返回 DeepSeek-Reasonix
heartbeat_store.go
根目录 / desktop / heartbeat_store.go
1 package main
2
3 import (
4 "context"
5 "crypto/sha256"
6 "encoding/json"
7 "errors"
8 "fmt"
9 "log"
10 "os"
11 "path/filepath"
12 "sort"
13 "time"
14
15 "reasonix/internal/fileutil"
16 filelock "reasonix/internal/identitylock"
17 )
18
19 func (e *HeartbeatEngine) readConfigSnapshot() (heartbeatConfigSnapshot, error) {
20 path := e.configPath()
21 b, err := readFileUTF8(path)
22 if err != nil {
23 if os.IsNotExist(err) {
24 return heartbeatConfigSnapshot{}, nil
25 }
26 return heartbeatConfigSnapshot{}, err
27 }
28 var cfg heartbeatConfig
29 if err := json.Unmarshal(b, &cfg); err != nil {
30 return heartbeatConfigSnapshot{}, fmt.Errorf("invalid config: %w", err)
31 }
32 // Reject future schemas on read as well as write: the scheduler must not
33 // execute tasks with scheduling or approval semantics this binary does not
34 // understand.
35 if cfg.SchemaVersion > heartbeatSchemaVersion {
36 return heartbeatConfigSnapshot{}, fmt.Errorf("heartbeat config schemaVersion %d is newer than this binary supports (%d); upgrade Reasonix", cfg.SchemaVersion, heartbeatSchemaVersion)
37 }
38 // Merge the run-history sidecar (execution journal kept outside the main
39 // config so an older binary cannot drop it on a full-table save). Union by
40 // execution timestamp and keep the newest maxRunHistory entries.
41 runs, err := e.readRunHistorySidecar(cfg)
42 if err != nil {
43 return heartbeatConfigSnapshot{}, err
44 }
45 if len(runs) > 0 {
46 for i := range cfg.Tasks {
47 if hist, ok := runs[cfg.Tasks[i].ID]; ok && len(hist) > 0 {
48 cfg.Tasks[i].RunHistory = mergeRunHistory(cfg.Tasks[i].RunHistory, hist)
49 }
50 }
51 }
52 snapshot := heartbeatConfigSnapshot{
53 cfg: cfg,
54 digest: sha256.Sum256(b),
55 exists: true,
56 }
57 return snapshot, nil
58 }
59
60 func (e *HeartbeatEngine) recordConfigSnapshotLocked(snapshot heartbeatConfigSnapshot) {
61 e.cfgRevision = snapshot.cfg.Revision
62 e.cfgDigest = snapshot.digest
63 e.cfgKnown = snapshot.exists
64 e.cfgInitialized = true
65 if snapshot.exists {
66 e.cfgDeleted = false
67 }
68 }
69
70 // adoptExternalEditsLocked compares the exact content digest so edits are
71 // detected even on filesystems with coarse mtimes.
72 func (e *HeartbeatEngine) adoptExternalEditsLocked() {
73 snapshot, err := e.readConfigSnapshot()
74 if err != nil {
75 log.Printf("[heartbeat] invalid external config: %v", err)
76 return
77 }
78 if !snapshot.exists {
79 // Treat deletion of a previously observed config as authoritative.
80 // Retaining the old tasks would execute them on the next tick and recreate
81 // the file from stale state.
82 if e.cfgInitialized && e.cfgKnown {
83 e.tasks = nil
84 e.pendingTopics = make(map[string]heartbeatPendingTopic)
85 e.cfgDeleted = true
86 e.recordConfigSnapshotLocked(snapshot)
87 }
88 return
89 }
90 if e.cfgKnown && snapshot.digest == e.cfgDigest {
91 return
92 }
93 e.recordConfigSnapshotLocked(snapshot)
94 e.tasks = snapshot.cfg.Tasks
95 e.prunePendingTopicsLocked(e.tasks)
96 }
97
98 func (e *HeartbeatEngine) writeTasks(tasks []HeartbeatTask, expected heartbeatConfigSnapshot, compare bool) error {
99 if tasks == nil {
100 tasks = []HeartbeatTask{}
101 }
102 path := e.configPath()
103 if err := os.MkdirAll(filepath.Dir(path), 0o755); err != nil {
104 return err
105 }
106 lockCtx, cancel := context.WithTimeout(context.Background(), 2*time.Second)
107 defer cancel()
108 release, err := filelock.Acquire(lockCtx, path+".lock")
109 if err != nil {
110 return err
111 }
112 defer release()
113
114 current, err := e.readConfigSnapshot()
115 if err != nil {
116 return err
117 }
118 // Forward protection: a config written by a future binary carries a
119 // schemaVersion this binary does not understand. Refuse to overwrite it
120 // with a full-table save instead of silently downgrading the schema.
