feat(infra): add scheduler and worker-generate standalone processes

Extract monitoring recovery/inspection workers and schedule dispatch
into a dedicated scheduler process. Add worker-generate process for
article generation and template-assist queue consumption. Introduce
shared/schedule runtime for cron-based worker lifecycle management.

Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-04-15 14:19:21 +08:00
parent b46cfaaa61
commit 3f5b5c1e01
11 changed files with 1588 additions and 1 deletions
+7 -1
View File
@@ -1,4 +1,4 @@
.PHONY: dev-init dev-api migrate-up migrate-down migrate-monitoring-up migrate-monitoring-down migrate-create sqlc-generate tenant-scope-guard lint test tidy seed
.PHONY: dev-init dev-api dev-worker-generate dev-scheduler migrate-up migrate-down migrate-monitoring-up migrate-monitoring-down migrate-create sqlc-generate tenant-scope-guard lint test tidy seed
DB_URL ?= postgres://geo:geo_dev@localhost:5432/geo?sslmode=disable
MONITORING_DB_URL ?= postgres://geo:geo_dev@localhost:5433/geo_monitoring?sslmode=disable
@@ -15,6 +15,12 @@ dev-init: ## Start infra, run migrations, seed data
dev-api: ## Run tenant-api server
go run ./cmd/tenant-api
dev-worker-generate: ## Run article generation worker
go run ./cmd/worker-generate
dev-scheduler: ## Run scheduler workers
go run ./cmd/scheduler
migrate-up: ## Run all pending migrations
$(MIGRATE_CMD) -path migrations -database "$(DB_URL)" up
+77
View File
@@ -0,0 +1,77 @@
package main
import (
"context"
"log"
"os"
"os/signal"
"syscall"
"github.com/geo-platform/tenant-api/internal/bootstrap"
internalscheduler "github.com/geo-platform/tenant-api/internal/scheduler"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
)
func main() {
configPath := "configs/config.yaml"
if p := os.Getenv("CONFIG_PATH"); p != "" {
configPath = p
}
app, err := bootstrap.New(configPath)
if err != nil {
log.Fatalf("bootstrap: %v", err)
}
defer app.Close()
generationCfg := app.Config.Generation
generationCfg.StreamEnabled = false
knowledgeSvc := tenantapp.NewKnowledgeService(
app.DB,
app.RetrievalProvider,
app.VectorStore,
app.ObjectStorage,
app.LLM,
app.Logger,
app.Config.Retrieval,
app.Config.LLM,
0,
0,
)
promptRuleSvc := tenantapp.NewPromptRuleGenerationService(
app.DB,
app.LLM,
app.RabbitMQ,
knowledgeSvc,
app.GenerationStreams,
generationCfg,
app.Config.LLM.MaxOutputTokens,
)
monitoringCallbackService := tenantapp.NewMonitoringCallbackService(
app.DB,
app.MonitoringDB,
app.RabbitMQ,
app.LLM,
app.Logger,
)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
internalscheduler.NewScheduleDispatchWorker(app.DB, app.RabbitMQ, promptRuleSvc, app.Logger, app.Config.Scheduler).Start(ctx)
internalscheduler.NewMonitoringResultRecoveryWorker(app.MonitoringDB, monitoringCallbackService, app.Logger).Start(ctx)
internalscheduler.NewMonitoringLeaseRecoveryWorker(app.MonitoringDB, app.Logger).Start(ctx)
internalscheduler.NewMonitoringReceivedInspectionWorker(app.MonitoringDB, app.Logger).Start(ctx)
internalscheduler.NewKnowledgeDeletedCleanupWorker(knowledgeSvc).Start(ctx)
app.Logger.Info("scheduler started")
quit := make(chan os.Signal, 1)
signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
<-quit
cancel()
app.Logger.Info("scheduler shutting down...")
}
+95
View File
@@ -0,0 +1,95 @@
package main
import (
"context"
"log"
"os"
"os/signal"
"syscall"
"github.com/geo-platform/tenant-api/internal/bootstrap"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
"github.com/geo-platform/tenant-api/internal/tenant/repository"
generateworker "github.com/geo-platform/tenant-api/internal/worker/generate"
)
func main() {
configPath := "configs/config.yaml"
if p := os.Getenv("CONFIG_PATH"); p != "" {
configPath = p
}
app, err := bootstrap.New(configPath)
if err != nil {
log.Fatalf("bootstrap: %v", err)
}
defer app.Close()
generationCfg := app.Config.Generation
generationCfg.StreamEnabled = false
knowledgeSvc := tenantapp.NewKnowledgeService(
app.DB,
app.RetrievalProvider,
app.VectorStore,
app.ObjectStorage,
app.LLM,
app.Logger,
app.Config.Retrieval,
app.Config.LLM,
0,
0,
)
templateSvc := tenantapp.NewTemplateService(
app.DB,
repository.NewCachedTemplateRepository(
repository.NewTemplateRepository(app.DB),
app.Cache,
),
app.LLM,
nil,
knowledgeSvc,
app.GenerationStreams,
generationCfg,
app.Config.LLM.MaxOutputTokens,
)
promptRuleSvc := tenantapp.NewPromptRuleGenerationService(
app.DB,
app.LLM,
nil,
knowledgeSvc,
app.GenerationStreams,
generationCfg,
app.Config.LLM.MaxOutputTokens,
)
ctx, cancel := context.WithCancel(context.Background())
defer cancel()
generateworker.NewArticleGenerationWorker(
app.DB,
app.RabbitMQ,
templateSvc,
promptRuleSvc,
app.Logger,
app.Config.Generation,
).Start(ctx)
generateworker.NewTemplateAssistWorker(
app.RabbitMQ,
templateSvc,
app.Logger,
app.Config.Generation.WorkerConcurrency,
app.Config.Generation.ArticleTimeout,
).Start(ctx)
app.Logger.Info("worker-generate started")
quit := make(chan os.Signal, 1)
signal.Notify(quit, syscall.SIGINT, syscall.SIGTERM)
<-quit
cancel()
app.Logger.Info("worker-generate shutting down...")
