git.giacomolli.uk - onechat/blob - schedule.go

   1 package main
   2
   3 import (
   4         "cmp"
   5         "context"
   6         "encoding/json"
   7         "errors"
   8         "fmt"
   9         "log/slog"
  10         "os"
  11         "slices"
  12         "strconv"
  13         "strings"
  14         "sync"
  15         "time"
  16 )
  17
  18 const (
  19         // scheduleFileName is the file of the schedules in the directory of the
  20         // agent.
  21         scheduleFileName = "schedules.json"
  22
  23         // scheduleTextBytes is the most bytes of the text of a schedule.
  24         scheduleTextBytes = 4096
  25
  26         // scheduleAtLayout is the layout of the time of a one-time schedule.
  27         scheduleAtLayout = "2006-01-02 15:04"
  28
  29         // scheduleTimeLayout is the layout of the times in the answers of the
  30         // tools and in the log.
  31         scheduleTimeLayout = "Monday 2006-01-02 15:04 -07:00"
  32
  33         // scheduleSource is the source of the work events of the runs.
  34         scheduleSource = "schedule"
  35
  36         // scheduleClockWaitTimeout is the longest wait of Run before it looks
  37         // at the times again, so that it sees a suspend of the machine or a
  38         // change of the clock.
  39         scheduleClockWaitTimeout = time.Minute
  40
  41         // scheduleLate is how long after its time a run is late.
  42         scheduleLate = time.Minute
  43
  44         // scheduleEmptyReply is the last line of the message of a run.
  45         scheduleEmptyReply = "If nothing in this task needs the user, end your turn with an empty reply."
  46
  47         // scheduleLateLine tells the agent that a run is late.
  48         scheduleLateLine = "This run is late: the daemon was stopped at that time."
  49 )
  50
  51 // Schedules holds the schedules of the agent: tasks that start a turn at one
  52 // time or at the times of a cron expression. Each change writes the file whole
  53 // before it returns, so a schedule survives a restart. The mutex protects the
  54 // schedules, because the tools change them from a turn while Run reads them.
  55 // changed tells Run that a set or a remove changed the times.
  56 type Schedules struct {
  57         path    string
  58         owner   string
  59         clock   func() time.Time
  60         logger  *slog.Logger
  61         changed chan struct{}
  62
  63         mutex sync.Mutex
  64         file  ScheduleFile
  65 }
  66
  67 // ScheduleFile is the content of the file of the schedules: the ID of the last
  68 // schedule, so that an ID is never used again, and the schedules.
  69 type ScheduleFile struct {
  70         Next      int        `json:"next"`
  71         Schedules []Schedule `json:"schedules"`
  72 }
  73
  74 // Schedule is one schedule: its owner, its ID, the task for the agent, the cron
  75 // expression of a repeating schedule, empty for a one-time schedule, and its
  76 // next time.
  77 type Schedule struct {
  78         Owner string    `json:"owner"`
  79         ID    int       `json:"id"`
  80         Text  string    `json:"text"`
  81         Cron  string    `json:"cron"`
  82         Due   time.Time `json:"due"`
  83 }
  84
  85 // openSchedules reads the schedules of the file at path. A missing file holds
  86 // none. The owner names its agent. Schedules without an owner take this name. A
  87 // different stored owner is an error. The clock gives the time and the time
  88 // zone of the schedules, and logger gets one line for each run.
