| 1 | package release |
| 2 | |
| 3 | import ( |
| 4 | "fmt" |
| 5 | "math/rand" |
| 6 | "os" |
| 7 | "sort" |
| 8 | "strconv" |
| 9 | "strings" |
| 10 | "sync" |
| 11 | "testing" |
| 12 | |
| 13 | "github.com/carlosframework/platform/internal/registry" |
| 14 | ) |
| 15 | |
| 16 | // TestUpdateRequestLifecycleSimulation runs random schedules of one |
| 17 | // instance's update life across three boxes (settings plan 4 Task 11): the |
| 18 | // channel head moves and bakes, the member presses "Update now", pins, |
| 19 | // unpins and holds, root and agent passes decide and repoint, wakes run |
| 20 | // builds against the database, boxes are lost and the instance moves. |
| 21 | // |
| 22 | // The decisions are production's: rootPass calls Target and NextConsumed; |
| 23 | // wake calls DecideHeldWake (the held wake check, ruling R-HELD-PURE) and |
| 24 | // enforces it as the activator's wakeCheck does (refuse, run the target, or |
| 25 | // wait for root), then HeldExecStale and NextBuildState (beforeExec, ruling |
| 26 | // R-MERGE); a graceful move calls CarryRequest (ruling R-SPEC-43A-2), and |
| 27 | // root's write of its record passes publishGate, the re-check root makes |
| 28 | // under the carry fence (ruling R-CARRY-FENCE). What |
| 29 | // surrounds them is not pure (settingsTarget's file and registry reads, |
| 30 | // adoptHost's bake, evidence and repoint, beforeExec's bucket write, the |
| 31 | // wake's wait) and is mirrored here in the smallest form that keeps its |
| 32 | // decisions. |
| 33 | // |
| 34 | // The invariants are checked against the model's own records, never against |
| 35 | // production helpers (a mutant of Watermark must not also move the oracle): |
| 36 | // |
| 37 | // 1. A completed request never runs again: once the watermark is >= R, no |
| 38 | // box resolves R again, rule 2 never selects R again, and no wake |
| 39 | // records R as newly completed. A wake that completes R does so because |
| 40 | // R's T ran, and a wake that runs the build state's T for a pending R |
| 41 | // completes it. The watermark is spec §4.3 rule 2's (ruling R-INV1): the |
| 42 | // highest completed id in the bucket's build state (the true object, |
| 43 | // whatever the box's copy says) or in the consumed file of the box doing |
| 44 | // the pass, and also every failed completion a box had recorded when the |
| 45 | // instance left it by a graceful move (ruling R-INV1-2: the move carries |
| 46 | // it; only a lost box's records are exempt). UPDATESIM_STRICT=1 reads |
| 47 | // every box's consumed file at once instead; that is documentation of a |
| 48 | // known gap, not the gate (below). |
| 49 | // 2. The build state names the build that last touched the database. |
| 50 | // 3. No install lacks evidence. At the decision: Target names pointer |
| 51 | // evidence only for the head, and asks for bake only on the head. At |
| 52 | // the repoint: the build is the head, has a promotion record, or is the |
| 53 | // box's own staged copy; that half holds by construction of the |
| 54 | // adoptHost mirror (its evidence switch), and is kept as a tripwire. |
| 55 | // 4. Failure records outrank: no repoint a Target decision makes for a |
| 56 | // route with settings lands on a build the box holds as failed (the |
| 57 | // model never sets GoodVersion, so no instance has served a failed |
| 58 | // build). A route with default settings keeps today's adoption, which |
| 59 | // never reads the book, consumed file or not (rulings R-T6-BOOK, |
| 60 | // R-T6-BOOK-2; adopt.go settingsTarget's !plain), so it is out of scope. |
| 61 | // 5. Unresolved requests block nothing: a held instance with an unresolved |
| 62 | // request whose head is baking recovers onto last_executed; and a held |
| 63 | // instance (no pin) never repoints anywhere but last_executed, a T its |
| 64 | // pending request was resolved to, or, never having run, the head. |
| 65 | // 6. A request resolves at its first eligible pass, pin or not: the world |
| 66 | // records firstEligible[R], the head at the first root pass where R is |
| 67 | // pending, the build state is read (R-TARGET-UNKNOWN) and the head is |
| 68 | // eligible on that box (or the request asked for fast track). That pass |
| 69 | // must resolve R to it; every resolution of R a box records, and every |
| 70 | // rule 2 install of R, names exactly that build. A resolution the |
| 71 | // instance can no longer see was never run (a run completes R in the |
| 72 | // build state), so it is spec §4.2's "Lost requests": firstEligible[R] |
| 73 | // is cleared and the next eligible pass resolves R again. A record the |
| 74 | // instance sees again (it returned to the box holding it) is R's |
| 75 | // resolution once more; when the build state and the box's file both |
| 76 | // hold one, the build state's is (ruling R-RESOLUTION-STATE-WINS), and |
| 77 | // root's next pass with the state read rewrites the file to agree. |
| 78 | // 7. A wake runs only what the rules allow (§4.3 rule 3, "The wake checks |
| 79 | // too"; ruling R-HELD-PURE), from the model's own records: a pinned |
| 80 | // instance runs its pin; one with a pending request resolved to T runs |
| 81 | // T; a held one runs the bucket's last_executed. Held is manual, or a |
| 82 | // major setting whose head, as of root's last decision on this box, has |
| 83 | // no tag or another major, or a major setting root has not decided for |
| 84 | // yet (a missing or stale decision never makes a wake unheld). |
| 85 | // |
| 86 | // Copies (the agent's registry copy root and the wake read) go stale the ways |
| 87 | // production keeps them stale: a refresh that fails keeps the last known |
| 88 | // state and records ReadErr; a box the instance returns to may still hold its |
| 89 | // row from the earlier stay, copy and all, its refreshes failed while the |
| 90 | // instance was away. Root reads such a copy as unread (BuildStateCopy.Read). |
| 91 | // The wake refreshes first and is refused when that fails (Astra round 1 |
| 92 | // finding 1). |
| 93 | // |
| 94 | // Strict mode (UPDATESIM_STRICT=1) checks invariants 1 and 6 across every |
| 95 | // box's consumed file, not only the one the instance is on. It is not the |
| 96 | // gate (ruling R-INV1): a LOST box's records are the spec's lost request |
| 97 | // (§4.2), and strict mode counts them. |
| 98 | // |
| 99 | // Seeds: UPDATESIM_SEEDS (default 200), spread over UPDATESIM_PAR goroutines |
| 100 | // (default 1); each seed has its own rand source and world. UPDATESIM_SEED=n |
| 101 | // replays seed n alone. A failure prints the seed, the step and the trace, |
| 102 | // and a line "INVARIANT <n> ..." per invariant any seed violated. |
| 103 | func TestUpdateRequestLifecycleSimulation(t *testing.T) { |
| 104 | t.Run("seed 531 shape, unpinned", testUpdateSimSeed531Unpinned) |
| 105 | t.Run("carry fence, a failed completion racing a cut", testUpdateSimCarryFence) |
| 106 | t.Run("a file and the build state disagree on R1", testUpdateSimDisagreement) |
| 107 | seeds := updateSimEnv(t, "UPDATESIM_SEEDS", 200) |
| 108 | par := updateSimEnv(t, "UPDATESIM_PAR", 1) |
| 109 | first := 0 |
| 110 | if v := os.Getenv("UPDATESIM_SEED"); v != "" { |
| 111 | n, err := strconv.Atoi(v) |
| 112 | if err != nil { |
| 113 | t.Fatal(err) |
| 114 | } |
| 115 | first, seeds = n, 1 |
| 116 | } |
| 117 | if par < 1 { |
| 118 | par = 1 |
| 119 | } |
| 120 | strict := os.Getenv("UPDATESIM_STRICT") == "1" |
| 121 | results := make([]updateSimResult, seeds) |
| 122 | jobs := make(chan int) |
| 123 | var wg sync.WaitGroup |
| 124 | for i := 0; i < par; i++ { |
| 125 | wg.Add(1) |
| 126 | go func() { |
| 127 | defer wg.Done() |
| 128 | for i := range jobs { |
| 129 | results[i] = runUpdateSim(int64(first+i), updateSimSteps, strict) |
| 130 | } |
| 131 | }() |
| 132 | } |
| 133 | for i := 0; i < seeds; i++ { |
| 134 | jobs <- i |
| 135 | } |
| 136 | close(jobs) |
| 137 | wg.Wait() |
| 138 | |
| 139 | var total updateSimStats |
| 140 | byInv := map[int][]*updateSimFailure{} |
| 141 | for i := range results { |
| 142 | total.add(results[i].stats) |
| 143 | if f := results[i].fail; f != nil { |
| 144 | seen := map[int]bool{} |
| 145 | for _, v := range f.viols { |
| 146 | if !seen[v.inv] { |
| 147 | seen[v.inv] = true |
| 148 | byInv[v.inv] = append(byInv[v.inv], f) |
| 149 | } |
| 150 | } |
| 151 | } |
| 152 | } |
| 153 | t.Logf("%d seeds x %d steps (par %d, strict %v): %s", seeds, updateSimSteps, par, strict, total) |
| 154 | invs := make([]int, 0, len(byInv)) |