121 if current.exists && current.cfg.SchemaVersion > heartbeatSchemaVersion {
122 return fmt.Errorf("heartbeat config schemaVersion %d is newer than this binary supports (%d); upgrade Reasonix before editing", current.cfg.SchemaVersion, heartbeatSchemaVersion)
123 }
124 if compare && (current.exists != expected.exists || current.digest != expected.digest || current.cfg.Revision != expected.cfg.Revision) {
125 return ErrHeartbeatConfigConflict
126 }
127 revision := current.cfg.Revision + 1
128 if !current.exists {
129 revision = 1
130 }
131 // Keep run history only in the sidecar so older full-table writers cannot
132 // drop it from heartbeat-tasks.json. The engine writes that owned state back
133 // through writeRunHistorySidecar.
134 sidecar := make(map[string][]HeartbeatRun, len(tasks))
135 mainTasks := make([]HeartbeatTask, len(tasks))
136 for i, t := range tasks {
137 mainTasks[i] = t
138 mainTasks[i].RunHistory = nil
139 if len(t.RunHistory) > 0 {
140 sidecar[t.ID] = t.RunHistory
141 }
142 }
143 cfg := heartbeatConfig{SchemaVersion: heartbeatSchemaVersion, Revision: revision, Tasks: mainTasks}
144 b, err := json.MarshalIndent(cfg, "", " ")
145 if err != nil {
146 return err
147 }
148 // Publish the sidecar first; the main config is the two-file commit marker.
149 // If the config write fails, restore the previous sidecar while the config
150 // lock remains held.
151 previousSidecar, sidecarReadErr := os.ReadFile(e.runHistoryPath())
152 previousSidecarExists := sidecarReadErr == nil
153 if sidecarReadErr != nil && !os.IsNotExist(sidecarReadErr) {
154 return sidecarReadErr
155 }
156 if previousSidecarExists {
157 var persistedSidecar heartbeatRunHistorySidecar
158 if err := json.Unmarshal(previousSidecar, &persistedSidecar); err == nil && persistedSidecar.SchemaVersion > heartbeatRunHistorySchemaVersion {
159 return fmt.Errorf("heartbeat run-history sidecar schemaVersion %d is newer than this binary supports (%d); upgrade Reasonix before editing", persistedSidecar.SchemaVersion, heartbeatRunHistorySchemaVersion)
160 }
161 }
162 var previousGeneration *heartbeatRunHistoryGeneration
163 if current.exists {
164 previousGeneration = &heartbeatRunHistoryGeneration{
165 Revision: current.cfg.Revision,
166 Runs: heartbeatRunHistoryByTask(current.cfg.Tasks),
167 }
168 }
169 if err := e.writeRunHistorySidecar(revision, sidecar, previousGeneration); err != nil {
170 return err
171 }
172 if err := fileutil.AtomicWriteFile(path, b, 0o644); err != nil {
173 var rollbackErr error
174 if previousSidecarExists {
175 rollbackErr = fileutil.AtomicWriteFile(e.runHistoryPath(), previousSidecar, 0o644)
176 } else if removeErr := os.Remove(e.runHistoryPath()); removeErr != nil && !os.IsNotExist(removeErr) {
177 rollbackErr = removeErr
178 }
179 if rollbackErr != nil {
180 return fmt.Errorf("write heartbeat config: %w (restore run-history sidecar: %w)", err, rollbackErr)
181 }
182 return err
183 }
184 return nil
185 }
186
187 func (e *HeartbeatEngine) mergeRunUpdatesLocked(updates map[string]HeartbeatTask) {
188 if len(updates) == 0 {
189 return
190 }
191 // Rebase onto the human/AI-editable disk list before saving. The engine owns
192 // only run-state fields; task definitions added, edited, or deleted outside
193 // this process remain authoritative.
194 for range 3 {
195 expected, err := e.readConfigSnapshot()
196 if err != nil {
197 log.Printf("[heartbeat] cannot read config before run-state merge: %v", err)
198 return
199 }
200 tasks := expected.cfg.Tasks
201 if !expected.exists {
202 switch {
203 case e.cfgDeleted || (e.cfgInitialized && e.cfgKnown):
204 // Observe deletion in this CAS loop too. A run can finish before the
205 // next scheduler tick adopts external edits; relying only on tick
206 // would let that completion recreate a file the user just removed.
