| 1 | package activator |
| 2 | |
| 3 | import ( |
| 4 | "context" |
| 5 | "errors" |
| 6 | "strings" |
| 7 | "testing" |
| 8 | "time" |
| 9 | ) |
| 10 | |
| 11 | func TestObserveSettle(t *testing.T) { |
| 12 | t.Parallel() |
| 13 | if r := observeSettle(context.Background(), &fakeProc{}, 20*time.Millisecond); r != settleOK { |
| 14 | t.Fatalf("a process that stayed up: %v", r) |
| 15 | } |
| 16 | // A process that dies inside the window: alive at the first check, dead |
| 17 | // at the observation's next poll (no sleep places the death). |
| 18 | p := &fakeProc{} |
| 19 | p.exitOnPoll(2) |
| 20 | if r := observeSettle(context.Background(), p, time.Second); r != settleExited { |
| 21 | t.Fatalf("a process that died: %v", r) |
| 22 | } |
| 23 | // An observation cancelled inside the window: the cancel happens at its |
| 24 | // first poll after the opening check. |
| 25 | ctx, cancel := context.WithCancel(context.Background()) |
| 26 | defer cancel() |
| 27 | if r := observeSettle(ctx, &cancelOnPoll{at: 2, cancel: cancel}, time.Second); r != settleCancelled { |
| 28 | t.Fatalf("a cancelled observation: %v", r) |
| 29 | } |
| 30 | // Cancellation wins over an exit seen at the same moment (spec §6.4). |
| 31 | dead := &fakeProc{} |
| 32 | dead.exit() |
| 33 | gone, stop := context.WithCancel(context.Background()) |
| 34 | stop() |
| 35 | if r := observeSettle(gone, dead, time.Second); r != settleCancelled { |
| 36 | t.Fatalf("cancelled and exited: %v, want cancelled", r) |
| 37 | } |
| 38 | } |
| 39 | |
| 40 | // cancelOnPoll is a live process whose at-th Exited() call cancels the |
| 41 | // observation watching it. |
| 42 | type cancelOnPoll struct { |
| 43 | polls int |
| 44 | at int |
| 45 | cancel context.CancelFunc |
| 46 | } |
| 47 | |
| 48 | func (p *cancelOnPoll) Stop(context.Context) error { return nil } |
| 49 | func (p *cancelOnPoll) Exited() bool { |
| 50 | p.polls++ |
| 51 | if p.polls == p.at { |
| 52 | p.cancel() |
| 53 | } |
| 54 | return false |
| 55 | } |
| 56 | |
| 57 | func TestSwapCanarySettle(t *testing.T) { |
| 58 | old := swapSettle |
| 59 | t.Cleanup(func() { swapSettle = old }) |
| 60 | |
| 61 | t.Run("exit during the settle fails the build", func(t *testing.T) { |
| 62 | swapSettle = 2 * time.Second |
| 63 | f := newSwapFixture(t) |
| 64 | f.a.genHealth = func(context.Context, Gen) error { |
| 65 | // Alive at the settle's opening check, dead at its next poll. |
| 66 | _, p := f.b.lastGen() |
| 67 | p.exitOnPoll(2) |
| 68 | return nil |
| 69 | } |
| 70 | at := f.attempt(ModeOverlap) |
| 71 | at.Canary = true |
| 72 | res := f.a.Swap(context.Background(), at, f.newest) |
| 73 | if res.Outcome != SwapFailed || !strings.Contains(res.Reason, errSettleExit.Error()) { |
| 74 | t.Fatalf("Swap = %+v", res) |
| 75 | } |
| 76 | if got := f.appSocket(t); got != f.incumbent.Socket { |
| 77 | t.Fatalf("AppSocket = %q, want the incumbent", got) |
| 78 | } |
| 79 | }) |
| 80 | t.Run("a canary waits out the settle", func(t *testing.T) { |
| 81 | swapSettle = 100 * time.Millisecond |
| 82 | f := newSwapFixture(t) |
| 83 | at := f.attempt(ModeOverlap) |
| 84 | at.Canary = true |
| 85 | start := time.Now() |
| 86 | if res := f.a.Swap(context.Background(), at, f.newest); res.Outcome != SwapDone { |
| 87 | t.Fatalf("Swap = %+v", res) |
| 88 | } |
| 89 | if took := time.Since(start); took < swapSettle { |