| 155 | for inv := range byInv { |
| 156 | invs = append(invs, inv) |
| 157 | } |
| 158 | sort.Ints(invs) |
| 159 | for _, inv := range invs { |
| 160 | fs := byInv[inv] |
| 161 | f := fs[0] |
| 162 | var msg string |
| 163 | for _, v := range f.viols { |
| 164 | if v.inv == inv { |
| 165 | msg = v.msg |
| 166 | break |
| 167 | } |
| 168 | } |
| 169 | t.Errorf("INVARIANT %d (%s) violated in %d of %d seeds; first: seed %d step %d (%s): %s\ntrace (replay with UPDATESIM_SEED=%d):\n%s", |
| 170 | inv, updateSimInvariants[inv], len(fs), seeds, f.seed, f.step, f.action, msg, f.seed, strings.Join(f.trace, "\n")) |
| 171 | } |
| 172 | } |
| 173 | |
| 174 | // testUpdateSimSeed531Unpinned replays the shape seed 531 found, through |
| 175 | // the model's own actions (so every invariant is checked at every step) and |
| 176 | // so through the production Target, NextConsumed, DecideHeldWake, |
| 177 | // CarryRequest, publishGate and NextBuildState. Box 0 runs a root pass that |
| 178 | // was already running when the cut began; before the carry fence its write |
| 179 | // landed after the cut carried box 0's record and stayed in box 0's file |
| 180 | // only. Under the fence (ruling R-CARRY-FENCE) it comes after the carry, so |
| 181 | // it meets the cut's hold and writes nothing. Box 1 resolves R1, to b3, |
| 182 | // under a pin set after the request, and its wake (running the pin) carries |
| 183 | // {R1, b3} into the build state; the instance returns to box 0, which holds |
| 184 | // no record of R1, and the pin is cleared. Rule 2 must then install the |
| 185 | // build state's b3, and the wake that runs it must complete R1. |
| 186 | func testUpdateSimSeed531Unpinned(t *testing.T) { |
| 187 | w := newUpdateSimWorld(0, false) |
| 188 | w.scripted = true |
| 189 | for i := range w.boxes { |
| 190 | w.boxes[i] = newSimBox(fmt.Sprintf("box%d", i)) |
| 191 | } |
| 192 | w.builds = []string{"b1", "b2", "b3"} |
| 193 | for i, b := range w.builds { |
| 194 | w.tags[b], w.major[b], w.promoted[b] = fmt.Sprintf("1.%d.0", i+1), 1, true |
| 195 | } |
| 196 | w.head = "b2" |
| 197 | for _, b := range w.boxes[:2] { |
| 198 | b.eligible["b1"], b.eligible["b2"] = true, true |
| 199 | } |
| 200 | w.state = &BuildState{Schema: 1, Host: updateSimHost, LastExecuted: "b1"} |
| 201 | w.lastRan, w.replica = "b1", true |
| 202 | w.at = 0 |
| 203 | w.boxes[0].live, w.boxes[0].staged["b1"] = "b1", true |
| 204 | w.boxes[0].copy = BuildStateCopy{Known: true, Replica: true, State: cloneSimState(w.state)} |
| 205 | w.settings = UpdateSettings{Updates: UpdatesManual, InstallRequest: 1} |
| 206 | |
| 207 | steps := []func() string{ |
| 208 | func() string { w.settings.Pin = "b1"; return "pin b1 (after the request)" }, |
| 209 | func() string { |
| 210 | return w.moveToRacing(1, true, func() string { return w.rootPass(w.boxes[0], true) }, true) |
| 211 | }, |
| 212 | func() string { w.head = "b3"; w.boxes[1].eligible["b3"] = true; return "promote b3, baked on box1" }, |
| 213 | func() string { return w.rootPass(w.boxes[1], true) }, |
| 214 | func() string { return w.wake(w.boxes[1]) }, |
| 215 | func() string { return w.moveTo(0, true) }, |
| 216 | func() string { w.settings.Pin = ""; return "unpin" }, |
| 217 | func() string { return w.rootPass(w.boxes[0], true) }, |
| 218 | func() string { return w.wake(w.boxes[0]) }, |
| 219 | } |
| 220 | if !runScriptedSim(t, "the scripted seed-531 shape", w, steps) { |
| 221 | return |
| 222 | } |
| 223 | if c := w.boxes[0].consumed; c != nil && c.Resolved != nil && c.Resolved.Target == "b2" { |
| 224 | t.Errorf("scripted seed-531 shape: box0's racing pass published {R1, b2} under the cut's hold\ntrace:\n%s", strings.Join(w.trace, "\n")) |
| 225 | } |
| 226 | if st := w.state; st == nil || st.Completed != 1 || st.CompletedFailed || w.boxes[0].live != "b3" { |
| 227 | t.Errorf("scripted seed-531 shape: after the unpinned pass and wake on box0, live %q, build state %+v; want b3 live and R1 completed by it\ntrace:\n%s", |
| 228 | w.boxes[0].live, w.state, strings.Join(w.trace, "\n")) |
| 229 | } |
| 230 | t.Logf("trace:\n%s", strings.Join(w.trace, "\n")) |
| 231 | } |
| 232 | |
| 233 | // runScriptedSim runs a scripted shape's steps, checking every invariant |
| 234 | // after each; false when one failed (reported). |
| 235 | func runScriptedSim(t *testing.T, name string, w *simWorld, steps []func() string) bool { |
| 236 | t.Helper() |
| 237 | for i, step := range steps { |
| 238 | w.viols = nil |
| 239 | action := step() |
| 240 | w.trace = append(w.trace, fmt.Sprintf("%3d %s %s", i+1, action, w.summary())) |
| 241 | w.check() |
| 242 | for _, v := range w.viols { |
| 243 | t.Errorf("INVARIANT %d (%s) violated in %s at step %d (%s): %s\ntrace:\n%s", |
| 244 | v.inv, updateSimInvariants[v.inv], name, i+1, action, v.msg, strings.Join(w.trace, "\n")) |
| 245 | } |
| 246 | if len(w.viols) > 0 { |
| 247 | return false |
| 248 | } |
| 249 | } |
| 250 | return true |
| 251 | } |
| 252 | |
| 253 | // testUpdateSimDisagreement gates resolvedFor's order (ruling |
| 254 | // R-RESOLUTION-STATE-WINS; ruling R-M9-SCRIPTED) on a state built directly: |
| 255 | // box 0's consumed file resolves the pending R1 to b1 while the build state |
| 256 | // resolves the same R1 to b3, unpinned and manual. Under the carry fence no |
| 257 | // graceful path in this model leaves that state behind any more. It is still |
| 258 | // reachable in production from a box running a pre-fence binary (a racing |
| 259 | // root pass wrote its record after the cut's carry), or from a registry copy |
| 260 | // that stays stale for one reconcile without a read error (root resolves R1 |
| 261 | // again from a copy that lacks the state's resolution). Root's rule 2 must |
| 262 | // install the build state's b3 (through the production Target and |
| 263 | // NextConsumed), and the wake that runs it must complete R1 (DecideHeldWake, |
| 264 | // HeldExecStale and NextBuildState). Following the file would run b1, which |
| 265 | // never completes R1. |
| 266 | func testUpdateSimDisagreement(t *testing.T) { |
| 267 | w := newUpdateSimWorld(0, false) |
| 268 | w.scripted = true |
| 269 | for i := range w.boxes { |
| 270 | w.boxes[i] = newSimBox(fmt.Sprintf("box%d", i)) |
| 271 | } |
| 272 | w.builds = []string{"b1", "b2", "b3"} |
| 273 | for i, b := range w.builds { |
| 274 | w.tags[b], w.major[b], w.promoted[b] = fmt.Sprintf("1.%d.0", i+1), 1, true |
| 275 | } |
| 276 | w.head = "b3" |
| 277 | for _, b := range w.boxes { |
| 278 | b.eligible["b1"], b.eligible["b2"], b.eligible["b3"] = true, true, true |
| 279 | } |
| 280 | w.state = &BuildState{Schema: 1, Host: updateSimHost, LastExecuted: "b2", Resolved: &ResolvedRequest{ID: 1, Target: "b3"}} |
| 281 | w.lastRan, w.replica = "b2", true |
| 282 | w.at = 0 |
| 283 | b0 := w.boxes[0] |
| 284 | b0.live, b0.staged["b2"] = "b2", true |
| 285 | b0.copy = BuildStateCopy{Known: true, Replica: true, State: cloneSimState(w.state)} |
| 286 | b0.consumed = &Consumed{Schema: 1, Host: updateSimHost, Resolved: &ResolvedRequest{ID: 1, Target: "b1"}} |
| 287 | w.settings = UpdateSettings{Updates: UpdatesManual, InstallRequest: 1} |
| 288 | // R1's resolution as the world records it: the build state's |
| 289 | // (R-RESOLUTION-STATE-WINS), so invariant 6 judges installs against b3. |
| 290 | w.syncResolutions() |
| 291 | steps := []func() string{ |
| 292 | func() string { return w.rootPass(b0, true) }, |
| 293 | func() string { return w.wake(b0) }, |
| 294 | } |
| 295 | const name = "the scripted disagreement (file {R1, b1}, build state {R1, b3})" |
| 296 | // The outcome is asserted even when an invariant stopped the steps early. |
| 297 | runScriptedSim(t, name, w, steps) |
| 298 | if st := w.state; b0.live != "b3" || st == nil || st.LastExecuted != "b3" || st.Completed != 1 || st.CompletedFailed { |
| 299 | t.Errorf("%s: live %q, build state %+v; want rule 2 to install the build state's b3 and its wake to complete R1\ntrace:\n%s", |
| 300 | name, b0.live, st, strings.Join(w.trace, "\n")) |
| 301 | } |
| 302 | t.Logf("%s, trace:\n%s", name, strings.Join(w.trace, "\n")) |
| 303 | } |
| 304 | |
| 305 | // testUpdateSimCarryFence is Astra round 2 finding 2's shape (ruling |
| 306 | // R-CARRY-FENCE), scripted both ways. Box 0 resolves R1 to b2 and installs |
| 307 | // it; b2 then fails on box 0, so root's next pass there records R1 |
| 308 | // completed, failed. That pass races a graceful cut to box 1. Early (its |
| 309 | // fence first) it publishes, and the cut carries the failed completion: box |
