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