]> git.giacomolli.uk - onechat/blob - backup_test.go
Keep test daemon sockets under the path limit
[onechat] / backup_test.go
1 package main
2
3 import (
4 "archive/tar"
5 "context"
6 "database/sql"
7 "io"
8 "os"
9 "path/filepath"
10 "strings"
11 "sync"
12 "testing"
13 "time"
14
15 "github.com/stretchr/testify/require"
16 )
17
18 func TestRunningDaemonBackupRestoresTheChatAndLinks(t *testing.T) {
19 dir := t.TempDir()
20 whatsappTestSession(t, dir, true)
21
22 files := map[string]string{
23 "auth.json": `{"access_token":"fake-secret"}`,
24 "telegram.json": `{"token":"fake-token","owner":123}`,
25 "config.json": `{"endpoint":"api"}`,
26 "agent/instructions.md": "Keep the instructions.\n",
27 "agent/telegram-offset.json": "23\n",
28 }
29 for name, data := range files {
30 path := filepath.Join(dir, name)
31 require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o700))
32 require.NoError(t, os.WriteFile(path, []byte(data), 0o600))
33 }
34
35 memory, whatsapp, subagents := backupStartDaemon(t, dir)
36 require.True(t, whatsapp.isLinked())
37
38 _, err := memory.Append("note", "A durable decision.")
39 require.NoError(t, err)
40
41 _, err = memory.Append("note", "The next decision.")
42 require.NoError(t, err)
43 require.NoError(t, memory.Settle(t.Context()))
44
45 child, err := subagents.Create(t.Context())
46 require.NoError(t, err)
47
48 _, err = child.Append("note", "subagent")
49 require.NoError(t, err)
50
51 _, err = child.Append("user", "A saved task.")
52 require.NoError(t, err)
53
54 childText, err := child.Show()
55 require.NoError(t, err)
56
57 for _, excluded := range []string{"probe/skip", "agent-backup-example/skip"} {
58 path := filepath.Join(dir, excluded)
59 require.NoError(t, os.MkdirAll(filepath.Dir(path), 0o700))
60 require.NoError(t, os.WriteFile(path, []byte("excluded"), 0o600))
61 }
62
63 archive := filepath.Join(t.TempDir(), "daily.tar.gz")
64 t.Setenv("ONECHAT_DIR", dir)
65
66 var answer strings.Builder
67
68 require.NoError(t, backup(t.Context(), archive, &answer))
69 require.Regexp(t, `^`+archive+` \([0-9]+ bytes\)\n$`, answer.String())
70
71 info, err := os.Stat(archive)
72 require.NoError(t, err)
73 require.Equal(t, os.FileMode(0o600), info.Mode().Perm())
74
75 restored := t.TempDir()
76 require.NoError(t, backupRestore(t.Context(), archive, restored))
77
78 for name, data := range files {
79 bytes, err := os.ReadFile(filepath.Join(restored, name))
80 require.NoError(t, err)
81 require.Equal(t, data, string(bytes))
82 }
83
84 for _, excluded := range []string{"probe", "agent-backup-example", socketFileName, "agent/lock"} {
85 _, err = os.Stat(filepath.Join(restored, excluded))
86 require.ErrorIs(t, err, os.ErrNotExist)
87 }
88
89 restoredMemory, restoredWhatsApp, restoredSubagents := backupStartDaemon(t, restored)
90 require.Equal(t, memory.Len(), restoredMemory.Len())
91 require.Equal(t, memory.View(), restoredMemory.View())
92
93 for id := range memory.Len() {
94 expected, err := memory.Message(id)
95 require.NoError(t, err)
96
97 actual, err := restoredMemory.Message(id)
98 require.NoError(t, err)
99 require.Equal(t, expected, actual)
100 }
101
102 require.True(t, restoredWhatsApp.isLinked())
103 require.Equal(t, whatsapp.link().own, restoredWhatsApp.link().own)
104
105 restoredChild, err := restoredSubagents.Get(t.Context(), child.name)
106 require.NoError(t, err)
107
108 restoredText, err := restoredChild.Show()
109 require.NoError(t, err)
110 require.Equal(t, childText, restoredText)
111 }
112
113 func TestBackupRestoresEachChildWithItsOwnState(t *testing.T) {
114 dir := t.TempDir()
115 daemonChildConfig(t, dir, "Garden")
116 daemonChildConfig(t, dir, "Library")
117
118 server := newModelFake(t)
119 clock := newTestClock()
120 agents, err := openDaemonAgents(t.Context(), dir, server.URL, ModelAPIKey(modelTestKey), defaultConfig().Compactor, clock.Now, discardLog())
121 require.NoError(t, err)
122
123 defer func() {
124 require.NoError(t, agents.Close())
125 }()
126
127 whatsapp, _, _ := whatsappStartTest(t, dir, clock)
128
129 var loops []*TurnLoop
130
131 for _, agent := range agents {
132 loop, err := agent.Loop(whatsapp, func(Event) {
133 })
134 require.NoError(t, err)
135
136 loops = append(loops, loop)
137 _, err = loop.Submit("parent", "Keep the decision of "+agent.agent.config.Name)