| 310 | // 1 never selects R1. Late (the cut's fence first) it meets the cut's hold |
| 311 | // and publishes nothing: R1 is still pending with T b2, which box 1 does not |
| 312 | // hold as failed, so box 1 installs b2 and its wake completes R1. A late |
| 313 | // record published past the hold (the fence's re-check removed) is left |
| 314 | // behind on box 0, and box 1 selects R1 after it completed (invariant 1). |
| 315 | func testUpdateSimCarryFence(t *testing.T) { |
| 316 | for _, late := range []bool{false, true} { |
| 317 | name := "the carry fence, root's pass before the hold" |
| 318 | if late { |
| 319 | name = "the carry fence, root's pass after the carry" |
| 320 | } |
| 321 | w := newUpdateSimWorld(0, false) |
| 322 | w.scripted = true |
| 323 | for i := range w.boxes { |
| 324 | w.boxes[i] = newSimBox(fmt.Sprintf("box%d", i)) |
| 325 | } |
| 326 | w.builds = []string{"b1", "b2"} |
| 327 | for i, b := range w.builds { |
| 328 | w.tags[b], w.major[b], w.promoted[b] = fmt.Sprintf("1.%d.0", i+1), 1, true |
| 329 | } |
| 330 | w.head = "b2" |
| 331 | for _, b := range w.boxes { |
| 332 | b.eligible["b1"], b.eligible["b2"] = true, true |
| 333 | } |
| 334 | w.state = &BuildState{Schema: 1, Host: updateSimHost, LastExecuted: "b1"} |
| 335 | w.lastRan, w.replica = "b1", true |
| 336 | w.at = 0 |
| 337 | w.boxes[0].live, w.boxes[0].staged["b1"] = "b1", true |
| 338 | w.boxes[0].copy = BuildStateCopy{Known: true, Replica: true, State: cloneSimState(w.state)} |
| 339 | w.settings = UpdateSettings{Updates: UpdatesManual, InstallRequest: 1} |
| 340 | steps := []func() string{ |
| 341 | func() string { return w.rootPass(w.boxes[0], true) }, |
| 342 | func() string { w.boxes[0].failed["b2"] = true; return "mark-failed b2 on box0" }, |
| 343 | func() string { |
| 344 | return w.moveToRacing(1, true, func() string { return w.rootPass(w.boxes[0], true) }, late) |
| 345 | }, |
| 346 | func() string { return w.rootPass(w.boxes[1], true) }, |
| 347 | func() string { return w.wake(w.boxes[1]) }, |
| 348 | } |
| 349 | if !runScriptedSim(t, name, w, steps) { |
| 350 | continue |
| 351 | } |
| 352 | st, c := w.state, w.boxes[0].consumed |
| 353 | switch { |
| 354 | case !late && (st == nil || st.Completed != 1 || !st.CompletedFailed || w.boxes[1].live == "b2"): |
| 355 | t.Errorf("%s: build state %+v, box1 live %q; want R1 carried completed, failed, and b2 never installed on box1\ntrace:\n%s", |
| 356 | name, st, w.boxes[1].live, strings.Join(w.trace, "\n")) |
| 357 | case late && (c == nil || c.Completed != 0 || st == nil || st.Completed != 1 || st.CompletedFailed || w.boxes[1].live != "b2"): |
| 358 | t.Errorf("%s: box0's file %+v, build state %+v, box1 live %q; want nothing published on box0, b2 installed on box1 and R1 completed by its run\ntrace:\n%s", |
| 359 | name, c, st, w.boxes[1].live, strings.Join(w.trace, "\n")) |
| 360 | } |
| 361 | t.Logf("%s, trace:\n%s", name, strings.Join(w.trace, "\n")) |
| 362 | } |
| 363 | } |
| 364 | |
| 365 | const ( |
| 366 | updateSimHost = "blog.example.com" |
| 367 | updateSimSteps = 80 |
| 368 | updateSimBoxes = 3 |
| 369 | ) |
| 370 | |
| 371 | var updateSimInvariants = map[int]string{ |
| 372 | 0: "model consistency", |
| 373 | 1: "a completed request never runs again", |
| 374 | 2: "the build state names the build that last touched the database", |
| 375 | 3: "no install lacks evidence", |
| 376 | 4: "failure records outrank", |
| 377 | 5: "unresolved requests block nothing", |
| 378 | 6: "a request resolves at its first eligible pass, pin or not", |
| 379 | 7: "a wake runs only what the rules allow", |
| 380 | } |
| 381 | |
| 382 | func updateSimEnv(t *testing.T, name string, def int) int { |
| 383 | v := os.Getenv(name) |
| 384 | if v == "" { |
| 385 | return def |
| 386 | } |
| 387 | n, err := strconv.Atoi(v) |
| 388 | if err != nil || n < 0 { |
| 389 | t.Fatalf("%s=%q: want a count", name, v) |
| 390 | } |
| 391 | return n |
| 392 | } |
| 393 | |
| 394 | // simBox is one box's view of the instance: root's consumed file, the live |
| 395 | // link, the failure book, the version store's staged copies, the bake clock |
| 396 | // (versions whose window has run here) and the agent's registry copy of the |
| 397 | // build state, which is what root reads. |
| 398 | type simBox struct { |
| 399 | name string |
| 400 | // status is the row's status on this box: "" (ready), or "moving" from |
| 401 | // the moment a cut holds it until the hand-over. |
| 402 | status string |
| 403 | consumed *Consumed |
| 404 | live string |
| 405 | failed map[string]bool |
| 406 | staged map[string]bool |
| 407 | eligible map[string]bool |
| 408 | copy BuildStateCopy |
| 409 | // The model's own record of root's last decision on this box (for |
| 410 | // invariant 7, never Consumed.Settings): the settings column it decided |
| 411 | // for and the head it saw. decided is false until root's first pass. |
| 412 | decided bool |
| 413 | decidedSettings string |
| 414 | decidedHead string |
| 415 | } |
| 416 | |
| 417 | // simWorld is the instance and everything around it. state is the bucket |
| 418 | // object (nil until the first pre-exec write). replica is whether the |
| 419 | // instance has a database (a legacy instance has one with no build state). |
| 420 | type simWorld struct { |
| 421 | boxes []*simBox |
| 422 | at int |
| 423 | state *BuildState |
| 424 | replica bool |
| 425 | settings UpdateSettings |
| 426 | head string |
| 427 | builds []string |
| 428 | tags map[string]string |
| 429 | major map[string]int // 0: the build has no semver tag |
| 430 | promoted map[string]bool |
| 431 | lastRan string |
| 432 | rng *rand.Rand |
| 433 | strict bool |
| 434 | |
| 435 | firstEligible map[uint64]string |
| 436 | completedByRun map[uint64]bool |
| 437 | // carried is the highest failed completion any box had recorded when |
| 438 | // the instance left it by a graceful move (ruling R-INV1-2): invariant |
| 439 | // 1's watermark includes it. Only a lost box's records are exempt. |
| 440 | carried uint64 |
| 441 | // scripted turns off the random faults (a failed refresh, a wait root |
| 442 | // does not answer), for the scripted shapes. |
| 443 | scripted bool |
| 444 | |
| 445 | trace []string |
| 446 | viols []updateSimViolation |
| 447 | stats updateSimStats |
| 448 | } |
| 449 | |
| 450 | type updateSimViolation struct { |
| 451 | inv int |
| 452 | msg string |
| 453 | } |
| 454 | |
| 455 | type updateSimFailure struct { |
| 456 | seed int64 |
| 457 | step int |
| 458 | action string |
| 459 | viols []updateSimViolation |
| 460 | trace []string |
| 461 | } |
| 462 | |
| 463 | type updateSimResult struct { |
| 464 | fail *updateSimFailure |
| 465 | stats updateSimStats |
| 466 | } |
| 467 | |
| 468 | // updateSimStats counts the scenarios the invariants need, so a clean run |
| 469 | // shows they were reached. |
| 470 | type updateSimStats struct { |
| 471 | resolutions, runCompletions, failedCompletions int |
| 472 | failedHeadResolutions int // R-TARGET-FAILEDHEAD |
| 473 | heldRecoveries int // invariant 5's scenario, repointing onto last_executed |
| 474 | heldBakingPasses int // invariant 5's scenario, any |
| 475 | unknownWithCompleted int // Known false, the current request already completed (R-TARGET-UNKNOWN) |
| 476 | unknownReturning int // Known false on a box whose consumed file names the current request |
| 477 | pinnedEligible int // invariant 6 under a pin |
| 478 | lostResolutions int |
| 479 | pickedUp int // a move or return made another record of R the one in force |
| 480 | leftBehind int // a resolution or rule 2 pick strict mode would refuse |
| 481 | moves, returns int |
| 482 | carriedRequests int // graceful moves whose cut carried a request record |
| 483 | carriedFailed int // ... a failed completion among them |
| 484 | unreadRefusals int // wakes refused: their refresh failed |
| 485 | askRoot int // wakes with no decision from root for the settings |
| 486 | askRootRuns int // ... that ran the installed build at once |
| 487 | rootWaits int // wakes that asked root and waited |
| 488 | heldRuns int // wakes that ran a held target |
| 489 | retainedRuns int // ... from a retained copy |
| 490 | keptCopies int // returns onto a kept, possibly stale, copy |
| 491 | racingEarly int // root passes racing a cut that published before its carry |
| 492 | racingFenced int // ... that met the move hold under the fence: nothing published |
| 493 | } |
| 494 | |
| 495 | func (s *updateSimStats) add(o updateSimStats) { |
| 496 | s.resolutions += o.resolutions |
