| 1 | package activator |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "strings" |
| 6 | "sync" |
| 7 | "testing" |
| 8 | "time" |
| 9 | ) |
| 10 | |
| 11 | // restoreInFlightForTest reports whether host's entry has a parent restore |
| 12 | // retry in flight. |
| 13 | func (a *Service) restoreInFlightForTest(host string) bool { |
| 14 | a.mu.Lock() |
| 15 | defer a.mu.Unlock() |
| 16 | e := a.entries[host] |
| 17 | return e != nil && e.ops.restoring |
| 18 | } |
| 19 | |
| 20 | // owedRestore leaves f's entry with a parent restore owed: an overlap whose |
| 21 | // last member's restore failed (F11), so the parent stays marked raised. |
| 22 | func owedRestore(t *testing.T, f *swapFixture) { |
| 23 | t.Helper() |
| 24 | f.b.gmu.Lock() |
| 25 | f.b.restoreFails = 1 |
| 26 | f.b.gmu.Unlock() |
| 27 | if res := f.a.Swap(context.Background(), f.attempt(ModeOverlap), f.newest); res.Outcome != SwapDone { |
| 28 | t.Fatalf("Swap = %+v", res) |
| 29 | } |
| 30 | if !f.a.parentRaisedForTest(f.host) { |
| 31 | t.Fatal("fixture: the failed restore should leave the parent marked raised") |
| 32 | } |
| 33 | } |
| 34 | |
| 35 | // gateRestores holds every later RestoreGenParent before its write and |
| 36 | // returns the gate and the channel each held call signals on. |
| 37 | func gateRestores(f *swapFixture) (gate, entered chan struct{}) { |
| 38 | gate, entered = make(chan struct{}), make(chan struct{}, 8) |
| 39 | f.b.gmu.Lock() |
| 40 | f.b.restoreGate, f.b.restoreEntered = gate, entered |
| 41 | f.b.gmu.Unlock() |
| 42 | return gate, entered |
| 43 | } |
| 44 | |
| 45 | // joinSignal installs the barrier's join hook and returns the channel it |
| 46 | // signals on when a stop waits on the operation set. |
| 47 | func joinSignal(f *swapFixture) chan struct{} { |
| 48 | joining := make(chan struct{}, 1) |
| 49 | f.a.mu.Lock() |
| 50 | f.a.joinWaiting = func(_, what string) { |
| 51 | if what == "operation set" { |
| 52 | select { |
| 53 | case joining <- struct{}{}: |
| 54 | default: |
| 55 | } |
| 56 | } |
| 57 | } |
| 58 | f.a.mu.Unlock() |
| 59 | return joining |
| 60 | } |
| 61 | |
| 62 | // Final review R2-2: a parent restore retry is a lifecycle operation of its |
| 63 | // entry. A terminal stop joins one in flight before its lease release; a |
| 64 | // stopping entry launches none; and a hold's stop joins it before its |
| 65 | // pinned replacement is installed, so the old retry can never write the |
| 66 | // old entry's single-generation tier after a later overlap raised the |
| 67 | // host's parent for two children. |
| 68 | func TestAParentRestoreRetryIsJoinedByTheEntrysStop(t *testing.T) { |
| 69 | t.Parallel() |
| 70 | t.Run("a terminal stop joins the retry in flight", func(t *testing.T) { |
| 71 | f := newSwapFixture(t) |
| 72 | owedRestore(t, f) |
| 73 | gate, entered := gateRestores(f) |
| 74 | f.a.sweepOnce() |
| 75 | <-entered // the retry is past its launch, before its write |
| 76 | joining := joinSignal(f) |
| 77 | rep := f.b.lastRep() |
| 78 | f.a.mu.Lock() |
| 79 | f.a.beginStopLocked(f.a.entries[f.host], StopIdle) |
| 80 | f.a.mu.Unlock() |
| 81 | waitFor(t, "the stop to join the retry or finish", func() bool { |
| 82 | select { |
| 83 | case <-joining: |
| 84 | return true |
| 85 | default: |
| 86 | } |
| 87 | return !f.a.entryPresent(f.host) |
| 88 | }) |
| 89 | if !f.a.entryPresent(f.host) || rep.isClosed() { |
| 90 | t.Errorf("the stop finished (entry present %v, lease released %v) while the parent restore retry was in flight", f.a.entryPresent(f.host), rep.isClosed()) |
| 91 | } |
| 92 | close(gate) |
| 93 | waitFor(t, "the stop to finish once the retry is done", func() bool { return !f.a.entryPresent(f.host) && rep.isClosed() }) |
| 94 | }) |
| 95 | |
| 96 | t.Run("a stopping entry launches no retry", func(t *testing.T) { |