138 require.NoError(t, err)
139 require.NoError(t, loop.turn(t.Context()))
140 require.NoError(t, agent.memory.Settle(t.Context()))
141
142 subagent, err := agent.subagents.Create(t.Context())
143 require.NoError(t, err)
144
145 _, err = subagent.Append("user", "Keep the task of "+agent.agent.config.Name)
146 require.NoError(t, err)
147 }
148
149 archive := filepath.Join(t.TempDir(), "children.tar.gz")
150 _, err = backupData(t.Context(), dir, archive, whatsapp, func(ctx context.Context, snapshot func() error) error {
151 return daemonSnapshot(ctx, loops, snapshot)
152 })
153 require.NoError(t, err)
154
155 restored := t.TempDir()
156 require.NoError(t, backupRestore(t.Context(), archive, restored))
157
158 restoredAgents, err := openDaemonAgents(t.Context(), restored, server.URL, ModelAPIKey(modelTestKey), defaultConfig().Compactor, clock.Now, discardLog())
159 require.NoError(t, err)
160
161 defer func() {
162 require.NoError(t, restoredAgents.Close())
163 }()
164
165 require.Len(t, restoredAgents, 3)
166
167 for i, agent := range agents {
168 other := restoredAgents[i]
169 require.Equal(t, agent.agent.config, other.agent.config)
170 require.Equal(t, agent.memory.View(), other.memory.View())
171 require.Equal(t, agent.memory.Len(), other.memory.Len())
172 require.Equal(t, modelReadUsage(t, agent.client.UsagePath()), modelReadUsage(t, other.client.UsagePath()))
173
174 for id := range agent.memory.Len() {
175 before, err := agent.memory.Message(id)
176 require.NoError(t, err)
177
178 after, err := other.memory.Message(id)
179 require.NoError(t, err)
180 require.Equal(t, before, after)
181 }
182
183 entries, err := os.ReadDir(filepath.Join(agent.agent.dir, subagentDirName))
184 require.NoError(t, err)
185 require.Len(t, entries, 1)
186
187 subagent, err := other.subagents.Get(t.Context(), entries[0].Name())
188 require.NoError(t, err)
189
190 text, err := subagent.Show()
191 require.NoError(t, err)
192 require.Contains(t, text, "Keep the task of "+agent.agent.config.Name)
193 }
194 }
195
196 func TestRestoreRefusesNonemptyDataWithoutChangingIt(t *testing.T) {
197 dir := t.TempDir()
198 path := filepath.Join(dir, "keep")
199 require.NoError(t, os.WriteFile(path, []byte("keep"), 0o600))
200
201 err := backupRestore(t.Context(), "missing.tar.gz", dir)
202 require.ErrorContains(t, err, "empty data directory")
203
204 entries, err := os.ReadDir(dir)
205 require.NoError(t, err)
206 require.Len(t, entries, 1)
207
208 data, err := os.ReadFile(path)
209 require.NoError(t, err)
210 require.Equal(t, "keep", string(data))
211 }
212
213 func TestRestoreRefusesTheDaemonTokenLock(t *testing.T) {
214 dir := t.TempDir()
215 tokens, err := lockTokens(dir)
216 require.NoError(t, err)
217
218 defer func() {
219 require.NoError(t, tokens.Close())
220 }()
221
222 err = backupRestore(t.Context(), "missing.tar.gz", dir)
223 require.ErrorIs(t, err, errTokensInUse)
224 }
225
226 func TestUnsafeRestoreInstallsNoArchiveFiles(t *testing.T) {
227 archive := backupTestArchive(t,
228 tar.Header{
229 Name: "safe",
230 Typeflag: tar.TypeReg,
231 Mode: 0o600,
232 },
233 tar.Header{
234 Name: "../escape",
235 Typeflag: tar.TypeReg,
236 Mode: 0o600,
237 },
238 )
239 dir := t.TempDir()
240 err := backupRestore(t.Context(), archive, dir)
241 require.ErrorContains(t, err, "unsafe archive path")
242
243 entries, err := os.ReadDir(dir)
244 require.NoError(t, err)
245 require.Len(t, entries, 1)
246 require.Equal(t, tokenLockFileName, entries[0].Name())
247 }
248
249 func TestBackupNeedsARunningDaemon(t *testing.T) {
250 t.Setenv("ONECHAT_DIR", t.TempDir())
251
252 err := backup(t.Context(), filepath.Join(t.TempDir(), "backup.tar.gz"), io.Discard)
253 require.EqualError(t, err, "backup needs a running daemon")
254 }
255
256 func TestWhatsAppSessionBackupIncludesCommittedWALWrites(t *testing.T) {
257 dir := t.TempDir()
258 whatsappTestSession(t, dir, true)
259
260 whatsapp, _, _ := whatsappStartTest(t, dir, newTestClock())
261 address, err := whatsappSessionAddress(filepath.Join(dir, whatsappDirName, whatsappSessionFileName))
262 require.NoError(t, err)
263
264 database, err := sql.Open("sqlite", address)
265 require.NoError(t, err)
266
267 defer func() {
268 require.NoError(t, database.Close())
269 }()
