| 1 | //go:build linux |
| 2 | |
| 3 | package hostfiles |
| 4 | |
| 5 | import ( |
| 6 | "context" |
| 7 | "os" |
| 8 | "path/filepath" |
| 9 | "sync/atomic" |
| 10 | "testing" |
| 11 | "time" |
| 12 | |
| 13 | "golang.org/x/sys/unix" |
| 14 | |
| 15 | "github.com/carlosframework/platform/internal/release" |
| 16 | ) |
| 17 | |
| 18 | // realWatcher returns a working inotify watcher, or skips the test where |
| 19 | // the kernel will not give one (a shared machine at its per-user limit). |
| 20 | // CI must not skip: see the plan's Task 7, step 10. |
| 21 | func realWatcher(t *testing.T, root string, skip func(string, bool) bool) watcher { |
| 22 | t.Helper() |
| 23 | w, err := newWatcher(root, skip) |
| 24 | if err != nil { |
| 25 | if os.Getenv("CARLOS_REQUIRE_INOTIFY") != "" { |
| 26 | t.Fatalf("inotify is required here: %v", err) |
| 27 | } |
| 28 | t.Skipf("no inotify instance available: %v", err) |
| 29 | } |
| 30 | t.Cleanup(func() { w.Close() }) |
| 31 | return w |
| 32 | } |
| 33 | |
| 34 | // startReal runs the loop on a real watcher with scans and polling off, |
| 35 | // and returns once the first full scan is done: everything after that is |
| 36 | // found by an inotify event or not at all. |
| 37 | func startReal(t *testing.T, s release.Store, dir string) { |
| 38 | t.Helper() |
| 39 | w := realWatcher(t, dir, nil) |
| 40 | p := open(t, s) |
| 41 | stop := startRun(t, p, dir, w, Options{Deadline: 50 * time.Millisecond, Scan: time.Hour, Poll: time.Hour}) |
| 42 | t.Cleanup(func() { stop() }) |
| 43 | } |
| 44 | |
| 45 | func TestWatcherPublishesAFileStillHeldOpen(t *testing.T) { |
| 46 | s, dir := release.NewDirStore(t.TempDir()), t.TempDir() |
| 47 | startReal(t, s, dir) |
| 48 | f, err := os.Create(filepath.Join(dir, "open.log")) |
| 49 | if err != nil { |
| 50 | t.Fatal(err) |
| 51 | } |
| 52 | defer f.Close() // deliberately not closed before the assertion |
| 53 | if _, err := f.WriteString("line one\n"); err != nil { |
| 54 | t.Fatal(err) |
| 55 | } |
| 56 | eventually(t, "open.log published while still open", func() bool { |
| 57 | return manifestHas(s, "open.log", sum([]byte("line one\n"))) |
| 58 | }) |
| 59 | // The first publication could have come from the file's creation. A |
| 60 | // second write through the same descriptor produces no create, no |
| 61 | // close and no rename: only IN_MODIFY can announce it. |
| 62 | if _, err := f.WriteString("line two\n"); err != nil { |
| 63 | t.Fatal(err) |
| 64 | } |
| 65 | eventually(t, "the second write published, still open", func() bool { |
| 66 | return manifestHas(s, "open.log", sum([]byte("line one\nline two\n"))) |
| 67 | }) |
| 68 | } |
| 69 | |
| 70 | func TestWatcherSeesAFileInADirectoryCreatedAfterItStarted(t *testing.T) { |
| 71 | s, dir := release.NewDirStore(t.TempDir()), t.TempDir() |
| 72 | startReal(t, s, dir) |
| 73 | write(t, dir, "new/deep/a.txt", "a") |
| 74 | eventually(t, "new/deep/a.txt published", func() bool { |
| 75 | return manifestHas(s, "new/deep/a.txt", sum([]byte("a"))) |
| 76 | }) |
