]> git.giacomolli.uk - onechat/blob - daemon_message.go
8415e49d4b7110d596eaed1013e79e6f29ce09b7
[onechat] / daemon_message.go
1 package main
2
3 import (
4 "context"
5 "encoding/json"
6 "errors"
7 "fmt"
8 "os"
9 "path/filepath"
10 "slices"
11 "strings"
12 "sync"
13 )
14
15 // openDaemonMessages connects the root and its children. It restores queued
16 // messages before the loops start. A queue entry stays until its recipient
17 // finishes the turn that takes it, so a stop can interrupt delivery or a turn.
18 func openDaemonMessages(loops []*TurnLoop) (*DaemonMessages, error) {
19 m := &DaemonMessages{
20 path: filepath.Join(loops[0].agent.dir, "messages.json"),
21 ordered: append([]*TurnLoop(nil), loops...),
22 loops: make(map[string]*TurnLoop),
23 }
24 for _, loop := range loops {
25 m.loops[loop.agent.config.Name] = loop
26 loop.messages = m
27 loop.tools.messages = m
28 if loop.agent.config.Parent == "" {
29 loop.logger = loop.logger.With("name", loop.agent.config.Name)
30 }
31 }
32
33 for _, child := range loops[1:] {
34 err := loops[0].runs.subagents.Reserve(child.agent.config.Name)
35 if err != nil {
36 return nil, fmt.Errorf("reserve child name: %w", err)
37 }
38 }
39
40 data, err := os.ReadFile(m.path)
41 if err != nil && !errors.Is(err, os.ErrNotExist) {
42 return nil, fmt.Errorf("read agent messages: %w", err)
43 }
44
45 if err == nil {
46 err = json.Unmarshal(data, &m.state)
47 if err != nil {
48 return nil, fmt.Errorf("decode agent messages: %w", err)
49 }
50 }
51
52 if m.state.Next < 0 {
53 return nil, errors.New("invalid agent message sequence")
54 }
55
56 previous := 0
57 for _, message := range m.state.Pending {
58 source := m.loops[message.Source]
59 target := m.loops[message.Target]
60 if source == nil || target == nil || message.ID <= previous || message.ID > m.state.Next || (message.LogID != nil && *message.LogID < 0) {
61 return nil, errors.New("invalid queued agent message")
62 }
63
64 if (message.Kind != "user" || target.agent.config.Parent != message.Source) && (message.Kind != "work" || source.agent.config.Parent != message.Target) {
65 return nil, errors.New("queued agents are not parent and child")
66 }
67
68 previous = message.ID
69 }
70
71 for i := range m.state.Pending {
72 err = m.deliver(&m.state.Pending[i])
73 if err != nil {
74 return nil, fmt.Errorf("restore agent message: %w", err)
75 }
76 }
77
78 return m, nil
79 }
80
81 // DaemonMessages owns the durable queue between the root and its children. The
82 // mutex serializes queue writes and delivery. Backups take it before the locks
83 // of the loops, memory, and subagents. A stored log ID makes delivery
84 // idempotent, including a stop between the log append and the wake.
85 type DaemonMessages struct {
86 mutex sync.Mutex
87 path string
88 loops map[string]*TurnLoop
89 ordered []*TurnLoop
90 state DaemonMessageFile
91 failure error
92 }
93
94 // deliver records one queued message and wakes its recipient. The caller holds
95 // the mutex, or owns the queue before the loops start.
96 func (m *DaemonMessages) deliver(message *DaemonMessage) error {
97 loop := m.loops[message.Target]
98 logged, err := loop.memory.AppendOnce(message.Kind, message.Text, message.LogID, func(id int) error {
99 message.LogID = &id
100 return m.save()
101 })
102 if err != nil {
103 return fmt.Errorf("record agent message: %w", err)
104 }
105
106 status := ""
107 if message.Kind == "work" {
108 status = subagentFinished
109 }
110
111 loop.notify(Event{
112 Kind: message.Kind,
113 ID: logged.ID,
114 Text: message.Text,
115 Source: message.Source,
116 Status: status,
117 })
118 loop.Wake(logged)
119 loop.logger.Info("message received", "from", message.Source, "id", message.ID)
120
121 return nil
122 }
123
124 // save replaces the queue file under the mutex, or before the loops start.
125 func (m *DaemonMessages) save() error {
126 data, err := json.Marshal(m.state)
127 if err != nil {
128 return fmt.Errorf("encode agent messages: %w", err)
129 }
130
131 err = writeFileAtomic(m.path, append(data, '\n'))
132 if err != nil {
133 return fmt.Errorf("save agent messages: %w", err)
134 }
135
136 return nil
137 }
138
139 // AddChild holds the queue lock while open makes a child's files and loop. The
140 // loop joins backups before its first task can be delivered.
141 func (m *DaemonMessages) AddChild(parent, name string, open func() (*TurnLoop, error)) (*TurnLoop, error) {
142 m.mutex.Lock()
143 defer m.mutex.Unlock()
144
145 if m.loops[name] != nil {
146 return nil, fmt.Errorf("name %q is already used", name)
147 }
148
149 root := m.loops[parent]
150 if root == nil || root.agent.config.Parent != "" {
151 return nil, errors.New("only the root can create children")
152 }
153
154 if m.failure != nil {
155 return nil, fmt.Errorf("agent message queue stopped: %w", m.failure)
156 }
157
158 loop, err := open()
159 if err != nil {
160 return nil, fmt.Errorf("open child: %w", err)
161 }
162
163 loop.messages = m
164 loop.tools.messages = m
165 m.loops[name] = loop
166 m.ordered = append(m.ordered, loop)
167
168 return loop, nil
169 }
170
171 // Child returns only a direct child of parent. Loops live until the daemon
172 // stops, so the returned loop stays open for the caller's turn.