207 e.tasks = nil
208 e.pendingTopics = make(map[string]heartbeatPendingTopic)
209 e.cfgDeleted = true
210 e.recordConfigSnapshotLocked(expected)
211 return
212 case !e.cfgInitialized:
213 // Only an engine that has never observed disk may bootstrap from an
214 // in-memory list. Once a config existed, deletion is authoritative.
215 tasks = append([]HeartbeatTask(nil), e.tasks...)
216 }
217 }
218 mergeHeartbeatRunUpdates(tasks, updates)
219 if err := e.writeTasks(tasks, expected, true); err != nil {
220 if errors.Is(err, ErrHeartbeatConfigConflict) {
221 continue
222 }
223 log.Printf("[heartbeat] run-state merge failed: %v", err)
224 return
225 }
226 latest, err := e.readConfigSnapshot()
227 if err != nil {
228 log.Printf("[heartbeat] reload after run-state merge: %v", err)
229 return
230 }
231 e.recordConfigSnapshotLocked(latest)
232 e.tasks = tasks
233 e.prunePendingTopicsLocked(tasks)
234 return
235 }
236 log.Printf("[heartbeat] run-state merge lost repeated config races; next tick will retry")
237 }
238
239 func mergeHeartbeatRunUpdates(tasks []HeartbeatTask, updates map[string]HeartbeatTask) {
240 for i := range tasks {
241 update, ok := updates[tasks[i].ID]
242 if !ok {
243 continue
244 }
245 // Run state is monotonic. A runtime that lost the cross-process lease may
246 // merge a stale snapshot later, but must not roll back the owner's timestamp
247 // or fresh-conversation topic.
248 newerRun := update.LastRunAt > tasks[i].LastRunAt
249 if update.TopicID != "" && (tasks[i].TopicID == "" || newerRun) {
250 tasks[i].TopicID = update.TopicID
251 }
252 if newerRun {
253 tasks[i].LastRunAt = update.LastRunAt
254 }
255 if tasks[i].CreatedAt == 0 && update.CreatedAt != 0 {
256 tasks[i].CreatedAt = update.CreatedAt
257 }
258 // Merge run history: the on-disk list may be a stale snapshot (external
259 // edits or a tick that raced), so union by execution timestamp and keep
260 // the most recent maxRunHistory entries.
261 if len(update.RunHistory) > 0 {
262 tasks[i].RunHistory = mergeRunHistory(tasks[i].RunHistory, update.RunHistory)
263 }
264 }
265 }
266
267 // mergeRunHistory unions two run-history lists by execution timestamp (At),
268 // dedupes, sorts oldest-first and keeps the newest maxRunHistory entries. Used
269 // by both the update-merge and the disk-protection paths below.
270 func mergeRunHistory(base, extra []HeartbeatRun) []HeartbeatRun {
271 merged := append([]HeartbeatRun(nil), base...)
272 seen := make(map[int64]bool, len(merged))
273 for _, r := range merged {
274 seen[r.At] = true
275 }
276 for _, r := range extra {
277 if !seen[r.At] {
278 merged = append(merged, r)
279 }
280 }
281 sort.Slice(merged, func(a, b int) bool { return merged[a].At < merged[b].At })
282 if len(merged) > maxRunHistory {
283 merged = merged[len(merged)-maxRunHistory:]
284 }
285 return merged
286 }
287
288 // mergeHeartbeatDiskRunHistory protects engine-owned execution state during a
289 // frontend full-list save. A stale snapshot must not roll back TopicID or
290 // LastRunAt, and run history is unioned by timestamp so both snapshots survive.
291 func mergeHeartbeatDiskRunHistory(submitted, disk []HeartbeatTask) []HeartbeatTask {
292 if len(disk) == 0 {
293 return submitted
294 }
295 diskByID := make(map[string]HeartbeatTask, len(disk))
296 for _, d := range disk {
297 diskByID[d.ID] = d
298 }
299 out := make([]HeartbeatTask, len(submitted))
300 copy(out, submitted)
301 for i := range out {
302 diskTask, ok := diskByID[out[i].ID]
303 if !ok {
304 continue
305 }
306 out[i].TopicID = diskTask.TopicID
307 out[i].LastRunAt = diskTask.LastRunAt
308 // Always union by At: once history reaches maxRunHistory, a new disk run
309 // replaces the oldest entry without changing length, so length comparison
310 // would incorrectly drop the engine's new run.
311 out[i].RunHistory = mergeRunHistory(out[i].RunHistory, diskTask.RunHistory)
312 }
313 return out
314 }
315
315 lines GO