| 1 | package activator |
| 2 | |
| 3 | import ( |
| 4 | "sync" |
| 5 | "testing" |
| 6 | "time" |
| 7 | |
| 8 | "github.com/carlosframework/platform/internal/registry" |
| 9 | ) |
| 10 | |
| 11 | // stopEventGate is an EventSink that holds a terminal stop at its first |
| 12 | // line ("instance stopping"), before it reads the sidecar join, until the |
| 13 | // test lets it go: the stop's read of the join can then be placed after |
| 14 | // anything the test arranges. |
| 15 | type stopEventGate struct { |
| 16 | once sync.Once |
| 17 | entered chan struct{} |
| 18 | release chan struct{} |
| 19 | } |
| 20 | |
| 21 | func newStopEventGate() *stopEventGate { |
| 22 | return &stopEventGate{entered: make(chan struct{}), release: make(chan struct{})} |
| 23 | } |
| 24 | |
| 25 | func (g *stopEventGate) App(string, string, string, string) {} |
| 26 | |
| 27 | func (g *stopEventGate) Event(_, _, _, _, msg string) { |
| 28 | if msg != "instance stopping" { |
| 29 | return |
| 30 | } |
| 31 | first := false |
| 32 | g.once.Do(func() { first = true }) |
| 33 | if !first { |
| 34 | return |
| 35 | } |
| 36 | close(g.entered) |
| 37 | <-g.release |
| 38 | } |
| 39 | |
| 40 | // shrinkSidecarJoinWait shortens a flag-off stop's bounded wait on an |
| 41 | // in-flight sidecar start for one test. |
| 42 | func shrinkSidecarJoinWait(t *testing.T) { |
| 43 | t.Helper() |
| 44 | prev := sidecarStartJoinWait |
| 45 | sidecarStartJoinWait = 50 * time.Millisecond |
| 46 | t.Cleanup(func() { sidecarStartJoinWait = prev }) |
| 47 | } |
| 48 | |
| 49 | // Fix wave 1, item 1 (Astra final review 1, finding 1): the missed-join |
| 50 | // window. A sidecar restart returns while its entry is stopping, and the |
| 51 | // stop reads the join only after the restart's critical section has run |
| 52 | // but while its cleanup is still stopping the sidecar it brought back. The |
| 53 | // stop must still see the join, and the lease goes only once that sidecar |
| 54 | // is stopped. |
| 55 | func TestAStopThatReadsTheJoinDuringTheRestartsCleanupWaitsForIt(t *testing.T) { |
| 56 | shrinkSidecarJoinWait(t) |
| 57 | a, b, _, _ := testActivator(t, time.Hour) |
| 58 | h := &hostEmptyFake{fakeBackend: b} |
| 59 | a.backend = h |
| 60 | g := newStopEventGate() |
| 61 | a.Logs = g |
| 62 | b.sidecarStaged = true |
| 63 | host := "missed-join.example.com" |
| 64 | |
| 65 | arrangeAwake(t, a, testRoute(host)) |
| 66 | h.reset() |
| 67 | b.lastSidecar().exit() // the sweeper now wants a restart |
| 68 | |
| 69 | startGate, entered := make(chan struct{}), make(chan struct{}) |
| 70 | scStopGate := make(chan struct{}) |
| 71 | b.mu.Lock() |
| 72 | b.sidecarGate, b.sidecarStartEntered = startGate, entered |
| 73 | b.sidecarStopGate = scStopGate // the returned sidecar's stop blocks |
| 74 | b.mu.Unlock() |
| 75 | a.sweepOnce() |
| 76 | <-entered |
| 77 | |
| 78 | rep := b.lastRep() |
| 79 | if !a.Bounce(host) { |
| 80 | t.Fatal("Bounce did not find the awake entry") |
| 81 | } |
| 82 | <-g.entered // the stop is running, and has not read the join yet |
| 83 | |
| 84 | close(startGate) // the restart returns onto a stopping entry |
| 85 | waitFor(t, "the restart to bring back its sidecar", func() bool { return b.sidecarStartCount() == 2 }) |
| 86 | sc := b.lastSidecar() |
| 87 | rep.setOnClose(func() { |
| 88 | if !sc.isStopped() { |
| 89 | t.Error("the lease was released while the sidecar the restart brought back was still being stopped") |
| 90 | } |
| 91 | }) |
| 92 | waitFor(t, "the restart's cleanup to begin stopping it", func() bool { return sc.stopEntered() >= 1 }) |
| 93 | close(g.release) // now the stop reads the join |
| 94 | |
| 95 | time.Sleep(4 * sidecarStartJoinWait) |
| 96 | for range 3 { |
| 97 | a.sweepOnce() |
| 98 | } |
| 99 | if rep.isClosed() { |
| 100 | t.Fatal("the lease went while the returned sidecar's stop was still in flight") |
| 101 | } |
| 102 | |
| 103 | close(scStopGate) |
| 104 | waitFor(t, "the stop to finish once the returned sidecar is stopped", func() bool { |
| 105 | a.sweepOnce() |
| 106 | return rep.isClosed() && a.entryCount() == 0 |
| 107 | }) |
| 108 | } |
| 109 | |
| 110 | // Fix wave 1, item 1: a nested instance's restart returns a sidecar whose |