| 97 | f := newSwapFixture(t) |
| 98 | owedRestore(t, f) |
| 99 | gate, _ := gateRestores(f) |
| 100 | defer close(gate) |
| 101 | stopGate := make(chan struct{}) |
| 102 | f.a.mu.Lock() |
| 103 | e := f.a.entries[f.host] |
| 104 | cur := e.proc.(*fakeProc) |
| 105 | f.a.mu.Unlock() |
| 106 | cur.setStopGate(stopGate) |
| 107 | f.a.mu.Lock() |
| 108 | f.a.beginStopLocked(e, StopIdle) |
| 109 | f.a.mu.Unlock() |
| 110 | // Past the barrier, inside the current generation's Stop. |
| 111 | waitFor(t, "the stop to reach the current generation", func() bool { return cur.stopEntered() > 0 }) |
| 112 | f.a.sweepOnce() |
| 113 | if f.a.restoreInFlightForTest(f.host) { |
| 114 | t.Error("the sweep launched a parent restore retry for a stopping entry") |
| 115 | } |
| 116 | close(stopGate) |
| 117 | waitFor(t, "the stop to finish", func() bool { return !f.a.entryPresent(f.host) }) |
| 118 | }) |
| 119 | |
| 120 | t.Run("a hold, its replacement and a later overlap", func(t *testing.T) { |
| 121 | f := newSwapFixture(t) |
| 122 | owedRestore(t, f) |
| 123 | gate, entered := gateRestores(f) |
| 124 | f.a.sweepOnce() |
| 125 | <-entered |
| 126 | joining := joinSignal(f) |
| 127 | f.a.mu.Lock() |
| 128 | oldE := f.a.entries[f.host] |
| 129 | f.a.mu.Unlock() |
| 130 | f.rev.Add(1) |
| 131 | results := make(chan SwapResult, 1) |
| 132 | go func() { results <- f.a.Swap(context.Background(), f.attempt(ModeHold), f.newest) }() |
| 133 | joined := false |
| 134 | waitFor(t, "the hold's stop to join the retry or install its replacement", func() bool { |
| 135 | select { |
| 136 | case <-joining: |
| 137 | joined = true |
| 138 | return true |
| 139 | default: |
| 140 | } |
| 141 | f.a.mu.Lock() |
| 142 | defer f.a.mu.Unlock() |
| 143 | return f.a.entries[f.host] != oldE |
| 144 | }) |
| 145 | var once sync.Once |
| 146 | release := func() { once.Do(func() { close(gate) }) } |
| 147 | defer release() |
| 148 | if !joined { |
| 149 | t.Error("the hold's stop installed its pinned replacement while the old entry's parent restore retry was in flight") |
| 150 | } else { |
| 151 | release() |
| 152 | } |
| 153 | select { |
| 154 | case res := <-results: |
| 155 | if res.Outcome != SwapDone { |
| 156 | t.Fatalf("hold Swap = %+v", res) |
| 157 | } |
| 158 | case <-time.After(10 * time.Second): |
| 159 | t.Fatal("the hold never finished") |
| 160 | } |
| 161 | waitFor(t, "the replacement to be awake", func() bool { return f.a.entryAwake(f.host) }) |
| 162 | // A later overlap on the replacement raises the host's parent for two |
| 163 | // children. An old retry still held lands now, if it can. |
| 164 | _, _, before := f.b.calls() |
| 165 | f.b.gmu.Lock() |
| 166 | f.b.onRaise = func() { |
| 167 | if joined { |
| 168 | return |
| 169 | } |
| 170 | release() |
| 171 | deadline := time.Now().Add(5 * time.Second) |
| 172 | for time.Now().Before(deadline) { |
| 173 | if _, _, n := f.b.calls(); n > before { |
| 174 | return |
| 175 | } |
| 176 | time.Sleep(time.Millisecond) |
| 177 | } |
| 178 | } |
| 179 | f.b.gmu.Unlock() |
| 180 | f.rev.Add(1) |
| 181 | if res := f.a.Swap(context.Background(), f.attempt(ModeOverlap), f.newest); res.Outcome != SwapDone { |
| 182 | t.Fatalf("later overlap Swap = %+v", res) |
| 183 | } |
| 184 | ev := f.b.events() |
| 185 | raise := -1 |
| 186 | for i, x := range ev { |
| 187 | if strings.HasPrefix(x, "raise:") { |
| 188 | raise = i |
| 189 | } |
| 190 | } |
| 191 | for _, x := range ev[raise+1:] { |
| 192 | if strings.HasPrefix(x, "switch:") { |
| 193 | break |
| 194 | } |
| 195 | if strings.HasPrefix(x, "restore:") { |
| 196 | t.Fatalf("a parent restore (%s) was written after the later overlap raised the parent and before its switch, while two generations ran: %v", x, ev[raise:]) |
| 197 | } |
| 198 | } |
| 199 | }) |
| 200 | } |
| 201 | |
amadan