| 497 | s.runCompletions += o.runCompletions |
| 498 | s.failedCompletions += o.failedCompletions |
| 499 | s.failedHeadResolutions += o.failedHeadResolutions |
| 500 | s.heldRecoveries += o.heldRecoveries |
| 501 | s.heldBakingPasses += o.heldBakingPasses |
| 502 | s.unknownWithCompleted += o.unknownWithCompleted |
| 503 | s.unknownReturning += o.unknownReturning |
| 504 | s.pinnedEligible += o.pinnedEligible |
| 505 | s.lostResolutions += o.lostResolutions |
| 506 | s.pickedUp += o.pickedUp |
| 507 | s.leftBehind += o.leftBehind |
| 508 | s.moves += o.moves |
| 509 | s.returns += o.returns |
| 510 | s.carriedRequests += o.carriedRequests |
| 511 | s.carriedFailed += o.carriedFailed |
| 512 | s.unreadRefusals += o.unreadRefusals |
| 513 | s.askRoot += o.askRoot |
| 514 | s.askRootRuns += o.askRootRuns |
| 515 | s.rootWaits += o.rootWaits |
| 516 | s.heldRuns += o.heldRuns |
| 517 | s.retainedRuns += o.retainedRuns |
| 518 | s.keptCopies += o.keptCopies |
| 519 | s.racingEarly += o.racingEarly |
| 520 | s.racingFenced += o.racingFenced |
| 521 | } |
| 522 | |
| 523 | func (s updateSimStats) String() string { |
| 524 | return fmt.Sprintf("resolutions %d (failed head %d, under a pin %d), run completions %d, failed completions %d, "+ |
| 525 | "held+baking passes %d (recoveries onto last_executed %d), unread state with the request completed %d, "+ |
| 526 | "unread state on a returning box %d, lost resolutions %d, records picked up by a move %d, "+ |
| 527 | "left-behind records acted past %d, moves %d (returns %d, onto a kept copy %d), "+ |
| 528 | "cuts that carried a request record %d (a failed completion %d), wakes refused unread %d, "+ |
| 529 | "wakes with no decision for the settings %d (ran the installed build %d), wakes that waited on root %d, "+ |
| 530 | "held runs %d (from a retained copy %d), root passes racing a cut %d (published before the carry %d, fenced %d)", |
| 531 | s.resolutions, s.failedHeadResolutions, s.pinnedEligible, s.runCompletions, s.failedCompletions, |
| 532 | s.heldBakingPasses, s.heldRecoveries, s.unknownWithCompleted, s.unknownReturning, s.lostResolutions, |
| 533 | s.pickedUp, s.leftBehind, s.moves, s.returns, s.keptCopies, s.carriedRequests, s.carriedFailed, |
| 534 | s.unreadRefusals, s.askRoot, s.askRootRuns, s.rootWaits, s.heldRuns, s.retainedRuns, |
| 535 | s.racingEarly+s.racingFenced, s.racingEarly, s.racingFenced) |
| 536 | } |
| 537 | |
| 538 | func newUpdateSimWorld(seed int64, strict bool) *simWorld { |
| 539 | w := &simWorld{ |
| 540 | rng: rand.New(rand.NewSource(seed)), |
| 541 | strict: strict, |
| 542 | settings: UpdateSettings{Updates: UpdatesAuto}, |
| 543 | tags: map[string]string{}, |
| 544 | major: map[string]int{}, |
| 545 | promoted: map[string]bool{}, |
| 546 | firstEligible: map[uint64]string{}, |
| 547 | completedByRun: map[uint64]bool{}, |
| 548 | } |
| 549 | for i := 0; i < updateSimBoxes; i++ { |
| 550 | w.boxes = append(w.boxes, newSimBox(fmt.Sprintf("box%d", i))) |
| 551 | } |
| 552 | w.head = "b1" |
| 553 | w.builds = []string{"b1"} |
| 554 | w.tags["b1"], w.major["b1"], w.promoted["b1"] = "1.1.0", 1, true |
| 555 | w.boxes[0].eligible["b1"] = true |
| 556 | if w.rng.Intn(100) < 30 { |
| 557 | // An instance from before this feature: a database, a live link, no |
| 558 | // build state. |
| 559 | w.replica = true |
| 560 | w.boxes[0].live = "b1" |
| 561 | w.boxes[0].staged["b1"] = true |
| 562 | } |
| 563 | if w.rng.Intn(100) < 25 { |
| 564 | w.settings.Updates = UpdatesManual |
| 565 | } |
| 566 | return w |
| 567 | } |
| 568 | |
| 569 | func newSimBox(name string) *simBox { |
| 570 | return &simBox{name: name, failed: map[string]bool{}, staged: map[string]bool{}, eligible: map[string]bool{}} |
| 571 | } |
| 572 | |
| 573 | func runUpdateSim(seed int64, steps int, strict bool) updateSimResult { |
| 574 | w := newUpdateSimWorld(seed, strict) |
| 575 | for step := 1; step <= steps; step++ { |
| 576 | w.viols = nil |
| 577 | action := w.act() |
| 578 | w.trace = append(w.trace, fmt.Sprintf("%3d %s %s", step, action, w.summary())) |
| 579 | w.check() |
| 580 | if len(w.viols) > 0 { |
| 581 | return updateSimResult{stats: w.stats, fail: &updateSimFailure{seed: seed, step: step, action: action, viols: w.viols, trace: w.trace}} |
| 582 | } |
| 583 | } |
| 584 | return updateSimResult{stats: w.stats} |
| 585 | } |
| 586 | |
| 587 | func (w *simWorld) fail(inv int, format string, args ...any) { |
| 588 | w.viols = append(w.viols, updateSimViolation{inv: inv, msg: fmt.Sprintf(format, args...)}) |
| 589 | } |
| 590 | |
| 591 | func (w *simWorld) cur() *simBox { return w.boxes[w.at] } |
| 592 | |
| 593 | // act picks one action at random and applies it. |
| 594 | func (w *simWorld) act() string { |
| 595 | weights := []int{8, 7, 9, 22, 5, 9, 14, 4, 2, 5, 5, 5, 3} |
| 596 | sum := 0 |
| 597 | for _, x := range weights { |
| 598 | sum += x |
| 599 | } |
| 600 | n := w.rng.Intn(sum) |
| 601 | i := 0 |
| 602 | for ; n >= weights[i]; i++ { |
| 603 | n -= weights[i] |
| 604 | } |
| 605 | switch i { |
| 606 | case 0: |
| 607 | return w.updateNow() |
| 608 | case 1: |
| 609 | return w.promote() |
| 610 | case 2: |
| 611 | return w.expireBake() |
| 612 | case 3: |
| 613 | return w.rootPass(w.cur(), true) |
| 614 | case 4: |
| 615 | return w.rootPass(w.cur(), false) |
| 616 | case 5: |
| 617 | return w.refresh() |
| 618 | case 6: |
| 619 | return w.wake(w.cur()) |
| 620 | case 7: |
| 621 | return w.move() |
| 622 | case 8: |
| 623 | return w.loseBox() |
| 624 | case 9: |
| 625 | return w.markFailed() |
| 626 | case 10: |
| 627 | return w.toggleManual() |
| 628 | case 11: |
| 629 | return w.pin() |
| 630 | default: |
| 631 | return w.setMajor() |
| 632 | } |
| 633 | } |
| 634 | |
| 635 | // updateNow is the console's "Update now": a new request id. The console |
| 636 | // refuses it on a pinned instance (console apiupdates.go); a pin set while a |
| 637 | // request is pending is allowed, so the model sets pins either way. |
| 638 | func (w *simWorld) updateNow() string { |
| 639 | if w.settings.Pin != "" { |
| 640 | return "update-now refused: pinned" |
| 641 | } |
| 642 | w.settings.InstallRequest++ |
| 643 | w.settings.InstallFastTrack = w.rng.Intn(100) < 20 |
| 644 | return fmt.Sprintf("update-now R%d fast=%v", w.settings.InstallRequest, w.settings.InstallFastTrack) |
| 645 | } |
| 646 | |
| 647 | // promote moves the channel head: a new build (no box has baked it yet), or |
| 648 | // a pull-back to an earlier one, which stays eligible wherever it baked. |
| 649 | func (w *simWorld) promote() string { |
| 650 | if len(w.builds) > 1 && w.rng.Intn(100) < 15 { |
| 651 | var older []string |
| 652 | for _, b := range w.builds { |
| 653 | if b != w.head { |
| 654 | older = append(older, b) |
| 655 | } |
| 656 | } |
| 657 | w.head = older[w.rng.Intn(len(older))] |
| 658 | return "promote pull-back " + w.head |
| 659 | } |
| 660 | n := len(w.builds) + 1 |
| 661 | b := fmt.Sprintf("b%d", n) |
| 662 | w.builds = append(w.builds, b) |
| 663 | w.head = b |
| 664 | w.promoted[b] = w.rng.Intn(100) < 85 |
| 665 | switch r := w.rng.Intn(100); { |
| 666 | case r < 10: |
| 667 | w.major[b], w.tags[b] = 0, "" |
| 668 | default: |
| 669 | m := 1 |
| 670 | if r >= 75 { |
| 671 | m = 2 |
| 672 | } |
| 673 | w.major[b], w.tags[b] = m, fmt.Sprintf("%d.%d.0", m, n) |
| 674 | } |
| 675 | return fmt.Sprintf("promote %s record=%v tag=%q", b, w.promoted[b], w.tags[b]) |
| 676 | } |
| 677 | |
| 678 | func (w *simWorld) expireBake() string { |
| 679 | b := w.cur() |
| 680 | if w.rng.Intn(100) >= 60 { |
| 681 | b = w.boxes[w.rng.Intn(len(w.boxes))] |
| 682 | } |
| 683 | b.eligible[w.head] = true |
| 684 | return fmt.Sprintf("bake-expired %s on %s", w.head, b.name) |
| 685 | } |
| 686 | |
| 687 | // refresh is the agent reading the bucket object into its registry copy. |
| 688 | // One that fails keeps the last known copy and records ReadErr, as |
| 689 | // buildStates.refresh does. |
| 690 | func (w *simWorld) refresh() string { |
| 691 | b := w.cur() |
| 692 | if w.refreshCopy(b) { |
| 693 | return "refresh " + b.name |
| 694 | } |
| 695 | return "refresh " + b.name + " FAILED (copy kept, read error)" |
| 696 | } |
| 697 | |
| 698 | // refreshCopy is buildStates.refresh on b: true when the read succeeded. |
| 699 | func (w *simWorld) refreshCopy(b *simBox) bool { |
| 700 | if !w.scripted && w.rng.Intn(100) < 15 { |
| 701 | b.copy.ReadErr = "s3: timeout" |
| 702 | return false |
| 703 | } |
| 704 | b.copy = BuildStateCopy{Known: true, Replica: w.replica, State: cloneSimState(w.state)} |
| 705 | return true |
