health_concurrency_test.go 4.0 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118
  1. package sdwan
  2. import (
  3. "sync"
  4. "testing"
  5. "time"
  6. )
  7. func TestProbeSnapshotsAndStaleResult(t *testing.T) {
  8. original := &SDWANMember{ID: "member", Interface: "eth0", TargetProbe: "192.0.2.1", Status: StatusHealthy}
  9. mgr := &Manager{groups: map[string]*SDWANGroup{"group": {ID: "group", Members: []*SDWANMember{original}}}}
  10. snap := mgr.List()
  11. snap[0].Members[0].Status = StatusDown
  12. if original.Status != StatusHealthy {
  13. t.Fatal("snapshot mutated live state")
  14. }
  15. mgr.mu.Lock()
  16. original.TargetProbe = "192.0.2.2"
  17. mgr.mu.Unlock()
  18. if mgr.applyProbeResults(snap) || original.Status != StatusHealthy {
  19. t.Fatal("stale probe applied")
  20. }
  21. snap = mgr.List()
  22. snap[0].Members[0].Status = StatusDown
  23. if !mgr.applyProbeResults(snap) || original.Status != StatusDown {
  24. t.Fatal("valid health transition not applied")
  25. }
  26. }
  27. func TestProbeRunsInParallelAndMergesSafely(t *testing.T) {
  28. mgr := &Manager{groups: map[string]*SDWANGroup{"g": {ID: "g", Members: []*SDWANMember{{ID: "a", Interface: "a"}, {ID: "b", Interface: "b"}}}}}
  29. h := NewHealthMonitor(nil)
  30. started := make(chan string, 2)
  31. release := make(chan struct{})
  32. h.probe = func(iface, target string) (float64, bool) { started <- iface; <-release; return 5, true }
  33. snapshot := mgr.List()
  34. done := make(chan struct{})
  35. go func() { defer close(done); h.runProbeIteration(snapshot); mgr.applyProbeResults(snapshot) }()
  36. for i := 0; i < 2; i++ {
  37. select {
  38. case <-started:
  39. case <-time.After(time.Second):
  40. close(release)
  41. t.Fatal("probes did not run concurrently")
  42. }
  43. }
  44. var wg sync.WaitGroup
  45. for i := 0; i < 4; i++ {
  46. wg.Add(1)
  47. go func() {
  48. defer wg.Done()
  49. for j := 0; j < 1000; j++ {
  50. mgr.List()
  51. mgr.GetTelemetrySummary()
  52. }
  53. }()
  54. }
  55. close(release)
  56. <-done
  57. wg.Wait()
  58. for _, m := range mgr.List()[0].Members {
  59. if m.Status != StatusHealthy || m.LastChecked.IsZero() {
  60. t.Fatalf("probe missing: %+v", m)
  61. }
  62. }
  63. }
  64. func TestFailoverProbeHysteresisAndRecovery(t *testing.T) {
  65. mgr := &Manager{groups: map[string]*SDWANGroup{"g": {ID: "g", Mode: ModeFailover, Members: []*SDWANMember{{ID: "a", Interface: "a", Gateway: "192.0.2.1", Priority: 1, Status: StatusHealthy}, {ID: "b", Interface: "b", Gateway: "192.0.2.2", Priority: 2, Status: StatusHealthy}}}}, engine: NewKernelEngine()}
  66. h := NewHealthMonitor(nil)
  67. primaryUp := false
  68. h.probe = func(iface, target string) (float64, bool) { return 5, iface != "a" || primaryUp }
  69. cycle := func() { snapshot := mgr.List(); h.runProbeIteration(snapshot); mgr.applyProbeResults(snapshot) }
  70. for i := 0; i < 4; i++ {
  71. cycle()
  72. hops, ok := mgr.ResolveGroupForRouting("g")
  73. if !ok || len(hops) != 1 || hops[0].Interface != "a" {
  74. t.Fatalf("failover too early at %d: %+v", i, hops)
  75. }
  76. }
  77. cycle()
  78. hops, ok := mgr.ResolveGroupForRouting("g")
  79. if !ok || len(hops) != 1 || hops[0].Interface != "b" {
  80. t.Fatalf("backup not selected: %+v", hops)
  81. }
  82. primaryUp = true
  83. cycle()
  84. hops, _ = mgr.ResolveGroupForRouting("g")
  85. if len(hops) != 1 || hops[0].Interface != "b" {
  86. t.Fatalf("recovered too early: %+v", hops)
  87. }
  88. cycle()
  89. hops, _ = mgr.ResolveGroupForRouting("g")
  90. if len(hops) != 1 || hops[0].Interface != "a" {
  91. t.Fatalf("primary not restored: %+v", hops)
  92. }
  93. }
  94. func TestProbeIntervalsPerGroup(t *testing.T) {
  95. now := time.Unix(1000, 0)
  96. last := map[string]time.Time{}
  97. groups := []*SDWANGroup{{ID: "fast", ProbeInterval: 3}, {ID: "slow", ProbeInterval: 10}, {ID: "disabled", Disabled: true}}
  98. if due := dueProbeGroups(groups, last, now); len(due) != 2 {
  99. t.Fatalf("initial probes: %d", len(due))
  100. }
  101. if due := dueProbeGroups(groups, last, now.Add(2*time.Second)); len(due) != 0 {
  102. t.Fatalf("probed too early: %d", len(due))
  103. }
  104. if due := dueProbeGroups(groups, last, now.Add(3*time.Second)); len(due) != 1 || due[0].ID != "fast" {
  105. t.Fatalf("3-second interval: %+v", due)
  106. }
  107. if due := dueProbeGroups(groups, last, now.Add(10*time.Second)); len(due) != 2 {
  108. t.Fatalf("10-second interval: %+v", due)
  109. }
  110. dueProbeGroups(groups[:1], last, now.Add(11*time.Second))
  111. if _, ok := last["slow"]; ok {
  112. t.Fatal("removed group retained schedule")
  113. }
  114. }