// Copyright (c) 2026 Jose Cely // Use of this source code is governed by MIT-style // license that can be found in the LICENSE file. package scheduler import ( "context" "fmt" "sync" "time" "git.smarteching.com/goffee/core/logger" "gorm.io/gorm" "gorm.io/gorm/clause" ) // ─── Status types ──────────────────────────────────────────────────────────── // QueueStatus represents the lifecycle state of a queued item. type QueueStatus string const ( StatusPending QueueStatus = "pending" StatusProcessing QueueStatus = "processing" // claimed by a scheduler tick ) // ProcessedStatus is the final outcome recorded in processed_items. type ProcessedStatus string const ( ProcessedOK ProcessedStatus = "ok" ProcessedError ProcessedStatus = "error" ) // ─── Models ────────────────────────────────────────────────────────────────── // QueueItem is a task waiting to be executed. // Any external program can insert a row into this table to enqueue work. // // INSERT INTO queue_items (task_type, payload, priority, max_runs, thread) VALUES (?, ?, ?, ?, ?); type QueueItem struct { ID uint `gorm:"primaryKey;autoIncrement"` TaskType string `gorm:"not null;index"` // e.g. "send_email", "sync_records" Payload string `gorm:"type:text"` // JSON or plain string, interpreted by the handler Status QueueStatus `gorm:"not null;default:'pending';index"` Priority int `gorm:"not null;default:0;index"` // higher = runs first within a tick MaxRuns int `gorm:"not null;default:1"` // 1 = run once, -1 = repeat indefinitely, N = run N times Thread int `gorm:"not null;default:0"` // 0 = sequential, 1 = concurrent per type CreatedAt time.Time UpdatedAt time.Time } // TableName overrides the GORM default. func (QueueItem) TableName() string { return "queue_items" } // ProcessedItem is an immutable record of a completed execution. // Rows are never updated — only inserted. type ProcessedItem struct { ID uint `gorm:"primaryKey;autoIncrement"` QueueItemID uint `gorm:"not null;index"` // FK → queue_items.id (soft ref) TaskType string `gorm:"not null;index"` Payload string `gorm:"type:text"` Status ProcessedStatus `gorm:"not null;index"` ErrorMsg string `gorm:"type:text"` // empty when Status == "ok" StartedAt time.Time `gorm:"not null"` FinishedAt time.Time `gorm:"not null"` DurationMs int64 `gorm:"not null"` // FinishedAt - StartedAt in ms CreatedAt time.Time } // TableName overrides the GORM default. func (ProcessedItem) TableName() string { return "processed_items" } // SchedulerMeta stores key-value metadata for the scheduler, such as the // semaphore state. This table persists across restarts so the scheduler can // recover its previous state. type SchedulerMeta struct { Key string `gorm:"primaryKey;size:64"` Value string `gorm:"size:16;not null"` } // TableName overrides the GORM default. func (SchedulerMeta) TableName() string { return "scheduler_meta" } // ─── Handler ───────────────────────────────────────────────────────────────── // HandlerFunc processes a single queue item. // Return a non-nil error to mark the item as processed with error status. type HandlerFunc func(ctx context.Context, item *QueueItem) error // Registry maps task types to their handlers. // Register all handlers before calling Scheduler.Start. type Registry struct { handlers map[string]HandlerFunc } // NewRegistry creates an empty handler registry. func NewRegistry() *Registry { return &Registry{handlers: make(map[string]HandlerFunc)} } // Register associates a task type with a handler. // Panics on duplicate registration to catch misconfiguration at startup. func (r *Registry) Register(taskType string, h HandlerFunc) { if _, exists := r.handlers[taskType]; exists { panic(fmt.Sprintf("scheduler: handler already registered for task type %q", taskType)) } r.handlers[taskType] = h } // get returns the handler for a task type, or a fallback error handler. func (r *Registry) get(taskType string) HandlerFunc { if h, ok := r.handlers[taskType]; ok { return h } return func(_ context.Context, item *QueueItem) error { return fmt.Errorf("no handler registered for task