| 1 | //go:build linux |
| 2 | |
| 3 | package hostexec |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "errors" |
| 8 | "testing" |
| 9 | "time" |
| 10 | ) |
| 11 | |
| 12 | func TestPlan14ExecutionShutdownJoinsAcquisition(t *testing.T) { |
| 13 | b, disk, remote, _, q, peer := brokerFixture(t) |
| 14 | entered, resume := make(chan struct{}), make(chan struct{}) |
| 15 | remote.onPut = func(int) { close(entered); <-resume } |
| 16 | result := make(chan error, 1) |
| 17 | go func() { _, err := b.Execute(t.Context(), peer, q); result <- err }() |
| 18 | <-entered |
| 19 | closing := make(chan error, 1) |
| 20 | go func() { closing <- b.Close(t.Context()) }() |
| 21 | var early bool |
| 22 | select { |
| 23 | case <-closing: |
| 24 | early = true |
| 25 | case <-time.After(50 * time.Millisecond): |
| 26 | } |
| 27 | probe, cancel := context.WithTimeout(t.Context(), 20*time.Millisecond) |
| 28 | unlock, lockErr := disk.Lock(probe, q.Host+".lock") |
| 29 | cancel() |
| 30 | if lockErr == nil { |
| 31 | unlock() |
| 32 | } |
| 33 | close(resume) |
| 34 | if err := <-result; err == nil { |
| 35 | t.Error("acquisition returned a permit during shutdown") |
| 36 | } |
| 37 | if !early { |
| 38 | if err := <-closing; err != nil { |
| 39 | t.Fatal(err) |
| 40 | } |
| 41 | } |
| 42 | if early || !errors.Is(lockErr, context.DeadlineExceeded) { |
| 43 | t.Fatalf("shutdown released ownership before acquisition joined: early=%v lock=%v", early, lockErr) |
| 44 | } |
| 45 | r, err := b.read(q.Host) |
| 46 | if err != nil || r.Status != "lost" || r.LeaseETag == "" { |
| 47 | t.Fatalf("late acquisition evidence was lost: %+v %v", r, err) |
| 48 | } |
| 49 | probe, cancel = context.WithTimeout(t.Context(), time.Second) |
| 50 | defer cancel() |
| 51 | unlock, err = disk.Lock(probe, q.Host+".lock") |
| 52 | if err != nil { |
| 53 | t.Fatal(err) |
| 54 | } |
| 55 | unlock() |
| 56 | } |
| 57 | |
| 58 | func TestPlan14ExecutionShutdownRevokesBeforeCleanup(t *testing.T) { |
| 59 | b, disk, _, _, q, peer := brokerFixture(t) |
| 60 | if _, err := b.Execute(t.Context(), peer, q); err != nil { |
| 61 | t.Fatal(err) |
| 62 | } |
| 63 | entered, resume := make(chan struct{}), make(chan struct{}) |
| 64 | b.cfg.Cleanup = func(context.Context, ExecutionRecord) error { close(entered); <-resume; return nil } |
| 65 | closing := make(chan error, 1) |
| 66 | go func() { closing <- b.Close(t.Context()) }() |
| 67 | <-entered |
| 68 | held, err := disk.LiveHeld(q.Host + ".live") |
| 69 | close(resume) |
| 70 | if closeErr := <-closing; closeErr != nil { |
| 71 | t.Fatal(closeErr) |
| 72 | } |
| 73 | if err != nil || held { |
| 74 | t.Fatalf("cleanup began before live permission was withdrawn: %v", err) |
| 75 | } |
| 76 | q.ID = "after-shutdown" |
| 77 | if _, err := b.Execute(t.Context(), peer, q); err == nil { |
| 78 | t.Fatal("closed broker accepted new work") |
| 79 | } |
| 80 | } |
| 81 | |
| 82 | func TestPlan14ExecutionShutdownTimeoutRetainsOwnership(t *testing.T) { |
| 83 | b, disk, remote, _, q, peer := brokerFixture(t) |
| 84 | entered, resume := make(chan struct{}), make(chan struct{}) |
| 85 | remote.onPut = func(int) { close(entered); <-resume } |
| 86 | result := make(chan error, 1) |
| 87 | go func() { _, err := b.Execute(t.Context(), peer, q); result <- err }() |
| 88 | <-entered |
| 89 | bounded, cancel := context.WithTimeout(t.Context(), 20*time.Millisecond) |
| 90 | err := b.Close(bounded) |
| 91 | cancel() |
| 92 | if !errors.Is(err, context.DeadlineExceeded) { |
| 93 | close(resume) |
| 94 | <-result |
| 95 | t.Fatalf("shutdown did not retain the outstanding acquisition: %v", err) |
| 96 | } |
| 97 | probe, cancel := context.WithTimeout(t.Context(), 20*time.Millisecond) |
| 98 | unlock, lockErr := disk.Lock(probe, q.Host+".lock") |
| 99 | cancel() |
| 100 | if lockErr == nil { |
| 101 | unlock() |
| 102 | } |
| 103 | close(resume) |
| 104 | if err := <-result; err == nil { |
| 105 | t.Fatal("late acquisition returned permission") |
| 106 | } |
| 107 | if err := b.Close(t.Context()); err != nil { |
| 108 | t.Fatal(err) |
| 109 | } |
| 110 | if !errors.Is(lockErr, context.DeadlineExceeded) { |
| 111 | t.Fatalf("timeout abandoned ownership: %v", lockErr) |
| 112 | } |
| 113 | } |
| 114 | |