| 1 | package cgroup |
| 2 | |
| 3 | import ( |
| 4 | "errors" |
| 5 | "os" |
| 6 | "os/exec" |
| 7 | "path/filepath" |
| 8 | "strconv" |
| 9 | "testing" |
| 10 | ) |
| 11 | |
| 12 | func TestExecNameAndIsExecName(t *testing.T) { |
| 13 | if got := ExecName("keymail.dev"); got != "carlos-exec@keymail.dev" { |
| 14 | t.Fatalf("ExecName = %q, want %q", got, "carlos-exec@keymail.dev") |
| 15 | } |
| 16 | for name, want := range map[string]bool{ |
| 17 | "carlos-exec@keymail.dev": true, |
| 18 | "carlos-exec@": false, // no host — not one of ours |
| 19 | "supervise": false, // the agent's own leaf |
| 20 | "carlos-app@x.service": false, |
| 21 | "cgroup.procs": false, |
| 22 | } { |
| 23 | if got := IsExecName(name); got != want { |
| 24 | t.Errorf("IsExecName(%q) = %v, want %v", name, got, want) |
| 25 | } |
| 26 | } |
| 27 | } |
| 28 | |
| 29 | // TestParseSelfCgroup pins the /proc/self/cgroup v2 line format: the "0::" |
| 30 | // line carries the unified-hierarchy path. v1-only content (no 0:: line) |
| 31 | // and the root cgroup ("0::/") both report not-ok — neither is a subtree |
| 32 | // systemd could have delegated to us. |
| 33 | func TestParseSelfCgroup(t *testing.T) { |
| 34 | cases := []struct { |
| 35 | name, data, want string |
| 36 | ok bool |
| 37 | }{ |
| 38 | {"systemd service", "0::/system.slice/carlos-edge.service\n", "system.slice/carlos-edge.service", true}, |
| 39 | {"hybrid box, v2 line among v1", "12:pids:/system.slice\n0::/system.slice/carlos-edge.service\n", "system.slice/carlos-edge.service", true}, |
| 40 | {"v1 only", "12:pids:/system.slice\n11:cpu:/system.slice\n", "", false}, |
| 41 | {"root cgroup", "0::/\n", "", false}, |
| 42 | {"empty", "", "", false}, |
| 43 | } |
| 44 | for _, c := range cases { |
| 45 | got, ok := ParseSelfCgroup(c.data) |
| 46 | if got != c.want || ok != c.ok { |
| 47 | t.Errorf("%s: ParseSelfCgroup = (%q, %v), want (%q, %v)", c.name, got, ok, c.want, c.ok) |
| 48 | } |
| 49 | } |
| 50 | } |
| 51 | |
| 52 | // newDelegatedRoot builds the temp-dir shape of a delegated cgroup: the two |
| 53 | // control files systemd chowns to the unit's user when Delegate= is set. |
| 54 | // Everything the dance does is plain file I/O, so a temp dir is a faithful |
| 55 | // fixture — the same seam listCarlosUnits uses in internal/host. |
| 56 | // forgetKills clears the process-wide kill marks (Linux builds set it). The |
| 57 | // marks are keyed by (device, inode), so a mark an earlier test left would |
| 58 | // name a fresh test directory on a filesystem that reuses inodes (ext4, |
| 59 | // xfs), and the start would remove a cgroup that never took a kill. |
| 60 | // cgroupfs does not reuse cgroup inode ids; a temp dir can. |
| 61 | var forgetKills = func() {} |
| 62 | |
| 63 | func newDelegatedRoot(t *testing.T) string { |
| 64 | t.Helper() |
| 65 | forgetKills() |
| 66 | root := t.TempDir() |
| 67 | for _, f := range []string{"cgroup.procs", "cgroup.subtree_control"} { |
| 68 | if err := os.WriteFile(filepath.Join(root, f), nil, 0o644); err != nil { |
| 69 | t.Fatal(err) |
| 70 | } |
| 71 | } |
| 72 | return root |
| 73 | } |
| 74 | |
| 75 | // TestSetupAtRefusesUndelegatedRoot: without Delegate= the control files are |
| 76 | // root-owned (here: absent) — SetupAt must error so Setup degrades to the |
| 77 | // nil manager instead of half-arming. |
| 78 | func TestSetupAtRefusesUndelegatedRoot(t *testing.T) { |
| 79 | if _, err := SetupAt(t.TempDir(), t.Logf); err == nil { |
| 80 | t.Fatal("SetupAt on an undelegated root succeeded, want error") |
| 81 | } |
| 82 | } |
| 83 | |