}
@@ -0,0 +1,43 @@
package scheduler
import (
"context"
"time"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
)
type KnowledgeDeletedCleanupWorker struct {
service *tenantapp.KnowledgeService
interval time.Duration
}
func NewKnowledgeDeletedCleanupWorker(service *tenantapp.KnowledgeService) *KnowledgeDeletedCleanupWorker {
return &KnowledgeDeletedCleanupWorker{
service: service,
interval: 15 * time.Second,
}
}
func (w *KnowledgeDeletedCleanupWorker) Start(ctx context.Context) {
if w == nil || w.service == nil {
return
}
go w.run(ctx)
}
func (w *KnowledgeDeletedCleanupWorker) run(ctx context.Context) {
w.service.RunDeletedCleanupSweep()
ticker := time.NewTicker(w.interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
w.service.RunDeletedCleanupSweep()
}
}
}
@@ -0,0 +1,88 @@
package scheduler
import (
"context"
"time"
"github.com/jackc/pgx/v5/pgxpool"
"go.uber.org/zap"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
)
const (
defaultMonitoringLeaseRecoveryInterval = 30 * time.Minute
defaultMonitoringLeaseRecoveryTimeout = 30 * time.Second
)
type MonitoringLeaseRecoveryWorker struct {
monitoringPool *pgxpool.Pool
logger *zap.Logger
interval time.Duration
timeout time.Duration
}
func NewMonitoringLeaseRecoveryWorker(monitoringPool *pgxpool.Pool, logger *zap.Logger) *MonitoringLeaseRecoveryWorker {
return &MonitoringLeaseRecoveryWorker{
monitoringPool: monitoringPool,
logger: logger,
interval: defaultMonitoringLeaseRecoveryInterval,
timeout: defaultMonitoringLeaseRecoveryTimeout,
}
}
func (w *MonitoringLeaseRecoveryWorker) Start(ctx context.Context) {
if w == nil || w.monitoringPool == nil {
return
}
go w.run(ctx)
}
func (w *MonitoringLeaseRecoveryWorker) run(ctx context.Context) {
w.runOnce(ctx)
ticker := time.NewTicker(w.interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
w.runOnce(ctx)
}
}
}
func (w *MonitoringLeaseRecoveryWorker) runOnce(parent context.Context) {
ctx, cancel := context.WithTimeout(parent, w.timeout)
defer cancel()
tx, err := w.monitoringPool.Begin(ctx)
if err != nil {
if w.logger != nil {
w.logger.Warn("monitoring lease recovery begin failed", zap.Error(err))
}
return
}
defer tx.Rollback(ctx)
expiredCount, err := tenantapp.ExpireMonitoringLeasedTasks(ctx, tx, nil, time.Now().UTC())
if err != nil {
if w.logger != nil {
w.logger.Warn("monitoring lease recovery failed", tenantapp.MonitoringWorkerErrorFields(err)...)
}
return
}
if err := tx.Commit(ctx); err != nil {
if w.logger != nil {
w.logger.Warn("monitoring lease recovery commit failed", zap.Error(err))
}
return
}
if expiredCount > 0 && w.logger != nil {
w.logger.Info("monitoring lease recovery completed", zap.Int64("expired_task_count", expiredCount))
}
}
@@ -0,0 +1,114 @@
package scheduler
import (
"context"
"time"
"github.com/jackc/pgx/v5/pgxpool"
"go.uber.org/zap"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
)
const (
defaultMonitoringReceivedInspectionInterval = 5 * time.Minute
defaultMonitoringReceivedInspectionTimeout = 30 * time.Second
defaultMonitoringReceivedAlertThreshold = 15 * time.Minute
defaultMonitoringReceivedInspectLimit = 100
)
type MonitoringReceivedInspectionWorker struct {
monitoringPool *pgxpool.Pool
logger *zap.Logger
interval time.Duration
timeout time.Duration
threshold time.Duration
limit int
}
func NewMonitoringReceivedInspectionWorker(monitoringPool *pgxpool.Pool, logger *zap.Logger) *MonitoringReceivedInspectionWorker {
return &MonitoringReceivedInspectionWorker{
monitoringPool: monitoringPool,
logger: logger,
interval: defaultMonitoringReceivedInspectionInterval,
timeout: defaultMonitoringReceivedInspectionTimeout,
threshold: defaultMonitoringReceivedAlertThreshold,
limit: defaultMonitoringReceivedInspectLimit,
}
}
func (w *MonitoringReceivedInspectionWorker) Start(ctx context.Context) {
if w == nil || w.monitoringPool == nil {
return
}
go w.run(ctx)
}
func (w *MonitoringReceivedInspectionWorker) run(ctx context.Context) {
w.runOnce(ctx)
ticker := time.NewTicker(w.interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
w.runOnce(ctx)
}
}
}
func (w *MonitoringReceivedInspectionWorker) runOnce(parent context.Context) {
ctx, cancel := context.WithTimeout(parent, w.timeout)
defer cancel()
tx, err := w.monitoringPool.Begin(ctx)
if err != nil {
if w.logger != nil {
w.logger.Warn("monitoring received watchdog begin failed", zap.Error(err))
}
return
}
defer tx.Rollback(ctx)
now := time.Now().UTC()
items, err := tenantapp.MarkMonitoringStaleReceivedTasks(ctx, tx, now, w.threshold, w.limit)
if err != nil {
if w.logger != nil {
w.logger.Warn("monitoring received watchdog failed", tenantapp.MonitoringWorkerErrorFields(err)...)
