| 1 | package agent |
| 2 | |
| 3 | import ( |
| 4 | "errors" |
| 5 | "os" |
| 6 | "strings" |
| 7 | "testing" |
| 8 | "time" |
| 9 | |
| 10 | "reasonix/internal/fileutil" |
| 11 | "reasonix/internal/provider" |
| 12 | "reasonix/internal/store" |
| 13 | ) |
| 14 | |
| 15 | // holdLockAndShortenWait makes the next save time out on the file lock the |
| 16 | // way a stalled peer process would cause it to. |
| 17 | func holdLockAndShortenWait(t *testing.T, path string) (release func()) { |
| 18 | t.Helper() |
| 19 | restore := SetSessionFileLockWaitForTest(40*time.Millisecond, 5*time.Millisecond) |
| 20 | t.Cleanup(restore) |
| 21 | held, err := HoldSessionFileLockForTest(path) |
| 22 | if err != nil { |
| 23 | t.Fatalf("hold session lock: %v", err) |
| 24 | } |
| 25 | released := false |
| 26 | release = func() { |
| 27 | if !released { |
| 28 | released = true |
| 29 | held() |
| 30 | } |
| 31 | } |
| 32 | t.Cleanup(release) |
| 33 | return release |
| 34 | } |
| 35 | |
| 36 | func headsByKind(t *testing.T, path string) (main, other SessionHead) { |
| 37 | t.Helper() |
| 38 | heads, err := ListSessionHeads(path) |
| 39 | if err != nil || len(heads) != 2 { |
| 40 | t.Fatalf("heads = %+v err=%v, want main plus one more", heads, err) |
| 41 | } |
| 42 | return heads[0], heads[1] |
| 43 | } |
| 44 | |
| 45 | func TestAppendForShutdownWithoutLockKeepsTailOnFreshHead(t *testing.T) { |
| 46 | path := dagTestSession(t) |
| 47 | s := dagSavedSession(t, path, "q1", "a1") |
| 48 | s.Add(provider.Message{Role: provider.RoleUser, Content: "q2"}) |
| 49 | s.Add(provider.Message{Role: provider.RoleAssistant, Content: "a2"}) |
| 50 | release := holdLockAndShortenWait(t, path) |
| 51 | if err := s.SaveSnapshot(path); !errors.Is(err, ErrSessionFileLockHeld) { |
| 52 | t.Fatalf("locked save err = %v, want ErrSessionFileLockHeld", err) |
| 53 | } |
| 54 | handled, err := s.AppendForShutdownWithoutLock(path, false) |
| 55 | if err != nil || !handled { |
| 56 | t.Fatalf("AppendForShutdownWithoutLock = handled %v err %v", handled, err) |
| 57 | } |
| 58 | assertNoTranscriptCopies(t, path) |
| 59 | main, tail := headsByKind(t, path) |
| 60 | if tail.Kind != HeadKindConcurrent || !tail.Selected || tail.MessageCount != 5 || tail.ParentHead != SessionMainHead { |
| 61 | t.Fatalf("shutdown head = %+v", tail) |
| 62 | } |
| 63 | if !main.Covered || main.MessageCount != 3 { |
| 64 | t.Fatalf("main head after unlocked append = %+v, want it covered by the shutdown head", main) |
| 65 | } |
| 66 | if ref, ok := s.Head(); !ok || ref.HeadID != tail.ID { |
| 67 | t.Fatalf("session head = %+v ok=%v, want the shutdown head", ref, ok) |
| 68 | } |
| 69 | events := s.DrainHeadEvents() |
| 70 | if len(events) != 1 || events[0].Kind != HeadEventForkedConcurrent || events[0].HeadID != tail.ID { |
| 71 | t.Fatalf("head events = %+v", events) |
| 72 | } |
| 73 | release() |
| 74 | // The next locked save continues on the shutdown head and finally refreshes |
| 75 | // the derived files it skipped. |
| 76 | if err := s.Save(path); err != nil { |
| 77 | t.Fatalf("locked save after shutdown append: %v", err) |
| 78 | } |
| 79 | if meta, _, _ := LoadBranchMeta(path); meta.HeadID != tail.ID || meta.HeadCount != 2 { |
| 80 | t.Fatalf("meta mirror after the next locked save = head %q count %d", meta.HeadID, meta.HeadCount) |
| 81 | } |
| 82 | s.Add(provider.Message{Role: provider.RoleUser, Content: "q3"}) |