| 77 | // And a later write inside the new directory is an event, not a scan. |
| 78 | time.Sleep(150 * time.Millisecond) |
| 79 | write(t, dir, "new/deep/b.txt", "b") |
| 80 | eventually(t, "new/deep/b.txt published", func() bool { |
| 81 | return manifestHas(s, "new/deep/b.txt", sum([]byte("b"))) |
| 82 | }) |
| 83 | } |
| 84 | |
| 85 | func TestWatcherPublishesARenameAsDeleteAndCreate(t *testing.T) { |
| 86 | s, dir := release.NewDirStore(t.TempDir()), t.TempDir() |
| 87 | write(t, dir, "tmp", "v") |
| 88 | startReal(t, s, dir) |
| 89 | if err := os.Rename(filepath.Join(dir, "tmp"), filepath.Join(dir, "final")); err != nil { |
| 90 | t.Fatal(err) |
| 91 | } |
| 92 | eventually(t, "final present and tmp tombstoned", func() bool { |
| 93 | m, _, _ := Load(context.Background(), s, pfx) |
| 94 | return m.Entries["tmp"].Deleted && m.Entries["final"].SHA256 == sum([]byte("v")) |
| 95 | }) |
| 96 | } |
| 97 | |
| 98 | func TestWatcherDoesNotWatchASkippedDirectory(t *testing.T) { |
| 99 | root := t.TempDir() |
| 100 | if err := os.MkdirAll(filepath.Join(root, "cache"), 0o755); err != nil { |
| 101 | t.Fatal(err) |
| 102 | } |
| 103 | w := realWatcher(t, root, func(rel string, isDir bool) bool { return isDir && rel == "cache" }) |
| 104 | write(t, root, "cache/x", "x") |
| 105 | select { |
| 106 | case ev := <-w.Events(): |
| 107 | t.Fatalf("event from a skipped directory: %+v", ev) |
| 108 | case <-time.After(150 * time.Millisecond): |
| 109 | } |
| 110 | } |
| 111 | |
| 112 | // unwatched builds a watcher with one directory that has no watch and no |
| 113 | // event on its way: the state an overflow leaves when it swallows that |
| 114 | // directory's creation. The directory exists before the watcher starts |
| 115 | // and is excluded while the watcher is built, so the kernel never queues |
| 116 | // an event that could install the watch later behind the test's back. |
| 117 | func unwatched(t *testing.T) (w *inotifyWatcher, root string) { |
| 118 | t.Helper() |
| 119 | root = t.TempDir() |
| 120 | if err := os.MkdirAll(filepath.Join(root, "lost"), 0o755); err != nil { |
| 121 | t.Fatal(err) |
| 122 | } |
| 123 | var building atomic.Bool |
| 124 | building.Store(true) |
| 125 | w = realWatcher(t, root, func(rel string, isDir bool) bool { |
| 126 | return building.Load() && isDir && rel == "lost" |
| 127 | }).(*inotifyWatcher) |
| 128 | building.Store(false) |
| 129 | return w, root |
| 130 | } |
| 131 | |
| 132 | // noEventFor fails if an event for rel arrives within 200ms. |
| 133 | func noEventFor(t *testing.T, w *inotifyWatcher, rel string) { |
| 134 | t.Helper() |
| 135 | deadline := time.After(200 * time.Millisecond) |
| 136 | for { |
| 137 | select { |
| 138 | case ev := <-w.Events(): |
| 139 | if ev.rel == rel { |
| 140 | t.Fatalf("an event for %s, which has no watch", rel) |
| 141 | } |
| 142 | case <-deadline: |
| 143 | return |
| 144 | } |
| 145 | } |
| 146 | } |
| 147 | |
| 148 | func eventFor(t *testing.T, w *inotifyWatcher, rel string) { |
| 149 | t.Helper() |
| 150 | deadline := time.After(2 * time.Second) |