}
return
}
if err := tx.Commit(ctx); err != nil {
if w.logger != nil {
w.logger.Warn("monitoring received watchdog commit failed", zap.Error(err))
}
return
}
if len(items) == 0 || w.logger == nil {
return
}
w.logger.Warn("monitoring received watchdog found overdue tasks",
zap.Int("task_count", len(items)),
zap.Duration("threshold", w.threshold),
)
for _, item := range items {
w.logger.Warn("monitoring task stuck in received state",
zap.Int64("task_id", item.ID),
zap.Int64("tenant_id", item.TenantID),
zap.Int64("brand_id", item.BrandID),
zap.Int64("question_id", item.QuestionID),
zap.String("collector_type", item.CollectorType),
zap.String("ai_platform_id", item.AIPlatformID),
zap.String("business_date", item.BusinessDate.Format("2006-01-02")),
zap.Time("callback_received_at", item.CallbackReceivedAt.UTC()),
zap.Duration("received_lag", now.Sub(item.CallbackReceivedAt.UTC())),
)
}
}
@@ -0,0 +1,205 @@
package scheduler
import (
"context"
"time"
"github.com/jackc/pgx/v5"
"github.com/jackc/pgx/v5/pgxpool"
"go.uber.org/zap"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
)
const (
defaultMonitoringResultRecoveryInterval = 1 * time.Minute
defaultMonitoringResultRecoveryTimeout = 30 * time.Second
defaultMonitoringResultRecoveryLimit = 100
)
type MonitoringResultRecoveryWorker struct {
monitoringPool *pgxpool.Pool
service *tenantapp.MonitoringCallbackService
logger *zap.Logger
interval time.Duration
timeout time.Duration
limit int
}
type monitoringRecoverableTaskResult struct {
ID int64
CallbackReceivedAt time.Time
RequestPayloadJSON []byte
}
func NewMonitoringResultRecoveryWorker(
monitoringPool *pgxpool.Pool,
service *tenantapp.MonitoringCallbackService,
logger *zap.Logger,
) *MonitoringResultRecoveryWorker {
return &MonitoringResultRecoveryWorker{
monitoringPool: monitoringPool,
service: service,
logger: logger,
interval: defaultMonitoringResultRecoveryInterval,
timeout: defaultMonitoringResultRecoveryTimeout,
limit: defaultMonitoringResultRecoveryLimit,
}
}
func (w *MonitoringResultRecoveryWorker) Start(ctx context.Context) {
if w == nil || w.monitoringPool == nil || w.service == nil {
return
}
go w.run(ctx)
}
func (w *MonitoringResultRecoveryWorker) run(ctx context.Context) {
w.runOnce(ctx)
ticker := time.NewTicker(w.interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
w.runOnce(ctx)
}
}
}
func (w *MonitoringResultRecoveryWorker) runOnce(parent context.Context) {
ctx, cancel := context.WithTimeout(parent, w.timeout)
defer cancel()
tx, err := w.monitoringPool.Begin(ctx)
if err != nil {
if w.logger != nil {
w.logger.Warn("monitoring result recovery begin failed", zap.Error(err))
}
return
}
defer tx.Rollback(ctx)
items, err := loadMonitoringRecoverableTaskResults(ctx, tx, w.limit)
if err != nil {
if w.logger != nil {
w.logger.Warn("monitoring result recovery failed", tenantapp.MonitoringWorkerErrorFields(err)...)
}
return
}
if err := tx.Commit(ctx); err != nil {
if w.logger != nil {
w.logger.Warn("monitoring result recovery commit failed", zap.Error(err))
}
return
}
if len(items) == 0 {
return
}
republishedCount := 0
for _, item := range items {
if w.tryRepublish(parent, item) {
republishedCount++
}
}
if republishedCount > 0 && w.logger != nil {
w.logger.Info("monitoring result recovery republished tasks", zap.Int("task_count", republishedCount))
}
}
func (w *MonitoringResultRecoveryWorker) tryRepublish(parent context.Context, item monitoringRecoverableTaskResult) bool {
if err := tenantapp.ValidateMonitoringTaskResultEnvelope(item.RequestPayloadJSON); err != nil {
w.service.PatchTaskResultQueueMeta(item.ID, item.CallbackReceivedAt, tenantapp.MonitoringTaskResultQueueMetaPatch{
QueueStatus: "invalid_payload",
QueueClaimedAt: nil,
QueueEnqueuedAt: nil,
LastQueuePublishError: stringPtr(err.Error()),
})
if w.logger != nil {
w.logger.Warn("monitoring result recovery skipped invalid payload", zap.Int64("task_id", item.ID), zap.Error(err))
}
return false
}
ctx, cancel := context.WithTimeout(parent, 5*time.Second)
defer cancel()
if err := w.service.PublishTaskResultEnvelope(ctx, item.RequestPayloadJSON); err != nil {
w.service.PatchTaskResultQueueMeta(item.ID, item.CallbackReceivedAt, tenantapp.MonitoringTaskResultQueueMetaPatch{
QueueStatus: "publish_failed",
QueueClaimedAt: nil,
QueueEnqueuedAt: nil,
LastQueuePublishError: stringPtr(err.Error()),
})
if w.logger != nil {
w.logger.Warn("monitoring result recovery republish failed", zap.Int64("task_id", item.ID), zap.Error(err))
}
return false
}
enqueuedAt := time.Now().UTC().Round(0)
w.service.PatchTaskResultQueueMeta(item.ID, item.CallbackReceivedAt, tenantapp.MonitoringTaskResultQueueMetaPatch{
QueueStatus: "queued",
QueueClaimedAt: nil,
QueueEnqueuedAt: &enqueuedAt,
LastQueuePublishError: nil,
})
return true
}