  89 func openSchedules(path, owner string, clock func() time.Time, logger *slog.Logger) (*Schedules, error) {
  90         s := &Schedules{
  91                 path:    path,
  92                 owner:   owner,
  93                 clock:   clock,
  94                 logger:  logger,
  95                 changed: make(chan struct{}, 1),
  96         }
  97
  98         data, err := os.ReadFile(path)
  99         if errors.Is(err, os.ErrNotExist) {
 100                 return s, nil
 101         }
 102
 103         if err != nil {
 104                 return nil, fmt.Errorf("read schedules: %w", err)
 105         }
 106
 107         err = json.Unmarshal(data, &s.file)
 108         if err != nil {
 109                 return nil, fmt.Errorf("%s: %w", path, err)
 110         }
 111
 112         changed := false
 113         for i := range s.file.Schedules {
 114                 schedule := &s.file.Schedules[i]
 115                 if schedule.Owner == "" {
 116                         schedule.Owner = owner
 117                         changed = true
 118                 }
 119
 120                 if schedule.Owner != owner {
 121                         return nil, fmt.Errorf("schedule %d belongs to %q, not %q", schedule.ID, schedule.Owner, owner)
 122                 }
 123
 124                 if schedule.Cron == "" {
 125                         continue
 126                 }
 127
 128                 _, err = scheduleParseCron(schedule.Cron)
 129                 if err != nil {
 130                         return nil, fmt.Errorf("%s: schedule %d: %w", path, schedule.ID, err)
 131                 }
 132         }
 133
 134         if changed {
 135                 err = s.save()
 136                 if err != nil {
 137                         return nil, fmt.Errorf("record schedule owner: %w", err)
 138                 }
 139         }
 140
 141         return s, nil
 142 }
 143
 144 // Run starts the runs of the due schedules in l until ctx ends. It waits until
 145 // the first due time, at most scheduleClockWaitTimeout at a time, and also
 146 // wakes at a set or a remove. So a schedule that came due while the daemon was
 147 // stopped runs at the start. It returns the error of a write of the log or of
 148 // the file.
 149 func (s *Schedules) Run(ctx context.Context, l *TurnLoop) error {
 150         for {
 151                 err := s.deliver(ctx, l)
 152                 if err != nil {
 153                         return err
 154                 }
 155
 156                 var (
 157                         timer   *time.Timer
 158                         timeout <-chan time.Time
 159                 )
 160
 161                 wait, found := s.Wait(s.clock())
 162                 if found {
 163                         timer = time.NewTimer(min(wait, scheduleClockWaitTimeout))
 164                         timeout = timer.C
 165                 }
 166
 167                 select {
 168                 case <-ctx.Done():
 169                         return nil
 170                 case <-s.changed:
 171                 case <-timeout:
 172                 }
 173
 174                 if timer != nil {
 175                         timer.Stop()
 176                 }
 177         }
 178 }
 179
 180 // Set adds a schedule with text and returns the answer of the tool. Exactly one
 181 // of at and cron is not nil. at is a local time in scheduleAtLayout after now,
 182 // and it goes through scheduleWallTime, so a time in the hour that the clock
 183 // repeats is its first occurrence. cron is an expression of scheduleParseCron.
 184 func (s *Schedules) Set(text string, at, cron *string) (string, error) {
 185         text = strings.TrimSpace(text)
 186         if text == "" || len(text) > scheduleTextBytes {
 187                 return "", fmt.Errorf("the text must have 1 to %d bytes", scheduleTextBytes)
 188         }
 189
 190         now := s.clock()
 191         schedule := Schedule{
 192                 Text:  text,
 193                 Owner: s.owner,
 194         }
 195
 196         switch {
 197         case (at == nil) == (cron == nil):
 198                 return "", errors.New("give either at or cron, and null for the other")
 199         case at != nil:
 200                 parsed, err := time.Parse(scheduleAtLayout, *at)
 201                 if err != nil {
 202                         return "", fmt.Errorf("at must be a local time such as %s", now.Format(scheduleAtLayout))
 203                 }
 204
 205                 schedule.Due = scheduleWallTime(parsed.Year(), parsed.Month(), parsed.Day(), parsed.Hour(), parsed.Minute(), now.Location())
 206                 if !schedule.Due.After(now) {
 207                         return "", fmt.Errorf("the time %s is not after now, %s", schedule.Due.Format(scheduleTimeLayout), now.Format(scheduleTimeLayout))
 208                 }
 209         default:
 210                 parsed, err := scheduleParseCron(*cron)
 211                 if err != nil {
 212                         return "", err
 213                 }
 214
 215                 next, found := parsed.Next(now)
 216                 if !found {
 217                         return "", fmt.Errorf("the cron expression %q matches no time in %d years", *cron, scheduleCronYears)
 218                 }
 219
 220                 schedule.Cron = parsed.String()
 221                 schedule.Due = next
 222         }
 223
 224         s.mutex.Lock()
 225         defer s.mutex.Unlock()
 226
 227         s.file.Next++
 228         schedule.ID = s.file.Next
 229         s.file.Schedules = append(s.file.Schedules, schedule)
 230         err := s.save()
 231         if err != nil {
 232                 return "", err
 233         }
 234
 235         s.change()
 236
 237         return fmt.Sprintf("Schedule %d set. Next: %s.", schedule.ID, schedule.Due.In(now.Location()).Format(scheduleTimeLayout)), nil
 238 }
 239
 240 // List returns the answer of the tool: one line for each schedule, by due time,
 241 // with its ID, its next time, its cron expression or once, and its text on one
 242 // line.
