]> git.giacomolli.uk - onechat/blob - rebuild.go
Resume queued appends at their saved log ID
[onechat] / rebuild.go
1 package main
2
3 import (
4 "context"
5 "encoding/json"
6 "errors"
7 "fmt"
8 "io"
9 "io/fs"
10 "log/slog"
11 "os"
12 "path/filepath"
13 "slices"
14 "strings"
15 "time"
16 )
17
18 // rebuildCallLimit is the most model calls of a rebuild. A rebuild summarizes
19 // the whole merged log, which can need more calls than a probe sample, and the
20 // quota stop of the build run still guards the account.
21 const rebuildCallLimit = 1500
22
23 // rebuild merges added messages with the log of agent, in date order, and
24 // builds a new memory beside it. The caller holds the token lock. The model
25 // uses the original agent prompt and cache key. A failed or stopped build
26 // removes the new directory and leaves the original directory unchanged.
27 // Success copies the other agent files, adds this run's usage, and swaps the
28 // directories, with the original kept as a timestamped backup. Message IDs are
29 // new positions in the merged log, but kind, text, and date stay exact.
30 func rebuild(ctx context.Context, agent *Agent, added []DatedMessage, client *ModelClient, summarizer Summarizer, out io.Writer) (backup string, err error) {
31 err = rebuildIdle(agent.dir)
32 if err != nil {
33 return "", fmt.Errorf("check rebuild work: %w", err)
34 }
35
36 old, err := rebuildReadLog(ctx, agent.dir)
37 if err != nil {
38 return "", fmt.Errorf("read original log: %w", err)
39 }
40
41 entries, err := rebuildMerge(old, added)
42 if err != nil {
43 return "", fmt.Errorf("merge rebuild messages: %w", err)
44 }
45
46 root, err := os.MkdirTemp(filepath.Dir(agent.dir), ".agent-rebuild-")
47 if err != nil {
48 return "", fmt.Errorf("create rebuild directory: %w", err)
49 }
50
51 defer func() {
52 closeErr := os.RemoveAll(root)
53 if closeErr != nil {
54 err = errors.Join(err, fmt.Errorf("remove rebuild directory: %w", closeErr))
55 }
56 }()
57
58 memory, err := openMemory(ctx, root, time.Now, summarizer, slog.New(slog.DiscardHandler))
59 if err != nil {
60 return "", fmt.Errorf("open rebuilt memory: %w", err)
61 }
62
63 client.LimitCalls(rebuildCallLimit)
64
65 run := BuildRun{
66 memory: memory,
67 client: client,
68 usagePath: client.UsagePath(),
69 }
70
71 appended, stop := run.Run(ctx, entries)
72 sizes, err := memory.buildModelLines()
73 if stop == "" {
74 err = errors.Join(err, memory.Settle(ctx), memory.rebuildComplete(), memory.Flush(ctx))
75 }
76
77 err = errors.Join(err, memory.Close())
78 records, usageErr := buildReadUsage(client.UsagePath())
79 err = errors.Join(err, usageErr, buildWriteReport(out, appended, stop, records, sizes))
80 if stop != "" {
81 err = errors.Join(err, fmt.Errorf("rebuild stopped: %s", stop))
82 }
83
84 if err != nil {
85 return "", fmt.Errorf("finish rebuilt memory: %w", err)
86 }
87
88 err = rebuildCopy(agent.dir, root)
89 if err != nil {
90 return "", fmt.Errorf("copy other agent files: %w", err)
91 }
92
93 usage, err := os.ReadFile(client.UsagePath())
94 if errors.Is(err, os.ErrNotExist) {
95 err = nil
96 }
97
98 if err != nil {
99 return "", fmt.Errorf("read rebuild usage: %w", err)
100 }
101
102 if len(usage) > 0 {
103 err = rebuildAddUsage(filepath.Join(root, "usage.jsonl"), usage)
104 if err != nil {
105 return "", fmt.Errorf("append rebuild usage: %w", err)
106 }
107 }
108
109 if ctx.Err() != nil {
110 return "", ctx.Err()
111 }
112
113 backup = agent.dir + "-backup-" + time.Now().Format("20060102-150405.000000000")
114 err = os.Rename(agent.dir, backup)
115 if err != nil {
116 return "", fmt.Errorf("back up agent: %w", err)
117 }
118
119 err = os.Rename(root, agent.dir)
120 if err != nil {
121 restoreErr := os.Rename(backup, agent.dir)
122 if restoreErr != nil {
123 err = errors.Join(err, fmt.Errorf("restore original agent: %w", restoreErr))
124 }
125
126 return "", fmt.Errorf("install rebuilt agent: %w", err)
127 }
128
129 return backup, nil
130 }
131
132 // rebuildIdle refuses an agent with unfinished subagents, including ones that a
133 // daemon restart must resume. A missing work file lists no work.
134 func rebuildIdle(root string) error {
135 data, err := os.ReadFile(filepath.Join(root, subagentWorkFileName))
136 if errors.Is(err, os.ErrNotExist) {
137 return nil
138 }
139
140 if err != nil {
141 return fmt.Errorf("read running subagents: %w", err)
142 }
143
144 var work SubagentWorkFile
145
146 err = json.Unmarshal(data, &work)
147 if err != nil {
148 return fmt.Errorf("decode running subagents: %w", err)
149 }
150
151 if len(work.Running) > 0 {
152 return errors.New("rebuild refused: subagents have unfinished work")
153 }
154
155 return nil
156 }
157
158 // rebuildReadLog reads and validates the day files without a write or an index
159 // repair. Even an incomplete final line is an error, not a repair.