173 func (m *DaemonMessages) Child(parent, name string) *TurnLoop {
174 m.mutex.Lock()
175 defer m.mutex.Unlock()
176
177 loop := m.loops[name]
178 if loop == nil || loop.agent.config.Parent != parent {
179 return nil
180 }
181
182 return loop
183 }
184
185 // ChildState lists the names and responsibilities of a parent's children in
186 // name order. It returns no heading when the parent has no children.
187 func (m *DaemonMessages) ChildState(parent string) string {
188 m.mutex.Lock()
189 defer m.mutex.Unlock()
190
191 var lines []string
192
193 for name, loop := range m.loops {
194 if loop.agent.config.Parent == parent {
195 lines = append(lines, name+": "+loop.agent.config.Responsibility)
196 }
197 }
198
199 if len(lines) == 0 {
200 return ""
201 }
202
203 slices.Sort(lines)
204
205 return "\n\nChildren:\n" + strings.Join(lines, "\n")
206 }
207
208 // SnapshotAgents locks the queue and all registered loops, including children
209 // made during this daemon run, before snapshot copies their files.
210 func (m *DaemonMessages) SnapshotAgents(ctx context.Context, snapshot func() error) error {
211 return m.Snapshot(func() error {
212 return daemonSnapshot(ctx, m.ordered, snapshot)
213 })
214 }
215
216 // Tell queues a user message from parent to its own child, then wakes the
217 // child. It is the daemon entry point for parent-to-child messages.
218 func (m *DaemonMessages) Tell(parent, child, text string) error {
219 m.mutex.Lock()
220 defer m.mutex.Unlock()
221
222 source := m.loops[parent]
223 target := m.loops[child]
224 if source == nil || target == nil || target.agent.config.Parent != parent {
225 return errors.New("the recipient is not a child of the sender")
226 }
227
228 if m.failure != nil {
229 return fmt.Errorf("agent message queue stopped: %w", m.failure)
230 }
231
232 m.add(parent, child, "user", text)
233
234 err := m.save()
235 if err == nil {
236 err = m.deliver(&m.state.Pending[len(m.state.Pending)-1])
237 }
238
239 if err != nil {
240 return fmt.Errorf("send agent message: %w", m.failed(err))
241 }
242
243 return nil
244 }
245
246 // add appends a queue entry under the mutex. Its ID is never reused.
247 func (m *DaemonMessages) add(source, target, kind, text string) {
248 m.state.Next++
249 m.state.Pending = append(m.state.Pending, DaemonMessage{
250 ID: m.state.Next,
251 Source: source,
252 Target: target,
253 Kind: kind,
254 Text: text,
255 })
256 }
257
258 // failed stops all loops after a queue failure. Further sends must not change
259 // the queue, because the outcome of the failed write is not known.
260 func (m *DaemonMessages) failed(err error) error {
261 m.failure = err
262 for _, loop := range m.loops {
263 loop.fail(err)
264 }
265
266 return err
267 }
268
269 // Snapshot holds the queue lock while snapshot locks and copies agent files.
270 // Its callback must take the memory locks after this lock, as delivery does.
271 func (m *DaemonMessages) Snapshot(snapshot func() error) error {
272 m.mutex.Lock()
273 defer m.mutex.Unlock()
274
275 err := snapshot()
276 if err != nil {
277 return fmt.Errorf("copy agent message snapshot: %w", err)
278 }
279
280 return nil
281 }
282
283 // Finish acknowledges the queued messages taken by a turn and queues its final
284 // report in the same file replacement. A child with no final text sends
285 // nothing. Call after a final reply, a failure, or a user interrupt, but not
286 // when a daemon stop interrupts a model request.
287 func (m *DaemonMessages) Finish(loop *TurnLoop, active []Message, reply string) error {
288 m.mutex.Lock()
289 defer m.mutex.Unlock()
290
291 if m.failure != nil {
292 return fmt.Errorf("agent message queue stopped: %w", m.failure)
293 }
294
295 ids := make(map[int]bool, len(active))
296 for _, message := range active {
297 ids[message.ID] = true
298 }
299
300 name := loop.agent.config.Name
301 before := len(m.state.Pending)
302 m.state.Pending = slices.DeleteFunc(m.state.Pending, func(message DaemonMessage) bool {
303 return message.Target == name && message.LogID != nil && ids[*message.LogID]
304 })
305
306 parent := loop.agent.config.Parent
307 report := parent != "" && strings.TrimSpace(reply) != ""
308 if report {
309 m.add(name, parent, "work", "["+name+"] "+reply)
310 }
311
312 if !report && len(m.state.Pending) == before {
313 return nil
314 }
315
316 err := m.save()
317 if err == nil && report {
318 err = m.deliver(&m.state.Pending[len(m.state.Pending)-1])
319 }
320
321 if err != nil {
322 return fmt.Errorf("finish agent messages: %w", m.failed(err))
323 }
324
325 return nil
326 }
327
328 // DaemonMessageFile holds the sequence and the unacknowledged deliveries.
329 type DaemonMessageFile struct {
330 Next int `json:"next"`
331 Pending []DaemonMessage `json:"pending"`
332 }
333
334 // DaemonMessage is one queued delivery. LogID is set before the log append. A
335 // nil LogID means that delivery has not started.
336 type DaemonMessage struct {
337 ID int `json:"id"`
338 Source string `json:"source"`
339 Target string `json:"target"`
340 Kind string `json:"kind"`
341 Text string `json:"text"`
342 LogID *int `json:"log_id,omitempty"`
343 }