| 84 | // TestSetupAtDance pins the dance's observable results: the agent's own PID |
| 85 | // moved into the supervise/ leaf (cgroup v2 forbids enabling controllers on |
| 86 | // a cgroup with member processes), then the four controllers enabled in |
| 87 | // subtree_control, and Root reporting the subtree. |
| 88 | func TestSetupAtDance(t *testing.T) { |
| 89 | root := newDelegatedRoot(t) |
| 90 | m, err := SetupAt(root, t.Logf) |
| 91 | if err != nil { |
| 92 | t.Fatal(err) |
| 93 | } |
| 94 | if m.Root() != root { |
| 95 | t.Errorf("Root = %q, want %q", m.Root(), root) |
| 96 | } |
| 97 | pid, err := os.ReadFile(filepath.Join(root, "supervise", "cgroup.procs")) |
| 98 | if err != nil { |
| 99 | t.Fatalf("supervise leaf not populated: %v", err) |
| 100 | } |
| 101 | if string(pid) != strconv.Itoa(os.Getpid()) { |
| 102 | t.Errorf("supervise/cgroup.procs = %q, want our own pid %d", pid, os.Getpid()) |
| 103 | } |
| 104 | ctrl, err := os.ReadFile(filepath.Join(root, "cgroup.subtree_control")) |
| 105 | if err != nil { |
| 106 | t.Fatal(err) |
| 107 | } |
| 108 | if string(ctrl) != "+cpu +memory +io +pids" { |
| 109 | t.Errorf("subtree_control = %q, want %q", ctrl, "+cpu +memory +io +pids") |
| 110 | } |
| 111 | } |
| 112 | |
| 113 | // TestSetupAtSweepsStaleExecDirs: a previous agent's per-host dirs are empty |
| 114 | // after its children died with the service; the dance removes them. A |
| 115 | // non-empty one (something still alive in it — shouldn't happen, but rmdir |
| 116 | // would fail anyway) and a foreign name are left alone. |
| 117 | func TestSetupAtSweepsStaleExecDirs(t *testing.T) { |
| 118 | root := newDelegatedRoot(t) |
| 119 | stale := filepath.Join(root, "carlos-exec@old.example.com") |
| 120 | if err := os.Mkdir(stale, 0o755); err != nil { |
| 121 | t.Fatal(err) |
| 122 | } |
| 123 | occupied := filepath.Join(root, "carlos-exec@busy.example.com") |
| 124 | if err := os.MkdirAll(filepath.Join(occupied, "child"), 0o755); err != nil { |
| 125 | t.Fatal(err) |
| 126 | } |
| 127 | foreign := filepath.Join(root, "not-ours") |
| 128 | if err := os.Mkdir(foreign, 0o755); err != nil { |
| 129 | t.Fatal(err) |
| 130 | } |
| 131 | |
| 132 | if _, err := SetupAt(root, t.Logf); err != nil { |
| 133 | t.Fatal(err) |
| 134 | } |
| 135 | if _, err := os.Stat(stale); !os.IsNotExist(err) { |
| 136 | t.Errorf("stale exec dir survived the sweep: %v", err) |
| 137 | } |
| 138 | if _, err := os.Stat(occupied); err != nil { |
| 139 | t.Errorf("non-empty exec dir must be tolerated, not force-removed: %v", err) |
| 140 | } |
| 141 | if _, err := os.Stat(foreign); err != nil { |
| 142 | t.Errorf("foreign dir must be untouched: %v", err) |
| 143 | } |
| 144 | } |
| 145 | |
| 146 | func TestNilManagerRoot(t *testing.T) { |
| 147 | var m *Manager |
| 148 | if got := m.Root(); got != "" { |
| 149 | t.Fatalf("nil Root = %q, want \"\"", got) |
| 150 | } |
| 151 | } |
| 152 | |
| 153 | // fakeStart swaps the package's spawn seam for a recorder, restoring it on |
| 154 | // cleanup. The fake never spawns anything: "success" fabricates cmd.Process |
| 155 | // from our own PID so the PID-move path has a real number to write. |
| 156 | func fakeStart(t *testing.T, fail func(cmd *exec.Cmd) error) *[]*exec.Cmd { |
| 157 | t.Helper() |
| 158 | var calls []*exec.Cmd |
| 159 | orig := startCmd |
| 160 | startCmd = func(cmd *exec.Cmd) error { |
| 161 | calls = append(calls, cmd) |
| 162 | if err := fail(cmd); err != nil { |
| 163 | return err |
| 164 | } |
| 165 | cmd.Process, _ = os.FindProcess(os.Getpid()) |
| 166 | return nil |