160 func rebuildReadLog(ctx context.Context, root string) ([]DatedMessage, error) {
161 paths, err := filepath.Glob(filepath.Join(root, "main", "*.jsonl"))
162 if err != nil {
163 return nil, fmt.Errorf("list message files: %w", err)
164 }
165
166 slices.Sort(paths)
167
168 var entries []DatedMessage
169
170 for _, path := range paths {
171 var day uint32
172
173 day, err = journalDayNumber(strings.TrimSuffix(filepath.Base(path), ".jsonl"))
174 if err != nil {
175 return nil, fmt.Errorf("parse message day: %w", err)
176 }
177
178 err = journalScanFile(ctx, path, day, 0, false, func(line []byte, _ Position) error {
179 var message Message
180
181 err = json.Unmarshal(line, &message)
182 if err != nil {
183 return fmt.Errorf("decode message: %w", err)
184 }
185
186 err = logValidateMessage(message)
187 if err != nil {
188 return fmt.Errorf("validate message: %w", err)
189 }
190
191 if message.ID != len(entries) {
192 return fmt.Errorf("expected message %d, found %d", len(entries), message.ID)
193 }
194
195 entries = append(entries, DatedMessage{
196 Kind: message.Kind,
197 Text: message.Text,
198 Date: message.Date,
199 })
200
201 return nil
202 })
203 if err != nil {
204 return nil, fmt.Errorf("read message file: %w", err)
205 }
206 }
207
208 return entries, nil
209 }
210
211 // rebuildMerge preserves the input order for equal dates, with old messages
212 // before added messages. Each date must be explicit and valid.
213 func rebuildMerge(old, added []DatedMessage) ([]DatedMessage, error) {
214 entries := append(slices.Clone(old), added...)
215 dates := make(map[string]time.Time)
216 for _, entry := range entries {
217 err := logValidateMessage(Message{
218 Kind: entry.Kind,
219 Text: entry.Text,
220 Size: len(entry.Text),
221 Date: entry.Date,
222 })
223 if err != nil {
224 return nil, fmt.Errorf("validate dated message: %w", err)
225 }
226
227 var date time.Time
228
229 date, err = time.Parse(time.RFC3339Nano, entry.Date)
230 if err != nil {
231 return nil, fmt.Errorf("parse merge date: %w", err)
232 }
233
234 dates[entry.Date] = date
235 }
236
237 slices.SortStableFunc(entries, func(a, b DatedMessage) int {
238 return dates[a.Date].Compare(dates[b.Date])
239 })
240
241 return entries, nil
242 }
243
244 // rebuildCopy copies the other agent directories and regular files. It excludes
245 // the memory and its lock, and refuses all other file types.
246 func rebuildCopy(source, target string) error {
247 err := filepath.WalkDir(source, func(path string, entry fs.DirEntry, err error) error {
248 if err != nil {
249 return fmt.Errorf("read agent entry: %w", err)
250 }
251
252 relative, err := filepath.Rel(source, path)
253 if err != nil {
254 return fmt.Errorf("find agent relative path: %w", err)
255 }
256
257 if relative == "." {
258 return nil
259 }
260
261 switch relative {
262 case "main", "tree", "index":
263 return filepath.SkipDir
264 case "lock", "view.json":
265 return nil
266 }
267
268 destination := filepath.Join(target, relative)
269 switch {
270 case entry.IsDir():
271 err = os.Mkdir(destination, 0o700)
272 if err != nil {
273 return fmt.Errorf("create agent directory: %w", err)
274 }
275 case entry.Type().IsRegular():
276 var data []byte
277
278 data, err = os.ReadFile(path)
279 if err != nil {
280 return fmt.Errorf("read agent file: %w", err)
281 }
282
283 err = os.WriteFile(destination, data, 0o600)
284 if err != nil {
285 return fmt.Errorf("copy agent file: %w", err)
286 }
287 default:
288 return fmt.Errorf("copy agent: unsupported file type for %s", relative)
289 }
290
291 return nil
292 })
293 if err != nil {
294 return fmt.Errorf("copy agent files: %w", err)
295 }
296
297 return nil
298 }
299
300 // rebuildAddUsage keeps both the original usage and the calls of this run.
301 func rebuildAddUsage(path string, data []byte) (err error) {
302 file, err := os.OpenFile(path, os.O_CREATE|os.O_WRONLY|os.O_APPEND, 0o600)
303 if err != nil {
304 return fmt.Errorf("open rebuilt usage: %w", err)
305 }
306
307 defer func() {
308 closeErr := file.Close()
309 if closeErr != nil {
310 err = errors.Join(err, fmt.Errorf("close rebuilt usage: %w", closeErr))
311 }
312 }()
313
314 _, err = file.Write(data)
315 if err != nil {
316 return fmt.Errorf("append rebuilt usage: %w", err)
317 }
318
319 return nil
320 }