| 706 | } |
| 707 | |
| 708 | // move hands the instance to another box with its replica and build state: |
| 709 | // a graceful move (a legacy cut, a Phase 1 cut, a rollback's return). The |
| 710 | // destination's registry row is new (copy unread) or carries the source's |
| 711 | // column (registry coldhandover copies build_state); a box the instance |
| 712 | // returns to may still hold its row from the earlier stay, old copy and all |
| 713 | // (a rolled-back move's source keeps its row). It still has its old consumed |
| 714 | // file and live link either way. |
| 715 | // |
| 716 | // A third of moves race a root pass on the source that was already running |
| 717 | // when the cut began (Astra round 2 finding 2). Under the carry fence it |
| 718 | // either published before the cut's carry, which takes its record, or meets |
| 719 | // the cut's hold and publishes nothing (moveToRacing). |
| 720 | func (w *simWorld) move() string { |
| 721 | j := w.rng.Intn(len(w.boxes) - 1) |
| 722 | if j >= w.at { |
| 723 | j++ |
| 724 | } |
| 725 | carried := w.rng.Intn(100) < 50 |
| 726 | if w.rng.Intn(100) < 33 { |
| 727 | src := w.cur() |
| 728 | return w.moveToRacing(j, carried, func() string { return w.rootPass(src, true) }, w.rng.Intn(100) < 50) |
| 729 | } |
| 730 | return w.moveTo(j, carried) |
| 731 | } |
| 732 | |
| 733 | // moveTo moves the instance to box j; carried is whether the destination's |
| 734 | // registry row carries the source's build-state copy. |
| 735 | func (w *simWorld) moveTo(j int, carried bool) string { |
| 736 | return w.moveToRacing(j, carried, nil, false) |
| 737 | } |
| 738 | |
| 739 | // moveToRacing is moveTo with race, when set, a root pass on the source that |
| 740 | // was already running when the cut began, modelled through the carry fence |
| 741 | // (ruling R-CARRY-FENCE). Root's publication and the cut's carry serialize |
| 742 | // on one lock, the cut takes it only once its hold is effective, and root |
| 743 | // re-checks the hold under it (publishGate, in rootPass). So the race runs |
| 744 | // either before the hold (early: its fence came first; it publishes, and |
| 745 | // the carry below takes the record) or after the carry (late: its fence |
| 746 | // came second; it meets the hold and publishes nothing). The cut holds the |
| 747 | // row ("moving"), then carries the source's request record into the build |
| 748 | // state under the lease (release.CarryRequest; ruling R-SPEC-43A-2), and the |
| 749 | // source's registry copy follows the write, as buildStates.write sets it. |
| 750 | // Invariant 1's carried watermark is the source's file at the hand-over, |
| 751 | // after both (Astra round 2's simulation note: never sampled before a |
| 752 | // racing callback). |
| 753 | func (w *simWorld) moveToRacing(j int, carried bool, race func() string, late bool) string { |
| 754 | src := w.cur() |
| 755 | dst := w.boxes[j] |
| 756 | note := "" |
| 757 | if race != nil && !late { |
| 758 | note += " | racing (before the hold) " + race() |
| 759 | w.stats.racingEarly++ |
| 760 | } |
| 761 | src.status = "moving" |
| 762 | next := CarryRequest(cloneSimState(w.state), cloneSimConsumed(src.consumed)) |
| 763 | if !SameRequestState(w.state, &next) { |
| 764 | next.Schema, next.Host = 1, updateSimHost |
| 765 | w.state = &next |
| 766 | src.copy = BuildStateCopy{Known: true, Replica: w.replica, State: cloneSimState(&next)} |
| 767 | w.stats.carriedRequests++ |
| 768 | note += " carried " + simRes(next.Resolved) + fmt.Sprintf(" done %d%s", next.Completed, simFailedMark(next.CompletedFailed)) |
| 769 | } |
| 770 | if race != nil && late { |
| 771 | note += " | racing (after the carry) " + race() |
| 772 | w.stats.racingFenced++ |
| 773 | } |
| 774 | if c := src.consumed; c != nil && c.CompletedFailed && c.Completed > w.carried { |
| 775 | w.carried = c.Completed |
| 776 | w.stats.carriedFailed++ |
| 777 | } |
| 778 | src.status = "" |
| 779 | kept := false |
| 780 | switch { |
| 781 | case carried: |
| 782 | dst.copy = cloneSimCopy(src.copy) |
| 783 | case dst.copy.Known && !w.scripted && w.rng.Intn(100) < 50: |
| 784 | kept = true // its row from the earlier stay |
| 785 | w.stats.keptCopies++ |
| 786 | default: |
| 787 | dst.copy = BuildStateCopy{} |
| 788 | } |
| 789 | if w.scripted || w.rng.Intn(100) < 60 { |
| 790 | src.copy = BuildStateCopy{} // the route retires here |
| 791 | } else if src.copy.Known { |
| 792 | // The source keeps its row for a return. Its refreshes while the |
| 793 | // instance is away read a location that is not this box's (another |
| 794 | // bucket: refused) and fail, so the copy keeps its last state with a |
| 795 | // read error; a refresh that succeeds would leave it fresh, which no |
| 796 | // rule can mistake. |
| 797 | src.copy.ReadErr = "s3: the instance is on another box" |
| 798 | } |
| 799 | w.at = j |
| 800 | w.stats.moves++ |
| 801 | if dst.consumed != nil || dst.live != "" { |
| 802 | w.stats.returns++ |
| 803 | } |
| 804 | w.syncResolutions() |
| 805 | return fmt.Sprintf("move %s->%s copy-carried=%v kept=%v%s", src.name, dst.name, carried, kept, note) |
| 806 | } |
| 807 | |
| 808 | // loseBox replaces a box with an empty one: its consumed file, live link, |
| 809 | // version store, failure book and bake clock are gone. If the instance was |
| 810 | // on it, its registry row is new too. |
| 811 | func (w *simWorld) loseBox() string { |
| 812 | k := w.rng.Intn(len(w.boxes)) |
| 813 | name := w.boxes[k].name |
| 814 | w.boxes[k] = newSimBox(name) |
| 815 | w.syncResolutions() |
| 816 | return "lose " + name |
| 817 | } |
| 818 | |
| 819 | // syncResolutions runs after a move or a lost box, for every request with a |
| 820 | // first eligible build. A resolution the instance can no longer see (no |
| 821 | // record in the build state or, strict mode aside, its box's consumed file) |
| 822 | // never ran, since a run completes the request in the build state: spec |
| 823 | // §4.2 "Lost requests", so the next eligible pass resolves it again. A |
| 824 | // record it can see is R's resolution from now on (the instance returned to |
| 825 | // a box holding one); of two, the build state's (R-RESOLUTION-STATE-WINS). |
| 826 | func (w *simWorld) syncResolutions() { |
| 827 | b := w.cur() |
| 828 | records := map[uint64]map[string]bool{} |
| 829 | note := func(r *ResolvedRequest) { |
| 830 | if r == nil { |
| 831 | return |
| 832 | } |
| 833 | if records[r.ID] == nil { |
| 834 | records[r.ID] = map[string]bool{} |
| 835 | } |
| 836 | records[r.ID][r.Target] = true |
| 837 | } |
| 838 | if w.state != nil { |
| 839 | note(w.state.Resolved) |
| 840 | } |
| 841 | for _, x := range w.boxes { |
| 842 | if (x == b || w.strict) && x.consumed != nil { |
| 843 | note(x.consumed.Resolved) |
| 844 | } |
| 845 | } |
| 846 | ids := map[uint64]bool{} |
| 847 | for r := range w.firstEligible { |
| 848 | ids[r] = true |
| 849 | } |
| 850 | for r := range records { |
| 851 | ids[r] = true |
| 852 | } |
| 853 | for r := range ids { |
| 854 | if r <= w.wm(b) { |
| 855 | continue |
| 856 | } |
| 857 | seen := records[r] |
| 858 | fe, had := w.firstEligible[r] |
| 859 | switch { |
| 860 | case len(seen) == 0: |
| 861 | delete(w.firstEligible, r) |
| 862 | w.stats.lostResolutions++ |
| 863 | case !w.strict: |
| 864 | // One record, or the build state's when both hold one |
| 865 | // (R-RESOLUTION-STATE-WINS). |
| 866 | t := "" |
| 867 | for x := range seen { |
| 868 | t = x |
| 869 | } |
| 870 | if w.state != nil && w.state.Resolved != nil && w.state.Resolved.ID == r { |
| 871 | t = w.state.Resolved.Target |
| 872 | } |
| 873 | if !had || t != fe { |
| 874 | w.firstEligible[r] = t |
| 875 | w.stats.pickedUp++ |
| 876 | } |
| 877 | } |
| 878 | } |
| 879 | } |
| 880 | |
| 881 | func (w *simWorld) markFailed() string { |
| 882 | b := w.cur() |
| 883 | var cands []string |
| 884 | add := func(v string) { |
| 885 | if v != "" { |
| 886 | cands = append(cands, v) |
| 887 | } |
| 888 | } |
| 889 | add(b.live) |
| 890 | add(w.head) |
| 891 | add(w.settings.Pin) |
| 892 | if w.state != nil { |
| 893 | add(w.state.LastExecuted) |
| 894 | if w.state.Resolved != nil { |
| 895 | add(w.state.Resolved.Target) |
| 896 | } |
| 897 | } |
| 898 | if b.consumed != nil && b.consumed.Resolved != nil { |
| 899 | add(b.consumed.Resolved.Target) |
| 900 | } |
| 901 | v := cands[w.rng.Intn(len(cands))] |
| 902 | b.failed[v] = true |
| 903 | return fmt.Sprintf("mark-failed %s on %s", v, b.name) |
| 904 | } |
| 905 | |
| 906 | func (w *simWorld) toggleManual() string { |
