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                 check:   scheduleCheck,
  62                 changed: make(chan struct{}, 1),
  63         }
  64
  65         data, err := os.ReadFile(path)
  66         if errors.Is(err, os.ErrNotExist) {
  67                 return s, nil
  68         }
  69
  70         if err != nil {
  71                 return nil, fmt.Errorf("read schedules: %w", err)
  72         }
  73
  74         err = json.Unmarshal(data, &s.file)
  75         if err != nil {
  76                 return nil, fmt.Errorf("%s: %w", path, err)
  77         }
  78
  79         changed := false
  80         for i := range s.file.Schedules {
  81                 schedule := &s.file.Schedules[i]
  82                 if schedule.Owner == "" {
  83                         schedule.Owner = owner
  84                         changed = true
  85                 }
  86
  87                 if schedule.Owner != owner {
  88                         return nil, fmt.Errorf("schedule %d belongs to %q, not %q", schedule.ID, schedule.Owner, owner)
  89                 }
  90
  91                 if schedule.Cron == "" {
  92                         continue
  93                 }
  94
  95                 _, err = scheduleParseCron(schedule.Cron)
  96                 if err != nil {
  97                         return nil, fmt.Errorf("%s: schedule %d: %w", path, schedule.ID, err)
  98                 }
  99         }
 100
 101         if changed {
 102                 err = s.save()
 103                 if err != nil {
 104                         return nil, fmt.Errorf("record schedule owner: %w", err)
 105                 }
 106         }
 107
 108         return s, nil
 109 }
 110
 111 // Schedules holds the schedules of the agent: tasks that start a turn at one
 112 // time or at the times of a cron expression. Each change writes the file whole
 113 // before it returns, so a schedule survives a restart. The mutex protects the
 114 // schedules, because the tools change them from a turn while Run reads them.
 115 // changed tells Run that a set or a remove changed the times.
 116 type Schedules struct {
 117         path    string
 118         owner   string
 119         clock   func() time.Time
 120         logger  *slog.Logger
 121         check   time.Duration
 122         changed chan struct{}
 123
 124         mutex sync.Mutex
 125         file  ScheduleFile
 126 }
 127
 128 // Run starts the runs of the due schedules in l until ctx ends. It waits until
 129 // the first due time, at most check at a time, and also wakes at a set or a
 130 // remove. So a schedule that came due while the daemon was stopped runs at the
 131 // start. It returns the error of a write of the log or of the file.
 132 func (s *Schedules) Run(ctx context.Context, l *TurnLoop) error {
 133         for {
 134                 err := s.deliver(ctx, l)
 135                 if err != nil {
 136                         return err
 137                 }
 138
 139                 var (
 140                         timer   *time.Timer
 141                         timeout <-chan time.Time
 142                 )
 143
 144                 wait, found := s.Wait(s.clock())
 145                 if found {
 146                         timer = time.NewTimer(min(wait, s.check))
 147                         timeout = timer.C
 148                 }
 149
 150                 select {
 151                 case <-ctx.Done():
 152                         return nil
 153                 case <-s.changed:
 154                 case <-timeout:
 155                 }
 156
 157                 if timer != nil {
 158                         timer.Stop()
 159                 }
 160         }
 161 }
 162
 163 // Set adds a schedule with text and returns the answer of the tool. Exactly one
 164 // of at and cron is not nil. at is a local time in scheduleAtLayout after now,
 165 // and it goes through scheduleWallTime, so a time in the hour that the clock
 166 // repeats is its first occurrence. cron is an expression of scheduleParseCron.
 167 func (s *Schedules) Set(text string, at, cron *string) (string, error) {
 168         text = strings.TrimSpace(text)
 169         if text == "" || len(text) > scheduleTextBytes {
 170                 return "", fmt.Errorf("the text must have 1 to %d bytes", scheduleTextBytes)
 171         }
 172
 173         now := s.clock()
 174         schedule := Schedule{
 175                 Text:  text,
 176                 Owner: s.owner,
 177         }
 178
 179         switch {
 180         case (at == nil) == (cron == nil):
 181                 return "", errors.New("give either at or cron, and null for the other")
 182         case at != nil:
 183                 parsed, err := time.Parse(scheduleAtLayout, *at)
 184                 if err != nil {
 185                         return "", fmt.Errorf("at must be a local time such as %s", now.Format(scheduleAtLayout))
 186                 }
 187
 188                 schedule.Due = scheduleWallTime(parsed.Year(), parsed.Month(), parsed.Day(), parsed.Hour(), parsed.Minute(), now.Location())
 189                 if !schedule.Due.After(now) {
 190                         return "", fmt.Errorf("the time %s is not after now, %s", schedule.Due.Format(scheduleTimeLayout), now.Format(scheduleTimeLayout))
 191                 }
 192         default:
 193                 parsed, err := scheduleParseCron(*cron)
 194                 if err != nil {
 195                         return "", err
 196                 }
 197
 198                 next, found := parsed.Next(now)
 199                 if !found {
 200                         return "", fmt.Errorf("the cron expression %q matches no time in %d years", *cron, scheduleCronYears)
 201                 }
 202
 203                 schedule.Cron = parsed.String()
 204                 schedule.Due = next
 205         }
 206
 207         s.mutex.Lock()
 208         defer s.mutex.Unlock()
 209
 210         s.file.Next++
 211         schedule.ID = s.file.Next
 212         s.file.Schedules = append(s.file.Schedules, schedule)
 213         err := s.save()
 214         if err != nil {
 215                 return "", err
 216         }
 217
 218         s.change()
 219
 220         return fmt.Sprintf("Schedule %d set. Next: %s.", schedule.ID, schedule.Due.In(now.Location()).Format(scheduleTimeLayout)), nil
 221 }
 222
 223 // List returns the answer of the tool: one line for each schedule, by due time,
 224 // with its ID, its next time, its cron expression or once, and its text on one
 225 // line.