| 167 | } |
| 168 | t.Cleanup(func() { startCmd = orig }) |
| 169 | return &calls |
| 170 | } |
| 171 | |
| 172 | func buildTrue() *exec.Cmd { return exec.Command("/bin/true") } |
| 173 | |
| 174 | // TestStartNilManagerSpawnsPlainly: the nil manager is the no-delegation |
| 175 | // state — one plain start, no SysProcAttr, no filesystem writes. |
| 176 | func TestStartNilManagerSpawnsPlainly(t *testing.T) { |
| 177 | calls := fakeStart(t, func(*exec.Cmd) error { return nil }) |
| 178 | var m *Manager |
| 179 | cmd, err := m.Start("h.example.com", Limit{}, buildTrue) |
| 180 | if err != nil || cmd == nil { |
| 181 | t.Fatalf("Start = (%v, %v), want started cmd", cmd, err) |
| 182 | } |
| 183 | if len(*calls) != 1 || (*calls)[0].SysProcAttr != nil { |
| 184 | t.Fatalf("nil manager must start exactly once with no SysProcAttr: %d calls", len(*calls)) |
| 185 | } |
| 186 | } |
| 187 | |
| 188 | // TestStartGenuineSpawnFailureIsReturned: when the start itself fails (here |
| 189 | // the PID-move start a temp dir's placement failure falls back to), the |
| 190 | // error propagates and the host dir this spawn made does not survive it. |
| 191 | func TestStartGenuineSpawnFailureIsReturned(t *testing.T) { |
| 192 | root := newDelegatedRoot(t) |
| 193 | m, err := SetupAt(root, t.Logf) |
| 194 | if err != nil { |
| 195 | t.Fatal(err) |
| 196 | } |
| 197 | fakeStart(t, func(*exec.Cmd) error { return errors.New("no such binary") }) |
| 198 | if _, err := m.Start("h.example.com", Limit{}, buildTrue); err == nil { |
| 199 | t.Fatal("Start with a failing spawn succeeded, want error") |
| 200 | } |
| 201 | if _, err := os.Stat(m.hostDir("h.example.com")); !os.IsNotExist(err) { |
| 202 | t.Errorf("ghost host dir survived a failed spawn: %v", err) |
| 203 | } |
| 204 | } |
| 205 | |
| 206 | // TestCleanupHost: removes an empty per-host dir, tolerates a non-empty |
| 207 | // one, and is nil-safe. |
| 208 | func TestCleanupHost(t *testing.T) { |
| 209 | root := newDelegatedRoot(t) |
| 210 | m, err := SetupAt(root, t.Logf) |
| 211 | if err != nil { |
| 212 | t.Fatal(err) |
| 213 | } |
| 214 | empty := filepath.Join(root, ExecName("gone.example.com")) |
| 215 | if err := os.Mkdir(empty, 0o755); err != nil { |
| 216 | t.Fatal(err) |
| 217 | } |
| 218 | m.CleanupHost("gone.example.com") |
| 219 | if _, err := os.Stat(empty); !os.IsNotExist(err) { |
| 220 | t.Errorf("empty host dir not removed: %v", err) |
| 221 | } |
| 222 | |
| 223 | busy := filepath.Join(root, ExecName("busy.example.com")) |
| 224 | if err := os.MkdirAll(filepath.Join(busy, "child"), 0o755); err != nil { |
| 225 | t.Fatal(err) |
| 226 | } |
| 227 | m.CleanupHost("busy.example.com") // must not panic or force-remove |
| 228 | if _, err := os.Stat(busy); err != nil { |
| 229 | t.Errorf("non-empty host dir must survive: %v", err) |
| 230 | } |
| 231 | |
| 232 | var nilM *Manager |
| 233 | nilM.CleanupHost("h.example.com") // must not panic |
| 234 | } |
| 235 | |
| 236 | // TestStartAndMoveRemovesGhostDirOnSpawnFailure: when the spawn itself |
| 237 | // never happens, the host cgroup dir must not survive it — a ghost dir |
| 238 | // would make the sampler report permanent zeros for a host with nothing |
| 239 | // running. |
| 240 | func TestStartAndMoveRemovesGhostDirOnSpawnFailure(t *testing.T) { |
| 241 | root := newDelegatedRoot(t) |
| 242 | m, err := SetupAt(root, t.Logf) |
| 243 | if err != nil { |
| 244 | t.Fatal(err) |
| 245 | } |
| 246 | dir := m.hostDir("h.example.com") |
| 247 | if err := os.MkdirAll(dir, 0o755); err != nil { |
| 248 | t.Fatal(err) |
| 249 | } |