| 907 | if w.settings.Updates == UpdatesManual { |
| 908 | w.settings.Updates = UpdatesAuto |
| 909 | } else { |
| 910 | w.settings.Updates = UpdatesManual |
| 911 | } |
| 912 | return "updates=" + w.settings.Updates |
| 913 | } |
| 914 | |
| 915 | func (w *simWorld) pin() string { |
| 916 | if w.settings.Pin != "" && w.rng.Intn(100) < 50 { |
| 917 | w.settings.Pin = "" |
| 918 | return "unpin" |
| 919 | } |
| 920 | w.settings.Pin = w.builds[w.rng.Intn(len(w.builds))] |
| 921 | return "pin " + w.settings.Pin |
| 922 | } |
| 923 | |
| 924 | func (w *simWorld) setMajor() string { |
| 925 | switch w.rng.Intn(3) { |
| 926 | case 0: |
| 927 | w.settings.Major = nil |
| 928 | return "major=none" |
| 929 | default: |
| 930 | m := 1 + w.rng.Intn(2) |
| 931 | w.settings.Major = &m |
| 932 | return fmt.Sprintf("major=%d", m) |
| 933 | } |
| 934 | } |
| 935 | |
| 936 | // heldNow is rule 3's condition from the model's own records. |
| 937 | func (w *simWorld) heldNow() bool { |
| 938 | u := w.settings |
| 939 | if u.Updates == UpdatesManual { |
| 940 | return true |
| 941 | } |
| 942 | return u.Major != nil && (w.major[w.head] == 0 || w.major[w.head] != *u.Major) |
| 943 | } |
| 944 | |
| 945 | // wm is invariant 1's watermark for a pass or wake on box b: the highest |
| 946 | // completed id in the bucket's build state or b's consumed file (spec §4.3 |
| 947 | // rule 2), or in every box's consumed file in strict mode. |
| 948 | func (w *simWorld) wm(b *simBox) uint64 { |
| 949 | if w.strict { |
| 950 | return w.watermarkAnywhere() |
| 951 | } |
| 952 | var m uint64 |
| 953 | if w.state != nil { |
| 954 | m = w.state.Completed |
| 955 | } |
| 956 | if b.consumed != nil && b.consumed.Completed > m { |
| 957 | m = b.consumed.Completed |
| 958 | } |
| 959 | if w.carried > m { |
| 960 | m = w.carried |
| 961 | } |
| 962 | return m |
| 963 | } |
| 964 | |
| 965 | // watermarkAnywhere is the highest completed id any record holds: the |
| 966 | // build state and every box's consumed file. |
| 967 | func (w *simWorld) watermarkAnywhere() uint64 { |
| 968 | var m uint64 |
| 969 | if w.state != nil { |
| 970 | m = w.state.Completed |
| 971 | } |
| 972 | for _, b := range w.boxes { |
| 973 | if b.consumed != nil && b.consumed.Completed > m { |
| 974 | m = b.consumed.Completed |
| 975 | } |
| 976 | } |
| 977 | return m |
| 978 | } |
| 979 | |
| 980 | // boxView is what a pass on b can see of request r: pending (above both |
| 981 | // sources' completed ids) and the T either source resolved it to, the build |
| 982 | // state's first (R-RESOLUTION-STATE-WINS). |
| 983 | func boxView(c *Consumed, st *BuildState, r uint64) (pending bool, t string) { |
| 984 | var m uint64 |
| 985 | if c != nil { |
| 986 | m = c.Completed |
| 987 | } |
| 988 | if st != nil && st.Completed > m { |
| 989 | m = st.Completed |
| 990 | } |
| 991 | if r == 0 || r <= m { |
| 992 | return false, "" |
| 993 | } |
| 994 | if st != nil && st.Resolved != nil && st.Resolved.ID == r { |
| 995 | return true, st.Resolved.Target |
| 996 | } |
| 997 | if c != nil && c.Resolved != nil && c.Resolved.ID == r { |
| 998 | return true, c.Resolved.Target |
| 999 | } |
| 1000 | return true, "" |
| 1001 | } |
| 1002 | |
| 1003 | // rootPass is settingsTarget then adoptHost for the instance on box b. |
| 1004 | // resolve is root's pass (it writes the consumed file); false is the agent's |
| 1005 | // pass, which stops when the decision needs root's resolution. |
| 1006 | func (w *simWorld) rootPass(b *simBox, resolve bool) string { |
| 1007 | who := "root" |
| 1008 | if !resolve { |
| 1009 | who = "agent" |
| 1010 | } |
| 1011 | u := w.settings |
| 1012 | head := w.head |
| 1013 | wmBefore := w.wm(b) |
| 1014 | held := w.heldNow() |
| 1015 | |
| 1016 | // settingsTarget: a route with default settings and no consumed file is |
| 1017 | // today's adoption (no Target, no book, no consumed write). |
| 1018 | viaTarget := !(u.Default() && b.consumed == nil) |
| 1019 | var dec TargetDecision |
| 1020 | var in TargetInput |
| 1021 | if !viaTarget { |
| 1022 | dec = TargetDecision{Version: head, Reason: ReasonAdopt, Evidence: EvidencePointer, Bake: true, Detail: head} |
| 1023 | } else { |
| 1024 | in = TargetInput{ |
| 1025 | Route: registry.Route{Host: updateSimHost, Kind: "instance", UpdateSettings: u.Column()}, |
| 1026 | Pointer: &ChannelPointer{Channel: "production", Version: head}, |
| 1027 | HeadTag: w.tags[head], |
| 1028 | Settings: u, |
| 1029 | Build: cloneSimCopy(b.copy), |
| 1030 | Consumed: cloneSimConsumed(b.consumed), |
| 1031 | Installed: b.live, |
| 1032 | } |
| 1033 | if !b.eligible[head] { |
| 1034 | in.HeadHeld = "in its supply-chain bake window" |
| 1035 | } |
| 1036 | // settingsTarget gives Target the failure book only for a route |
| 1037 | // with settings (!plain); default settings keep today's adoption, |
| 1038 | // which never read it, even with a consumed file (R-T6-BOOK-2). |
| 1039 | if !u.Default() { |
| 1040 | failed := b.failed |
| 1041 | in.Failed = func(v string) bool { return failed[v] } |
| 1042 | } |
| 1043 | dec = Target(in) |
| 1044 | if resolve { |
| 1045 | // publishConsumed: root's record is written under the carry |
| 1046 | // fence, and only for a row no move holds (ruling |
| 1047 | // R-CARRY-FENCE). A refused pass publishes nothing and installs |
| 1048 | // nothing for the route: the decision never happened. |
| 1049 | if err := publishGate(®istry.Route{Host: updateSimHost, Kind: "instance", Status: b.status}); err != nil { |
| 1050 | return fmt.Sprintf("%s@%s %s: fenced: nothing published or installed", who, b.name, simDec(dec)) |
| 1051 | } |
| 1052 | } |
| 1053 | w.checkEvidence(dec, head) |
| 1054 | w.checkDecision(b, in, dec, resolve, wmBefore, held) |
| 1055 | if resolve { |
| 1056 | next := NextConsumed(in, dec) |
| 1057 | b.consumed = &next |
| 1058 | b.decided, b.decidedSettings, b.decidedHead = true, u.Column(), head |
| 1059 | // Invariant 6: with the build state read, root's file agrees with |
| 1060 | // it on the pending request's T once the pass is written. |
| 1061 | if r := u.InstallRequest; in.Build.Read() && r > w.wm(b) && w.state != nil && w.state.Resolved != nil && w.state.Resolved.ID == r && |
| 1062 | next.Resolved != nil && next.Resolved.ID == r && next.Resolved.Target != w.state.Resolved.Target { |
| 1063 | w.fail(6, "after root's pass R%d has two resolutions: the build state says %s, %s's consumed file %s", |
| 1064 | r, w.state.Resolved.Target, b.name, next.Resolved.Target) |
| 1065 | } |
| 1066 | } else if dec.Resolve != nil && dec.Reason == ReasonRequest { |
| 1067 | return fmt.Sprintf("%s@%s %s: stops, relays to root", who, b.name, simDec(dec)) |
| 1068 | } |
| 1069 | } |
| 1070 | desc := fmt.Sprintf("%s@%s %s", who, b.name, simDec(dec)) |
| 1071 | |
| 1072 | // Invariant 5's scenario: a held instance, no pin, the build state read |
| 1073 | // with a last_executed, an unresolved request whose head bakes here. |
| 1074 | recovery := "" |
| 1075 | if viaTarget && held && u.Pin == "" && in.Build.Read() && in.Build.State != nil && in.Build.State.LastExecuted != "" { |
| 1076 | le := in.Build.State.LastExecuted |
| 1077 | pending, t := boxView(in.Consumed, in.Build.State, u.InstallRequest) |
| 1078 | if pending && t == "" && in.HeadHeld != "" && !u.InstallFastTrack && !b.failed[le] && (w.promoted[le] || b.staged[le]) { |
| 1079 | recovery = le |
| 1080 | w.stats.heldBakingPasses++ |
| 1081 | } |
| 1082 | } |
| 1083 | defer func() { |
| 1084 | if recovery != "" && b.live != recovery { |
| 1085 | w.fail(5, "held, request R%d unresolved behind baking head %s, last_executed %s authorized: live is %q after the pass (%s)", |
| 1086 | u.InstallRequest, head, recovery, b.live, simDec(dec)) |
| 1087 | } |
| 1088 | }() |
| 1089 | |
| 1090 | // adoptHost. |
| 1091 | ver := dec.Version |
| 1092 | if ver == "" { |
| 1093 | return desc + ": nothing installed" |
| 1094 | } |
| 1095 | if b.live == ver { |
| 1096 | return desc + ": up to date" |
| 1097 | } |
| 1098 | if (!viaTarget || dec.Bake) && !b.eligible[ver] { |
| 1099 | return desc + ": " + ver + " baking" |
| 1100 | } |
| 1101 | stagedBefore := b.staged[ver] |
| 1102 | switch { |
| 1103 | case !viaTarget || dec.Evidence == EvidencePointer: |
| 1104 | if ver != head { |
| 1105 | return desc + ": not the signed head, not staged" |
| 1106 | } |
| 1107 | case dec.Evidence == EvidenceRecord: |
| 1108 | if !w.promoted[ver] && !stagedBefore { |
| 1109 | return desc + ": no promotion record, no staged copy: refused" |
| 1110 | } |
| 1111 | default: |
| 1112 | return desc + ": no evidence" |