| 83 | if err := s.Save(path); err != nil { |
| 84 | t.Fatal(err) |
| 85 | } |
| 86 | st := dagReplay(t, path) |
| 87 | if got := strings.Join(dagChain(st, tail.ID), ","); got != "sys,q1,a1,q2,a2,q3" { |
| 88 | t.Fatalf("shutdown head chain = %s", got) |
| 89 | } |
| 90 | if len(st.heads) != 2 { |
| 91 | t.Fatalf("a later locked save must not fork again: %d heads", len(st.heads)) |
| 92 | } |
| 93 | reloaded, err := LoadSession(path) |
| 94 | if err != nil { |
| 95 | t.Fatal(err) |
| 96 | } |
| 97 | if ref, _ := reloaded.Head(); ref.HeadID != tail.ID || len(reloaded.Messages) != 6 { |
| 98 | t.Fatalf("reload = head %q with %d messages", ref.HeadID, len(reloaded.Messages)) |
| 99 | } |
| 100 | } |
| 101 | |
| 102 | func TestAppendForShutdownWithoutLockAfterTornTail(t *testing.T) { |
| 103 | path := dagTestSession(t) |
| 104 | s := dagSavedSession(t, path, "q1", "a1") |
| 105 | s.Add(provider.Message{Role: provider.RoleUser, Content: "q2"}) |
| 106 | logPath := store.SessionEventLog(path) |
| 107 | f, err := os.OpenFile(logPath, os.O_WRONLY|os.O_APPEND, 0o600) |
| 108 | if err != nil { |
| 109 | t.Fatal(err) |
| 110 | } |
| 111 | if _, err := f.WriteString(`{"schema_version":2,"type":"message","id":"torn-by-a-crash","head":"main","msgs":[{"role":"user","con`); err != nil { |
| 112 | t.Fatal(err) |
| 113 | } |
| 114 | _ = f.Close() |
| 115 | release := holdLockAndShortenWait(t, path) |
| 116 | if err := s.SaveSnapshot(path); !errors.Is(err, ErrSessionFileLockHeld) { |
| 117 | t.Fatalf("locked save err = %v", err) |
| 118 | } |
| 119 | if handled, err := s.AppendForShutdownWithoutLock(path, false); err != nil || !handled { |
| 120 | t.Fatalf("AppendForShutdownWithoutLock = handled %v err %v", handled, err) |
| 121 | } |
| 122 | st := dagReplay(t, path) |
| 123 | if st.damaged || st.holes != 1 { |
| 124 | t.Fatalf("replay after append behind a torn tail: damaged=%v holes=%d", st.damaged, st.holes) |
| 125 | } |
| 126 | _, tail := headsByKind(t, path) |
| 127 | if got := strings.Join(dagChain(st, tail.ID), ","); got != "sys,q1,a1,q2" { |
| 128 | t.Fatalf("shutdown head chain = %s", got) |
| 129 | } |
| 130 | raw, _ := os.ReadFile(logPath) |
| 131 | if !strings.Contains(string(raw), "torn-by-a-crash") { |
| 132 | t.Fatal("the torn line must stay in the log until a rotation reclaims it") |
| 133 | } |
| 134 | release() |
| 135 | reloaded, err := LoadSession(path) |
| 136 | if err != nil || len(reloaded.Messages) != 4 || reloaded.Messages[3].Content != "q2" { |
| 137 | t.Fatalf("reload = %d messages err=%v", len(reloaded.Messages), err) |
| 138 | } |
| 139 | // With the lease held (the single-writer proof), the next locked save |
| 140 | // rotates the hole away even though it has nothing new to append. |
| 141 | lease, err := TryAcquireSessionLease(path) |
| 142 | if err != nil { |
| 143 | t.Fatalf("TryAcquireSessionLease: %v", err) |
| 144 | } |
| 145 | defer lease.Release() |
| 146 | if err := reloaded.Save(path); err != nil { |
| 147 | t.Fatalf("locked save after a hole: %v", err) |
| 148 | } |
| 149 | if again := dagReplay(t, path); again.holes != 0 || again.generation != 2 { |
| 150 | t.Fatalf("the next locked save must rotate the hole away: holes=%d generation=%d", again.holes, again.generation) |
| 151 | } |
| 152 | } |
| 153 | |
| 154 | func TestAppendForShutdownWithoutLockWritesMarkersOnTheSameHead(t *testing.T) { |
| 155 | path := dagTestSession(t) |