| 111 | // stop cannot prove its exit (ErrStillRunning). Nested rows skip the |
| 112 | // host-cgroup proof, so the sidecar's own stop proof is the only one: the |
| 113 | // stop stays pending with the lease, the sweep re-stops the sidecar, and |
| 114 | // the lease goes only after it has exited. |
| 115 | func TestANestedStopKeepsTheLeaseOverAReturnedSidecarItCannotProve(t *testing.T) { |
| 116 | shrinkSidecarJoinWait(t) |
| 117 | a, b, _, _ := testActivator(t, time.Hour) |
| 118 | h := &hostEmptyFake{fakeBackend: b} |
| 119 | a.backend = h |
| 120 | b.layoutFn = func(registry.Route) (Layout, string) { return LayoutNested, "" } |
| 121 | b.sidecarStaged = true |
| 122 | host := "nested-stuck.example.com" |
| 123 | |
| 124 | arrangeAwake(t, a, testRoute(host)) |
| 125 | h.reset() |
| 126 | b.lastSidecar().exit() |
| 127 | |
| 128 | startGate, entered := make(chan struct{}), make(chan struct{}) |
| 129 | b.mu.Lock() |
| 130 | b.sidecarGate, b.sidecarStartEntered = startGate, entered |
| 131 | b.sidecarStuck = true // the returned sidecar's stop answers ErrStillRunning |
| 132 | b.mu.Unlock() |
| 133 | a.sweepOnce() |
| 134 | <-entered |
| 135 | |
| 136 | rep, inst := b.lastRep(), b.lastProc() |
| 137 | if !a.Bounce(host) { |
| 138 | t.Fatal("Bounce did not find the awake entry") |
| 139 | } |
| 140 | waitFor(t, "the instance to stop", inst.isStopped) |
| 141 | close(startGate) |
| 142 | waitFor(t, "the restart to bring back its sidecar", func() bool { return b.sidecarStartCount() == 2 }) |
| 143 | sc := b.lastSidecar() |
| 144 | rep.setOnClose(func() { |
| 145 | if !sc.Exited() { |
| 146 | t.Error("the lease was released while the returned sidecar had not exited") |
| 147 | } |
| 148 | }) |
| 149 | waitFor(t, "the restart's stop of it", func() bool { return sc.stopCalls() >= 1 }) |
| 150 | time.Sleep(4 * sidecarStartJoinWait) |
| 151 | for range 3 { |
| 152 | a.sweepOnce() |
| 153 | } |
| 154 | if rep.isClosed() || a.entryCount() != 1 { |
| 155 | t.Fatalf("closed=%v entries=%d: the lease went over a sidecar whose stop was unproven", rep.isClosed(), a.entryCount()) |
| 156 | } |
| 157 | waitFor(t, "the sweep to re-stop the unproven sidecar", func() bool { |
| 158 | a.sweepOnce() |
| 159 | return sc.stopCalls() >= 2 |
| 160 | }) |
| 161 | |
| 162 | sc.exit() |
| 163 | waitFor(t, "the stop to finish once the sidecar exited", func() bool { |
| 164 | a.sweepOnce() |
| 165 | return rep.isClosed() && a.entryCount() == 0 |
| 166 | }) |
| 167 | if n := h.count(); n != 0 { |
| 168 | t.Fatalf("host proofs = %d: a nested stop has no host-cgroup proof", n) |
| 169 | } |
| 170 | } |
| 171 | |
| 172 | // Fix wave 1, item 1: the co-stop's own sidecar, flag off and nested, whose |
| 173 | // stop cannot prove its exit, is held to the same rule: the lease is kept |
| 174 | // until it has exited. |
| 175 | func TestANestedStopKeepsTheLeaseOverItsCoStoppedSidecarUntilItExits(t *testing.T) { |
| 176 | t.Parallel() |
| 177 | a, b, _, _ := testActivator(t, time.Hour) |
| 178 | h := &hostEmptyFake{fakeBackend: b} |
| 179 | a.backend = h |
| 180 | b.layoutFn = func(registry.Route) (Layout, string) { return LayoutNested, "" } |
| 181 | b.sidecarStaged = true |
| 182 | b.sidecarStuck = true |
| 183 | host := "nested-costop.example.com" |
| 184 | |
| 185 | arrangeAwake(t, a, testRoute(host)) |
| 186 | sc := b.lastSidecar() |
| 187 | rep, inst := b.lastRep(), b.lastProc() |
| 188 | rep.setOnClose(func() { |
| 189 | if !sc.Exited() { |
| 190 | t.Error("the lease was released while the co-stopped sidecar had not exited") |
| 191 | } |
| 192 | }) |
| 193 | if !a.Bounce(host) { |
| 194 | t.Fatal("Bounce did not find the awake entry") |
| 195 | } |
| 196 | waitFor(t, "the instance to stop", inst.isStopped) |
| 197 | for range 3 { |
| 198 | a.sweepOnce() |
| 199 | } |
| 200 | if rep.isClosed() || a.entryCount() != 1 { |
| 201 | t.Fatalf("closed=%v entries=%d: the lease went over a sidecar whose stop was unproven", rep.isClosed(), a.entryCount()) |
| 202 | } |
| 203 | sc.exit() |
| 204 | waitFor(t, "the stop to finish once the sidecar exited", func() bool { |
| 205 | a.sweepOnce() |
| 206 | return rep.isClosed() && a.entryCount() == 0 |
| 207 | }) |
| 208 | } |
| 209 | |
amadan