedClaim(t, s, "b.customer.test", "kass") |
| 231 | if got, err := c.List(ctx, s); err != nil || len(got) != 2 { |
| 232 | t.Fatalf("second fold = %+v, %v; want both claims", got, err) |
| 233 | } |
| 234 | } |
| 235 | |
| 236 | // stamplessStore hides DirStore's ListMod behind a plain Store, the way the |
| 237 | // fleet door's client and the package's test fakes do. |
| 238 | type stamplessStore struct{ inner release.Store } |
| 239 | |
| 240 | func (s stamplessStore) Get(ctx context.Context, key string) ([]byte, string, error) { |
| 241 | return s.inner.Get(ctx, key) |
| 242 | } |
| 243 | func (s stamplessStore) Put(ctx context.Context, key string, body []byte, ifMatch string) (string, error) { |
| 244 | return s.inner.Put(ctx, key, body, ifMatch) |
| 245 | } |
| 246 | func (s stamplessStore) List(ctx context.Context, prefix string) ([]string, error) { |
| 247 | return s.inner.List(ctx, prefix) |
| 248 | } |
| 249 | func (s stamplessStore) ListDirs(ctx context.Context, prefix string) ([]string, error) { |
| 250 | return s.inner.ListDirs(ctx, prefix) |
| 251 | } |
| 252 | func (s stamplessStore) Delete(ctx context.Context, key string) error { |
| 253 | return s.inner.Delete(ctx, key) |
| 254 | } |
| 255 | |
| 256 | // A nil cache is legal and means "no caching", so a caller with none to |
| 257 | // offer passes nil rather than every call site growing a branch. |
| 258 | func TestNilClaimCacheFoldsWithoutCaching(t *testing.T) { |
| 259 | s := release.NewDirStore(t.TempDir()) |
| 260 | ctx := context.Background() |
| 261 | seedClaim(t, s, "a.customer.test", "kass") |
| 262 | var c *ClaimCache |
| 263 | got, err := c.List(ctx, s) |
| 264 | if err != nil || len(got) != 1 { |
| 265 | t.Fatalf("nil cache fold = %+v, %v", got, err) |
| 266 | } |
| 267 | c.Reset() // must not panic |
| 268 | } |
| 269 | |
| 270 | // A transient read failure must not be remembered. Remembering it would |
| 271 | // hide a perfectly good claim until something moved its bucket stamp — and |
| 272 | // nothing edits a settled claim, so for that hostname "until" means never. |
| 273 | // A decode refusal IS remembered, because the same bytes refuse the same way |
| 274 | // every pass and re-reading them forever buys nothing. |
| 275 | func TestClaimCacheNeverRemembersATransientReadFailure(t *testing.T) { |
| 276 | dir := t.TempDir() |
| 277 | inner := release.NewDirStore(dir) |
| 278 | s := &flakyStore{Store: inner} |
| 279 | ctx := context.Background() |
| 280 | seedClaim(t, inner, "a.customer.test", "kass") |
| 281 | backdate(t, dir, "a.customer.test") |
| 282 | |
| 283 | var c ClaimCache |
| 284 | s.failNext = true |
| 285 | got, err := c.List(ctx, s) |
| 286 | if err != nil { |
| 287 | t.Fatal(err) |
| 288 | } |
| 289 | if len(got) != 0 { |
| 290 | t.Fatalf("fold during the blip = %+v, want nothing readable", got) |
| 291 | } |
| 292 | // The blip is over. The claim must come back on the very next fold, |
| 293 | // with no stamp change to prompt it. |
| 294 | got, err = c.List(ctx, s) |
| 295 | if err != nil { |
| 296 | t.Fatal(err) |
| 297 | } |
| 298 | if len(got) != 1 || got[0].Host != "a.customer.test" { |
| 299 | t.Fatalf("after the blip = %+v, want the claim back — a read failure was cached", got) |
| 300 | } |
| 301 | } |
| 302 | |
| 303 | type flakyStore struct { |
| 304 | release.Store |
| 305 | failNext bool |
| 306 | } |
| 307 | |
| 308 | func (f *flakyStore) Get(ctx context.Context, key string) ([]byte, string, error) { |
| 309 | if f.failNext { |
| 310 | f.failNext = false |
| 311 | return nil, "", errFlaky |
| 312 | } |
| 313 | return f.Store.Get(ctx, key) |
| 314 | } |
| 315 | |
| 316 | func (f *flakyStore) ListMod(ctx context.Context, prefix string) ([]release.KeyMod, error) { |
| 317 | return f.Store.(release.ModLister).ListMod(ctx, prefix) |
| 318 | } |
| 319 | |
| 320 | var errFlaky = errors.New("bucket blip") |
| 321 | |
| 322 | // An object that will never decode is remembered, so a deployment carrying |
| 323 | // one piece of litter under domains/ does not pay a GET for it on every |
| 324 | // fold, on every box, forever. |
| 325 | func TestClaimCacheRemembersAnUndecodableObject(t *testing.T) { |
| 326 | dir := t.TempDir() |
| 327 | s := &countingStore{Store: release.NewDirStore(dir)} |
| 328 | ctx := context.Background() |
| 329 | if _, err := s.Put(ctx, ClaimKey("junk.customer.test"), []byte("not json"), ""); err != nil { |
| 330 | t.Fatal(err) |
| 331 | } |
| 332 | backdate(t, dir, "junk.customer.test") |
| 333 | var c ClaimCache |
| 334 | if got, err := c.List(ctx, s); err != nil || len(got) != 0 { |
| 335 | t.Fatalf("fold = %+v, %v; want nothing", got, err) |
| 336 | } |
| 337 | s.gets = 0 |
| 338 | if _, err := c.List(ctx, s); err != nil { |
| 339 | t.Fatal(err) |
| 340 | } |
| 341 | if s.gets != 0 { |
| 342 | t.Errorf("spent %d gets re-reading an object that will never decode", s.gets) |
| 343 | } |
| 344 | } |
| 345 | |
| 346 | // The entry map is replaced whole rather than merged, so a deployment that |
| 347 | // churns through hostnames does not accumulate an entry per hostname it has |
| 348 | // ever held. Reaching into the field is white-box on purpose: there is no |
| 349 | // behaviour to observe here, only unbounded growth, and unbounded growth |
| 350 | // with no test is how a leak reaches production. |
| 351 | func TestClaimCacheDoesNotGrowWithChurnedHostnames(t *testing.T) { |
| 352 | dir := t.TempDir() |
| 353 | s := release.NewDirStore(dir) |
| 354 | ctx := context.Background() |
| 355 | var c ClaimCache |
| 356 | for i, host := range []string{"one.customer.test", "two.customer.test", "three.customer.test"} { |
| 357 | seedClaim(t, s, host, "kass") |
| 358 | backdate(t, dir, host) |
| 359 | if _, err := c.List(ctx, s); err != nil { |
| 360 | t.Fatal(err) |
| 361 | } |
| 362 | if _, err := DetachClaim(ctx, s, host, "acct", "kass"); err != nil { |
| 363 | t.Fatal(err) |
| 364 | } |
| 365 | if _, err := c.List(ctx, s); err != nil { |
| 366 | t.Fatal(err) |
| 367 | } |
| 368 | c.mu.Lock() |
| 369 | n := len(c.entries) |
| 370 | c.mu.Unlock() |
| 371 | if n != 0 { |
| 372 | t.Fatalf("after churning %d hostname(s) the cache holds %d entries, want 0", i+1, n) |
| 373 | } |
| 374 | } |
| 375 | } |
| 376 | |
amadan