| 156 | s := dagSavedSession(t, path, "q1") |
| 157 | if !s.QueueTurnBegin("turn-1", false) { |
| 158 | t.Fatal("QueueTurnBegin refused") |
| 159 | } |
| 160 | s.Add(provider.Message{Role: provider.RoleAssistant, Content: "a1"}) |
| 161 | if err := s.Save(path); err != nil { |
| 162 | t.Fatal(err) |
| 163 | } |
| 164 | if !s.QueueTurnEnd("turn-1") { |
| 165 | t.Fatal("QueueTurnEnd refused") |
| 166 | } |
| 167 | release := holdLockAndShortenWait(t, path) |
| 168 | if err := s.SaveSnapshot(path); !errors.Is(err, ErrSessionFileLockHeld) { |
| 169 | t.Fatalf("locked save err = %v", err) |
| 170 | } |
| 171 | if handled, err := s.AppendForShutdownWithoutLock(path, false); err != nil || !handled { |
| 172 | t.Fatalf("AppendForShutdownWithoutLock = handled %v err %v", handled, err) |
| 173 | } |
| 174 | release() |
| 175 | st := dagReplay(t, path) |
| 176 | if len(st.heads) != 1 || st.heads[SessionMainHead].openTurn != nil { |
| 177 | t.Fatalf("a marker-only batch must close the turn on the same head: heads=%d openTurn=%+v", len(st.heads), st.heads[SessionMainHead].openTurn) |
| 178 | } |
| 179 | if got := dagEntryTypes(t, path); got[len(got)-1] != sessionDAGTypeTurnEnd { |
| 180 | t.Fatalf("entries = %v, want the turn_end appended last", got) |
| 181 | } |
| 182 | } |
| 183 | |
| 184 | func TestAppendForShutdownWithoutLockLeavesSchemaOneToRecovery(t *testing.T) { |
| 185 | useSchemaOneLog(t) |
| 186 | path := dagTestSession(t) |
| 187 | s := dagSavedSession(t, path, "q1") |
| 188 | s.Add(provider.Message{Role: provider.RoleAssistant, Content: "a1"}) |
| 189 | holdLockAndShortenWait(t, path) |
| 190 | if handled, err := s.AppendForShutdownWithoutLock(path, false); handled || err != nil { |
| 191 | t.Fatalf("schema-1 session = handled %v err %v, want the recovery-copy path", handled, err) |
| 192 | } |
| 193 | } |
| 194 | |
| 195 | func TestReplaySkipsTornLineBetweenEntries(t *testing.T) { |
| 196 | path := dagTestSession(t) |
| 197 | ids, base := dagLinearLog(t, path) |
| 198 | logPath := store.SessionEventLog(path) |
| 199 | f, err := os.OpenFile(logPath, os.O_WRONLY|os.O_APPEND, 0o600) |
| 200 | if err != nil { |
| 201 | t.Fatal(err) |
| 202 | } |
| 203 | if _, err := f.WriteString(`{"schema_version":2,"type":"message","id":"torn","head":"ma`); err != nil { |
| 204 | t.Fatal(err) |
| 205 | } |
| 206 | _ = f.Close() |
| 207 | torn := dagReplay(t, path) |
| 208 | if !torn.damaged || torn.holes != 0 || len(torn.nodes) != len(ids) { |
| 209 | t.Fatalf("torn tail must stay damaged: damaged=%v holes=%d nodes=%d", torn.damaged, torn.holes, len(torn.nodes)) |
| 210 | } |
| 211 | f, err = os.OpenFile(logPath, os.O_WRONLY|os.O_APPEND, 0o600) |
| 212 | if err != nil { |
| 213 | t.Fatal(err) |
| 214 | } |
| 215 | if _, err := f.WriteString("\n"); err != nil { |
| 216 | t.Fatal(err) |
| 217 | } |
| 218 | _ = f.Close() |
| 219 | dagAppend(t, path, dagMessageEntry(t, SessionMainHead, ids[len(ids)-1], "", dagMsg(provider.RoleAssistant, "after the hole", "after"), base.Add(time.Minute))) |
| 220 | st := dagReplay(t, path) |
| 221 | if st.damaged || st.holes != 1 || len(st.nodes) != len(ids)+1 { |
| 222 | t.Fatalf("replay past a hole: damaged=%v holes=%d nodes=%d", st.damaged, st.holes, len(st.nodes)) |
| 223 | } |
| 224 | if got := dagChain(st, SessionMainHead); got[len(got)-1] != "after the hole" { |
| 225 | t.Fatalf("chain = %v", got) |
| 226 | } |