| 250 | fakeStart(t, func(*exec.Cmd) error { return errors.New("no such binary") }) |
| 251 | if _, err := m.startAndMove("h.example.com", buildTrue); err == nil { |
| 252 | t.Fatal("startAndMove with a failing spawn succeeded, want error") |
| 253 | } |
| 254 | if _, err := os.Stat(dir); !os.IsNotExist(err) { |
| 255 | t.Errorf("ghost host dir survived a failed spawn: %v", err) |
| 256 | } |
| 257 | } |
| 258 | |
| 259 | // TestStartAndMoveToleratesBrokenSeam: a startCmd seam that reports success |
| 260 | // without ever setting cmd.Process must not panic while writing the PID — |
| 261 | // this package must never panic the agent over a broken seam. |
| 262 | func TestStartAndMoveToleratesBrokenSeam(t *testing.T) { |
| 263 | root := newDelegatedRoot(t) |
| 264 | m, err := SetupAt(root, t.Logf) |
| 265 | if err != nil { |
| 266 | t.Fatal(err) |
| 267 | } |
| 268 | if err := os.MkdirAll(m.hostDir("h.example.com"), 0o755); err != nil { |
| 269 | t.Fatal(err) |
| 270 | } |
| 271 | orig := startCmd |
| 272 | startCmd = func(cmd *exec.Cmd) error { return nil } // never sets cmd.Process |
| 273 | t.Cleanup(func() { startCmd = orig }) |
| 274 | if _, err := m.startAndMove("h.example.com", buildTrue); err != nil { |
| 275 | t.Fatal(err) |
| 276 | } |
| 277 | } |
| 278 | |
| 279 | // TestStartWritesMemoryLimit: a non-zero Limit lands in the host cgroup as |
| 280 | // memory.max and memory.high (high = 90% of max, so a tenant is throttled |
| 281 | // and reclaimed before it is killed). The temp-dir fixture has no kernel |
| 282 | // behind it, so the assertion is on the bytes written. |
| 283 | func TestStartWritesMemoryLimit(t *testing.T) { |
| 284 | root := newDelegatedRoot(t) |
| 285 | m, err := SetupAt(root, t.Logf) |
| 286 | if err != nil { |
| 287 | t.Fatal(err) |
| 288 | } |
| 289 | fakeStart(t, func(*exec.Cmd) error { return nil }) |
| 290 | if _, err := m.Start("h.example.com", Limit{MemoryMax: 1 << 30}, buildTrue); err != nil { |
| 291 | t.Fatal(err) |
| 292 | } |
| 293 | dir := m.hostDir("h.example.com") |
| 294 | for file, want := range map[string]string{ |
| 295 | "memory.max": "1073741824", |
| 296 | "memory.high": "966367638", // max/10*9, integer arithmetic |
| 297 | "memory.oom.group": "1", |
| 298 | "memory.swap.max": "1073741824", // = max: the cap is total, not RAM-only |
| 299 | } { |
| 300 | got, err := os.ReadFile(filepath.Join(dir, file)) |
| 301 | if err != nil { |
| 302 | t.Fatalf("%s not written: %v", file, err) |
| 303 | } |
| 304 | if string(got) != want { |
| 305 | t.Errorf("%s = %q, want %q", file, got, want) |
| 306 | } |
| 307 | } |
| 308 | } |
| 309 | |
| 310 | // TestStartZeroLimitWritesNothing: the zero Limit is "no cap" — the |
| 311 | // pre-spec behaviour, byte for byte, for any caller that has no tier. |
| 312 | func TestStartZeroLimitWritesNothing(t *testing.T) { |
| 313 | root := newDelegatedRoot(t) |
| 314 | m, err := SetupAt(root, t.Logf) |
| 315 | if err != nil { |
| 316 | t.Fatal(err) |
| 317 | } |
| 318 | fakeStart(t, func(*exec.Cmd) error { return nil }) |
| 319 | if _, err := m.Start("h.example.com", Limit{}, buildTrue); err != nil { |
| 320 | t.Fatal(err) |
| 321 | } |
| 322 | for _, file := range []string{"memory.max", "memory.high"} { |
| 323 | if _, err := os.Stat(filepath.Join(m.hostDir("h.example.com"), file)); !os.IsNotExist(err) { |
| 324 | t.Errorf("%s exists for a zero limit", file) |
| 325 | } |
| 326 | } |
| 327 | } |
| 328 | |
| 329 | // TestStartLimitWriteFailureStillSpawns: a cgroup that refuses the limit |
| 330 | // (read-only control file, an old kernel) costs one log line and an |