| 1113 | } |
| 1114 | |
| 1115 | // The repoint. Invariant 3's repoint half holds by construction of the |
| 1116 | // evidence switch above; it stays as a tripwire for the mirror. |
| 1117 | if !(ver == head || w.promoted[ver] || stagedBefore) { |
| 1118 | w.fail(3, "repointed onto %s: not the head %s, no promotion record, not staged on %s", ver, head, b.name) |
| 1119 | } |
| 1120 | if viaTarget && !u.Default() && b.failed[ver] { |
| 1121 | w.fail(4, "%s repointed onto %s, which it holds as failed (%s)", b.name, ver, simDec(dec)) |
| 1122 | } |
| 1123 | if viaTarget && held && u.Pin == "" { |
| 1124 | ok := !w.replica && ver == head |
| 1125 | if w.state != nil && w.state.LastExecuted == ver { |
| 1126 | ok = true |
| 1127 | } |
| 1128 | if r := u.InstallRequest; r > w.wm(b) { |
| 1129 | if _, t := boxView(b.consumed, w.state, r); t == ver { |
| 1130 | ok = true |
| 1131 | } |
| 1132 | } |
| 1133 | if !ok { |
| 1134 | w.fail(5, "held instance repointed onto %s: not last_executed %q, not a T of pending R%d, not a new instance's head (%s)", |
| 1135 | ver, simLE(w.state), u.InstallRequest, simDec(dec)) |
| 1136 | } |
| 1137 | } |
| 1138 | if recovery != "" && ver == recovery { |
| 1139 | w.stats.heldRecoveries++ |
| 1140 | } |
| 1141 | prev := b.live |
| 1142 | b.staged[ver] = true |
| 1143 | b.live = ver |
| 1144 | return fmt.Sprintf("%s: repoint %q->%s", desc, prev, ver) |
| 1145 | } |
| 1146 | |
| 1147 | // checkEvidence is invariant 3 at the decision: only the head rests on the |
| 1148 | // signed pointer, and only the head bakes (§4.1, §4.3: a pin, a resolved T |
| 1149 | // and a recovery waive bake and need a record or a staged copy). |
| 1150 | func (w *simWorld) checkEvidence(dec TargetDecision, head string) { |
| 1151 | if dec.Version == "" || dec.Version == head { |
| 1152 | return |
| 1153 | } |
| 1154 | if dec.Evidence == EvidencePointer { |
| 1155 | w.fail(3, "decision %s rests on the pointer for %s, not the head %s", simDec(dec), dec.Version, head) |
| 1156 | } |
| 1157 | if dec.Bake { |
| 1158 | w.fail(3, "decision %s bakes %s, not the head %s", simDec(dec), dec.Version, head) |
| 1159 | } |
| 1160 | } |
| 1161 | |
| 1162 | // checkDecision checks invariants 1 and 6 on one Target decision, before |
| 1163 | // root writes it. |
| 1164 | func (w *simWorld) checkDecision(b *simBox, in TargetInput, dec TargetDecision, resolve bool, wmBefore uint64, held bool) { |
| 1165 | u := in.Settings |
| 1166 | r := u.InstallRequest |
| 1167 | head := in.Pointer.Version |
| 1168 | |
| 1169 | if !in.Build.Read() && r != 0 { |
| 1170 | if w.state != nil && w.state.Completed >= r { |
| 1171 | w.stats.unknownWithCompleted++ |
| 1172 | } |
| 1173 | if in.Consumed != nil && in.Consumed.Resolved != nil && in.Consumed.Resolved.ID == r { |
| 1174 | w.stats.unknownReturning++ |
| 1175 | } |
| 1176 | } |
| 1177 | |
| 1178 | if all := w.watermarkAnywhere(); (dec.Reason == ReasonRequest && r <= all && r > wmBefore) || |
| 1179 | (resolve && dec.Resolve != nil && dec.Resolve.ID <= all && dec.Resolve.ID > wmBefore) { |
| 1180 | w.stats.leftBehind++ |
| 1181 | } |
| 1182 | |
| 1183 | // Invariant 1: rule 2 never selects a completed request. |
| 1184 | if dec.Reason == ReasonRequest && r <= wmBefore { |
| 1185 | w.fail(1, "rule 2 selected R%d (%s), already completed (watermark %d)", r, dec.Version, wmBefore) |
| 1186 | } |
| 1187 | if !resolve { |
| 1188 | return |
| 1189 | } |
| 1190 | if dec.Resolve != nil && dec.Resolve.ID <= wmBefore { |
| 1191 | w.fail(1, "%s resolved R%d again (to %s) after it completed (watermark %d)", b.name, dec.Resolve.ID, dec.Resolve.Target, wmBefore) |
| 1192 | } |
| 1193 | |
| 1194 | // Invariant 6: the first eligible pass resolves, pin or not (with the |
| 1195 | // build state read: never read, or a refresh that failed, is unread). |
| 1196 | if in.Build.Read() && r != 0 { |
| 1197 | pending, t := boxView(in.Consumed, in.Build.State, r) |
| 1198 | eligible := in.HeadHeld == "" || u.InstallFastTrack |
| 1199 | if pending && t == "" && eligible && r > wmBefore { |
| 1200 | if _, ok := w.firstEligible[r]; !ok { |
| 1201 | w.firstEligible[r] = head |
| 1202 | } |
| 1203 | if u.Pin != "" { |
| 1204 | w.stats.pinnedEligible++ |
| 1205 | } |
| 1206 | want := w.firstEligible[r] |
| 1207 | if dec.Resolve == nil || dec.Resolve.ID != r || dec.Resolve.Target != want { |
| 1208 | w.fail(6, "R%d pending and %s eligible on %s (first eligible %s), pin %q: the pass resolved %s", r, head, b.name, want, u.Pin, simRes(dec.Resolve)) |
| 1209 | } |
| 1210 | } |
| 1211 | } |
| 1212 | if dec.Resolve != nil { |
| 1213 | want, ok := w.firstEligible[dec.Resolve.ID] |
| 1214 | if !ok || want != dec.Resolve.Target { |
| 1215 | w.fail(6, "%s resolved R%d to %s; first eligible %q", b.name, dec.Resolve.ID, dec.Resolve.Target, want) |
| 1216 | } |
| 1217 | w.stats.resolutions++ |
| 1218 | if b.failed[dec.Resolve.Target] { |
| 1219 | w.stats.failedHeadResolutions++ |
| 1220 | } |
| 1221 | } |
| 1222 | if dec.Reason == ReasonRequest { |
| 1223 | if want := w.firstEligible[r]; dec.Version != want { |
| 1224 | w.fail(6, "rule 2 installs %s for R%d; first eligible %q", dec.Version, r, want) |
| 1225 | } |
| 1226 | } |
| 1227 | if dec.Fail != 0 { |
| 1228 | w.stats.failedCompletions++ |
| 1229 | } |
| 1230 | } |
| 1231 | |
| 1232 | // wake is the agent's wake on box b, as production runs it with no planner |
| 1233 | // and an identity it can exec by path: |
| 1234 | // |
| 1235 | // - for a route whose settings could hold it, the copy is refreshed first |
| 1236 | // (buildStates.wakeTarget); a refresh that fails leaves ReadErr in it; |
| 1237 | // - the held wake check's decision is production's (DecideHeldWake, ruling |
| 1238 | // R-HELD-PURE), enforced as the activator's wakeCheck does: not held runs |
| 1239 | // the live link (today's wake); unread, unknown, or a target root reports |
| 1240 | // failed is refused; a held target runs when it is installed, or from |
| 1241 | // this box's retained copy; otherwise root is asked and the wake waits |
| 1242 | // for the install, and a changed answer ends the wait with nothing run. |
| 1243 | // With no decision from root for the current settings (AskRoot) the |
| 1244 | // build state's build runs only if it is installed; otherwise the wake |
| 1245 | // asks root for a pass and waits for its decision, then decides again. |
| 1246 | // The live link never runs blindly; |
| 1247 | // - a wait is root's pass, run inline (the stage relay), or, half the |
| 1248 | // time, no answer within the bound: the wake ends with nothing run; |
| 1249 | // - then beforeExec: HeldExecStale against the bucket object read under |
| 1250 | // the lease, and the merge, NextBuildState; then the run. |
| 1251 | func (w *simWorld) wake(b *simBox) string { |
| 1252 | u := w.settings |
| 1253 | col := u.Column() |
| 1254 | desc := "wake@" + b.name |
| 1255 | if !u.Default() { |
| 1256 | if !w.refreshCopy(b) { |
| 1257 | desc += " (refresh failed)" |
| 1258 | } |
| 1259 | } |
| 1260 | decide := func() HeldWakeDecision { |
| 1261 | return DecideHeldWake(HeldWakeInput{Settings: col, Build: cloneSimCopy(b.copy), Consumed: cloneSimConsumed(b.consumed), Installed: b.live}) |
| 1262 | } |
| 1263 | waitRoot := func() bool { |
| 1264 | w.stats.rootWaits++ |
| 1265 | if !w.scripted && w.rng.Intn(100) < 50 { |
| 1266 | return false |
| 1267 | } |
| 1268 | desc += " [waits; " + w.rootPass(b, true) + "]" |
| 1269 | return true |
| 1270 | } |
| 1271 | d := decide() |
| 1272 | if d.Held && d.AskRoot && !d.Failed { |
| 1273 | w.stats.askRoot++ |
| 1274 | if d.Target != "" && b.live == d.Target { |
| 1275 | w.stats.askRootRuns++ |
| 1276 | } else { |
| 1277 | if !waitRoot() { |
| 1278 | return desc + ": no decision from root for its settings within the bound, nothing run" |
| 1279 | } |
| 1280 | if d = decide(); d.AskRoot && (d.Target == "" || b.live != d.Target) { |
| 1281 | return desc + ": root has not decided for its settings, nothing run" |
| 1282 | } |
| 1283 | } |
| 1284 | } |
| 1285 | run, retained := "", false |
| 1286 | switch { |
| 1287 | case !d.Held: |
| 1288 | run = b.live // today's wake |
| 1289 | if run == "" { |
| 1290 | return desc + ": nothing installed" |
| 1291 | } |
| 1292 | case d.Unread: |
| 1293 | w.stats.unreadRefusals++ |
| 1294 | return desc + ": refused, its build state could not be read" |
| 1295 | case d.Unknown || d.Target == "": |
| 1296 | return desc + ": refused, build unknown" |
| 1297 | case d.Failed: |
| 1298 | return desc + ": refused, its target " + d.Target + " failed here" |