type %q", item.TaskType) } } // ─── Store ─────────────────────────────────────────────────────────────────── // Store handles all database operations for the scheduler. type Store struct { db *gorm.DB } // NewStore creates a Store. // Tables (queue_items, processed_items) func NewStore(db *gorm.DB) (*Store, error) { if db == nil { return nil, fmt.Errorf("scheduler: db is nil") } return &Store{db: db}, nil } // DB returns the underlying *gorm.DB used by the store. func (s *Store) DB() *gorm.DB { return s.db } // Enqueue inserts a new pending task. // maxRuns controls how many times the task executes: 1 = once (default), -1 = repeat indefinitely. // thread controls concurrency: 0 = sequential per type, 1 = concurrent per type. // Any value of maxRuns < -1 is clamped to 1. func (s *Store) Enqueue(ctx context.Context, taskType, payload string, priority, maxRuns, thread int) (*QueueItem, error) { if maxRuns < -1 { maxRuns = 1 } item := &QueueItem{ TaskType: taskType, Payload: payload, Status: StatusPending, Priority: priority, MaxRuns: maxRuns, Thread: thread, } if err := s.db.WithContext(ctx).Create(item).Error; err != nil { return nil, fmt.Errorf("enqueue %s: %w", taskType, err) } return item, nil } // ClaimBatch atomically claims up to `limit` pending tasks and marks them // as "processing" so that concurrent scheduler instances never double-execute. func (s *Store) ClaimBatch(ctx context.Context, limit int) ([]*QueueItem, error) { var items []*QueueItem err := s.db.WithContext(ctx).Transaction(func(tx *gorm.DB) error { query := tx. Clauses(clause.Locking{Strength: "UPDATE", Options: "SKIP LOCKED"}). Where("status = ?", StatusPending). Order("priority DESC, id ASC"). Limit(limit). Find(&items) if query.Error != nil { return query.Error } if len(items) == 0 { return nil } ids := make([]uint, len(items)) for i, it := range items { ids[i] = it.ID } return tx.Model(&QueueItem{}). Where("id IN ?", ids). Update("status", StatusProcessing). Error }) if err != nil { return nil, fmt.Errorf("claim batch: %w", err) } return items, nil } // RecordResult writes a ProcessedItem, then either deletes the QueueItem (if // MaxRuns is exhausted) or decrements MaxRuns and resets the status to pending // for re-execution on the next tick. A maxRuns of -1 means repeat indefinitely. func (s *Store) RecordResult(ctx context.Context, item *QueueItem, startedAt time.Time, execErr error) error { finishedAt := time.Now() status := ProcessedOK errMsg := "" if execErr != nil { status = ProcessedError errMsg = execErr.Error() } processed := &ProcessedItem{ QueueItemID: item.ID, TaskType: item.TaskType, Payload: item.Payload, Status: status, ErrorMsg: errMsg, StartedAt: startedAt, FinishedAt: finishedAt, DurationMs: finishedAt.Sub(startedAt).Milliseconds(), } return s.db.WithContext(ctx).Transaction(func(tx *gorm.DB) error { // 1. Insert the immutable result record. if err := tx.Create(processed).Error; err != nil { return fmt.Errorf("insert processed_item: %w", err) } // 2. Decide: delete the queue item or re-queue it. // MaxRuns == 1 means "run once" — delete. // MaxRuns == -1 means "repeat indefinitely" — always re-queue. // MaxRuns > 1 means "run N times" — decrement and re-queue. if item.MaxRuns == 1 { // Exhausted — delete. if err := tx.Delete(&QueueItem{}, item.ID).Error; err != nil { return fmt.Errorf("delete queue_item %d: %w", item.ID, err) } } else { // Decrement MaxRuns (if not -1/infinite) and set back to pending. newMaxRuns := item.MaxRuns if newMaxRuns > 1 { newMaxRuns-- } if err := tx.Model(&QueueItem{}). Where("id = ?", item.ID). Updates(map[string]interface{}{ "status": StatusPending, "max_runs": newMaxRuns, }).Error; err != nil { return fmt.Errorf("re-queue queue_item %d: %w", item.ID, err) } } return nil }) } // SemaphoreKey is the key used in scheduler_meta for the semaphore state. const SemaphoreKey = "semaphore" // SemaphoreValueGreen is the stored value for a green semaphore. const SemaphoreValueGreen = "green" // SemaphoreValueRed is the stored value for a red semaphore. const SemaphoreValueRed = "red" // SetSemaphore persists the semaphore state to the database. func (s *Store) SetSemaphore(ctx context.Context, green bool) error { value := SemaphoreValueRed if green { value = SemaphoreValueGreen } return s.db.WithContext(ctx).Transaction(func(tx *gorm.DB) error { var count int64 tx.Model(&SchedulerMeta{}).Where("key = ?", SemaphoreKey).Count(&count) if count == 0 { return tx.Create(&SchedulerMeta{Key: SemaphoreKey, Value: value}).Error } return tx.Model(&SchedulerMeta{}). Where("key = ?", SemaphoreKey). Update("value", value).Error }) } // GetSemaphore reads the persisted semaphore state from the database. // Returns true if green, false if red or not found. func (s *Store) GetSemaphore(ctx context.Context) bool { var meta SchedulerMeta err := s.db.WithContext(ctx).Where("key = ?", SemaphoreKey).First(&meta).Error if err != nil || meta.Value != SemaphoreValueGreen { return false } return true } // PendingCount returns the number of tasks currently in pending or processing state. func (s *Store) PendingCount(ctx context.Context) (int64, error) { var count int64 err := s.db.WithContext(ctx). Model(&QueueItem{}). Where("status IN ?", []QueueStatus{StatusPending, StatusProcessing}). Count(&count).Error return count, err } // RecentProcessed returns the last `n` processed items ordered by newest first. func (s *Store) RecentProcessed(ctx context.Context, n int) ([]*ProcessedItem, error) { var items []*ProcessedItem err := s.db.WithContext(ctx). Order("id DESC"). Limit(n). Find(&items).Error return items, err } // RecoverStuck resets any items stuck in "processing" state (e.g. after a // crash) back to "pending" so they can be re-claimed on the next tick. // Call this once at startup. func (s *Store) RecoverStuck(ctx context.Context) (int64, error) { result := s.db.WithContext(ctx). Model(&QueueItem{}). Where("status = ?", StatusProcessing). Update("status", StatusPending) return result.RowsAffected, result.Error } // ─── Semaphore ─────────────────────────────────────────────────────────────── // Semaphore is the interface for the scheduler kill switch. // Green (true) means tasks can run; Red (false) means tasks are blocked. type Semaphore interface { SetGreen() SetRed() IsGreen() bool } // inMemorySemaphore is the default channel-based implementation. type inMemorySemaphore struct { green chan bool } // NewSemaphore creates a Semaphore. Pass true to start green, false to start red. func NewSemaphore(startGreen bool) Semaphore { s := &inMemorySemaphore{green: make(chan bool, 1)} s.green <- startGreen return s } func (s *inMemorySemaphore) SetGreen() { <-s.green s.green <- true } func (s *inMemorySemaphore) SetRed() { <-s.green s.green <- false } func (s *inMemorySemaphore) IsGreen() bool { val := <-s.green s.green <- val return val } // PersistedSemaphore implements Semaphore and persists every state change to // the database via the Store, so the state survives application restarts. type PersistedSemaphore struct { inner Semaphore store *Store ctx context.Context } // NewPersistedSemaphore creates a PersistedSemaphore with an initial state. func NewPersistedSemaphore(store *Store, startGreen bool) *PersistedSemaphore { return &PersistedSemaphore{ inner: NewSemaphore(startGreen), store: store, ctx: context.Background(), } } // SetGreen allows task execution and persists the state. func (ps *PersistedSemaphore) SetGreen() { ps.inner.SetGreen() _ = ps.store.SetSemaphore(ps.ctx, true) } // SetRed blocks task execution and persists the state. func (ps *PersistedSemaphore) SetRed() { ps.inner.SetRed() _ = ps.store.SetSemaphore(ps.ctx, false) } // IsGreen reports whether execution is currently allowed. func (ps *PersistedSemaphore) IsGreen() bool { return ps.inner.IsGreen() } // ─── Scheduler ─────────────────────────────────────────────────────────────── // Scheduler polls the database every interval and executes pending tasks. type Scheduler struct { interval time.Duration rateLimit int // max tasks per tick; 0 = unlimited semaphore Semaphore store *Store registry *Registry stop chan