func loadMonitoringRecoverableTaskResults(ctx context.Context, tx pgx.Tx, limit int) ([]monitoringRecoverableTaskResult, error) {
if limit <= 0 {
limit = defaultMonitoringResultRecoveryLimit
}
rows, err := tx.Query(ctx, `
SELECT
id,
callback_received_at,
request_payload_json
FROM monitoring_collect_tasks
WHERE collector_type = $1
AND status = 'received'
AND callback_received_at IS NOT NULL
AND request_payload_json IS NOT NULL
AND jsonb_typeof(request_payload_json) = 'object'
AND COALESCE(NULLIF(request_payload_json ->> 'queue_status', ''), 'pending') IN ('pending', 'publish_failed')
ORDER BY callback_received_at ASC, id ASC
FOR UPDATE SKIP LOCKED
LIMIT $2
`, "plugin", limit)
if err != nil {
return nil, tenantapp.InternalMonitoringError(50041, "result_recovery_query_failed", "failed to load recoverable monitoring results", err)
}
defer rows.Close()
result := make([]monitoringRecoverableTaskResult, 0, limit)
for rows.Next() {
var item monitoringRecoverableTaskResult
if scanErr := rows.Scan(&item.ID, &item.CallbackReceivedAt, &item.RequestPayloadJSON); scanErr != nil {
return nil, tenantapp.InternalMonitoringError(50041, "result_recovery_scan_failed", "failed to parse recoverable monitoring results", scanErr)
}
result = append(result, item)
}
if err := rows.Err(); err != nil {
return nil, tenantapp.InternalMonitoringError(50041, "result_recovery_scan_failed", "failed to iterate recoverable monitoring results", err)
}
return result, nil
}
func stringPtr(value string) *string {
if value == "" {
return nil
}
result := value
return &result
}
@@ -0,0 +1,467 @@
package scheduler
import (
"context"
"sync"
"time"
"github.com/jackc/pgx/v5/pgtype"
"github.com/jackc/pgx/v5/pgxpool"
"go.uber.org/zap"
"github.com/geo-platform/tenant-api/internal/shared/config"
"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
sharedschedule "github.com/geo-platform/tenant-api/internal/shared/schedule"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
)
const (
defaultScheduleDispatchInterval = 15 * time.Second
defaultScheduleDispatchTimeout = 30 * time.Second
defaultScheduleDispatchBatch = 20
)
type ScheduleDispatchWorker struct {
pool *pgxpool.Pool
rabbitMQ *rabbitmq.Client
promptRuleService *tenantapp.PromptRuleGenerationService
logger *zap.Logger
interval time.Duration
timeout time.Duration
batchSize int
dispatchWorkers int
queueBackpressure int
}
type dueScheduleTask struct {
ID int64
TenantID int64
OperatorID *int64
PromptRuleID int64
BrandID *int64
Name string
CronExpr string
TargetPlatform *string
EnableWebSearch bool
GenerateCount int
StartAt *time.Time
EndAt *time.Time
ScheduledFor time.Time
}
func NewScheduleDispatchWorker(
pool *pgxpool.Pool,
rabbitMQClient *rabbitmq.Client,
promptRuleService *tenantapp.PromptRuleGenerationService,
logger *zap.Logger,
cfg config.SchedulerConfig,
) *ScheduleDispatchWorker {
return &ScheduleDispatchWorker{
pool: pool,
rabbitMQ: rabbitMQClient,
promptRuleService: promptRuleService,
logger: logger,
interval: schedulerDispatchInterval(cfg),
timeout: schedulerDispatchTimeout(cfg),
batchSize: schedulerDispatchBatchSize(cfg),
dispatchWorkers: schedulerDispatchConcurrency(cfg),
queueBackpressure: cfg.GenerationQueueBackpressureLimit,
}
}
func (w *ScheduleDispatchWorker) Start(ctx context.Context) {
if w == nil || w.pool == nil || w.promptRuleService == nil {
return
}
go w.run(ctx)
}
func (w *ScheduleDispatchWorker) run(ctx context.Context) {
w.runOnce(ctx)
ticker := time.NewTicker(w.interval)
defer ticker.Stop()
for {
select {
case <-ctx.Done():
return
case <-ticker.C:
w.runOnce(ctx)
}
}
}
func (w *ScheduleDispatchWorker) runOnce(parent context.Context) {
ctx, cancel := context.WithTimeout(parent, w.timeout)
defer cancel()
if count, err := w.hydrateMissingNextRuns(ctx); err != nil {
if w.logger != nil {
w.logger.Warn("schedule dispatch hydrate missing next_run_at failed", zap.Error(err))
}
} else if count > 0 && w.logger != nil {
w.logger.Info("schedule dispatch hydrated next_run_at", zap.Int("task_count", count))
}
if paused, stats, err := w.shouldPauseDispatch(ctx); err != nil {
if w.logger != nil {
w.logger.Warn("schedule dispatch inspect generation queue failed", zap.Error(err))
}
} else if paused {
if w.logger != nil {
w.logger.Info("schedule dispatch paused by generation queue backpressure",
zap.Int("queue_depth", stats.Messages),
zap.Int("queue_consumers", stats.Consumers),
zap.Int("queue_limit", w.queueBackpressure),
)
}
return
}
tasks, err := w.claimDueSchedules(ctx)
if err != nil {
if w.logger != nil {
w.logger.Warn("schedule dispatch claim due schedules failed", zap.Error(err))
}
return
}
w.dispatchBatch(parent, tasks)
}
func (w *ScheduleDispatchWorker) hydrateMissingNextRuns(ctx context.Context) (int, error) {
tx, err := w.pool.Begin(ctx)
if err != nil {
return 0, err
}
defer tx.Rollback(ctx)
rows, err := tx.Query(ctx, `