 226 func (s *Schedules) List() string {
 227         location := s.clock().Location()
 228         schedules := s.sorted()
 229         if len(schedules) == 0 {
 230                 return "No schedules."
 231         }
 232
 233         lines := make([]string, 0, len(schedules))
 234         for _, schedule := range schedules {
 235                 lines = append(lines, strconv.Itoa(schedule.ID)+" | "+schedule.Due.In(location).Format(scheduleTimeLayout)+" | "+cmp.Or(schedule.Cron, "once")+" | "+strings.Join(strings.Fields(schedule.Text), " "))
 236         }
 237
 238         return strings.Join(lines, "\n")
 239 }
 240
 241 // Len returns the number of schedules.
 242 func (s *Schedules) Len() int {
 243         s.mutex.Lock()
 244         defer s.mutex.Unlock()
 245
 246         return len(s.file.Schedules)
 247 }
 248
 249 // Remove removes the schedule id and returns the answer of the tool.
 250 func (s *Schedules) Remove(id int) (string, error) {
 251         s.mutex.Lock()
 252         defer s.mutex.Unlock()
 253
 254         index := slices.IndexFunc(s.file.Schedules, func(schedule Schedule) bool {
 255                 return schedule.ID == id
 256         })
 257
 258         if index < 0 {
 259                 return "", fmt.Errorf("no schedule %d", id)
 260         }
 261
 262         s.file.Schedules = slices.Delete(s.file.Schedules, index, index+1)
 263         err := s.save()
 264         if err != nil {
 265                 return "", err
 266         }
 267
 268         s.change()
 269
 270         return fmt.Sprintf("Schedule %d removed.", id), nil
 271 }
 272
 273 // deliver logs the message of each due schedule in l, in due order, as a work
 274 // message of scheduleSource, and wakes the loop once with all of them. A
 275 // schedule ends its run with Done after its message is durable in the log. A
 276 // crash between the two makes the run repeat once. Each run logs one line with
 277 // the ID of the schedule and, for a late run, how late it is. The line holds no
 278 // text of the schedule.
 279 func (s *Schedules) deliver(ctx context.Context, l *TurnLoop) error {
 280         if l.agent.config.Name != s.owner {
 281                 return fmt.Errorf("schedule owner %q does not match agent %q", s.owner, l.agent.config.Name)
 282         }
 283
 284         now := s.clock()
 285
 286         var turns []Message
 287
 288         for _, schedule := range s.Due(now) {
 289                 message, err := l.Work(scheduleSource, schedule.Message(now), "")
 290                 if err != nil {
 291                         return fmt.Errorf("log a schedule: %w", err)
 292                 }
 293
 294                 // A stop of the daemon during the flush is no failure: the
 295                 // schedule stays due and runs at the next start.
 296                 err = l.Flush(ctx)
 297                 if ctx.Err() != nil {
 298                         return nil
 299                 }
 300
 301                 if err != nil {
 302                         return fmt.Errorf("log a schedule: %w", err)
 303                 }
 304
 305                 err = s.Done(schedule, now)
 306                 if err != nil {
 307                         return err
 308                 }
 309
 310                 attrs := []any{"id", schedule.ID}
 311                 if schedule.late(now) {
 312                         attrs = append(attrs, "late", now.Sub(schedule.Due))
 313                 }
 314
 315                 s.logger.Info("schedule ran", attrs...)
 316
 317                 turns = append(turns, message)
 318         }
 319
 320         l.Wake(turns...)
 321
 322         return nil
 323 }
 324
 325 // Due returns the schedules whose time is not after now, by due time.
 326 func (s *Schedules) Due(now time.Time) []Schedule {
 327         var due []Schedule
 328
 329         for _, schedule := range s.sorted() {
 330                 if !schedule.Due.After(now) {
 331                         due = append(due, schedule)
 332                 }
 333         }
 334
 335         return due
 336 }
 337
 338 // Done ends the run of schedule at now: a one-time schedule goes, and a
 339 // repeating schedule gets the next time after now. So a repeating schedule that
 340 // missed several times runs once. When a remove took the schedule first, Done
 341 // changes nothing.