| 227 | if repaired, err := repairSessionDAGTail(path, st, time.Now().Add(time.Hour)); repaired || err != nil { |
| 228 | t.Fatalf("a hole is not a torn tail: repaired=%v err=%v", repaired, err) |
| 229 | } |
| 230 | } |
| 231 | |
| 232 | func TestRotationCarriesAppendsThatLandedDuringReplace(t *testing.T) { |
| 233 | path := dagTestSession(t) |
| 234 | s := dagSavedSession(t, path, "q1", "a1") |
| 235 | st := dagReplay(t, path) |
| 236 | leaf := st.heads[SessionMainHead].leaf |
| 237 | sessionDAGRotateBeforeReplace = func(sessionPath string) { |
| 238 | dagAppend(t, sessionPath, dagMessageEntry(t, SessionMainHead, leaf, "", dagMsg(provider.RoleUser, "landed mid-rotation", "late"), time.Now().UTC())) |
| 239 | } |
| 240 | t.Cleanup(func() { sessionDAGRotateBeforeReplace = nil }) |
| 241 | if err := rotateSessionDAG(path, st, time.Now().UTC()); err != nil { |
| 242 | t.Fatalf("rotate: %v", err) |
| 243 | } |
| 244 | sessionDAGRotateBeforeReplace = nil |
| 245 | fresh := dagReplay(t, path) |
| 246 | if fresh.generation != 2 { |
| 247 | t.Fatalf("generation after rotation = %d", fresh.generation) |
| 248 | } |
| 249 | if got := strings.Join(dagChain(fresh, SessionMainHead), ","); got != "sys,q1,a1,landed mid-rotation" { |
| 250 | t.Fatalf("chain after rotation = %s, want the late append carried over", got) |
| 251 | } |
| 252 | _ = s |
| 253 | } |
| 254 | |
| 255 | func TestUnlockedAppendWaitsForRotationMarkerThenReappends(t *testing.T) { |
| 256 | path := dagTestSession(t) |
| 257 | dagSavedSession(t, path, "q1", "a1") |
| 258 | st := dagReplay(t, path) |
| 259 | leaf := st.heads[SessionMainHead].leaf |
| 260 | rotated, err := buildRotatedSessionDAG(st, time.Now().UTC()) |
| 261 | if err != nil { |
| 262 | t.Fatal(err) |
| 263 | } |
| 264 | data, err := encodeSessionDAGEntries(rotated, time.Now().UTC()) |
| 265 | if err != nil { |
| 266 | t.Fatal(err) |
| 267 | } |
| 268 | marker := store.SessionEventLogRotating(path) |
| 269 | if err := os.WriteFile(marker, nil, 0o600); err != nil { |
| 270 | t.Fatal(err) |
| 271 | } |
| 272 | // A rotation that already read the log publishes its replacement while the |
| 273 | // appender is waiting on the marker, then clears the marker. |
| 274 | published := make(chan struct{}) |
| 275 | go func() { |
| 276 | defer close(published) |
| 277 | time.Sleep(150 * time.Millisecond) |
| 278 | if err := fileutil.AtomicWriteFileStrict(store.SessionEventLog(path), data, 0o600); err != nil { |
| 279 | t.Error(err) |
| 280 | } |
| 281 | _ = os.Remove(marker) |
| 282 | }() |
| 283 | entry := dagMessageEntry(t, SessionMainHead, leaf, "", dagMsg(provider.RoleUser, "appended around the rotation", "around"), time.Now().UTC()) |
| 284 | started := time.Now() |
| 285 | if _, err := appendSessionDAGEntriesUnlocked(path, []sessionDAGEntry{entry}); err != nil { |
| 286 | t.Fatalf("unlocked append: %v", err) |
| 287 | } |
| 288 | <-published |
| 289 | if time.Since(started) < 150*time.Millisecond { |
| 290 | t.Fatal("the append returned before the rotation marker cleared") |
| 291 | } |
| 292 | fresh := dagReplay(t, path) |
| 293 | if fresh.generation != 2 { |
| 294 | t.Fatalf("generation = %d, want the rotated log", fresh.generation) |
| 295 | } |
| 296 | if got := strings.Join(dagChain(fresh, SessionMainHead), ","); got != "sys,q1,a1,appended around the rotation" { |
| 297 | t.Fatalf("chain after the rotation = %s, want the append re-landed in the new log", got) |
| 298 | } |
| 299 | } |
| 300 |