SELECT id, cron_expr, start_at, end_at, status
FROM schedule_tasks
WHERE deleted_at IS NULL
AND status = 'enabled'
AND next_run_at IS NULL
ORDER BY created_at ASC, id ASC
FOR UPDATE SKIP LOCKED
LIMIT $1
`, w.batchSize)
if err != nil {
return 0, err
}
now := time.Now()
count := 0
updates := make([]struct {
id int64
nextRunAt *time.Time
}, 0, w.batchSize)
for rows.Next() {
var (
id int64
cronExpr string
status string
startAt pgtype.Timestamptz
endAt pgtype.Timestamptz
)
if err := rows.Scan(&id, &cronExpr, &startAt, &endAt, &status); err != nil {
rows.Close()
return 0, err
}
var nextRunAt *time.Time
if status == "enabled" {
nextRunAt, err = sharedschedule.ComputeNextRun(cronExpr, timePtrFromTimestamp(startAt), timePtrFromTimestamp(endAt), now)
if err != nil && w.logger != nil {
w.logger.Warn("schedule dispatch skipped invalid cron while hydrating next_run_at",
zap.Int64("schedule_task_id", id),
zap.String("cron_expr", cronExpr),
zap.Error(err),
)
}
}
updates = append(updates, struct {
id int64
nextRunAt *time.Time
}{
id: id,
nextRunAt: nextRunAt,
})
}
if err := rows.Err(); err != nil {
rows.Close()
return 0, err
}
rows.Close()
for _, update := range updates {
if _, err := tx.Exec(ctx, `
UPDATE schedule_tasks
SET next_run_at = $1,
updated_at = NOW()
WHERE id = $2
`, update.nextRunAt, update.id); err != nil {
return 0, err
}
count++
}
if err := tx.Commit(ctx); err != nil {
return 0, err
}
return count, nil
}
func (w *ScheduleDispatchWorker) claimDueSchedules(ctx context.Context) ([]dueScheduleTask, error) {
tx, err := w.pool.Begin(ctx)
if err != nil {
return nil, err
}
defer tx.Rollback(ctx)
rows, err := tx.Query(ctx, `
SELECT id, tenant_id, operator_id, prompt_rule_id, brand_id, name, cron_expr, target_platform,
enable_web_search, generate_count, start_at, end_at, next_run_at
FROM schedule_tasks
WHERE deleted_at IS NULL
AND status = 'enabled'
AND next_run_at IS NOT NULL
AND next_run_at <= $1
ORDER BY next_run_at ASC, id ASC
FOR UPDATE SKIP LOCKED
LIMIT $2
`, time.Now(), w.batchSize)
if err != nil {
return nil, err
}
now := time.Now()
tasks := make([]dueScheduleTask, 0, w.batchSize)
updates := make([]struct {
id int64
nextRun *time.Time
}, 0, w.batchSize)
for rows.Next() {
var (
task dueScheduleTask
operatorID pgtype.Int8
startAt pgtype.Timestamptz
endAt pgtype.Timestamptz
nextRunAt pgtype.Timestamptz
)
if err := rows.Scan(
&task.ID,
&task.TenantID,
&operatorID,
&task.PromptRuleID,
&task.BrandID,
&task.Name,
&task.CronExpr,
&task.TargetPlatform,
&task.EnableWebSearch,
&task.GenerateCount,
&startAt,
&endAt,
&nextRunAt,
); err != nil {
rows.Close()
return nil, err
}
task.StartAt = timePtrFromTimestamp(startAt)
task.EndAt = timePtrFromTimestamp(endAt)
task.OperatorID = int64PtrFromInt8(operatorID)
if task.GenerateCount <= 0 {
task.GenerateCount = 1
}
scheduledFor := timePtrFromTimestamp(nextRunAt)
if scheduledFor == nil {
continue
}
task.ScheduledFor = *scheduledFor
nextRun, nextErr := sharedschedule.ComputeNextRun(task.CronExpr, task.StartAt, task.EndAt, now.Add(time.Nanosecond))
if nextErr != nil && w.logger != nil {
w.logger.Warn("schedule dispatch encountered invalid cron while claiming due schedule",
zap.Int64("schedule_task_id", task.ID),
zap.String("cron_expr", task.CronExpr),
zap.Error(nextErr),
)
nextRun = nil
}
updates = append(updates, struct {
id int64
nextRun *time.Time
}{
id: task.ID,
nextRun: nextRun,
})
if sharedschedule.IsRunAllowed(task.ScheduledFor, task.StartAt, task.EndAt) {
tasks = append(tasks, task)
}
}
if err := rows.Err(); err != nil {
rows.Close()
return nil, err
}
rows.Close()
for _, update := range updates {
if _, err := tx.Exec(ctx, `
UPDATE schedule_tasks
SET next_run_at = $1,
updated_at = NOW()
WHERE id = $2
`, update.nextRun, update.id); err != nil {
return nil, err
}
}
if err := tx.Commit(ctx); err != nil {
return nil, err
}
return tasks, nil
}
func (w *ScheduleDispatchWorker) dispatch(parent context.Context, task dueScheduleTask) {
generateCount := task.GenerateCount
if generateCount <= 0 {
generateCount = 1
}
for idx := 0; idx < generateCount; idx++ {
ctx, cancel := context.WithTimeout(parent, w.timeout)
resp, err := w.promptRuleService.EnqueueScheduledGeneration(ctx, tenantapp.SchedulePromptRuleGenerationInput{
ScheduleTaskID: task.ID,
OperatorID: task.OperatorID,
TenantID: task.TenantID,
PromptRuleID: task.PromptRuleID,
BrandID: task.BrandID,
Name: task.Name,
TargetPlatform: task.TargetPlatform,
EnableWebSearch: task.EnableWebSearch,
GenerateCount: generateCount,
ScheduledFor: task.ScheduledFor,
})
cancel()
if err != nil {
if w.logger != nil {
w.logger.Warn("schedule dispatch enqueue generation failed",
zap.Int64("schedule_task_id", task.ID),
zap.Int64("tenant_id", task.TenantID),
zap.Int64("prompt_rule_id", task.PromptRuleID),
zap.Int("generate_index", idx+1),
zap.Int("generate_count", generateCount),