| 1299 | case b.live == d.Target: |
| 1300 | run = d.Target |
| 1301 | case !d.AskRoot && b.staged[d.Target]: |
| 1302 | run, retained = d.Target, true |
| 1303 | default: |
| 1304 | if !waitRoot() { |
| 1305 | return desc + ": waited for root to install " + d.Target + ", nothing run" |
| 1306 | } |
| 1307 | if b.live != d.Target { |
| 1308 | return desc + ": " + d.Target + " not installed when the wait ended, nothing run" |
| 1309 | } |
| 1310 | run = d.Target |
| 1311 | } |
| 1312 | if d.Held { |
| 1313 | w.stats.heldRuns++ |
| 1314 | if retained { |
| 1315 | w.stats.retainedRuns++ |
| 1316 | desc += " (retained copy)" |
| 1317 | } |
| 1318 | } |
| 1319 | // beforeExec: the held target revalidated against the object under the |
| 1320 | // lease, then the merge. A refused record runs nothing. |
| 1321 | fresh := DecideHeldWake(HeldWakeInput{Settings: col, Build: BuildStateCopy{Known: true, Replica: true, State: cloneSimState(w.state)}, |
| 1322 | Consumed: cloneSimConsumed(b.consumed), Installed: b.live}) |
| 1323 | if err := HeldExecStale(fresh, w.state, run, ""); err != nil { // the model's wake has no planner |
| 1324 | return desc + ": record refused: " + err.Error() |
| 1325 | } |
| 1326 | w.checkWakeRun(b, run) |
| 1327 | wmBefore := w.wm(b) |
| 1328 | cur := cloneSimState(w.state) |
| 1329 | next := NextBuildState(cur, cloneSimConsumed(b.consumed), run) |
| 1330 | next.Schema, next.Host = 1, updateSimHost |
| 1331 | |
| 1332 | if next.LastExecuted != run { |
| 1333 | w.fail(2, "wake ran %s; the build state says last_executed %q", run, next.LastExecuted) |
| 1334 | } |
| 1335 | var prev uint64 |
| 1336 | if cur != nil { |
| 1337 | prev = cur.Completed |
| 1338 | } |
| 1339 | // Only the current request: a newer one supersedes an older resolution. |
| 1340 | if r := w.settings.InstallRequest; cur != nil && cur.Resolved != nil && cur.Resolved.ID == r && cur.Resolved.Target == run && r > w.wm(b) && next.Completed < r { |
| 1341 | w.fail(1, "wake ran %s, the build state's T for pending R%d, and R%d stays pending (completed %d)", |
| 1342 | run, cur.Resolved.ID, cur.Resolved.ID, next.Completed) |
| 1343 | } |
| 1344 | if next.Completed > prev && !next.CompletedFailed { |
| 1345 | r := next.Completed |
| 1346 | if r <= wmBefore || w.completedByRun[r] { |
| 1347 | w.fail(1, "wake running %s records R%d newly completed; already completed (watermark %d)", run, r, wmBefore) |
| 1348 | } |
| 1349 | if next.Resolved == nil || next.Resolved.ID != r || next.Resolved.Target != run { |
| 1350 | w.fail(1, "wake running %s records R%d completed, but R%d's T is %s", run, r, r, simRes(next.Resolved)) |
| 1351 | } |
| 1352 | w.completedByRun[r] = true |
| 1353 | w.stats.runCompletions++ |
| 1354 | } |
| 1355 | w.state = &next |
| 1356 | w.lastRan = run |
| 1357 | w.replica = true |
| 1358 | b.copy = BuildStateCopy{Known: true, Replica: true, State: cloneSimState(&next)} |
| 1359 | return fmt.Sprintf("%s runs %s", desc, run) |
| 1360 | } |
| 1361 | |
| 1362 | // checkWakeRun is invariant 7 on a wake about to run build run on box b, from |
| 1363 | // the model's own records (never Consumed.Settings or Reason). |
| 1364 | func (w *simWorld) checkWakeRun(b *simBox, run string) { |
| 1365 | u := w.settings |
| 1366 | if u.Default() { |
| 1367 | return |
| 1368 | } |
| 1369 | if u.Pin != "" { |
| 1370 | if run != u.Pin { |
| 1371 | w.fail(7, "pinned on %s, the wake on %s ran %s", u.Pin, b.name, run) |
| 1372 | } |
| 1373 | return |
| 1374 | } |
| 1375 | if r := u.InstallRequest; r != 0 && r > w.wm(b) { |
| 1376 | t := "" |
| 1377 | if w.state != nil && w.state.Resolved != nil && w.state.Resolved.ID == r { |
| 1378 | t = w.state.Resolved.Target |
| 1379 | } else if b.consumed != nil && b.consumed.Resolved != nil && b.consumed.Resolved.ID == r { |
| 1380 | t = b.consumed.Resolved.Target |
| 1381 | } |
| 1382 | if t != "" { |
| 1383 | if run != t { |
| 1384 | w.fail(7, "R%d pending with T %s, the wake on %s ran %s", r, t, b.name, run) |
| 1385 | } |
| 1386 | return |
| 1387 | } |
| 1388 | } |
| 1389 | held := u.Updates == UpdatesManual |
| 1390 | if !held && u.Major != nil { |
| 1391 | if !b.decided || b.decidedSettings != u.Column() { |
| 1392 | held = true // root has not decided for these settings: the wake holds |
| 1393 | } else { |
| 1394 | m := w.major[b.decidedHead] |
| 1395 | held = m == 0 || m != *u.Major |
| 1396 | } |
| 1397 | } |
| 1398 | if le := simLE(w.state); held && le != "" && run != le { |
| 1399 | w.fail(7, "held (updates %s, major %v; root decided for these settings: %v), the wake on %s ran %s, not last_executed %s", |
| 1400 | u.Updates, simMajor(u.Major), b.decided && b.decidedSettings == u.Column(), b.name, run, le) |
| 1401 | } |
| 1402 | } |
| 1403 | |
| 1404 | func simMajor(m *int) string { |
| 1405 | if m == nil { |
| 1406 | return "-" |
| 1407 | } |
| 1408 | return strconv.Itoa(*m) |
| 1409 | } |
| 1410 | |
| 1411 | // check is the per-step state check (invariant 2); event checks run inside |
| 1412 | // the actions. A known copy that differs from the bucket is not a model |
| 1413 | // error: production keeps one after a failed refresh, and on a box the |
| 1414 | // instance returns to, until the next refresh (Astra round 1's note on this |
| 1415 | // simulation). The wake refreshes before it decides and is refused when |
| 1416 | // that fails; root's passes read the copy as it is. |
| 1417 | func (w *simWorld) check() { |
| 1418 | if w.lastRan != "" && (w.state == nil || w.state.LastExecuted != w.lastRan) { |
| 1419 | w.fail(2, "the database last ran %s; the build state says %q", w.lastRan, simLE(w.state)) |
| 1420 | } |
| 1421 | } |
| 1422 | |
| 1423 | func (w *simWorld) summary() string { |
| 1424 | b := w.cur() |
| 1425 | u := w.settings |
| 1426 | s := fmt.Sprintf("[at %s live %q head %s req %d pin %q upd %s", b.name, b.live, w.head, u.InstallRequest, u.Pin, u.Updates) |
| 1427 | if u.Major != nil { |
| 1428 | s += fmt.Sprintf(" major %d", *u.Major) |
| 1429 | } |
| 1430 | if w.state != nil { |
| 1431 | s += fmt.Sprintf(" | state le %s res %s done %d%s", w.state.LastExecuted, simRes(w.state.Resolved), w.state.Completed, simFailedMark(w.state.CompletedFailed)) |
| 1432 | } |
| 1433 | if b.consumed != nil { |
| 1434 | s += fmt.Sprintf(" | consumed res %s done %d%s", simRes(b.consumed.Resolved), b.consumed.Completed, simFailedMark(b.consumed.CompletedFailed)) |
| 1435 | } |
| 1436 | if !b.copy.Known { |
| 1437 | s += " | copy unread" |
| 1438 | } |
| 1439 | return s + "]" |
| 1440 | } |
| 1441 | |
| 1442 | func simFailedMark(f bool) string { |
| 1443 | if f { |
| 1444 | return " failed" |
| 1445 | } |
| 1446 | return "" |
| 1447 | } |
| 1448 | |
| 1449 | func simDec(d TargetDecision) string { |
| 1450 | s := fmt.Sprintf("{%s %q", d.Reason, d.Version) |
| 1451 | if d.Resolve != nil { |
| 1452 | s += " resolve " + simRes(d.Resolve) |
| 1453 | } |
| 1454 | if d.Fail != 0 { |
| 1455 | s += fmt.Sprintf(" fail R%d", d.Fail) |
| 1456 | } |
| 1457 | return s + "}" |
| 1458 | } |
| 1459 | |
| 1460 | func simRes(r *ResolvedRequest) string { |
| 1461 | if r == nil { |
| 1462 | return "-" |
| 1463 | } |
| 1464 | return fmt.Sprintf("R%d->%s", r.ID, r.Target) |
| 1465 | } |
| 1466 | |
| 1467 | func simLE(s *BuildState) string { |
| 1468 | if s == nil { |
| 1469 | return "" |
| 1470 | } |
| 1471 | return s.LastExecuted |
| 1472 | } |
| 1473 | |
| 1474 | func simStatesEqual(a, b *BuildState) bool { |
| 1475 | if a == nil || b == nil { |
| 1476 | return a == nil && b == nil |
| 1477 | } |
| 1478 | ra, rb := simRes(a.Resolved), simRes(b.Resolved) |
| 1479 | return a.LastExecuted == b.LastExecuted && ra == rb && a.Completed == b.Completed && a.CompletedFailed == b.CompletedFailed |
| 1480 | } |
| 1481 | |
| 1482 | func cloneSimState(s *BuildState) *BuildState { |
| 1483 | if s == nil { |
| 1484 | return nil |
| 1485 | } |
| 1486 | v := *s |
| 1487 | v.Resolved = cloneSimResolved(s.Resolved) |
| 1488 | return &v |
| 1489 | } |
| 1490 | |
| 1491 | func cloneSimConsumed(c *Consumed) *Consumed { |
| 1492 | if c == nil { |
| 1493 | return nil |
| 1494 | } |
| 1495 | v := *c |
| 1496 | v.Resolved = cloneSimResolved(c.Resolved) |
| 1497 | return &v |
| 1498 | } |
| 1499 | |
| 1500 | func cloneSimCopy(c BuildStateCopy) BuildStateCopy { |
| 1501 | c.State = cloneSimState(c.State) |
| 1502 | return c |
| 1503 | } |
| 1504 | |
| 1505 | func cloneSimResolved(r *ResolvedRequest) *ResolvedRequest { |
| 1506 | if r == nil { |
| 1507 | return nil |
| 1508 | } |
| 1509 | v := *r |
| 1510 | return &v |
| 1511 | } |
| 1512 | |
amadan