| 331 | // uncapped child — never a failed spawn. Same discipline as every other |
| 332 | // accounting failure in this package. |
| 333 | func TestStartLimitWriteFailureStillSpawns(t *testing.T) { |
| 334 | root := newDelegatedRoot(t) |
| 335 | m, err := SetupAt(root, t.Logf) |
| 336 | if err != nil { |
| 337 | t.Fatal(err) |
| 338 | } |
| 339 | dir := m.hostDir("h.example.com") |
| 340 | if err := os.MkdirAll(dir, 0o755); err != nil { |
| 341 | t.Fatal(err) |
| 342 | } |
| 343 | // A directory where the file should go: WriteFile fails with EISDIR. |
| 344 | if err := os.Mkdir(filepath.Join(dir, "memory.max"), 0o755); err != nil { |
| 345 | t.Fatal(err) |
| 346 | } |
| 347 | fakeStart(t, func(*exec.Cmd) error { return nil }) |
| 348 | cmd, err := m.Start("h.example.com", TierDefault, buildTrue) |
| 349 | if err != nil || cmd == nil { |
| 350 | t.Fatalf("Start = (%v, %v), want a started cmd despite the limit failure", cmd, err) |
| 351 | } |
| 352 | // max failed, so high must NOT have been written: a high with no max |
| 353 | // is a throttle with no wall, and the doc comment promises the short |
| 354 | // circuit. |
| 355 | if _, err := os.Stat(filepath.Join(dir, "memory.high")); !os.IsNotExist(err) { |
| 356 | t.Errorf("memory.high written after memory.max failed: %v", err) |
| 357 | } |
| 358 | } |
| 359 | |
| 360 | // TestStartBiasesChildOOMScore: a started child gets ExecOOMScoreAdj |
| 361 | // written to its oom_score_adj (redirected into a temp dir here), on both |
| 362 | // spawn paths. Without it the agent — the largest process in the edge |
| 363 | // tree — is the memcg OOM killer's first pick when the outer wall fires. |
| 364 | func TestStartBiasesChildOOMScore(t *testing.T) { |
| 365 | root := newDelegatedRoot(t) |
| 366 | m, err := SetupAt(root, t.Logf) |
| 367 | if err != nil { |
| 368 | t.Fatal(err) |
| 369 | } |
| 370 | adj := t.TempDir() |
| 371 | orig := oomScoreAdjPath |
| 372 | oomScoreAdjPath = func(pid int) string { return filepath.Join(adj, strconv.Itoa(pid)) } |
| 373 | t.Cleanup(func() { oomScoreAdjPath = orig }) |
| 374 | // A real start: /bin/true gives the child a pid to write for. |
| 375 | cmd, err := m.startAndMove("h.example.com", buildTrue) |
| 376 | if err != nil { |
| 377 | t.Fatal(err) |
| 378 | } |
| 379 | _ = cmd.Wait() |
| 380 | got, err := os.ReadFile(filepath.Join(adj, strconv.Itoa(cmd.Process.Pid))) |
| 381 | if err != nil { |
| 382 | t.Fatalf("oom_score_adj not written: %v", err) |
| 383 | } |
| 384 | if string(got) != ExecOOMScoreAdj { |
| 385 | t.Errorf("oom_score_adj = %q, want %q", got, ExecOOMScoreAdj) |
| 386 | } |
| 387 | } |
| 388 | |
| 389 | func TestTiers(t *testing.T) { |
| 390 | if TierDefault.MemoryMax != 512<<20 { |
| 391 | t.Errorf("TierDefault = %d, want 512M — the carlos-app@ template's number", TierDefault.MemoryMax) |
| 392 | } |
| 393 | if TierChildren.MemoryMax != 1<<30 { |
| 394 | t.Errorf("TierChildren = %d, want 1G — caps/render.go's children tier", TierChildren.MemoryMax) |
| 395 | } |
| 396 | } |
| 397 | |
| 398 | func TestSetupAtRetainingPreservesClaimEvidence(t *testing.T) { |
| 399 | root := newDelegatedRoot(t) |
| 400 | retained := filepath.Join(root, "carlos-exec@claim-old.local") |
| 401 | if err := os.Mkdir(retained, 0755); err != nil { |
| 402 | t.Fatal(err) |
| 403 | } |
| 404 | if _, err := SetupAtRetaining(root, t.Logf); err != nil { |
| 405 | t.Fatal(err) |
| 406 | } |
| 407 | if _, err := os.Stat(retained); err != nil { |
| 408 | t.Fatalf("retained claim tree removed: %v", err) |
| 409 | } |
| 410 | } |
| 411 | |
amadan