zap.Time("scheduled_for", task.ScheduledFor.UTC()),
zap.Error(err),
)
}
continue
}
if w.logger != nil {
w.logger.Info("schedule dispatch enqueued generation",
zap.Int64("schedule_task_id", task.ID),
zap.Int64("tenant_id", task.TenantID),
zap.Int64("prompt_rule_id", task.PromptRuleID),
zap.Int64("article_id", resp.ArticleID),
zap.Int64("generation_task_id", resp.TaskID),
zap.Int("generate_index", idx+1),
zap.Int("generate_count", generateCount),
zap.Time("scheduled_for", task.ScheduledFor.UTC()),
)
}
}
}
func (w *ScheduleDispatchWorker) dispatchBatch(parent context.Context, tasks []dueScheduleTask) {
if len(tasks) == 0 {
return
}
if w.dispatchWorkers <= 1 {
for _, task := range tasks {
if parent.Err() != nil {
return
}
w.dispatch(parent, task)
}
return
}
sem := make(chan struct{}, w.dispatchWorkers)
var wg sync.WaitGroup
for _, task := range tasks {
if parent.Err() != nil {
break
}
task := task
sem <- struct{}{}
wg.Add(1)
go func() {
defer wg.Done()
defer func() { <-sem }()
w.dispatch(parent, task)
}()
}
wg.Wait()
}
func (w *ScheduleDispatchWorker) shouldPauseDispatch(ctx context.Context) (bool, rabbitmq.QueueStats, error) {
if w == nil || w.rabbitMQ == nil || w.queueBackpressure <= 0 {
return false, rabbitmq.QueueStats{}, nil
}
stats, err := w.rabbitMQ.InspectGenerationQueue(ctx)
if err != nil {
return false, rabbitmq.QueueStats{}, err
}
return stats.Messages >= w.queueBackpressure, stats, nil
}
func schedulerDispatchInterval(cfg config.SchedulerConfig) time.Duration {
if cfg.DispatchInterval > 0 {
return cfg.DispatchInterval
}
return defaultScheduleDispatchInterval
}
func schedulerDispatchTimeout(cfg config.SchedulerConfig) time.Duration {
if cfg.DispatchTimeout > 0 {
return cfg.DispatchTimeout
}
return defaultScheduleDispatchTimeout
}
func schedulerDispatchBatchSize(cfg config.SchedulerConfig) int {
if cfg.DispatchBatchSize > 0 {
return cfg.DispatchBatchSize
}
return defaultScheduleDispatchBatch
}
func schedulerDispatchConcurrency(cfg config.SchedulerConfig) int {
if cfg.DispatchConcurrency > 0 {
return cfg.DispatchConcurrency
}
return 1
}
func timePtrFromTimestamp(value pgtype.Timestamptz) *time.Time {
if !value.Valid {
return nil
}
result := value.Time.Round(0)
return &result
}
func int64PtrFromInt8(value pgtype.Int8) *int64 {
if !value.Valid {
return nil
}
result := value.Int64
return &result
}
@@ -0,0 +1,68 @@
package schedule
import (
"fmt"
"strings"
"time"
cron "github.com/robfig/cron/v3"
)
var cronParser = cron.NewParser(
cron.SecondOptional |
cron.Minute |
cron.Hour |
cron.Dom |
cron.Month |
cron.Dow |
cron.Descriptor,
)
func Parse(cronExpr string) (cron.Schedule, error) {
trimmed := strings.TrimSpace(cronExpr)
if trimmed == "" {
return nil, fmt.Errorf("cron expression is required")
}
schedule, err := cronParser.Parse(trimmed)
if err != nil {
return nil, fmt.Errorf("parse cron expression %q: %w", trimmed, err)
}
return schedule, nil
}
func ComputeNextRun(cronExpr string, startAt, endAt *time.Time, now time.Time) (*time.Time, error) {
schedule, err := Parse(cronExpr)
if err != nil {
return nil, err
}
notBefore := now.In(time.Local)
if startAt != nil {
start := startAt.In(time.Local)
if start.After(notBefore) {
notBefore = start
}
}
next := schedule.Next(notBefore.Add(-time.Nanosecond)).Round(0)
if next.IsZero() {
return nil, nil
}
if endAt != nil && next.After(endAt.In(time.Local)) {
return nil, nil
}
return &next, nil
}
func IsRunAllowed(runAt time.Time, startAt, endAt *time.Time) bool {
runAt = runAt.In(time.Local)
if startAt != nil && runAt.Before(startAt.In(time.Local)) {
return false
}
if endAt != nil && runAt.After(endAt.In(time.Local)) {
return false
}
return true
}
@@ -0,0 +1,261 @@
package generate
import (
"context"
"database/sql"
"encoding/json"
"errors"
"fmt"
"os"
"time"
"github.com/jackc/pgx/v5"
"github.com/jackc/pgx/v5/pgxpool"
amqp "github.com/rabbitmq/amqp091-go"
"go.uber.org/zap"
"github.com/geo-platform/tenant-api/internal/shared/config"
"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
)
const defaultArticleGenerationWorkerRetryInterval = 5 * time.Second
var errArticleGenerationConsumerClosed = errors.New("article generation consumer channel closed")
type ArticleGenerationWorker struct {
pool *pgxpool.Pool
rabbitMQ *rabbitmq.Client
templateService *tenantapp.TemplateService
promptRuleService *tenantapp.PromptRuleGenerationService
logger *zap.Logger
retryInterval time.Duration
processTimeout time.Duration
consumerPrefix string
workerConcurrency int
}
func NewArticleGenerationWorker(
pool *pgxpool.Pool,
rabbitMQClient *rabbitmq.Client,
templateService *tenantapp.TemplateService,
promptRuleService *tenantapp.PromptRuleGenerationService,
logger *zap.Logger,
cfg config.GenerationConfig,
) *ArticleGenerationWorker {
workerConcurrency := cfg.WorkerConcurrency
if workerConcurrency <= 0 {