 342 func (s *Schedules) Done(schedule Schedule, now time.Time) error {
 343         s.mutex.Lock()
 344         defer s.mutex.Unlock()
 345
 346         index := slices.IndexFunc(s.file.Schedules, func(held Schedule) bool {
 347                 return held.ID == schedule.ID
 348         })
 349
 350         if index < 0 {
 351                 return nil
 352         }
 353
 354         if schedule.Cron == "" {
 355                 s.file.Schedules = slices.Delete(s.file.Schedules, index, index+1)
 356
 357                 return s.save()
 358         }
 359
 360         cron, err := scheduleParseCron(schedule.Cron)
 361         if err != nil {
 362                 return fmt.Errorf("schedule %d: %w", schedule.ID, err)
 363         }
 364
 365         next, found := cron.Next(now)
 366         if !found {
 367                 s.file.Schedules = slices.Delete(s.file.Schedules, index, index+1)
 368
 369                 return s.save()
 370         }
 371
 372         s.file.Schedules[index].Due = next
 373
 374         return s.save()
 375 }
 376
 377 // Wait returns how long after now the first schedule is due, or false when
 378 // there is no schedule.
 379 func (s *Schedules) Wait(now time.Time) (time.Duration, bool) {
 380         schedules := s.sorted()
 381         if len(schedules) == 0 {
 382                 return 0, false
 383         }
 384
 385         return max(schedules[0].Due.Sub(now), 0), true
 386 }
 387
 388 // change wakes Run after a set or a remove.
 389 func (s *Schedules) change() {
 390         select {
 391         case s.changed <- struct{}{}:
 392         default:
 393         }
 394 }
 395
 396 // sorted returns a copy of the schedules by due time, then by ID.
 397 func (s *Schedules) sorted() []Schedule {
 398         s.mutex.Lock()
 399
 400         schedules := slices.Clone(s.file.Schedules)
 401         s.mutex.Unlock()
 402
 403         slices.SortFunc(schedules, func(a, b Schedule) int {
 404                 return cmp.Or(a.Due.Compare(b.Due), cmp.Compare(a.ID, b.ID))
 405         })
 406
 407         return schedules
 408 }
 409
 410 // save writes the file whole, with mode 0600. The caller holds the mutex.
 411 func (s *Schedules) save() error {
 412         data, err := json.Marshal(s.file)
 413         if err == nil {
 414                 err = writeFileAtomic(s.path, append(data, '\n'))
 415         }
 416
 417         if err != nil {
 418                 return fmt.Errorf("save schedules: %w", err)
 419         }
 420
 421         return nil
 422 }
 423
 424 // ScheduleFile is the content of the file of the schedules: the ID of the last
 425 // schedule, so that an ID is never used again, and the schedules.
 426 type ScheduleFile struct {
 427         Next      int        `json:"next"`
 428         Schedules []Schedule `json:"schedules"`
 429 }
 430
 431 // Schedule is one schedule: its owner, its ID, the task for the agent, the cron
 432 // expression of a repeating schedule, empty for a one-time schedule, and its
 433 // next time.
 434 type Schedule struct {
 435         Owner string    `json:"owner"`
 436         ID    int       `json:"id"`
 437         Text  string    `json:"text"`
 438         Cron  string    `json:"cron"`
 439         Due   time.Time `json:"due"`
 440 }
 441
 442 // Message returns the text of a run at now for the log, with the times in the
 443 // location of now. The first line starts with [Schedule ID] and gives the due
 444 // time and once or the cron expression. Then come the text, the line about a
 445 // late run when now is more than scheduleLate after the due time, and the line
 446 // about the empty reply.
 447 func (s Schedule) Message(now time.Time) string {
 448         lines := []string{
 449                 "[Schedule " + strconv.Itoa(s.ID) + "] " + s.Due.In(now.Location()).Format(scheduleTimeLayout) + ", " + s.repeats(),
 450                 s.Text,
 451         }
 452
 453         if s.late(now) {
 454                 lines = append(lines, scheduleLateLine)
 455         }
 456
 457         return strings.Join(append(lines, scheduleEmptyReply), "\n")
 458 }
 459
 460 // late tells if a run at now is late: more than scheduleLate after the due
 461 // time.
 462 func (s Schedule) late(now time.Time) bool {
 463         return now.Sub(s.Due) > scheduleLate
 464 }
 465
 466 // repeats returns once for a one-time schedule, or else repeats and the cron
 467 // expression.
 468 func (s Schedule) repeats() string {
 469         if s.Cron == "" {
 470                 return "once"
 471         }
 472
 473         return "repeats " + s.Cron
 474 }