 243 func (s *Schedules) List() string {
 244         location := s.clock().Location()
 245         schedules := s.sorted()
 246         if len(schedules) == 0 {
 247                 return "No schedules."
 248         }
 249
 250         lines := make([]string, 0, len(schedules))
 251         for _, schedule := range schedules {
 252                 lines = append(lines, strconv.Itoa(schedule.ID)+" | "+schedule.Due.In(location).Format(scheduleTimeLayout)+" | "+cmp.Or(schedule.Cron, "once")+" | "+strings.Join(strings.Fields(schedule.Text), " "))
 253         }
 254
 255         return strings.Join(lines, "\n")
 256 }
 257
 258 // Len returns the number of schedules.
 259 func (s *Schedules) Len() int {
 260         s.mutex.Lock()
 261         defer s.mutex.Unlock()
 262
 263         return len(s.file.Schedules)
 264 }
 265
 266 // Remove removes the schedule id and returns the answer of the tool.
 267 func (s *Schedules) Remove(id int) (string, error) {
 268         s.mutex.Lock()
 269         defer s.mutex.Unlock()
 270
 271         index := slices.IndexFunc(s.file.Schedules, func(schedule Schedule) bool {
 272                 return schedule.ID == id
 273         })
 274
 275         if index < 0 {
 276                 return "", fmt.Errorf("no schedule %d", id)
 277         }
 278
 279         s.file.Schedules = slices.Delete(s.file.Schedules, index, index+1)
 280         err := s.save()
 281         if err != nil {
 282                 return "", err
 283         }
 284
 285         s.change()
 286
 287         return fmt.Sprintf("Schedule %d removed.", id), nil
 288 }
 289
 290 // deliver logs the message of each due schedule in l, in due order, as a work
 291 // message of scheduleSource, and wakes the loop once with all of them. A
 292 // schedule ends its run with Done after its message is durable in the log. A
 293 // crash between the two makes the run repeat once. Each run logs one line with
 294 // the ID of the schedule and, for a late run, how late it is. The line holds no
 295 // text of the schedule.
 296 func (s *Schedules) deliver(ctx context.Context, l *TurnLoop) error {
 297         if l.agent.config.Name != s.owner {
 298                 return fmt.Errorf("schedule owner %q does not match agent %q", s.owner, l.agent.config.Name)
 299         }
 300
 301         now := s.clock()
 302
 303         var turns []Message
 304
 305         for _, schedule := range s.Due(now) {
 306                 message, err := l.Work(scheduleSource, schedule.Message(now), "")
 307                 if err != nil {
 308                         return fmt.Errorf("log a schedule: %w", err)
 309                 }
 310
 311                 // A stop of the daemon during the flush is no failure: the
 312                 // schedule stays due and runs at the next start.
 313                 err = l.Flush(ctx)
 314                 if ctx.Err() != nil {
 315                         return nil
 316                 }
 317
 318                 if err != nil {
 319                         return fmt.Errorf("log a schedule: %w", err)
 320                 }
 321
 322                 err = s.Done(schedule, now)
 323                 if err != nil {
 324                         return err
 325                 }
 326
 327                 attrs := []any{"id", schedule.ID}
 328                 if schedule.late(now) {
 329                         attrs = append(attrs, "late", now.Sub(schedule.Due))
 330                 }
 331
 332                 s.logger.Info("schedule ran", attrs...)
 333
 334                 turns = append(turns, message)
 335         }
 336
 337         l.Wake(turns...)
 338
 339         return nil
 340 }
 341
 342 // Due returns the schedules whose time is not after now, by due time.