workerConcurrency = 1
}
processTimeout := cfg.ArticleTimeout + 2*time.Minute
if processTimeout <= 2*time.Minute {
processTimeout = 10 * time.Minute
}
return &ArticleGenerationWorker{
pool: pool,
rabbitMQ: rabbitMQClient,
templateService: templateService,
promptRuleService: promptRuleService,
logger: logger,
retryInterval: defaultArticleGenerationWorkerRetryInterval,
processTimeout: processTimeout,
consumerPrefix: fmt.Sprintf("worker-generate-%d", os.Getpid()),
workerConcurrency: workerConcurrency,
}
}
func (w *ArticleGenerationWorker) Start(ctx context.Context) {
if w == nil || w.pool == nil || w.rabbitMQ == nil {
return
}
for i := 0; i < w.workerConcurrency; i++ {
consumerName := fmt.Sprintf("%s-%d", w.consumerPrefix, i+1)
go w.run(ctx, consumerName)
}
}
func (w *ArticleGenerationWorker) run(ctx context.Context, consumerName string) {
for {
if err := w.consumeOnce(ctx, consumerName); err != nil && ctx.Err() == nil && w.logger != nil {
if errors.Is(err, errArticleGenerationConsumerClosed) {
w.logger.Info("article generation consumer channel closed, retrying",
zap.String("consumer", consumerName),
)
} else {
w.logger.Warn("article generation worker stopped unexpectedly",
zap.Error(err),
zap.String("consumer", consumerName),
)
}
}
select {
case <-ctx.Done():
return
case <-time.After(w.retryInterval):
}
}
}
func (w *ArticleGenerationWorker) consumeOnce(ctx context.Context, consumerName string) error {
deliveries, ch, err := w.rabbitMQ.ConsumeGenerationTask(consumerName)
if err != nil {
return err
}
defer ch.Close()
for {
select {
case <-ctx.Done():
return nil
case delivery, ok := <-deliveries:
if !ok {
return errArticleGenerationConsumerClosed
}
w.handleDelivery(ctx, delivery)
}
}
}
func (w *ArticleGenerationWorker) handleDelivery(parent context.Context, delivery amqp.Delivery) {
envelope, err := tenantapp.DecodeArticleGenerationTaskEnvelope(delivery.Body)
if err != nil {
w.rejectDelivery(delivery, false, err, 0)
return
}
ctx, cancel := context.WithTimeout(parent, w.processTimeout)
defer cancel()
task, terminal, err := w.loadTask(ctx, envelope.TaskID)
if err != nil {
if errors.Is(err, pgx.ErrNoRows) {
if ackErr := delivery.Ack(false); ackErr != nil && w.logger != nil {
w.logger.Warn("article generation ack failed for missing task",
zap.Error(ackErr),
zap.Int64("task_id", envelope.TaskID),
)
}
return
}
if task.ID > 0 {
_ = tenantapp.HandleClaimedGenerationTaskFailure(ctx, w.pool, w.templateService, w.promptRuleService, task, err)
if ackErr := delivery.Ack(false); ackErr != nil && w.logger != nil {
w.logger.Warn("article generation ack failed after task load failure",
zap.Error(ackErr),
zap.Int64("task_id", envelope.TaskID),
)
}
return
}
requeue := !delivery.Redelivered
w.rejectDelivery(delivery, requeue, err, envelope.TaskID)
return
}
if terminal {
if err := delivery.Ack(false); err != nil && w.logger != nil {
w.logger.Warn("article generation ack failed for terminal task",
zap.Error(err),
zap.Int64("task_id", envelope.TaskID),
)
}
return
}
switch {
case task.ArticleID <= 0 || task.QuotaReservationID <= 0:
_ = tenantapp.HandleClaimedGenerationTaskFailure(ctx, w.pool, w.templateService, w.promptRuleService, task, fmt.Errorf("generation task references are incomplete"))
case task.TaskType == "template":
w.templateService.ProcessQueuedTask(ctx, task)
case task.TaskType == "custom_generation":
w.promptRuleService.ProcessQueuedTask(ctx, task)
default:
_ = tenantapp.HandleClaimedGenerationTaskFailure(ctx, w.pool, w.templateService, w.promptRuleService, task, fmt.Errorf("unsupported generation task type: %s", task.TaskType))
}
if err := delivery.Ack(false); err != nil && w.logger != nil {
w.logger.Warn("article generation ack failed",
zap.Error(err),
zap.Int64("task_id", task.ID),
zap.Int64("article_id", task.ArticleID),
)
}
}
func (w *ArticleGenerationWorker) loadTask(ctx context.Context, taskID int64) (tenantapp.ClaimedGenerationTask, bool, error) {
var (
task tenantapp.ClaimedGenerationTask
status string
operatorID sql.NullInt64
articleID sql.NullInt64
reservationID sql.NullInt64
inputParamsJSON []byte
)
err := w.pool.QueryRow(ctx, `
SELECT id, tenant_id, operator_id, article_id, quota_reservation_id, task_type, input_params_json, status
FROM generation_tasks
WHERE id = $1
`, taskID).Scan(
&task.ID,
&task.TenantID,
&operatorID,
&articleID,
&reservationID,
&task.TaskType,
&inputParamsJSON,
&status,
)
if err != nil {
return tenantapp.ClaimedGenerationTask{}, false, err
}
if status == "completed" || status == "failed" {
return tenantapp.ClaimedGenerationTask{}, true, nil
}
task.OperatorID = operatorID.Int64
task.ArticleID = articleID.Int64
task.QuotaReservationID = reservationID.Int64
params, err := parseJSONMap(inputParamsJSON)
if err != nil {
task.InputParams = map[string]interface{}{}
return task, false, fmt.Errorf("decode generation input params: %w", err)