| 151 | for { |
| 152 | select { |
| 153 | case ev := <-w.Events(): |
| 154 | if ev.rel == rel { |
| 155 | return |
| 156 | } |
| 157 | case <-deadline: |
| 158 | t.Fatalf("no event for %s", rel) |
| 159 | } |
| 160 | } |
| 161 | } |
| 162 | |
| 163 | // Plan round 1 finding 8: a kernel overflow may have swallowed a |
| 164 | // directory's creation, so coverage is uncertain until Repair has walked |
| 165 | // the tree again. Only Repair can restore it here: remove the walk from |
| 166 | // Repair and the last assertion fails. |
| 167 | func TestWatcherOverflowMarksCoverageUncertainUntilRepaired(t *testing.T) { |
| 168 | w, root := unwatched(t) |
| 169 | if w.Degraded() { |
| 170 | t.Fatal("degraded before anything went wrong") |
| 171 | } |
| 172 | write(t, root, "lost/before.txt", "b") |
| 173 | noEventFor(t, w, "lost/before.txt") // proves the directory really is unwatched |
| 174 | |
| 175 | w.handle(0, unix.IN_Q_OVERFLOW, "") |
| 176 | if !w.Degraded() { |
| 177 | t.Fatal("an overflow left coverage marked complete") |
| 178 | } |
| 179 | if err := w.Repair(); err != nil { |
| 180 | t.Fatal(err) |
| 181 | } |
| 182 | if w.Degraded() { |
| 183 | t.Fatal("still degraded after a successful repair") |
| 184 | } |
| 185 | write(t, root, "lost/after.txt", "a") |
| 186 | eventFor(t, w, "lost/after.txt") |
| 187 | } |
| 188 | |
| 189 | // Plan round 2 finding 5: a failure that arrives while Repair is walking |
| 190 | // may concern a directory the walk has already passed. That Repair must |
| 191 | // not clear it. |
| 192 | func TestWatcherRepairDoesNotEraseAFailureThatArrivedDuringItsWalk(t *testing.T) { |
| 193 | w, _ := unwatched(t) |
| 194 | w.handle(0, unix.IN_Q_OVERFLOW, "") |
| 195 | orig := afterRepairWalk |
| 196 | t.Cleanup(func() { afterRepairWalk = orig }) |
| 197 | afterRepairWalk = func() { |
| 198 | afterRepairWalk = func() {} |
| 199 | w.handle(0, unix.IN_Q_OVERFLOW, "") // a second overflow, after the walk |
| 200 | } |
| 201 | if err := w.Repair(); err != nil { |
| 202 | t.Fatal(err) |
| 203 | } |
| 204 | if !w.Degraded() { |
| 205 | t.Fatal("the repair cleared a failure newer than its walk") |
| 206 | } |
| 207 | if err := w.Repair(); err != nil { |
| 208 | t.Fatal(err) |
| 209 | } |
| 210 | if w.Degraded() { |
| 211 | t.Fatal("a repair with nothing new during its walk left the watcher degraded") |
| 212 | } |
| 213 | } |
| 214 | |
| 215 | func TestAStoppedWatcherCannotBeRepairedAndClosesItsChannel(t *testing.T) { |
| 216 | w := realWatcher(t, t.TempDir(), nil).(*inotifyWatcher) |
| 217 | w.Close() |
| 218 | select { |
| 219 | case _, open := <-w.Events(): |
| 220 | if open { |
| 221 | t.Fatal("an event after Close") |
| 222 | } |
| 223 | case <-time.After(2 * time.Second): |
| 224 | t.Fatal("the channel did not close") |
| 225 | } |
| 226 | if err := w.Repair(); err == nil { |
| 227 | t.Fatal("a stopped watcher reported a successful repair") |
| 228 | } |
| 229 | if !w.Degraded() { |
| 230 | t.Fatal("a stopped watcher is not degraded") |
| 231 | } |
| 232 | } |
| 233 | |
amadan