struct{} done chan struct{} } // SchedulerConfig holds constructor options. type SchedulerConfig struct { Interval time.Duration RateLimit int // 0 = unlimited } // NewScheduler creates a Scheduler. func NewScheduler(cfg SchedulerConfig, sem Semaphore, st *Store, reg *Registry) *Scheduler { return &Scheduler{ interval: cfg.Interval, rateLimit: cfg.RateLimit, semaphore: sem, store: st, registry: reg, stop: make(chan struct{}), done: make(chan struct{}), } } // Start launches the scheduler goroutine. func (s *Scheduler) Start(ctx context.Context) { go s.run(ctx) } // Stop signals the scheduler and waits for it to exit cleanly. func (s *Scheduler) Stop() { close(s.stop) <-s.done } // run is the main event loop. func (s *Scheduler) run(ctx context.Context) { defer close(s.done) ticker := time.NewTicker(s.interval) defer ticker.Stop() for { select { case <-ctx.Done(): logger.ResolveLogger().Info("[scheduler] context cancelled, shutting down") return case <-s.stop: logger.ResolveLogger().Info("[scheduler] stop requested, shutting down") return case t := <-ticker.C: s.tick(ctx, t) } } } // tick runs on every ticker event. func (s *Scheduler) tick(ctx context.Context, t time.Time) { ts := t.Format(time.TimeOnly) if !s.semaphore.IsGreen() { logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] tick %s — semaphore RED, skipping", ts)) return } limit := s.rateLimit if limit == 0 { limit = 1000 } items, err := s.store.ClaimBatch(ctx, limit) if err != nil { logger.ResolveLogger().Error(fmt.Sprintf("[scheduler] tick %s — claim error: %v", ts, err)) return } if len(items) == 0 { return } logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] tick %s — executing %d task(s)", ts, len(items))) // Split items into sequential vs concurrent groups. var sequential []*QueueItem concurrentGroups := make(map[string][]*QueueItem) // taskType -> items for _, item := range items { if item.Thread != 0 { concurrentGroups[item.TaskType] = append(concurrentGroups[item.TaskType], item) } else { sequential = append(sequential, item) } } // Execute sequential items one at a time. for i, item := range sequential { s.execute(ctx, item, i+1, len(items)) } // Execute concurrent groups in parallel per type. for _, group := range concurrentGroups { s.executeGroup(ctx, group, len(items)) } if remaining, err := s.store.PendingCount(ctx); err == nil { logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] tick %s — done (%d task(s) still queued)", ts, remaining)) } } // execute runs a task's handler one or more times within the same tick. // When Thread=0 (sequential), runs execute one after another in a loop. // When Thread=1 (concurrent), runs fire simultaneously in goroutines. // The number of runs per tick is capped by the rate limit, and the item // is re-queued with remaining runs or deleted when exhausted. func (s *Scheduler) execute(ctx context.Context, item *QueueItem, pos, total int) { // Determine how many runs to fire this tick. desiredRuns := item.MaxRuns if desiredRuns < 1 { desiredRuns = 1 // -1 (infinite): one batch per tick } // Cap by the rate limit so one item cannot exhaust the tick budget. budget := s.rateLimit if budget <= 0 { budget = 1000 } if desiredRuns > budget { desiredRuns = budget } remainingRuns := 0 if item.MaxRuns == -1 { remainingRuns = -1 } else { remainingRuns = item.MaxRuns - desiredRuns if remainingRuns < 0 { remainingRuns = 0 } } isConcurrent := item.Thread != 0 if isConcurrent { logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] [%d/%d] task id=%d type=%s firing %d/%d concurrent run(s)", pos, total, item.ID, item.TaskType, desiredRuns, item.MaxRuns)) } else { logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] [%d/%d] task id=%d type=%s firing %d/%d sequential run(s)", pos, total, item.ID, item.TaskType, desiredRuns, item.MaxRuns)) } if isConcurrent { // ── Concurrent: all runs in parallel goroutines ── var wg sync.WaitGroup wg.Add(desiredRuns) for i := 0; i < desiredRuns; i++ { go func(runIdx int) { defer wg.Done() startedAt := time.Now() handler := s.registry.get(item.TaskType) execErr := handler(ctx, item) finishedAt := time.Now() elapsed := finishedAt.Sub(startedAt).Milliseconds() status := ProcessedOK errMsg := "" if execErr != nil { status = ProcessedError errMsg = execErr.Error() } processed := &ProcessedItem{ QueueItemID: item.ID, TaskType: item.TaskType, Payload: item.Payload, Status: status, ErrorMsg: errMsg, StartedAt: startedAt, FinishedAt: finishedAt, DurationMs: elapsed, } if err := s.store.db.WithContext(ctx).Create(processed).Error; err != nil { logger.ResolveLogger().Error(fmt.Sprintf("[scheduler] [%d/%d] failed to record processed_item for id=%d run=%d: %v", pos, total, item.ID, runIdx+1, err)) } if execErr != nil { logger.ResolveLogger().Error(fmt.Sprintf("[scheduler] [%d/%d] task id=%d run=%d FAILED after %dms: %v", pos, total, item.ID, runIdx+1, elapsed, execErr)) } else { logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] [%d/%d] task id=%d run=%d OK (%dms)", pos, total, item.ID, runIdx+1, elapsed)) } }(i) } wg.Wait() } else { // ── Sequential: all runs one after another ── for i := 0; i < desiredRuns; i++ { startedAt := time.Now() handler := s.registry.get(item.TaskType) execErr := handler(ctx, item) finishedAt := time.Now() elapsed := finishedAt.Sub(startedAt).Milliseconds() status := ProcessedOK errMsg := "" if execErr != nil { status = ProcessedError errMsg = execErr.Error() } processed := &ProcessedItem{ QueueItemID: item.ID, TaskType: item.TaskType, Payload: item.Payload, Status: status, ErrorMsg: errMsg, StartedAt: startedAt, FinishedAt: finishedAt, DurationMs: elapsed, } if err := s.store.db.WithContext(ctx).Create(processed).Error; err != nil { logger.ResolveLogger().Error(fmt.Sprintf("[scheduler] [%d/%d] failed to record processed_item for id=%d run=%d: %v", pos, total, item.ID, i+1, err)) } if execErr != nil { logger.ResolveLogger().Error(fmt.Sprintf("[scheduler] [%d/%d] task id=%d run=%d FAILED after %dms: %v", pos, total, item.ID, i+1, elapsed, execErr)) } else { logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] [%d/%d] task id=%d run=%d OK (%dms)", pos, total, item.ID, i+1, elapsed)) } } } // ── After all runs: decide the queue item's fate ── if remainingRuns == -1 { if err := s.store.db.WithContext(ctx). Model(&QueueItem{}). Where("id = ?", item.ID). Updates(map[string]interface{}{ "status": StatusPending, "max_runs": -1, }).Error; err != nil { logger.ResolveLogger().Error(fmt.Sprintf("[scheduler] [%d/%d] failed to re-queue infinite item id=%d: %v", pos, total, item.ID, err)) } logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] [%d/%d] task id=%d re-queued for next tick (infinite)", pos, total, item.ID)) } else if remainingRuns > 0 { if err := s.store.db.WithContext(ctx). Model(&QueueItem{}). Where("id = ?", item.ID). Updates(map[string]interface{}{ "status": StatusPending, "max_runs": remainingRuns, }).Error; err != nil { logger.ResolveLogger().Error(fmt.Sprintf("[scheduler] [%d/%d] failed to re-queue item id=%d with %d remaining runs: %v", pos, total, item.ID, remainingRuns, err)) } logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] [%d/%d] task id=%d re-queued with %d run(s) remaining", pos, total, item.ID, remainingRuns)) } else { if err := s.store.db.WithContext(ctx).Delete(&QueueItem{}, item.ID).Error; err != nil { logger.ResolveLogger().Error(fmt.Sprintf("[scheduler] [%d/%d] failed to delete finished item id=%d: %v", pos, total, item.ID, err)) } logger.ResolveLogger().Info(fmt.Sprintf("[scheduler] [%d/%d] task id=%d all runs exhausted, deleted", pos, total, item.ID)) } } // executeGroup runs all items in a group concurrently using goroutines. // Each item's result is recorded independently. func (s *Scheduler) executeGroup(ctx context.Context, items []*QueueItem, total int) { var wg sync.WaitGroup wg.Add(len(items)) for _, item := range items { item := item // capture range variable go func() { defer wg.Done() // Find the item's position among all items for logging context. // Since we don't have global position here, use a local index. s.execute(ctx, item, 0, total) }() } wg.Wait() }