| 90 | t.Fatalf("switched after %s, inside the %s settle", took, swapSettle) |
| 91 | } |
| 92 | }) |
| 93 | t.Run("a swap after the canary does not settle", func(t *testing.T) { |
| 94 | swapSettle = time.Hour |
| 95 | f := newSwapFixture(t) |
| 96 | done := make(chan SwapResult, 1) |
| 97 | go func() { done <- f.a.Swap(context.Background(), f.attempt(ModeOverlap), f.newest) }() |
| 98 | select { |
| 99 | case res := <-done: |
| 100 | if res.Outcome != SwapDone { |
| 101 | t.Fatalf("Swap = %+v", res) |
| 102 | } |
| 103 | case <-time.After(5 * time.Second): |
| 104 | t.Fatal("a non-canary swap waited for a settle") |
| 105 | } |
| 106 | }) |
| 107 | } |
| 108 | |
| 109 | // An intentional stop of the entry during the settle cancels the |
| 110 | // observation and is not a failure (spec §6.4). The successor is stopped |
| 111 | // before the incumbent, through the barrier. |
| 112 | func TestSwapIntentionalStopDuringSettleIsNotAFailure(t *testing.T) { |
| 113 | old := swapSettle |
| 114 | swapSettle = 5 * time.Second |
| 115 | t.Cleanup(func() { swapSettle = old }) |
| 116 | for _, tc := range []struct { |
| 117 | name string |
| 118 | stop func(f *swapFixture) |
| 119 | }{ |
| 120 | {"restart", func(f *swapFixture) { f.a.Bounce(f.host) }}, |
| 121 | {"hibernate", func(f *swapFixture) { f.clk.advance(2 * time.Hour); f.a.sweepOnce() }}, |
| 122 | {"lease lost", func(f *swapFixture) { f.b.lastRep().setErr(errors.New("renewal failed")); f.a.sweepOnce() }}, |
| 123 | {"maintenance fence", func(f *swapFixture) { |
| 124 | go func() { |
| 125 | ctx, cancel := context.WithTimeout(context.Background(), 10*time.Second) |
| 126 | defer cancel() |
| 127 | f.a.FenceAndStop(ctx, f.host, 1) //nolint:errcheck |
| 128 | }() |
| 129 | }}, |
| 130 | {"move or park retire", func(f *swapFixture) { f.a.RetireExcept(map[string]bool{}) }}, |
| 131 | {"shutdown", func(f *swapFixture) { go f.a.Shutdown(context.Background()) }}, |
| 132 | } { |
| 133 | t.Run(tc.name, func(t *testing.T) { |
| 134 | f := newSwapFixture(t) |
| 135 | healthy := make(chan struct{}) |
| 136 | f.a.genHealth = func(context.Context, Gen) error { close(healthy); return nil } |
| 137 | rep := f.b.lastRep() |
| 138 | at := f.attempt(ModeOverlap) |
| 139 | at.Canary = true |
| 140 | results := make(chan SwapResult, 1) |
| 141 | go func() { results <- f.a.Swap(context.Background(), at, f.newest) }() |
| 142 | <-healthy |
| 143 | // The settle has begun once it has polled the successor (nothing |
| 144 | // else asks it between the health check and the settle). |
| 145 | _, succ := f.b.lastGen() |
| 146 | waitFor(t, "the settle's first poll", func() bool { return succ.pollCount() > 0 }) |
| 147 | tc.stop(f) |
| 148 | var res SwapResult |
| 149 | select { |
| 150 | case res = <-results: |
| 151 | case <-time.After(10 * time.Second): |
| 152 | t.Fatal("Swap did not return after the stop") |
| 153 | } |
| 154 | if res.Outcome != SwapAbandoned || !strings.Contains(res.Reason, errSwapStopped.Error()) { |
| 155 | t.Fatalf("Swap = %+v, want abandoned by the stop", res) |
| 156 | } |
| 157 | waitFor(t, "the lease release", rep.isClosed) |
| 158 | ev := f.b.events() |
| 159 | if i, j := indexPrefix(ev, "stop:gen:"), indexPrefix(ev, "stop:instance"); i < 0 || j < 0 || i > j { |
| 160 | t.Fatalf("order %v: the successor must stop before the incumbent", ev) |
| 161 | } |
| 162 | }) |
| 163 | } |
| 164 | } |
| 165 | |