 343 func (s *Schedules) Due(now time.Time) []Schedule {
 344         var due []Schedule
 345
 346         for _, schedule := range s.sorted() {
 347                 if !schedule.Due.After(now) {
 348                         due = append(due, schedule)
 349                 }
 350         }
 351
 352         return due
 353 }
 354
 355 // Done ends the run of schedule at now: a one-time schedule goes, and a
 356 // repeating schedule gets the next time after now. So a repeating schedule that
 357 // missed several times runs once. When a remove took the schedule first, Done
 358 // changes nothing.
 359 func (s *Schedules) Done(schedule Schedule, now time.Time) error {
 360         s.mutex.Lock()
 361         defer s.mutex.Unlock()
 362
 363         index := slices.IndexFunc(s.file.Schedules, func(held Schedule) bool {
 364                 return held.ID == schedule.ID
 365         })
 366
 367         if index < 0 {
 368                 return nil
 369         }
 370
 371         if schedule.Cron == "" {
 372                 s.file.Schedules = slices.Delete(s.file.Schedules, index, index+1)
 373
 374                 return s.save()
 375         }
 376
 377         cron, err := scheduleParseCron(schedule.Cron)
 378         if err != nil {
 379                 return fmt.Errorf("schedule %d: %w", schedule.ID, err)
 380         }
 381
 382         next, found := cron.Next(now)
 383         if !found {
 384                 s.file.Schedules = slices.Delete(s.file.Schedules, index, index+1)
 385
 386                 return s.save()
 387         }
 388
 389         s.file.Schedules[index].Due = next
 390
 391         return s.save()
 392 }
 393
 394 // Wait returns how long after now the first schedule is due, or false when
 395 // there is no schedule.
 396 func (s *Schedules) Wait(now time.Time) (time.Duration, bool) {
 397         schedules := s.sorted()
 398         if len(schedules) == 0 {
 399                 return 0, false
 400         }
 401
 402         return max(schedules[0].Due.Sub(now), 0), true
 403 }
 404
 405 // change wakes Run after a set or a remove.
 406 func (s *Schedules) change() {
 407         select {
 408         case s.changed <- struct{}{}:
 409         default:
 410         }
 411 }
 412
 413 // sorted returns a copy of the schedules by due time, then by ID.
 414 func (s *Schedules) sorted() []Schedule {
 415         s.mutex.Lock()
 416
 417         schedules := slices.Clone(s.file.Schedules)
 418         s.mutex.Unlock()
 419
 420         slices.SortFunc(schedules, func(a, b Schedule) int {
 421                 return cmp.Or(a.Due.Compare(b.Due), cmp.Compare(a.ID, b.ID))
 422         })
 423
 424         return schedules
 425 }
 426
 427 // save writes the file whole, with mode 0600. The caller holds the mutex.
 428 func (s *Schedules) save() error {
 429         data, err := json.Marshal(s.file)
 430         if err == nil {
 431                 err = writeFileAtomic(s.path, append(data, '\n'))
 432         }
 433
 434         if err != nil {
 435                 return fmt.Errorf("save schedules: %w", err)
 436         }
 437
 438         return nil
 439 }
 440
 441 // Message returns the text of a run at now for the log, with the times in the
 442 // location of now. The first line starts with [Schedule ID] and gives the due
 443 // time and once or the cron expression. Then come the text, the line about a
 444 // late run when now is more than scheduleLate after the due time, and the line
 445 // about the empty reply.
 446 func (s Schedule) Message(now time.Time) string {
 447         lines := []string{
 448                 "[Schedule " + strconv.Itoa(s.ID) + "] " + s.Due.In(now.Location()).Format(scheduleTimeLayout) + ", " + s.repeats(),
 449                 s.Text,
 450         }
 451
 452         if s.late(now) {
 453                 lines = append(lines, scheduleLateLine)
 454         }
 455
 456         return strings.Join(append(lines, scheduleEmptyReply), "\n")
 457 }
 458
 459 // late tells if a run at now is late: more than scheduleLate after the due
 460 // time.
 461 func (s Schedule) late(now time.Time) bool {
 462         return now.Sub(s.Due) > scheduleLate
 463 }
 464
 465 // repeats returns once for a one-time schedule, or else repeats and the cron
 466 // expression.
 467 func (s Schedule) repeats() string {
 468         if s.Cron == "" {
 469                 return "once"
 470         }
 471
 472         return "repeats " + s.Cron
 473 }