}
task.InputParams = params
return task, false, nil
}
func (w *ArticleGenerationWorker) rejectDelivery(delivery amqp.Delivery, requeue bool, err error, taskID int64) {
if nackErr := delivery.Nack(false, requeue); nackErr != nil && w.logger != nil {
w.logger.Warn("article generation nack failed",
zap.Error(nackErr),
zap.Int64("task_id", taskID),
)
}
if w.logger != nil {
w.logger.Warn("article generation processing failed",
zap.Error(err),
zap.Bool("requeue", requeue),
zap.Bool("redelivered", delivery.Redelivered),
zap.Int64("task_id", taskID),
)
}
}
func parseJSONMap(raw []byte) (map[string]interface{}, error) {
if len(raw) == 0 {
return map[string]interface{}{}, nil
}
payload := make(map[string]interface{})
if err := json.Unmarshal(raw, &payload); err != nil {
return nil, err
}
return payload, nil
}
@@ -0,0 +1,163 @@
package generate
import (
"context"
"errors"
"fmt"
"os"
"time"
amqp "github.com/rabbitmq/amqp091-go"
"go.uber.org/zap"
"github.com/geo-platform/tenant-api/internal/shared/messaging/rabbitmq"
tenantapp "github.com/geo-platform/tenant-api/internal/tenant/app"
)
var errTemplateAssistConsumerClosed = errors.New("template assist consumer channel closed")
type TemplateAssistWorker struct {
rabbitMQ *rabbitmq.Client
templateService *tenantapp.TemplateService
logger *zap.Logger
retryInterval time.Duration
processTimeout time.Duration
consumerPrefix string
workerConcurrency int
}
func NewTemplateAssistWorker(
rabbitMQClient *rabbitmq.Client,
templateService *tenantapp.TemplateService,
logger *zap.Logger,
workerConcurrency int,
processTimeout time.Duration,
) *TemplateAssistWorker {
if workerConcurrency <= 0 {
workerConcurrency = 1
}
if processTimeout <= 0 {
processTimeout = 5 * time.Minute
}
return &TemplateAssistWorker{
rabbitMQ: rabbitMQClient,
templateService: templateService,
logger: logger,
retryInterval: 5 * time.Second,
processTimeout: processTimeout,
consumerPrefix: fmt.Sprintf("template-assist-%d", os.Getpid()),
workerConcurrency: workerConcurrency,
}
}
func (w *TemplateAssistWorker) Start(ctx context.Context) {
if w == nil || w.rabbitMQ == nil || w.templateService == nil {
return
}
for i := 0; i < w.workerConcurrency; i++ {
consumerName := fmt.Sprintf("%s-%d", w.consumerPrefix, i+1)
go w.run(ctx, consumerName)
}
}
func (w *TemplateAssistWorker) run(ctx context.Context, consumerName string) {
for {
if err := w.consumeOnce(ctx, consumerName); err != nil && ctx.Err() == nil && w.logger != nil {
if errors.Is(err, errTemplateAssistConsumerClosed) {
w.logger.Info("template assist consumer channel closed, retrying",
zap.String("consumer", consumerName),
)
} else {
w.logger.Warn("template assist worker stopped unexpectedly",
zap.Error(err),
zap.String("consumer", consumerName),
)
}
}
select {
case <-ctx.Done():
return
case <-time.After(w.retryInterval):
}
}
}
func (w *TemplateAssistWorker) consumeOnce(ctx context.Context, consumerName string) error {
deliveries, ch, err := w.rabbitMQ.ConsumeTemplateAssistTask(consumerName)
if err != nil {
return err
}
defer ch.Close()
for {
select {
case <-ctx.Done():
return nil
case delivery, ok := <-deliveries:
if !ok {
return errTemplateAssistConsumerClosed
}
w.handleDelivery(ctx, delivery)
}
}
}
func (w *TemplateAssistWorker) handleDelivery(parent context.Context, delivery amqp.Delivery) {
job, err := tenantapp.DecodeTemplateAssistTask(delivery.Body)
if err != nil {
w.rejectDelivery(delivery, false, err, "", "")
return
}
ctx, cancel := context.WithTimeout(parent, w.processTimeout)
defer cancel()
terminal, err := w.templateService.IsAssistTaskTerminal(ctx, *job)
if err != nil {
requeue := !delivery.Redelivered
w.rejectDelivery(delivery, requeue, err, job.TaskID, job.TaskType)
return
}
if terminal {
if err := delivery.Ack(false); err != nil && w.logger != nil {
w.logger.Warn("template assist ack failed for terminal task",
zap.Error(err),
zap.String("task_id", job.TaskID),
zap.String("task_type", job.TaskType),
)
}
return
}
w.templateService.ProcessQueuedAssistJob(ctx, *job)
if err := delivery.Ack(false); err != nil && w.logger != nil {
w.logger.Warn("template assist ack failed",
zap.Error(err),
zap.String("task_id", job.TaskID),
zap.String("task_type", job.TaskType),
)
}
}
func (w *TemplateAssistWorker) rejectDelivery(delivery amqp.Delivery, requeue bool, err error, taskID, taskType string) {
if nackErr := delivery.Nack(false, requeue); nackErr != nil && w.logger != nil {
w.logger.Warn("template assist nack failed",
zap.Error(nackErr),
zap.String("task_id", taskID),
zap.String("task_type", taskType),
)
}
if w.logger != nil {
w.logger.Warn("template assist processing failed",
zap.Error(err),
zap.Bool("requeue", requeue),
zap.Bool("redelivered", delivery.Redelivered),
zap.String("task_id", taskID),
zap.String("task_type", taskType),
)
}
}