270
271 _, err = database.ExecContext(t.Context(), "PRAGMA journal_mode=WAL; CREATE TABLE backup_sample (value TEXT); INSERT INTO backup_sample VALUES ('saved');")
272 require.NoError(t, err)
273
274 target := filepath.Join(t.TempDir(), "copy", whatsappSessionFileName)
275 require.NoError(t, whatsapp.BackupSession(t.Context(), target))
276
277 address, err = whatsappSessionAddress(target)
278 require.NoError(t, err)
279
280 restored, err := sql.Open("sqlite", address)
281 require.NoError(t, err)
282
283 defer func() {
284 require.NoError(t, restored.Close())
285 }()
286
287 var value string
288
289 err = restored.QueryRowContext(t.Context(), "SELECT value FROM backup_sample").Scan(&value)
290 require.NoError(t, err)
291 require.Equal(t, "saved", value)
292
293 info, err := os.Stat(target)
294 require.NoError(t, err)
295 require.Equal(t, os.FileMode(0o600), info.Mode().Perm())
296 }
297
298 func TestSessionBackupWaitsForRemovalToFinish(t *testing.T) {
299 dir := t.TempDir()
300 whatsappTestSession(t, dir, true)
301
302 whatsapp, fake, _ := whatsappStartTest(t, dir, newTestClock())
303 disconnect := &BackupDisconnect{
304 WhatsAppClient: fake,
305 entered: make(chan struct{}),
306 release: make(chan struct{}),
307 }
308
309 var release sync.Once
310
311 defer release.Do(func() {
312 close(disconnect.release)
313 })
314
315 whatsapp.mutex.Lock()
316
317 whatsapp.client = disconnect
318 whatsapp.mutex.Unlock()
319
320 removed := make(chan error, 1)
321 go func() {
322 removed <- whatsapp.removeSession()
323 }()
324
325 <-disconnect.entered
326
327 target := filepath.Join(t.TempDir(), "copy", whatsappSessionFileName)
328 backedUp := make(chan error, 1)
329 started := make(chan struct{})
330 go func() {
331 close(started)
332
333 backedUp <- whatsapp.BackupSession(t.Context(), target)
334 }()
335
336 <-started
337 require.Never(t, func() bool {
338 return len(backedUp) != 0
339 }, 50*time.Millisecond, time.Millisecond)
340 release.Do(func() {
341 close(disconnect.release)
342 })
343 require.NoError(t, <-removed)
344 require.NoError(t, <-backedUp)
345
346 for _, path := range []string{whatsapp.sessionPath(), target} {
347 _, err := os.Stat(path)
348 require.ErrorIs(t, err, os.ErrNotExist)
349 }
350 }
351
352 // BackupDisconnect holds session removal in its unlocked client cleanup.
353 type BackupDisconnect struct {
354 WhatsAppClient
355 entered chan struct{}
356 release chan struct{}
357 }
358
359 func (d *BackupDisconnect) Disconnect() {
360 close(d.entered)
361 <-d.release
362 }
363
364 // backupStartDaemon serves fake model and WhatsApp clients on the socket of
365 // dir. It keeps all state files under dir and holds the daemon token lock.
366 func backupStartDaemon(t *testing.T, dir string) (*Memory, *WhatsApp, *Subagents) {
367 t.Helper()
368
369 tokens, err := lockTokens(dir)
370 require.NoError(t, err)
371
372 clock := newTestClock()
373 agent, err := openAgent(filepath.Join(dir, daemonAgentDirName))
374 require.NoError(t, err)
375
376 memory, err := openMemory(t.Context(), agent.dir, clock.Now, newFakeSummarizer(), discardLog())
377 require.NoError(t, err)
378
379 whatsapp, _, _ := whatsappStartTest(t, dir, clock)
380 subagents := newSubagents(filepath.Join(agent.dir, subagentDirName), clock.Now, subagentNameRandom)
381 schedules, err := openSchedules(filepath.Join(agent.dir, scheduleFileName), "OneChat", clock.Now, discardLog())
382 require.NoError(t, err)
383
384 server := newModelFake(t)
385 client := newModelClient(server.URL, ModelAPIKey(modelTestKey), agent.CacheKey(), filepath.Join(agent.dir, "usage.jsonl"))
386 listener, err := daemonListenSocket(filepath.Join(dir, socketFileName))
387 require.NoError(t, err)
388
389 ctx, cancel := context.WithCancel(t.Context())
390 done := make(chan error, 1)
391 go func() {
392 done <- daemonServe(ctx, listener, nil, whatsapp, schedules, func(notify func(Event)) (*TurnLoop, error) {
393 return newTurnLoop(agent, memory, subagents, whatsapp, schedules, client, clock.Now, discardLog(), notify)
394 })
395 }()
396
397 t.Cleanup(func() {
398 cancel()
399 require.NoError(t, <-done)
400 require.NoError(t, subagents.Close())
401 require.NoError(t, memory.Close())
402 require.NoError(t, tokens.Close())
403 })
404
405 return memory, whatsapp, subagents
406 }